Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
Tim Shimmin | 87c199c | 2006-06-09 14:56:16 +1000 | [diff] [blame] | 2 | * Copyright (c) 2000-2006 Silicon Graphics, Inc. |
Nathan Scott | 7b71876 | 2005-11-02 14:58:39 +1100 | [diff] [blame] | 3 | * All Rights Reserved. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4 | * |
Nathan Scott | 7b71876 | 2005-11-02 14:58:39 +1100 | [diff] [blame] | 5 | * This program is free software; you can redistribute it and/or |
| 6 | * modify it under the terms of the GNU General Public License as |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7 | * published by the Free Software Foundation. |
| 8 | * |
Nathan Scott | 7b71876 | 2005-11-02 14:58:39 +1100 | [diff] [blame] | 9 | * This program is distributed in the hope that it would be useful, |
| 10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 12 | * GNU General Public License for more details. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 13 | * |
Nathan Scott | 7b71876 | 2005-11-02 14:58:39 +1100 | [diff] [blame] | 14 | * You should have received a copy of the GNU General Public License |
| 15 | * along with this program; if not, write the Free Software Foundation, |
| 16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 17 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 18 | #include "xfs.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 19 | #include "xfs_fs.h" |
Dave Chinner | 70a9883c | 2013-10-23 10:36:05 +1100 | [diff] [blame] | 20 | #include "xfs_shared.h" |
Dave Chinner | 239880e | 2013-10-23 10:50:10 +1100 | [diff] [blame] | 21 | #include "xfs_format.h" |
| 22 | #include "xfs_log_format.h" |
| 23 | #include "xfs_trans_resv.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 24 | #include "xfs_bit.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 25 | #include "xfs_sb.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 26 | #include "xfs_mount.h" |
Dave Chinner | 5706278 | 2013-10-15 09:17:51 +1100 | [diff] [blame] | 27 | #include "xfs_da_format.h" |
Dave Chinner | 9a2cc41 | 2014-12-04 09:43:17 +1100 | [diff] [blame] | 28 | #include "xfs_da_btree.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 29 | #include "xfs_inode.h" |
Dave Chinner | 239880e | 2013-10-23 10:50:10 +1100 | [diff] [blame] | 30 | #include "xfs_trans.h" |
Dave Chinner | 239880e | 2013-10-23 10:50:10 +1100 | [diff] [blame] | 31 | #include "xfs_log.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 32 | #include "xfs_log_priv.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 33 | #include "xfs_log_recover.h" |
Dave Chinner | a4fbe6a | 2013-10-23 10:51:50 +1100 | [diff] [blame] | 34 | #include "xfs_inode_item.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 35 | #include "xfs_extfree_item.h" |
| 36 | #include "xfs_trans_priv.h" |
Dave Chinner | a4fbe6a | 2013-10-23 10:51:50 +1100 | [diff] [blame] | 37 | #include "xfs_alloc.h" |
| 38 | #include "xfs_ialloc.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 39 | #include "xfs_quota.h" |
Christoph Hellwig | 0e446be | 2012-11-12 22:54:24 +1100 | [diff] [blame] | 40 | #include "xfs_cksum.h" |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 41 | #include "xfs_trace.h" |
Dave Chinner | 33479e0 | 2012-10-08 21:56:11 +1100 | [diff] [blame] | 42 | #include "xfs_icache.h" |
Dave Chinner | a4fbe6a | 2013-10-23 10:51:50 +1100 | [diff] [blame] | 43 | #include "xfs_bmap_btree.h" |
Dave Chinner | a4fbe6a | 2013-10-23 10:51:50 +1100 | [diff] [blame] | 44 | #include "xfs_error.h" |
Dave Chinner | 2b9ab5a | 2013-08-12 20:49:37 +1000 | [diff] [blame] | 45 | #include "xfs_dir2.h" |
Darrick J. Wong | 9e88b5d | 2016-08-03 12:09:48 +1000 | [diff] [blame] | 46 | #include "xfs_rmap_item.h" |
Brian Foster | 60a4a22 | 2016-09-26 08:34:27 +1000 | [diff] [blame] | 47 | #include "xfs_buf_item.h" |
Darrick J. Wong | f997ee2 | 2016-10-03 09:11:21 -0700 | [diff] [blame] | 48 | #include "xfs_refcount_item.h" |
Darrick J. Wong | 77d61fe | 2016-10-03 09:11:26 -0700 | [diff] [blame] | 49 | #include "xfs_bmap_item.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 50 | |
Dave Chinner | fc06c6d | 2013-08-12 20:49:22 +1000 | [diff] [blame] | 51 | #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1) |
| 52 | |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 53 | STATIC int |
| 54 | xlog_find_zeroed( |
| 55 | struct xlog *, |
| 56 | xfs_daddr_t *); |
| 57 | STATIC int |
| 58 | xlog_clear_stale_blocks( |
| 59 | struct xlog *, |
| 60 | xfs_lsn_t); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 61 | #if defined(DEBUG) |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 62 | STATIC void |
| 63 | xlog_recover_check_summary( |
| 64 | struct xlog *); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 65 | #else |
| 66 | #define xlog_recover_check_summary(log) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 67 | #endif |
Brian Foster | 7088c41 | 2016-01-05 07:40:16 +1100 | [diff] [blame] | 68 | STATIC int |
| 69 | xlog_do_recovery_pass( |
| 70 | struct xlog *, xfs_daddr_t, xfs_daddr_t, int, xfs_daddr_t *); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 71 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 72 | /* |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 73 | * This structure is used during recovery to record the buf log items which |
| 74 | * have been canceled and should not be replayed. |
| 75 | */ |
| 76 | struct xfs_buf_cancel { |
| 77 | xfs_daddr_t bc_blkno; |
| 78 | uint bc_len; |
| 79 | int bc_refcount; |
| 80 | struct list_head bc_list; |
| 81 | }; |
| 82 | |
| 83 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 84 | * Sector aligned buffer routines for buffer create/read/write/access |
| 85 | */ |
| 86 | |
Alex Elder | ff30a62 | 2010-04-13 15:22:58 +1000 | [diff] [blame] | 87 | /* |
| 88 | * Verify the given count of basic blocks is valid number of blocks |
| 89 | * to specify for an operation involving the given XFS log buffer. |
| 90 | * Returns nonzero if the count is valid, 0 otherwise. |
| 91 | */ |
| 92 | |
| 93 | static inline int |
| 94 | xlog_buf_bbcount_valid( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 95 | struct xlog *log, |
Alex Elder | ff30a62 | 2010-04-13 15:22:58 +1000 | [diff] [blame] | 96 | int bbcount) |
| 97 | { |
| 98 | return bbcount > 0 && bbcount <= log->l_logBBsize; |
| 99 | } |
| 100 | |
Alex Elder | 36adecf | 2010-04-13 15:21:13 +1000 | [diff] [blame] | 101 | /* |
| 102 | * Allocate a buffer to hold log data. The buffer needs to be able |
| 103 | * to map to a range of nbblks basic blocks at any valid (basic |
| 104 | * block) offset within the log. |
| 105 | */ |
Eric Sandeen | 5d77c0d | 2009-11-19 15:52:00 +0000 | [diff] [blame] | 106 | STATIC xfs_buf_t * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 107 | xlog_get_bp( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 108 | struct xlog *log, |
Dave Chinner | 3228149 | 2009-01-22 15:37:47 +1100 | [diff] [blame] | 109 | int nbblks) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 110 | { |
Christoph Hellwig | c8da0fa | 2011-07-08 14:36:25 +0200 | [diff] [blame] | 111 | struct xfs_buf *bp; |
| 112 | |
Alex Elder | ff30a62 | 2010-04-13 15:22:58 +1000 | [diff] [blame] | 113 | if (!xlog_buf_bbcount_valid(log, nbblks)) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 114 | xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer", |
Alex Elder | ff30a62 | 2010-04-13 15:22:58 +1000 | [diff] [blame] | 115 | nbblks); |
| 116 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp); |
Dave Chinner | 3228149 | 2009-01-22 15:37:47 +1100 | [diff] [blame] | 117 | return NULL; |
| 118 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 119 | |
Alex Elder | 36adecf | 2010-04-13 15:21:13 +1000 | [diff] [blame] | 120 | /* |
| 121 | * We do log I/O in units of log sectors (a power-of-2 |
| 122 | * multiple of the basic block size), so we round up the |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 123 | * requested size to accommodate the basic blocks required |
Alex Elder | 36adecf | 2010-04-13 15:21:13 +1000 | [diff] [blame] | 124 | * for complete log sectors. |
| 125 | * |
| 126 | * In addition, the buffer may be used for a non-sector- |
| 127 | * aligned block offset, in which case an I/O of the |
| 128 | * requested size could extend beyond the end of the |
| 129 | * buffer. If the requested size is only 1 basic block it |
| 130 | * will never straddle a sector boundary, so this won't be |
| 131 | * an issue. Nor will this be a problem if the log I/O is |
| 132 | * done in basic blocks (sector size 1). But otherwise we |
| 133 | * extend the buffer by one extra log sector to ensure |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 134 | * there's space to accommodate this possibility. |
Alex Elder | 36adecf | 2010-04-13 15:21:13 +1000 | [diff] [blame] | 135 | */ |
Alex Elder | 69ce58f | 2010-04-20 17:09:59 +1000 | [diff] [blame] | 136 | if (nbblks > 1 && log->l_sectBBsize > 1) |
| 137 | nbblks += log->l_sectBBsize; |
| 138 | nbblks = round_up(nbblks, log->l_sectBBsize); |
Alex Elder | 36adecf | 2010-04-13 15:21:13 +1000 | [diff] [blame] | 139 | |
Dave Chinner | e70b73f | 2012-04-23 15:58:49 +1000 | [diff] [blame] | 140 | bp = xfs_buf_get_uncached(log->l_mp->m_logdev_targp, nbblks, 0); |
Christoph Hellwig | c8da0fa | 2011-07-08 14:36:25 +0200 | [diff] [blame] | 141 | if (bp) |
| 142 | xfs_buf_unlock(bp); |
| 143 | return bp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 144 | } |
| 145 | |
Eric Sandeen | 5d77c0d | 2009-11-19 15:52:00 +0000 | [diff] [blame] | 146 | STATIC void |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 147 | xlog_put_bp( |
| 148 | xfs_buf_t *bp) |
| 149 | { |
| 150 | xfs_buf_free(bp); |
| 151 | } |
| 152 | |
Alex Elder | 48389ef | 2010-04-20 17:10:21 +1000 | [diff] [blame] | 153 | /* |
| 154 | * Return the address of the start of the given block number's data |
| 155 | * in a log buffer. The buffer covers a log sector-aligned region. |
| 156 | */ |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 157 | STATIC char * |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 158 | xlog_align( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 159 | struct xlog *log, |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 160 | xfs_daddr_t blk_no, |
| 161 | int nbblks, |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 162 | struct xfs_buf *bp) |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 163 | { |
Christoph Hellwig | fdc07f4 | 2010-05-10 17:28:14 +0000 | [diff] [blame] | 164 | xfs_daddr_t offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1); |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 165 | |
Dave Chinner | 4e94b71 | 2012-04-23 15:58:51 +1000 | [diff] [blame] | 166 | ASSERT(offset + nbblks <= bp->b_length); |
Chandra Seetharaman | 6292604 | 2011-07-22 23:40:15 +0000 | [diff] [blame] | 167 | return bp->b_addr + BBTOB(offset); |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 168 | } |
| 169 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 170 | |
| 171 | /* |
| 172 | * nbblks should be uint, but oh well. Just want to catch that 32-bit length. |
| 173 | */ |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 174 | STATIC int |
| 175 | xlog_bread_noalign( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 176 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 177 | xfs_daddr_t blk_no, |
| 178 | int nbblks, |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 179 | struct xfs_buf *bp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 180 | { |
| 181 | int error; |
| 182 | |
Alex Elder | ff30a62 | 2010-04-13 15:22:58 +1000 | [diff] [blame] | 183 | if (!xlog_buf_bbcount_valid(log, nbblks)) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 184 | xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer", |
Alex Elder | ff30a62 | 2010-04-13 15:22:58 +1000 | [diff] [blame] | 185 | nbblks); |
| 186 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 187 | return -EFSCORRUPTED; |
Dave Chinner | 3228149 | 2009-01-22 15:37:47 +1100 | [diff] [blame] | 188 | } |
| 189 | |
Alex Elder | 69ce58f | 2010-04-20 17:09:59 +1000 | [diff] [blame] | 190 | blk_no = round_down(blk_no, log->l_sectBBsize); |
| 191 | nbblks = round_up(nbblks, log->l_sectBBsize); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 192 | |
| 193 | ASSERT(nbblks > 0); |
Dave Chinner | 4e94b71 | 2012-04-23 15:58:51 +1000 | [diff] [blame] | 194 | ASSERT(nbblks <= bp->b_length); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 195 | |
| 196 | XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no); |
Dave Chinner | 0cac682 | 2016-02-10 15:01:11 +1100 | [diff] [blame] | 197 | bp->b_flags |= XBF_READ; |
Dave Chinner | aa0e883 | 2012-04-23 15:58:52 +1000 | [diff] [blame] | 198 | bp->b_io_length = nbblks; |
Dave Chinner | 0e95f19 | 2012-04-23 15:58:46 +1000 | [diff] [blame] | 199 | bp->b_error = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 200 | |
Dave Chinner | 595bff7 | 2014-10-02 09:05:14 +1000 | [diff] [blame] | 201 | error = xfs_buf_submit_wait(bp); |
| 202 | if (error && !XFS_FORCED_SHUTDOWN(log->l_mp)) |
Christoph Hellwig | 901796a | 2011-10-10 16:52:49 +0000 | [diff] [blame] | 203 | xfs_buf_ioerror_alert(bp, __func__); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 204 | return error; |
| 205 | } |
| 206 | |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 207 | STATIC int |
| 208 | xlog_bread( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 209 | struct xlog *log, |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 210 | xfs_daddr_t blk_no, |
| 211 | int nbblks, |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 212 | struct xfs_buf *bp, |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 213 | char **offset) |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 214 | { |
| 215 | int error; |
| 216 | |
| 217 | error = xlog_bread_noalign(log, blk_no, nbblks, bp); |
| 218 | if (error) |
| 219 | return error; |
| 220 | |
| 221 | *offset = xlog_align(log, blk_no, nbblks, bp); |
| 222 | return 0; |
| 223 | } |
| 224 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 225 | /* |
Dave Chinner | 4439647 | 2011-04-21 09:34:27 +0000 | [diff] [blame] | 226 | * Read at an offset into the buffer. Returns with the buffer in it's original |
| 227 | * state regardless of the result of the read. |
| 228 | */ |
| 229 | STATIC int |
| 230 | xlog_bread_offset( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 231 | struct xlog *log, |
Dave Chinner | 4439647 | 2011-04-21 09:34:27 +0000 | [diff] [blame] | 232 | xfs_daddr_t blk_no, /* block to read from */ |
| 233 | int nbblks, /* blocks to read */ |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 234 | struct xfs_buf *bp, |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 235 | char *offset) |
Dave Chinner | 4439647 | 2011-04-21 09:34:27 +0000 | [diff] [blame] | 236 | { |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 237 | char *orig_offset = bp->b_addr; |
Dave Chinner | 4e94b71 | 2012-04-23 15:58:51 +1000 | [diff] [blame] | 238 | int orig_len = BBTOB(bp->b_length); |
Dave Chinner | 4439647 | 2011-04-21 09:34:27 +0000 | [diff] [blame] | 239 | int error, error2; |
| 240 | |
Chandra Seetharaman | 02fe03d | 2011-07-22 23:40:22 +0000 | [diff] [blame] | 241 | error = xfs_buf_associate_memory(bp, offset, BBTOB(nbblks)); |
Dave Chinner | 4439647 | 2011-04-21 09:34:27 +0000 | [diff] [blame] | 242 | if (error) |
| 243 | return error; |
| 244 | |
| 245 | error = xlog_bread_noalign(log, blk_no, nbblks, bp); |
| 246 | |
| 247 | /* must reset buffer pointer even on error */ |
Chandra Seetharaman | 02fe03d | 2011-07-22 23:40:22 +0000 | [diff] [blame] | 248 | error2 = xfs_buf_associate_memory(bp, orig_offset, orig_len); |
Dave Chinner | 4439647 | 2011-04-21 09:34:27 +0000 | [diff] [blame] | 249 | if (error) |
| 250 | return error; |
| 251 | return error2; |
| 252 | } |
| 253 | |
| 254 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 255 | * Write out the buffer at the given block for the given number of blocks. |
| 256 | * The buffer is kept locked across the write and is returned locked. |
| 257 | * This can only be used for synchronous log writes. |
| 258 | */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 259 | STATIC int |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 260 | xlog_bwrite( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 261 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 262 | xfs_daddr_t blk_no, |
| 263 | int nbblks, |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 264 | struct xfs_buf *bp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 265 | { |
| 266 | int error; |
| 267 | |
Alex Elder | ff30a62 | 2010-04-13 15:22:58 +1000 | [diff] [blame] | 268 | if (!xlog_buf_bbcount_valid(log, nbblks)) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 269 | xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer", |
Alex Elder | ff30a62 | 2010-04-13 15:22:58 +1000 | [diff] [blame] | 270 | nbblks); |
| 271 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 272 | return -EFSCORRUPTED; |
Dave Chinner | 3228149 | 2009-01-22 15:37:47 +1100 | [diff] [blame] | 273 | } |
| 274 | |
Alex Elder | 69ce58f | 2010-04-20 17:09:59 +1000 | [diff] [blame] | 275 | blk_no = round_down(blk_no, log->l_sectBBsize); |
| 276 | nbblks = round_up(nbblks, log->l_sectBBsize); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 277 | |
| 278 | ASSERT(nbblks > 0); |
Dave Chinner | 4e94b71 | 2012-04-23 15:58:51 +1000 | [diff] [blame] | 279 | ASSERT(nbblks <= bp->b_length); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 280 | |
| 281 | XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no); |
Chandra Seetharaman | 72790aa | 2011-07-22 23:40:04 +0000 | [diff] [blame] | 282 | xfs_buf_hold(bp); |
Christoph Hellwig | 0c842ad | 2011-07-08 14:36:19 +0200 | [diff] [blame] | 283 | xfs_buf_lock(bp); |
Dave Chinner | aa0e883 | 2012-04-23 15:58:52 +1000 | [diff] [blame] | 284 | bp->b_io_length = nbblks; |
Dave Chinner | 0e95f19 | 2012-04-23 15:58:46 +1000 | [diff] [blame] | 285 | bp->b_error = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 286 | |
Christoph Hellwig | c2b006c | 2011-08-23 08:28:07 +0000 | [diff] [blame] | 287 | error = xfs_bwrite(bp); |
Christoph Hellwig | 901796a | 2011-10-10 16:52:49 +0000 | [diff] [blame] | 288 | if (error) |
| 289 | xfs_buf_ioerror_alert(bp, __func__); |
Christoph Hellwig | c2b006c | 2011-08-23 08:28:07 +0000 | [diff] [blame] | 290 | xfs_buf_relse(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 291 | return error; |
| 292 | } |
| 293 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 294 | #ifdef DEBUG |
| 295 | /* |
| 296 | * dump debug superblock and log record information |
| 297 | */ |
| 298 | STATIC void |
| 299 | xlog_header_check_dump( |
| 300 | xfs_mount_t *mp, |
| 301 | xlog_rec_header_t *head) |
| 302 | { |
Eric Sandeen | 08e96e1 | 2013-10-11 20:59:05 -0500 | [diff] [blame] | 303 | xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d", |
Joe Perches | 03daa57 | 2009-12-14 18:01:10 -0800 | [diff] [blame] | 304 | __func__, &mp->m_sb.sb_uuid, XLOG_FMT); |
Eric Sandeen | 08e96e1 | 2013-10-11 20:59:05 -0500 | [diff] [blame] | 305 | xfs_debug(mp, " log : uuid = %pU, fmt = %d", |
Joe Perches | 03daa57 | 2009-12-14 18:01:10 -0800 | [diff] [blame] | 306 | &head->h_fs_uuid, be32_to_cpu(head->h_fmt)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 307 | } |
| 308 | #else |
| 309 | #define xlog_header_check_dump(mp, head) |
| 310 | #endif |
| 311 | |
| 312 | /* |
| 313 | * check log record header for recovery |
| 314 | */ |
| 315 | STATIC int |
| 316 | xlog_header_check_recover( |
| 317 | xfs_mount_t *mp, |
| 318 | xlog_rec_header_t *head) |
| 319 | { |
Christoph Hellwig | 69ef921 | 2011-07-08 14:36:05 +0200 | [diff] [blame] | 320 | ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 321 | |
| 322 | /* |
| 323 | * IRIX doesn't write the h_fmt field and leaves it zeroed |
| 324 | * (XLOG_FMT_UNKNOWN). This stops us from trying to recover |
| 325 | * a dirty log created in IRIX. |
| 326 | */ |
Christoph Hellwig | 69ef921 | 2011-07-08 14:36:05 +0200 | [diff] [blame] | 327 | if (unlikely(head->h_fmt != cpu_to_be32(XLOG_FMT))) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 328 | xfs_warn(mp, |
| 329 | "dirty log written in incompatible format - can't recover"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 330 | xlog_header_check_dump(mp, head); |
| 331 | XFS_ERROR_REPORT("xlog_header_check_recover(1)", |
| 332 | XFS_ERRLEVEL_HIGH, mp); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 333 | return -EFSCORRUPTED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 334 | } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 335 | xfs_warn(mp, |
| 336 | "dirty log entry has mismatched uuid - can't recover"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 337 | xlog_header_check_dump(mp, head); |
| 338 | XFS_ERROR_REPORT("xlog_header_check_recover(2)", |
| 339 | XFS_ERRLEVEL_HIGH, mp); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 340 | return -EFSCORRUPTED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 341 | } |
| 342 | return 0; |
| 343 | } |
| 344 | |
| 345 | /* |
| 346 | * read the head block of the log and check the header |
| 347 | */ |
| 348 | STATIC int |
| 349 | xlog_header_check_mount( |
| 350 | xfs_mount_t *mp, |
| 351 | xlog_rec_header_t *head) |
| 352 | { |
Christoph Hellwig | 69ef921 | 2011-07-08 14:36:05 +0200 | [diff] [blame] | 353 | ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 354 | |
| 355 | if (uuid_is_nil(&head->h_fs_uuid)) { |
| 356 | /* |
| 357 | * IRIX doesn't write the h_fs_uuid or h_fmt fields. If |
| 358 | * h_fs_uuid is nil, we assume this log was last mounted |
| 359 | * by IRIX and continue. |
| 360 | */ |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 361 | xfs_warn(mp, "nil uuid in log - IRIX style log"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 362 | } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 363 | xfs_warn(mp, "log has mismatched uuid - can't recover"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 364 | xlog_header_check_dump(mp, head); |
| 365 | XFS_ERROR_REPORT("xlog_header_check_mount", |
| 366 | XFS_ERRLEVEL_HIGH, mp); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 367 | return -EFSCORRUPTED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 368 | } |
| 369 | return 0; |
| 370 | } |
| 371 | |
| 372 | STATIC void |
| 373 | xlog_recover_iodone( |
| 374 | struct xfs_buf *bp) |
| 375 | { |
Chandra Seetharaman | 5a52c2a58 | 2011-07-22 23:39:51 +0000 | [diff] [blame] | 376 | if (bp->b_error) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 377 | /* |
| 378 | * We're not going to bother about retrying |
| 379 | * this during recovery. One strike! |
| 380 | */ |
Dave Chinner | 595bff7 | 2014-10-02 09:05:14 +1000 | [diff] [blame] | 381 | if (!XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) { |
| 382 | xfs_buf_ioerror_alert(bp, __func__); |
| 383 | xfs_force_shutdown(bp->b_target->bt_mount, |
| 384 | SHUTDOWN_META_IO_ERROR); |
| 385 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 386 | } |
Brian Foster | 60a4a22 | 2016-09-26 08:34:27 +1000 | [diff] [blame] | 387 | |
| 388 | /* |
| 389 | * On v5 supers, a bli could be attached to update the metadata LSN. |
| 390 | * Clean it up. |
| 391 | */ |
| 392 | if (bp->b_fspriv) |
| 393 | xfs_buf_item_relse(bp); |
| 394 | ASSERT(bp->b_fspriv == NULL); |
| 395 | |
Christoph Hellwig | cb669ca | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 396 | bp->b_iodone = NULL; |
Dave Chinner | e8aaba9 | 2014-10-02 09:04:22 +1000 | [diff] [blame] | 397 | xfs_buf_ioend(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 398 | } |
| 399 | |
| 400 | /* |
| 401 | * This routine finds (to an approximation) the first block in the physical |
| 402 | * log which contains the given cycle. It uses a binary search algorithm. |
| 403 | * Note that the algorithm can not be perfect because the disk will not |
| 404 | * necessarily be perfect. |
| 405 | */ |
David Chinner | a8272ce | 2007-11-23 16:28:09 +1100 | [diff] [blame] | 406 | STATIC int |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 407 | xlog_find_cycle_start( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 408 | struct xlog *log, |
| 409 | struct xfs_buf *bp, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 410 | xfs_daddr_t first_blk, |
| 411 | xfs_daddr_t *last_blk, |
| 412 | uint cycle) |
| 413 | { |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 414 | char *offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 415 | xfs_daddr_t mid_blk; |
Alex Elder | e3bb2e3 | 2010-04-15 18:17:30 +0000 | [diff] [blame] | 416 | xfs_daddr_t end_blk; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 417 | uint mid_cycle; |
| 418 | int error; |
| 419 | |
Alex Elder | e3bb2e3 | 2010-04-15 18:17:30 +0000 | [diff] [blame] | 420 | end_blk = *last_blk; |
| 421 | mid_blk = BLK_AVG(first_blk, end_blk); |
| 422 | while (mid_blk != first_blk && mid_blk != end_blk) { |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 423 | error = xlog_bread(log, mid_blk, 1, bp, &offset); |
| 424 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 425 | return error; |
Christoph Hellwig | 03bea6f | 2007-10-12 10:58:05 +1000 | [diff] [blame] | 426 | mid_cycle = xlog_get_cycle(offset); |
Alex Elder | e3bb2e3 | 2010-04-15 18:17:30 +0000 | [diff] [blame] | 427 | if (mid_cycle == cycle) |
| 428 | end_blk = mid_blk; /* last_half_cycle == mid_cycle */ |
| 429 | else |
| 430 | first_blk = mid_blk; /* first_half_cycle == mid_cycle */ |
| 431 | mid_blk = BLK_AVG(first_blk, end_blk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 432 | } |
Alex Elder | e3bb2e3 | 2010-04-15 18:17:30 +0000 | [diff] [blame] | 433 | ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) || |
| 434 | (mid_blk == end_blk && mid_blk-1 == first_blk)); |
| 435 | |
| 436 | *last_blk = end_blk; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 437 | |
| 438 | return 0; |
| 439 | } |
| 440 | |
| 441 | /* |
Alex Elder | 3f943d8 | 2010-04-15 18:17:34 +0000 | [diff] [blame] | 442 | * Check that a range of blocks does not contain stop_on_cycle_no. |
| 443 | * Fill in *new_blk with the block offset where such a block is |
| 444 | * found, or with -1 (an invalid block number) if there is no such |
| 445 | * block in the range. The scan needs to occur from front to back |
| 446 | * and the pointer into the region must be updated since a later |
| 447 | * routine will need to perform another test. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 448 | */ |
| 449 | STATIC int |
| 450 | xlog_find_verify_cycle( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 451 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 452 | xfs_daddr_t start_blk, |
| 453 | int nbblks, |
| 454 | uint stop_on_cycle_no, |
| 455 | xfs_daddr_t *new_blk) |
| 456 | { |
| 457 | xfs_daddr_t i, j; |
| 458 | uint cycle; |
| 459 | xfs_buf_t *bp; |
| 460 | xfs_daddr_t bufblks; |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 461 | char *buf = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 462 | int error = 0; |
| 463 | |
Alex Elder | 6881a22 | 2010-04-13 15:22:29 +1000 | [diff] [blame] | 464 | /* |
| 465 | * Greedily allocate a buffer big enough to handle the full |
| 466 | * range of basic blocks we'll be examining. If that fails, |
| 467 | * try a smaller size. We need to be able to read at least |
| 468 | * a log sector, or we're out of luck. |
| 469 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 470 | bufblks = 1 << ffs(nbblks); |
Dave Chinner | 81158e0 | 2012-04-27 19:45:22 +1000 | [diff] [blame] | 471 | while (bufblks > log->l_logBBsize) |
| 472 | bufblks >>= 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 473 | while (!(bp = xlog_get_bp(log, bufblks))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 474 | bufblks >>= 1; |
Alex Elder | 69ce58f | 2010-04-20 17:09:59 +1000 | [diff] [blame] | 475 | if (bufblks < log->l_sectBBsize) |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 476 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 477 | } |
| 478 | |
| 479 | for (i = start_blk; i < start_blk + nbblks; i += bufblks) { |
| 480 | int bcount; |
| 481 | |
| 482 | bcount = min(bufblks, (start_blk + nbblks - i)); |
| 483 | |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 484 | error = xlog_bread(log, i, bcount, bp, &buf); |
| 485 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 486 | goto out; |
| 487 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 488 | for (j = 0; j < bcount; j++) { |
Christoph Hellwig | 03bea6f | 2007-10-12 10:58:05 +1000 | [diff] [blame] | 489 | cycle = xlog_get_cycle(buf); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 490 | if (cycle == stop_on_cycle_no) { |
| 491 | *new_blk = i+j; |
| 492 | goto out; |
| 493 | } |
| 494 | |
| 495 | buf += BBSIZE; |
| 496 | } |
| 497 | } |
| 498 | |
| 499 | *new_blk = -1; |
| 500 | |
| 501 | out: |
| 502 | xlog_put_bp(bp); |
| 503 | return error; |
| 504 | } |
| 505 | |
| 506 | /* |
| 507 | * Potentially backup over partial log record write. |
| 508 | * |
| 509 | * In the typical case, last_blk is the number of the block directly after |
| 510 | * a good log record. Therefore, we subtract one to get the block number |
| 511 | * of the last block in the given buffer. extra_bblks contains the number |
| 512 | * of blocks we would have read on a previous read. This happens when the |
| 513 | * last log record is split over the end of the physical log. |
| 514 | * |
| 515 | * extra_bblks is the number of blocks potentially verified on a previous |
| 516 | * call to this routine. |
| 517 | */ |
| 518 | STATIC int |
| 519 | xlog_find_verify_log_record( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 520 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 521 | xfs_daddr_t start_blk, |
| 522 | xfs_daddr_t *last_blk, |
| 523 | int extra_bblks) |
| 524 | { |
| 525 | xfs_daddr_t i; |
| 526 | xfs_buf_t *bp; |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 527 | char *offset = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 528 | xlog_rec_header_t *head = NULL; |
| 529 | int error = 0; |
| 530 | int smallmem = 0; |
| 531 | int num_blks = *last_blk - start_blk; |
| 532 | int xhdrs; |
| 533 | |
| 534 | ASSERT(start_blk != 0 || *last_blk != start_blk); |
| 535 | |
| 536 | if (!(bp = xlog_get_bp(log, num_blks))) { |
| 537 | if (!(bp = xlog_get_bp(log, 1))) |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 538 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 539 | smallmem = 1; |
| 540 | } else { |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 541 | error = xlog_bread(log, start_blk, num_blks, bp, &offset); |
| 542 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 543 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 544 | offset += ((num_blks - 1) << BBSHIFT); |
| 545 | } |
| 546 | |
| 547 | for (i = (*last_blk) - 1; i >= 0; i--) { |
| 548 | if (i < start_blk) { |
| 549 | /* valid log record not found */ |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 550 | xfs_warn(log->l_mp, |
| 551 | "Log inconsistent (didn't find previous header)"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 552 | ASSERT(0); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 553 | error = -EIO; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 554 | goto out; |
| 555 | } |
| 556 | |
| 557 | if (smallmem) { |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 558 | error = xlog_bread(log, i, 1, bp, &offset); |
| 559 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 560 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 561 | } |
| 562 | |
| 563 | head = (xlog_rec_header_t *)offset; |
| 564 | |
Christoph Hellwig | 69ef921 | 2011-07-08 14:36:05 +0200 | [diff] [blame] | 565 | if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 566 | break; |
| 567 | |
| 568 | if (!smallmem) |
| 569 | offset -= BBSIZE; |
| 570 | } |
| 571 | |
| 572 | /* |
| 573 | * We hit the beginning of the physical log & still no header. Return |
| 574 | * to caller. If caller can handle a return of -1, then this routine |
| 575 | * will be called again for the end of the physical log. |
| 576 | */ |
| 577 | if (i == -1) { |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 578 | error = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 579 | goto out; |
| 580 | } |
| 581 | |
| 582 | /* |
| 583 | * We have the final block of the good log (the first block |
| 584 | * of the log record _before_ the head. So we check the uuid. |
| 585 | */ |
| 586 | if ((error = xlog_header_check_mount(log->l_mp, head))) |
| 587 | goto out; |
| 588 | |
| 589 | /* |
| 590 | * We may have found a log record header before we expected one. |
| 591 | * last_blk will be the 1st block # with a given cycle #. We may end |
| 592 | * up reading an entire log record. In this case, we don't want to |
| 593 | * reset last_blk. Only when last_blk points in the middle of a log |
| 594 | * record do we update last_blk. |
| 595 | */ |
Eric Sandeen | 6211870 | 2008-03-06 13:44:28 +1100 | [diff] [blame] | 596 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 597 | uint h_size = be32_to_cpu(head->h_size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 598 | |
| 599 | xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE; |
| 600 | if (h_size % XLOG_HEADER_CYCLE_SIZE) |
| 601 | xhdrs++; |
| 602 | } else { |
| 603 | xhdrs = 1; |
| 604 | } |
| 605 | |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 606 | if (*last_blk - i + extra_bblks != |
| 607 | BTOBB(be32_to_cpu(head->h_len)) + xhdrs) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 608 | *last_blk = i; |
| 609 | |
| 610 | out: |
| 611 | xlog_put_bp(bp); |
| 612 | return error; |
| 613 | } |
| 614 | |
| 615 | /* |
| 616 | * Head is defined to be the point of the log where the next log write |
Zhi Yong Wu | 0a94da2 | 2013-08-07 10:11:08 +0000 | [diff] [blame] | 617 | * could go. This means that incomplete LR writes at the end are |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 618 | * eliminated when calculating the head. We aren't guaranteed that previous |
| 619 | * LR have complete transactions. We only know that a cycle number of |
| 620 | * current cycle number -1 won't be present in the log if we start writing |
| 621 | * from our current block number. |
| 622 | * |
| 623 | * last_blk contains the block number of the first block with a given |
| 624 | * cycle number. |
| 625 | * |
| 626 | * Return: zero if normal, non-zero if error. |
| 627 | */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 628 | STATIC int |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 629 | xlog_find_head( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 630 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 631 | xfs_daddr_t *return_head_blk) |
| 632 | { |
| 633 | xfs_buf_t *bp; |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 634 | char *offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 635 | xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk; |
| 636 | int num_scan_bblks; |
| 637 | uint first_half_cycle, last_half_cycle; |
| 638 | uint stop_on_cycle; |
| 639 | int error, log_bbnum = log->l_logBBsize; |
| 640 | |
| 641 | /* Is the end of the log device zeroed? */ |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 642 | error = xlog_find_zeroed(log, &first_blk); |
| 643 | if (error < 0) { |
| 644 | xfs_warn(log->l_mp, "empty log check failed"); |
| 645 | return error; |
| 646 | } |
| 647 | if (error == 1) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 648 | *return_head_blk = first_blk; |
| 649 | |
| 650 | /* Is the whole lot zeroed? */ |
| 651 | if (!first_blk) { |
| 652 | /* Linux XFS shouldn't generate totally zeroed logs - |
| 653 | * mkfs etc write a dummy unmount record to a fresh |
| 654 | * log so we can store the uuid in there |
| 655 | */ |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 656 | xfs_warn(log->l_mp, "totally zeroed log"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 657 | } |
| 658 | |
| 659 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 660 | } |
| 661 | |
| 662 | first_blk = 0; /* get cycle # of 1st block */ |
| 663 | bp = xlog_get_bp(log, 1); |
| 664 | if (!bp) |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 665 | return -ENOMEM; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 666 | |
| 667 | error = xlog_bread(log, 0, 1, bp, &offset); |
| 668 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 669 | goto bp_err; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 670 | |
Christoph Hellwig | 03bea6f | 2007-10-12 10:58:05 +1000 | [diff] [blame] | 671 | first_half_cycle = xlog_get_cycle(offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 672 | |
| 673 | last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */ |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 674 | error = xlog_bread(log, last_blk, 1, bp, &offset); |
| 675 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 676 | goto bp_err; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 677 | |
Christoph Hellwig | 03bea6f | 2007-10-12 10:58:05 +1000 | [diff] [blame] | 678 | last_half_cycle = xlog_get_cycle(offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 679 | ASSERT(last_half_cycle != 0); |
| 680 | |
| 681 | /* |
| 682 | * If the 1st half cycle number is equal to the last half cycle number, |
| 683 | * then the entire log is stamped with the same cycle number. In this |
| 684 | * case, head_blk can't be set to zero (which makes sense). The below |
| 685 | * math doesn't work out properly with head_blk equal to zero. Instead, |
| 686 | * we set it to log_bbnum which is an invalid block number, but this |
| 687 | * value makes the math correct. If head_blk doesn't changed through |
| 688 | * all the tests below, *head_blk is set to zero at the very end rather |
| 689 | * than log_bbnum. In a sense, log_bbnum and zero are the same block |
| 690 | * in a circular file. |
| 691 | */ |
| 692 | if (first_half_cycle == last_half_cycle) { |
| 693 | /* |
| 694 | * In this case we believe that the entire log should have |
| 695 | * cycle number last_half_cycle. We need to scan backwards |
| 696 | * from the end verifying that there are no holes still |
| 697 | * containing last_half_cycle - 1. If we find such a hole, |
| 698 | * then the start of that hole will be the new head. The |
| 699 | * simple case looks like |
| 700 | * x | x ... | x - 1 | x |
| 701 | * Another case that fits this picture would be |
| 702 | * x | x + 1 | x ... | x |
Nathan Scott | c41564b | 2006-03-29 08:55:14 +1000 | [diff] [blame] | 703 | * In this case the head really is somewhere at the end of the |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 704 | * log, as one of the latest writes at the beginning was |
| 705 | * incomplete. |
| 706 | * One more case is |
| 707 | * x | x + 1 | x ... | x - 1 | x |
| 708 | * This is really the combination of the above two cases, and |
| 709 | * the head has to end up at the start of the x-1 hole at the |
| 710 | * end of the log. |
| 711 | * |
| 712 | * In the 256k log case, we will read from the beginning to the |
| 713 | * end of the log and search for cycle numbers equal to x-1. |
| 714 | * We don't worry about the x+1 blocks that we encounter, |
| 715 | * because we know that they cannot be the head since the log |
| 716 | * started with x. |
| 717 | */ |
| 718 | head_blk = log_bbnum; |
| 719 | stop_on_cycle = last_half_cycle - 1; |
| 720 | } else { |
| 721 | /* |
| 722 | * In this case we want to find the first block with cycle |
| 723 | * number matching last_half_cycle. We expect the log to be |
| 724 | * some variation on |
Alex Elder | 3f943d8 | 2010-04-15 18:17:34 +0000 | [diff] [blame] | 725 | * x + 1 ... | x ... | x |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 726 | * The first block with cycle number x (last_half_cycle) will |
| 727 | * be where the new head belongs. First we do a binary search |
| 728 | * for the first occurrence of last_half_cycle. The binary |
| 729 | * search may not be totally accurate, so then we scan back |
| 730 | * from there looking for occurrences of last_half_cycle before |
| 731 | * us. If that backwards scan wraps around the beginning of |
| 732 | * the log, then we look for occurrences of last_half_cycle - 1 |
| 733 | * at the end of the log. The cases we're looking for look |
| 734 | * like |
Alex Elder | 3f943d8 | 2010-04-15 18:17:34 +0000 | [diff] [blame] | 735 | * v binary search stopped here |
| 736 | * x + 1 ... | x | x + 1 | x ... | x |
| 737 | * ^ but we want to locate this spot |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 738 | * or |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 739 | * <---------> less than scan distance |
Alex Elder | 3f943d8 | 2010-04-15 18:17:34 +0000 | [diff] [blame] | 740 | * x + 1 ... | x ... | x - 1 | x |
| 741 | * ^ we want to locate this spot |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 742 | */ |
| 743 | stop_on_cycle = last_half_cycle; |
| 744 | if ((error = xlog_find_cycle_start(log, bp, first_blk, |
| 745 | &head_blk, last_half_cycle))) |
| 746 | goto bp_err; |
| 747 | } |
| 748 | |
| 749 | /* |
| 750 | * Now validate the answer. Scan back some number of maximum possible |
| 751 | * blocks and make sure each one has the expected cycle number. The |
| 752 | * maximum is determined by the total possible amount of buffering |
| 753 | * in the in-core log. The following number can be made tighter if |
| 754 | * we actually look at the block size of the filesystem. |
| 755 | */ |
| 756 | num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log); |
| 757 | if (head_blk >= num_scan_bblks) { |
| 758 | /* |
| 759 | * We are guaranteed that the entire check can be performed |
| 760 | * in one buffer. |
| 761 | */ |
| 762 | start_blk = head_blk - num_scan_bblks; |
| 763 | if ((error = xlog_find_verify_cycle(log, |
| 764 | start_blk, num_scan_bblks, |
| 765 | stop_on_cycle, &new_blk))) |
| 766 | goto bp_err; |
| 767 | if (new_blk != -1) |
| 768 | head_blk = new_blk; |
| 769 | } else { /* need to read 2 parts of log */ |
| 770 | /* |
| 771 | * We are going to scan backwards in the log in two parts. |
| 772 | * First we scan the physical end of the log. In this part |
| 773 | * of the log, we are looking for blocks with cycle number |
| 774 | * last_half_cycle - 1. |
| 775 | * If we find one, then we know that the log starts there, as |
| 776 | * we've found a hole that didn't get written in going around |
| 777 | * the end of the physical log. The simple case for this is |
| 778 | * x + 1 ... | x ... | x - 1 | x |
| 779 | * <---------> less than scan distance |
| 780 | * If all of the blocks at the end of the log have cycle number |
| 781 | * last_half_cycle, then we check the blocks at the start of |
| 782 | * the log looking for occurrences of last_half_cycle. If we |
| 783 | * find one, then our current estimate for the location of the |
| 784 | * first occurrence of last_half_cycle is wrong and we move |
| 785 | * back to the hole we've found. This case looks like |
| 786 | * x + 1 ... | x | x + 1 | x ... |
| 787 | * ^ binary search stopped here |
| 788 | * Another case we need to handle that only occurs in 256k |
| 789 | * logs is |
| 790 | * x + 1 ... | x ... | x+1 | x ... |
| 791 | * ^ binary search stops here |
| 792 | * In a 256k log, the scan at the end of the log will see the |
| 793 | * x + 1 blocks. We need to skip past those since that is |
| 794 | * certainly not the head of the log. By searching for |
| 795 | * last_half_cycle-1 we accomplish that. |
| 796 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 797 | ASSERT(head_blk <= INT_MAX && |
Alex Elder | 3f943d8 | 2010-04-15 18:17:34 +0000 | [diff] [blame] | 798 | (xfs_daddr_t) num_scan_bblks >= head_blk); |
| 799 | start_blk = log_bbnum - (num_scan_bblks - head_blk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 800 | if ((error = xlog_find_verify_cycle(log, start_blk, |
| 801 | num_scan_bblks - (int)head_blk, |
| 802 | (stop_on_cycle - 1), &new_blk))) |
| 803 | goto bp_err; |
| 804 | if (new_blk != -1) { |
| 805 | head_blk = new_blk; |
Alex Elder | 9db127e | 2010-04-15 18:17:26 +0000 | [diff] [blame] | 806 | goto validate_head; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 807 | } |
| 808 | |
| 809 | /* |
| 810 | * Scan beginning of log now. The last part of the physical |
| 811 | * log is good. This scan needs to verify that it doesn't find |
| 812 | * the last_half_cycle. |
| 813 | */ |
| 814 | start_blk = 0; |
| 815 | ASSERT(head_blk <= INT_MAX); |
| 816 | if ((error = xlog_find_verify_cycle(log, |
| 817 | start_blk, (int)head_blk, |
| 818 | stop_on_cycle, &new_blk))) |
| 819 | goto bp_err; |
| 820 | if (new_blk != -1) |
| 821 | head_blk = new_blk; |
| 822 | } |
| 823 | |
Alex Elder | 9db127e | 2010-04-15 18:17:26 +0000 | [diff] [blame] | 824 | validate_head: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 825 | /* |
| 826 | * Now we need to make sure head_blk is not pointing to a block in |
| 827 | * the middle of a log record. |
| 828 | */ |
| 829 | num_scan_bblks = XLOG_REC_SHIFT(log); |
| 830 | if (head_blk >= num_scan_bblks) { |
| 831 | start_blk = head_blk - num_scan_bblks; /* don't read head_blk */ |
| 832 | |
| 833 | /* start ptr at last block ptr before head_blk */ |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 834 | error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0); |
| 835 | if (error == 1) |
| 836 | error = -EIO; |
| 837 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 838 | goto bp_err; |
| 839 | } else { |
| 840 | start_blk = 0; |
| 841 | ASSERT(head_blk <= INT_MAX); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 842 | error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0); |
| 843 | if (error < 0) |
| 844 | goto bp_err; |
| 845 | if (error == 1) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 846 | /* We hit the beginning of the log during our search */ |
Alex Elder | 3f943d8 | 2010-04-15 18:17:34 +0000 | [diff] [blame] | 847 | start_blk = log_bbnum - (num_scan_bblks - head_blk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 848 | new_blk = log_bbnum; |
| 849 | ASSERT(start_blk <= INT_MAX && |
| 850 | (xfs_daddr_t) log_bbnum-start_blk >= 0); |
| 851 | ASSERT(head_blk <= INT_MAX); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 852 | error = xlog_find_verify_log_record(log, start_blk, |
| 853 | &new_blk, (int)head_blk); |
| 854 | if (error == 1) |
| 855 | error = -EIO; |
| 856 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 857 | goto bp_err; |
| 858 | if (new_blk != log_bbnum) |
| 859 | head_blk = new_blk; |
| 860 | } else if (error) |
| 861 | goto bp_err; |
| 862 | } |
| 863 | |
| 864 | xlog_put_bp(bp); |
| 865 | if (head_blk == log_bbnum) |
| 866 | *return_head_blk = 0; |
| 867 | else |
| 868 | *return_head_blk = head_blk; |
| 869 | /* |
| 870 | * When returning here, we have a good block number. Bad block |
| 871 | * means that during a previous crash, we didn't have a clean break |
| 872 | * from cycle number N to cycle number N-1. In this case, we need |
| 873 | * to find the first block with cycle number N-1. |
| 874 | */ |
| 875 | return 0; |
| 876 | |
| 877 | bp_err: |
| 878 | xlog_put_bp(bp); |
| 879 | |
| 880 | if (error) |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 881 | xfs_warn(log->l_mp, "failed to find log head"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 882 | return error; |
| 883 | } |
| 884 | |
| 885 | /* |
Brian Foster | eed6b46 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 886 | * Seek backwards in the log for log record headers. |
| 887 | * |
| 888 | * Given a starting log block, walk backwards until we find the provided number |
| 889 | * of records or hit the provided tail block. The return value is the number of |
| 890 | * records encountered or a negative error code. The log block and buffer |
| 891 | * pointer of the last record seen are returned in rblk and rhead respectively. |
| 892 | */ |
| 893 | STATIC int |
| 894 | xlog_rseek_logrec_hdr( |
| 895 | struct xlog *log, |
| 896 | xfs_daddr_t head_blk, |
| 897 | xfs_daddr_t tail_blk, |
| 898 | int count, |
| 899 | struct xfs_buf *bp, |
| 900 | xfs_daddr_t *rblk, |
| 901 | struct xlog_rec_header **rhead, |
| 902 | bool *wrapped) |
| 903 | { |
| 904 | int i; |
| 905 | int error; |
| 906 | int found = 0; |
| 907 | char *offset = NULL; |
| 908 | xfs_daddr_t end_blk; |
| 909 | |
| 910 | *wrapped = false; |
| 911 | |
| 912 | /* |
| 913 | * Walk backwards from the head block until we hit the tail or the first |
| 914 | * block in the log. |
| 915 | */ |
| 916 | end_blk = head_blk > tail_blk ? tail_blk : 0; |
| 917 | for (i = (int) head_blk - 1; i >= end_blk; i--) { |
| 918 | error = xlog_bread(log, i, 1, bp, &offset); |
| 919 | if (error) |
| 920 | goto out_error; |
| 921 | |
| 922 | if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) { |
| 923 | *rblk = i; |
| 924 | *rhead = (struct xlog_rec_header *) offset; |
| 925 | if (++found == count) |
| 926 | break; |
| 927 | } |
| 928 | } |
| 929 | |
| 930 | /* |
| 931 | * If we haven't hit the tail block or the log record header count, |
| 932 | * start looking again from the end of the physical log. Note that |
| 933 | * callers can pass head == tail if the tail is not yet known. |
| 934 | */ |
| 935 | if (tail_blk >= head_blk && found != count) { |
| 936 | for (i = log->l_logBBsize - 1; i >= (int) tail_blk; i--) { |
| 937 | error = xlog_bread(log, i, 1, bp, &offset); |
| 938 | if (error) |
| 939 | goto out_error; |
| 940 | |
| 941 | if (*(__be32 *)offset == |
| 942 | cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) { |
| 943 | *wrapped = true; |
| 944 | *rblk = i; |
| 945 | *rhead = (struct xlog_rec_header *) offset; |
| 946 | if (++found == count) |
| 947 | break; |
| 948 | } |
| 949 | } |
| 950 | } |
| 951 | |
| 952 | return found; |
| 953 | |
| 954 | out_error: |
| 955 | return error; |
| 956 | } |
| 957 | |
| 958 | /* |
Brian Foster | 7088c41 | 2016-01-05 07:40:16 +1100 | [diff] [blame] | 959 | * Seek forward in the log for log record headers. |
| 960 | * |
| 961 | * Given head and tail blocks, walk forward from the tail block until we find |
| 962 | * the provided number of records or hit the head block. The return value is the |
| 963 | * number of records encountered or a negative error code. The log block and |
| 964 | * buffer pointer of the last record seen are returned in rblk and rhead |
| 965 | * respectively. |
| 966 | */ |
| 967 | STATIC int |
| 968 | xlog_seek_logrec_hdr( |
| 969 | struct xlog *log, |
| 970 | xfs_daddr_t head_blk, |
| 971 | xfs_daddr_t tail_blk, |
| 972 | int count, |
| 973 | struct xfs_buf *bp, |
| 974 | xfs_daddr_t *rblk, |
| 975 | struct xlog_rec_header **rhead, |
| 976 | bool *wrapped) |
| 977 | { |
| 978 | int i; |
| 979 | int error; |
| 980 | int found = 0; |
| 981 | char *offset = NULL; |
| 982 | xfs_daddr_t end_blk; |
| 983 | |
| 984 | *wrapped = false; |
| 985 | |
| 986 | /* |
| 987 | * Walk forward from the tail block until we hit the head or the last |
| 988 | * block in the log. |
| 989 | */ |
| 990 | end_blk = head_blk > tail_blk ? head_blk : log->l_logBBsize - 1; |
| 991 | for (i = (int) tail_blk; i <= end_blk; i++) { |
| 992 | error = xlog_bread(log, i, 1, bp, &offset); |
| 993 | if (error) |
| 994 | goto out_error; |
| 995 | |
| 996 | if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) { |
| 997 | *rblk = i; |
| 998 | *rhead = (struct xlog_rec_header *) offset; |
| 999 | if (++found == count) |
| 1000 | break; |
| 1001 | } |
| 1002 | } |
| 1003 | |
| 1004 | /* |
| 1005 | * If we haven't hit the head block or the log record header count, |
| 1006 | * start looking again from the start of the physical log. |
| 1007 | */ |
| 1008 | if (tail_blk > head_blk && found != count) { |
| 1009 | for (i = 0; i < (int) head_blk; i++) { |
| 1010 | error = xlog_bread(log, i, 1, bp, &offset); |
| 1011 | if (error) |
| 1012 | goto out_error; |
| 1013 | |
| 1014 | if (*(__be32 *)offset == |
| 1015 | cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) { |
| 1016 | *wrapped = true; |
| 1017 | *rblk = i; |
| 1018 | *rhead = (struct xlog_rec_header *) offset; |
| 1019 | if (++found == count) |
| 1020 | break; |
| 1021 | } |
| 1022 | } |
| 1023 | } |
| 1024 | |
| 1025 | return found; |
| 1026 | |
| 1027 | out_error: |
| 1028 | return error; |
| 1029 | } |
| 1030 | |
| 1031 | /* |
| 1032 | * Check the log tail for torn writes. This is required when torn writes are |
| 1033 | * detected at the head and the head had to be walked back to a previous record. |
| 1034 | * The tail of the previous record must now be verified to ensure the torn |
| 1035 | * writes didn't corrupt the previous tail. |
| 1036 | * |
| 1037 | * Return an error if CRC verification fails as recovery cannot proceed. |
| 1038 | */ |
| 1039 | STATIC int |
| 1040 | xlog_verify_tail( |
| 1041 | struct xlog *log, |
| 1042 | xfs_daddr_t head_blk, |
| 1043 | xfs_daddr_t tail_blk) |
| 1044 | { |
| 1045 | struct xlog_rec_header *thead; |
| 1046 | struct xfs_buf *bp; |
| 1047 | xfs_daddr_t first_bad; |
| 1048 | int count; |
| 1049 | int error = 0; |
| 1050 | bool wrapped; |
| 1051 | xfs_daddr_t tmp_head; |
| 1052 | |
| 1053 | bp = xlog_get_bp(log, 1); |
| 1054 | if (!bp) |
| 1055 | return -ENOMEM; |
| 1056 | |
| 1057 | /* |
| 1058 | * Seek XLOG_MAX_ICLOGS + 1 records past the current tail record to get |
| 1059 | * a temporary head block that points after the last possible |
| 1060 | * concurrently written record of the tail. |
| 1061 | */ |
| 1062 | count = xlog_seek_logrec_hdr(log, head_blk, tail_blk, |
| 1063 | XLOG_MAX_ICLOGS + 1, bp, &tmp_head, &thead, |
| 1064 | &wrapped); |
| 1065 | if (count < 0) { |
| 1066 | error = count; |
| 1067 | goto out; |
| 1068 | } |
| 1069 | |
| 1070 | /* |
| 1071 | * If the call above didn't find XLOG_MAX_ICLOGS + 1 records, we ran |
| 1072 | * into the actual log head. tmp_head points to the start of the record |
| 1073 | * so update it to the actual head block. |
| 1074 | */ |
| 1075 | if (count < XLOG_MAX_ICLOGS + 1) |
| 1076 | tmp_head = head_blk; |
| 1077 | |
| 1078 | /* |
| 1079 | * We now have a tail and temporary head block that covers at least |
| 1080 | * XLOG_MAX_ICLOGS records from the tail. We need to verify that these |
| 1081 | * records were completely written. Run a CRC verification pass from |
| 1082 | * tail to head and return the result. |
| 1083 | */ |
| 1084 | error = xlog_do_recovery_pass(log, tmp_head, tail_blk, |
| 1085 | XLOG_RECOVER_CRCPASS, &first_bad); |
| 1086 | |
| 1087 | out: |
| 1088 | xlog_put_bp(bp); |
| 1089 | return error; |
| 1090 | } |
| 1091 | |
| 1092 | /* |
| 1093 | * Detect and trim torn writes from the head of the log. |
| 1094 | * |
| 1095 | * Storage without sector atomicity guarantees can result in torn writes in the |
| 1096 | * log in the event of a crash. Our only means to detect this scenario is via |
| 1097 | * CRC verification. While we can't always be certain that CRC verification |
| 1098 | * failure is due to a torn write vs. an unrelated corruption, we do know that |
| 1099 | * only a certain number (XLOG_MAX_ICLOGS) of log records can be written out at |
| 1100 | * one time. Therefore, CRC verify up to XLOG_MAX_ICLOGS records at the head of |
| 1101 | * the log and treat failures in this range as torn writes as a matter of |
| 1102 | * policy. In the event of CRC failure, the head is walked back to the last good |
| 1103 | * record in the log and the tail is updated from that record and verified. |
| 1104 | */ |
| 1105 | STATIC int |
| 1106 | xlog_verify_head( |
| 1107 | struct xlog *log, |
| 1108 | xfs_daddr_t *head_blk, /* in/out: unverified head */ |
| 1109 | xfs_daddr_t *tail_blk, /* out: tail block */ |
| 1110 | struct xfs_buf *bp, |
| 1111 | xfs_daddr_t *rhead_blk, /* start blk of last record */ |
| 1112 | struct xlog_rec_header **rhead, /* ptr to last record */ |
| 1113 | bool *wrapped) /* last rec. wraps phys. log */ |
| 1114 | { |
| 1115 | struct xlog_rec_header *tmp_rhead; |
| 1116 | struct xfs_buf *tmp_bp; |
| 1117 | xfs_daddr_t first_bad; |
| 1118 | xfs_daddr_t tmp_rhead_blk; |
| 1119 | int found; |
| 1120 | int error; |
| 1121 | bool tmp_wrapped; |
| 1122 | |
| 1123 | /* |
Brian Foster | 82ff6cc | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1124 | * Check the head of the log for torn writes. Search backwards from the |
| 1125 | * head until we hit the tail or the maximum number of log record I/Os |
| 1126 | * that could have been in flight at one time. Use a temporary buffer so |
| 1127 | * we don't trash the rhead/bp pointers from the caller. |
Brian Foster | 7088c41 | 2016-01-05 07:40:16 +1100 | [diff] [blame] | 1128 | */ |
| 1129 | tmp_bp = xlog_get_bp(log, 1); |
| 1130 | if (!tmp_bp) |
| 1131 | return -ENOMEM; |
| 1132 | error = xlog_rseek_logrec_hdr(log, *head_blk, *tail_blk, |
| 1133 | XLOG_MAX_ICLOGS, tmp_bp, &tmp_rhead_blk, |
| 1134 | &tmp_rhead, &tmp_wrapped); |
| 1135 | xlog_put_bp(tmp_bp); |
| 1136 | if (error < 0) |
| 1137 | return error; |
| 1138 | |
| 1139 | /* |
| 1140 | * Now run a CRC verification pass over the records starting at the |
| 1141 | * block found above to the current head. If a CRC failure occurs, the |
| 1142 | * log block of the first bad record is saved in first_bad. |
| 1143 | */ |
| 1144 | error = xlog_do_recovery_pass(log, *head_blk, tmp_rhead_blk, |
| 1145 | XLOG_RECOVER_CRCPASS, &first_bad); |
| 1146 | if (error == -EFSBADCRC) { |
| 1147 | /* |
| 1148 | * We've hit a potential torn write. Reset the error and warn |
| 1149 | * about it. |
| 1150 | */ |
| 1151 | error = 0; |
| 1152 | xfs_warn(log->l_mp, |
| 1153 | "Torn write (CRC failure) detected at log block 0x%llx. Truncating head block from 0x%llx.", |
| 1154 | first_bad, *head_blk); |
| 1155 | |
| 1156 | /* |
| 1157 | * Get the header block and buffer pointer for the last good |
| 1158 | * record before the bad record. |
| 1159 | * |
| 1160 | * Note that xlog_find_tail() clears the blocks at the new head |
| 1161 | * (i.e., the records with invalid CRC) if the cycle number |
| 1162 | * matches the the current cycle. |
| 1163 | */ |
| 1164 | found = xlog_rseek_logrec_hdr(log, first_bad, *tail_blk, 1, bp, |
| 1165 | rhead_blk, rhead, wrapped); |
| 1166 | if (found < 0) |
| 1167 | return found; |
| 1168 | if (found == 0) /* XXX: right thing to do here? */ |
| 1169 | return -EIO; |
| 1170 | |
| 1171 | /* |
| 1172 | * Reset the head block to the starting block of the first bad |
| 1173 | * log record and set the tail block based on the last good |
| 1174 | * record. |
| 1175 | * |
| 1176 | * Bail out if the updated head/tail match as this indicates |
| 1177 | * possible corruption outside of the acceptable |
| 1178 | * (XLOG_MAX_ICLOGS) range. This is a job for xfs_repair... |
| 1179 | */ |
| 1180 | *head_blk = first_bad; |
| 1181 | *tail_blk = BLOCK_LSN(be64_to_cpu((*rhead)->h_tail_lsn)); |
| 1182 | if (*head_blk == *tail_blk) { |
| 1183 | ASSERT(0); |
| 1184 | return 0; |
| 1185 | } |
| 1186 | |
| 1187 | /* |
| 1188 | * Now verify the tail based on the updated head. This is |
| 1189 | * required because the torn writes trimmed from the head could |
| 1190 | * have been written over the tail of a previous record. Return |
| 1191 | * any errors since recovery cannot proceed if the tail is |
| 1192 | * corrupt. |
| 1193 | * |
| 1194 | * XXX: This leaves a gap in truly robust protection from torn |
| 1195 | * writes in the log. If the head is behind the tail, the tail |
| 1196 | * pushes forward to create some space and then a crash occurs |
| 1197 | * causing the writes into the previous record's tail region to |
| 1198 | * tear, log recovery isn't able to recover. |
| 1199 | * |
| 1200 | * How likely is this to occur? If possible, can we do something |
| 1201 | * more intelligent here? Is it safe to push the tail forward if |
| 1202 | * we can determine that the tail is within the range of the |
| 1203 | * torn write (e.g., the kernel can only overwrite the tail if |
| 1204 | * it has actually been pushed forward)? Alternatively, could we |
| 1205 | * somehow prevent this condition at runtime? |
| 1206 | */ |
| 1207 | error = xlog_verify_tail(log, *head_blk, *tail_blk); |
| 1208 | } |
| 1209 | |
| 1210 | return error; |
| 1211 | } |
| 1212 | |
| 1213 | /* |
Brian Foster | 65b99a0 | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1214 | * Check whether the head of the log points to an unmount record. In other |
| 1215 | * words, determine whether the log is clean. If so, update the in-core state |
| 1216 | * appropriately. |
| 1217 | */ |
| 1218 | static int |
| 1219 | xlog_check_unmount_rec( |
| 1220 | struct xlog *log, |
| 1221 | xfs_daddr_t *head_blk, |
| 1222 | xfs_daddr_t *tail_blk, |
| 1223 | struct xlog_rec_header *rhead, |
| 1224 | xfs_daddr_t rhead_blk, |
| 1225 | struct xfs_buf *bp, |
| 1226 | bool *clean) |
| 1227 | { |
| 1228 | struct xlog_op_header *op_head; |
| 1229 | xfs_daddr_t umount_data_blk; |
| 1230 | xfs_daddr_t after_umount_blk; |
| 1231 | int hblks; |
| 1232 | int error; |
| 1233 | char *offset; |
| 1234 | |
| 1235 | *clean = false; |
| 1236 | |
| 1237 | /* |
| 1238 | * Look for unmount record. If we find it, then we know there was a |
| 1239 | * clean unmount. Since 'i' could be the last block in the physical |
| 1240 | * log, we convert to a log block before comparing to the head_blk. |
| 1241 | * |
| 1242 | * Save the current tail lsn to use to pass to xlog_clear_stale_blocks() |
| 1243 | * below. We won't want to clear the unmount record if there is one, so |
| 1244 | * we pass the lsn of the unmount record rather than the block after it. |
| 1245 | */ |
| 1246 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
| 1247 | int h_size = be32_to_cpu(rhead->h_size); |
| 1248 | int h_version = be32_to_cpu(rhead->h_version); |
| 1249 | |
| 1250 | if ((h_version & XLOG_VERSION_2) && |
| 1251 | (h_size > XLOG_HEADER_CYCLE_SIZE)) { |
| 1252 | hblks = h_size / XLOG_HEADER_CYCLE_SIZE; |
| 1253 | if (h_size % XLOG_HEADER_CYCLE_SIZE) |
| 1254 | hblks++; |
| 1255 | } else { |
| 1256 | hblks = 1; |
| 1257 | } |
| 1258 | } else { |
| 1259 | hblks = 1; |
| 1260 | } |
| 1261 | after_umount_blk = rhead_blk + hblks + BTOBB(be32_to_cpu(rhead->h_len)); |
| 1262 | after_umount_blk = do_mod(after_umount_blk, log->l_logBBsize); |
| 1263 | if (*head_blk == after_umount_blk && |
| 1264 | be32_to_cpu(rhead->h_num_logops) == 1) { |
| 1265 | umount_data_blk = rhead_blk + hblks; |
| 1266 | umount_data_blk = do_mod(umount_data_blk, log->l_logBBsize); |
| 1267 | error = xlog_bread(log, umount_data_blk, 1, bp, &offset); |
| 1268 | if (error) |
| 1269 | return error; |
| 1270 | |
| 1271 | op_head = (struct xlog_op_header *)offset; |
| 1272 | if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) { |
| 1273 | /* |
| 1274 | * Set tail and last sync so that newly written log |
| 1275 | * records will point recovery to after the current |
| 1276 | * unmount record. |
| 1277 | */ |
| 1278 | xlog_assign_atomic_lsn(&log->l_tail_lsn, |
| 1279 | log->l_curr_cycle, after_umount_blk); |
| 1280 | xlog_assign_atomic_lsn(&log->l_last_sync_lsn, |
| 1281 | log->l_curr_cycle, after_umount_blk); |
| 1282 | *tail_blk = after_umount_blk; |
| 1283 | |
| 1284 | *clean = true; |
| 1285 | } |
| 1286 | } |
| 1287 | |
| 1288 | return 0; |
| 1289 | } |
| 1290 | |
Brian Foster | 717bc0e | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1291 | static void |
| 1292 | xlog_set_state( |
| 1293 | struct xlog *log, |
| 1294 | xfs_daddr_t head_blk, |
| 1295 | struct xlog_rec_header *rhead, |
| 1296 | xfs_daddr_t rhead_blk, |
| 1297 | bool bump_cycle) |
| 1298 | { |
| 1299 | /* |
| 1300 | * Reset log values according to the state of the log when we |
| 1301 | * crashed. In the case where head_blk == 0, we bump curr_cycle |
| 1302 | * one because the next write starts a new cycle rather than |
| 1303 | * continuing the cycle of the last good log record. At this |
| 1304 | * point we have guaranteed that all partial log records have been |
| 1305 | * accounted for. Therefore, we know that the last good log record |
| 1306 | * written was complete and ended exactly on the end boundary |
| 1307 | * of the physical log. |
| 1308 | */ |
| 1309 | log->l_prev_block = rhead_blk; |
| 1310 | log->l_curr_block = (int)head_blk; |
| 1311 | log->l_curr_cycle = be32_to_cpu(rhead->h_cycle); |
| 1312 | if (bump_cycle) |
| 1313 | log->l_curr_cycle++; |
| 1314 | atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn)); |
| 1315 | atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn)); |
| 1316 | xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle, |
| 1317 | BBTOB(log->l_curr_block)); |
| 1318 | xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle, |
| 1319 | BBTOB(log->l_curr_block)); |
| 1320 | } |
| 1321 | |
Brian Foster | 65b99a0 | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1322 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1323 | * Find the sync block number or the tail of the log. |
| 1324 | * |
| 1325 | * This will be the block number of the last record to have its |
| 1326 | * associated buffers synced to disk. Every log record header has |
| 1327 | * a sync lsn embedded in it. LSNs hold block numbers, so it is easy |
| 1328 | * to get a sync block number. The only concern is to figure out which |
| 1329 | * log record header to believe. |
| 1330 | * |
| 1331 | * The following algorithm uses the log record header with the largest |
| 1332 | * lsn. The entire log record does not need to be valid. We only care |
| 1333 | * that the header is valid. |
| 1334 | * |
| 1335 | * We could speed up search by using current head_blk buffer, but it is not |
| 1336 | * available. |
| 1337 | */ |
Eric Sandeen | 5d77c0d | 2009-11-19 15:52:00 +0000 | [diff] [blame] | 1338 | STATIC int |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1339 | xlog_find_tail( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 1340 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1341 | xfs_daddr_t *head_blk, |
Eric Sandeen | 65be605 | 2006-01-11 15:34:19 +1100 | [diff] [blame] | 1342 | xfs_daddr_t *tail_blk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1343 | { |
| 1344 | xlog_rec_header_t *rhead; |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 1345 | char *offset = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1346 | xfs_buf_t *bp; |
Brian Foster | 7088c41 | 2016-01-05 07:40:16 +1100 | [diff] [blame] | 1347 | int error; |
Brian Foster | 7088c41 | 2016-01-05 07:40:16 +1100 | [diff] [blame] | 1348 | xfs_daddr_t rhead_blk; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1349 | xfs_lsn_t tail_lsn; |
Brian Foster | eed6b46 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 1350 | bool wrapped = false; |
Brian Foster | 65b99a0 | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1351 | bool clean = false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1352 | |
| 1353 | /* |
| 1354 | * Find previous log record |
| 1355 | */ |
| 1356 | if ((error = xlog_find_head(log, head_blk))) |
| 1357 | return error; |
Brian Foster | 82ff6cc | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1358 | ASSERT(*head_blk < INT_MAX); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1359 | |
| 1360 | bp = xlog_get_bp(log, 1); |
| 1361 | if (!bp) |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1362 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1363 | if (*head_blk == 0) { /* special case */ |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1364 | error = xlog_bread(log, 0, 1, bp, &offset); |
| 1365 | if (error) |
Alex Elder | 9db127e | 2010-04-15 18:17:26 +0000 | [diff] [blame] | 1366 | goto done; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1367 | |
Christoph Hellwig | 03bea6f | 2007-10-12 10:58:05 +1000 | [diff] [blame] | 1368 | if (xlog_get_cycle(offset) == 0) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1369 | *tail_blk = 0; |
| 1370 | /* leave all other log inited values alone */ |
Alex Elder | 9db127e | 2010-04-15 18:17:26 +0000 | [diff] [blame] | 1371 | goto done; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1372 | } |
| 1373 | } |
| 1374 | |
| 1375 | /* |
Brian Foster | 82ff6cc | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1376 | * Search backwards through the log looking for the log record header |
| 1377 | * block. This wraps all the way back around to the head so something is |
| 1378 | * seriously wrong if we can't find it. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1379 | */ |
Brian Foster | 82ff6cc | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1380 | error = xlog_rseek_logrec_hdr(log, *head_blk, *head_blk, 1, bp, |
| 1381 | &rhead_blk, &rhead, &wrapped); |
| 1382 | if (error < 0) |
| 1383 | return error; |
| 1384 | if (!error) { |
| 1385 | xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__); |
| 1386 | return -EIO; |
| 1387 | } |
| 1388 | *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn)); |
| 1389 | |
| 1390 | /* |
Brian Foster | 717bc0e | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1391 | * Set the log state based on the current head record. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1392 | */ |
Brian Foster | 717bc0e | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1393 | xlog_set_state(log, *head_blk, rhead, rhead_blk, wrapped); |
Brian Foster | 65b99a0 | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1394 | tail_lsn = atomic64_read(&log->l_tail_lsn); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1395 | |
| 1396 | /* |
Brian Foster | 65b99a0 | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1397 | * Look for an unmount record at the head of the log. This sets the log |
| 1398 | * state to determine whether recovery is necessary. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1399 | */ |
Brian Foster | 65b99a0 | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1400 | error = xlog_check_unmount_rec(log, head_blk, tail_blk, rhead, |
| 1401 | rhead_blk, bp, &clean); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1402 | if (error) |
| 1403 | goto done; |
| 1404 | |
| 1405 | /* |
Brian Foster | 7f6aff3a | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1406 | * Verify the log head if the log is not clean (e.g., we have anything |
| 1407 | * but an unmount record at the head). This uses CRC verification to |
| 1408 | * detect and trim torn writes. If discovered, CRC failures are |
| 1409 | * considered torn writes and the log head is trimmed accordingly. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1410 | * |
Brian Foster | 7f6aff3a | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1411 | * Note that we can only run CRC verification when the log is dirty |
| 1412 | * because there's no guarantee that the log data behind an unmount |
| 1413 | * record is compatible with the current architecture. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1414 | */ |
Brian Foster | 7f6aff3a | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1415 | if (!clean) { |
| 1416 | xfs_daddr_t orig_head = *head_blk; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1417 | |
Brian Foster | 7f6aff3a | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1418 | error = xlog_verify_head(log, head_blk, tail_blk, bp, |
| 1419 | &rhead_blk, &rhead, &wrapped); |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1420 | if (error) |
Alex Elder | 9db127e | 2010-04-15 18:17:26 +0000 | [diff] [blame] | 1421 | goto done; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1422 | |
Brian Foster | 7f6aff3a | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1423 | /* update in-core state again if the head changed */ |
| 1424 | if (*head_blk != orig_head) { |
| 1425 | xlog_set_state(log, *head_blk, rhead, rhead_blk, |
| 1426 | wrapped); |
| 1427 | tail_lsn = atomic64_read(&log->l_tail_lsn); |
| 1428 | error = xlog_check_unmount_rec(log, head_blk, tail_blk, |
| 1429 | rhead, rhead_blk, bp, |
| 1430 | &clean); |
| 1431 | if (error) |
| 1432 | goto done; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1433 | } |
| 1434 | } |
| 1435 | |
| 1436 | /* |
Brian Foster | 65b99a0 | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1437 | * Note that the unmount was clean. If the unmount was not clean, we |
| 1438 | * need to know this to rebuild the superblock counters from the perag |
| 1439 | * headers if we have a filesystem using non-persistent counters. |
| 1440 | */ |
| 1441 | if (clean) |
| 1442 | log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1443 | |
| 1444 | /* |
| 1445 | * Make sure that there are no blocks in front of the head |
| 1446 | * with the same cycle number as the head. This can happen |
| 1447 | * because we allow multiple outstanding log writes concurrently, |
| 1448 | * and the later writes might make it out before earlier ones. |
| 1449 | * |
| 1450 | * We use the lsn from before modifying it so that we'll never |
| 1451 | * overwrite the unmount record after a clean unmount. |
| 1452 | * |
| 1453 | * Do this only if we are going to recover the filesystem |
| 1454 | * |
| 1455 | * NOTE: This used to say "if (!readonly)" |
| 1456 | * However on Linux, we can & do recover a read-only filesystem. |
| 1457 | * We only skip recovery if NORECOVERY is specified on mount, |
| 1458 | * in which case we would not be here. |
| 1459 | * |
| 1460 | * But... if the -device- itself is readonly, just skip this. |
| 1461 | * We can't recover this device anyway, so it won't matter. |
| 1462 | */ |
Alex Elder | 9db127e | 2010-04-15 18:17:26 +0000 | [diff] [blame] | 1463 | if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1464 | error = xlog_clear_stale_blocks(log, tail_lsn); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1465 | |
Alex Elder | 9db127e | 2010-04-15 18:17:26 +0000 | [diff] [blame] | 1466 | done: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1467 | xlog_put_bp(bp); |
| 1468 | |
| 1469 | if (error) |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 1470 | xfs_warn(log->l_mp, "failed to locate log tail"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1471 | return error; |
| 1472 | } |
| 1473 | |
| 1474 | /* |
| 1475 | * Is the log zeroed at all? |
| 1476 | * |
| 1477 | * The last binary search should be changed to perform an X block read |
| 1478 | * once X becomes small enough. You can then search linearly through |
| 1479 | * the X blocks. This will cut down on the number of reads we need to do. |
| 1480 | * |
| 1481 | * If the log is partially zeroed, this routine will pass back the blkno |
| 1482 | * of the first block with cycle number 0. It won't have a complete LR |
| 1483 | * preceding it. |
| 1484 | * |
| 1485 | * Return: |
| 1486 | * 0 => the log is completely written to |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1487 | * 1 => use *blk_no as the first block of the log |
| 1488 | * <0 => error has occurred |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1489 | */ |
David Chinner | a8272ce | 2007-11-23 16:28:09 +1100 | [diff] [blame] | 1490 | STATIC int |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1491 | xlog_find_zeroed( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 1492 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1493 | xfs_daddr_t *blk_no) |
| 1494 | { |
| 1495 | xfs_buf_t *bp; |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 1496 | char *offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1497 | uint first_cycle, last_cycle; |
| 1498 | xfs_daddr_t new_blk, last_blk, start_blk; |
| 1499 | xfs_daddr_t num_scan_bblks; |
| 1500 | int error, log_bbnum = log->l_logBBsize; |
| 1501 | |
Nathan Scott | 6fdf8cc | 2006-06-28 10:13:52 +1000 | [diff] [blame] | 1502 | *blk_no = 0; |
| 1503 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1504 | /* check totally zeroed log */ |
| 1505 | bp = xlog_get_bp(log, 1); |
| 1506 | if (!bp) |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1507 | return -ENOMEM; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1508 | error = xlog_bread(log, 0, 1, bp, &offset); |
| 1509 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1510 | goto bp_err; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1511 | |
Christoph Hellwig | 03bea6f | 2007-10-12 10:58:05 +1000 | [diff] [blame] | 1512 | first_cycle = xlog_get_cycle(offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1513 | if (first_cycle == 0) { /* completely zeroed log */ |
| 1514 | *blk_no = 0; |
| 1515 | xlog_put_bp(bp); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1516 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1517 | } |
| 1518 | |
| 1519 | /* check partially zeroed log */ |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1520 | error = xlog_bread(log, log_bbnum-1, 1, bp, &offset); |
| 1521 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1522 | goto bp_err; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1523 | |
Christoph Hellwig | 03bea6f | 2007-10-12 10:58:05 +1000 | [diff] [blame] | 1524 | last_cycle = xlog_get_cycle(offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1525 | if (last_cycle != 0) { /* log completely written to */ |
| 1526 | xlog_put_bp(bp); |
| 1527 | return 0; |
| 1528 | } else if (first_cycle != 1) { |
| 1529 | /* |
| 1530 | * If the cycle of the last block is zero, the cycle of |
| 1531 | * the first block must be 1. If it's not, maybe we're |
| 1532 | * not looking at a log... Bail out. |
| 1533 | */ |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 1534 | xfs_warn(log->l_mp, |
| 1535 | "Log inconsistent or not a log (last==0, first!=1)"); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1536 | error = -EINVAL; |
Eric Sandeen | 5d0a654 | 2013-07-31 20:32:30 -0500 | [diff] [blame] | 1537 | goto bp_err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1538 | } |
| 1539 | |
| 1540 | /* we have a partially zeroed log */ |
| 1541 | last_blk = log_bbnum-1; |
| 1542 | if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0))) |
| 1543 | goto bp_err; |
| 1544 | |
| 1545 | /* |
| 1546 | * Validate the answer. Because there is no way to guarantee that |
| 1547 | * the entire log is made up of log records which are the same size, |
| 1548 | * we scan over the defined maximum blocks. At this point, the maximum |
| 1549 | * is not chosen to mean anything special. XXXmiken |
| 1550 | */ |
| 1551 | num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log); |
| 1552 | ASSERT(num_scan_bblks <= INT_MAX); |
| 1553 | |
| 1554 | if (last_blk < num_scan_bblks) |
| 1555 | num_scan_bblks = last_blk; |
| 1556 | start_blk = last_blk - num_scan_bblks; |
| 1557 | |
| 1558 | /* |
| 1559 | * We search for any instances of cycle number 0 that occur before |
| 1560 | * our current estimate of the head. What we're trying to detect is |
| 1561 | * 1 ... | 0 | 1 | 0... |
| 1562 | * ^ binary search ends here |
| 1563 | */ |
| 1564 | if ((error = xlog_find_verify_cycle(log, start_blk, |
| 1565 | (int)num_scan_bblks, 0, &new_blk))) |
| 1566 | goto bp_err; |
| 1567 | if (new_blk != -1) |
| 1568 | last_blk = new_blk; |
| 1569 | |
| 1570 | /* |
| 1571 | * Potentially backup over partial log record write. We don't need |
| 1572 | * to search the end of the log because we know it is zero. |
| 1573 | */ |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1574 | error = xlog_find_verify_log_record(log, start_blk, &last_blk, 0); |
| 1575 | if (error == 1) |
| 1576 | error = -EIO; |
| 1577 | if (error) |
| 1578 | goto bp_err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1579 | |
| 1580 | *blk_no = last_blk; |
| 1581 | bp_err: |
| 1582 | xlog_put_bp(bp); |
| 1583 | if (error) |
| 1584 | return error; |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1585 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1586 | } |
| 1587 | |
| 1588 | /* |
| 1589 | * These are simple subroutines used by xlog_clear_stale_blocks() below |
| 1590 | * to initialize a buffer full of empty log record headers and write |
| 1591 | * them into the log. |
| 1592 | */ |
| 1593 | STATIC void |
| 1594 | xlog_add_record( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 1595 | struct xlog *log, |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 1596 | char *buf, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1597 | int cycle, |
| 1598 | int block, |
| 1599 | int tail_cycle, |
| 1600 | int tail_block) |
| 1601 | { |
| 1602 | xlog_rec_header_t *recp = (xlog_rec_header_t *)buf; |
| 1603 | |
| 1604 | memset(buf, 0, BBSIZE); |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 1605 | recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM); |
| 1606 | recp->h_cycle = cpu_to_be32(cycle); |
| 1607 | recp->h_version = cpu_to_be32( |
Eric Sandeen | 6211870 | 2008-03-06 13:44:28 +1100 | [diff] [blame] | 1608 | xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1); |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 1609 | recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block)); |
| 1610 | recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block)); |
| 1611 | recp->h_fmt = cpu_to_be32(XLOG_FMT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1612 | memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t)); |
| 1613 | } |
| 1614 | |
| 1615 | STATIC int |
| 1616 | xlog_write_log_records( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 1617 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1618 | int cycle, |
| 1619 | int start_block, |
| 1620 | int blocks, |
| 1621 | int tail_cycle, |
| 1622 | int tail_block) |
| 1623 | { |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 1624 | char *offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1625 | xfs_buf_t *bp; |
| 1626 | int balign, ealign; |
Alex Elder | 69ce58f | 2010-04-20 17:09:59 +1000 | [diff] [blame] | 1627 | int sectbb = log->l_sectBBsize; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1628 | int end_block = start_block + blocks; |
| 1629 | int bufblks; |
| 1630 | int error = 0; |
| 1631 | int i, j = 0; |
| 1632 | |
Alex Elder | 6881a22 | 2010-04-13 15:22:29 +1000 | [diff] [blame] | 1633 | /* |
| 1634 | * Greedily allocate a buffer big enough to handle the full |
| 1635 | * range of basic blocks to be written. If that fails, try |
| 1636 | * a smaller size. We need to be able to write at least a |
| 1637 | * log sector, or we're out of luck. |
| 1638 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1639 | bufblks = 1 << ffs(blocks); |
Dave Chinner | 81158e0 | 2012-04-27 19:45:22 +1000 | [diff] [blame] | 1640 | while (bufblks > log->l_logBBsize) |
| 1641 | bufblks >>= 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1642 | while (!(bp = xlog_get_bp(log, bufblks))) { |
| 1643 | bufblks >>= 1; |
Alex Elder | 69ce58f | 2010-04-20 17:09:59 +1000 | [diff] [blame] | 1644 | if (bufblks < sectbb) |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1645 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1646 | } |
| 1647 | |
| 1648 | /* We may need to do a read at the start to fill in part of |
| 1649 | * the buffer in the starting sector not covered by the first |
| 1650 | * write below. |
| 1651 | */ |
Alex Elder | 5c17f53 | 2010-04-13 15:22:48 +1000 | [diff] [blame] | 1652 | balign = round_down(start_block, sectbb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1653 | if (balign != start_block) { |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1654 | error = xlog_bread_noalign(log, start_block, 1, bp); |
| 1655 | if (error) |
| 1656 | goto out_put_bp; |
| 1657 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1658 | j = start_block - balign; |
| 1659 | } |
| 1660 | |
| 1661 | for (i = start_block; i < end_block; i += bufblks) { |
| 1662 | int bcount, endcount; |
| 1663 | |
| 1664 | bcount = min(bufblks, end_block - start_block); |
| 1665 | endcount = bcount - j; |
| 1666 | |
| 1667 | /* We may need to do a read at the end to fill in part of |
| 1668 | * the buffer in the final sector not covered by the write. |
| 1669 | * If this is the same sector as the above read, skip it. |
| 1670 | */ |
Alex Elder | 5c17f53 | 2010-04-13 15:22:48 +1000 | [diff] [blame] | 1671 | ealign = round_down(end_block, sectbb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1672 | if (j == 0 && (start_block + endcount > ealign)) { |
Chandra Seetharaman | 6292604 | 2011-07-22 23:40:15 +0000 | [diff] [blame] | 1673 | offset = bp->b_addr + BBTOB(ealign - start_block); |
Dave Chinner | 4439647 | 2011-04-21 09:34:27 +0000 | [diff] [blame] | 1674 | error = xlog_bread_offset(log, ealign, sectbb, |
| 1675 | bp, offset); |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1676 | if (error) |
| 1677 | break; |
| 1678 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1679 | } |
| 1680 | |
| 1681 | offset = xlog_align(log, start_block, endcount, bp); |
| 1682 | for (; j < endcount; j++) { |
| 1683 | xlog_add_record(log, offset, cycle, i+j, |
| 1684 | tail_cycle, tail_block); |
| 1685 | offset += BBSIZE; |
| 1686 | } |
| 1687 | error = xlog_bwrite(log, start_block, endcount, bp); |
| 1688 | if (error) |
| 1689 | break; |
| 1690 | start_block += endcount; |
| 1691 | j = 0; |
| 1692 | } |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1693 | |
| 1694 | out_put_bp: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1695 | xlog_put_bp(bp); |
| 1696 | return error; |
| 1697 | } |
| 1698 | |
| 1699 | /* |
| 1700 | * This routine is called to blow away any incomplete log writes out |
| 1701 | * in front of the log head. We do this so that we won't become confused |
| 1702 | * if we come up, write only a little bit more, and then crash again. |
| 1703 | * If we leave the partial log records out there, this situation could |
| 1704 | * cause us to think those partial writes are valid blocks since they |
| 1705 | * have the current cycle number. We get rid of them by overwriting them |
| 1706 | * with empty log records with the old cycle number rather than the |
| 1707 | * current one. |
| 1708 | * |
| 1709 | * The tail lsn is passed in rather than taken from |
| 1710 | * the log so that we will not write over the unmount record after a |
| 1711 | * clean unmount in a 512 block log. Doing so would leave the log without |
| 1712 | * any valid log records in it until a new one was written. If we crashed |
| 1713 | * during that time we would not be able to recover. |
| 1714 | */ |
| 1715 | STATIC int |
| 1716 | xlog_clear_stale_blocks( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 1717 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1718 | xfs_lsn_t tail_lsn) |
| 1719 | { |
| 1720 | int tail_cycle, head_cycle; |
| 1721 | int tail_block, head_block; |
| 1722 | int tail_distance, max_distance; |
| 1723 | int distance; |
| 1724 | int error; |
| 1725 | |
| 1726 | tail_cycle = CYCLE_LSN(tail_lsn); |
| 1727 | tail_block = BLOCK_LSN(tail_lsn); |
| 1728 | head_cycle = log->l_curr_cycle; |
| 1729 | head_block = log->l_curr_block; |
| 1730 | |
| 1731 | /* |
| 1732 | * Figure out the distance between the new head of the log |
| 1733 | * and the tail. We want to write over any blocks beyond the |
| 1734 | * head that we may have written just before the crash, but |
| 1735 | * we don't want to overwrite the tail of the log. |
| 1736 | */ |
| 1737 | if (head_cycle == tail_cycle) { |
| 1738 | /* |
| 1739 | * The tail is behind the head in the physical log, |
| 1740 | * so the distance from the head to the tail is the |
| 1741 | * distance from the head to the end of the log plus |
| 1742 | * the distance from the beginning of the log to the |
| 1743 | * tail. |
| 1744 | */ |
| 1745 | if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) { |
| 1746 | XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)", |
| 1747 | XFS_ERRLEVEL_LOW, log->l_mp); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1748 | return -EFSCORRUPTED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1749 | } |
| 1750 | tail_distance = tail_block + (log->l_logBBsize - head_block); |
| 1751 | } else { |
| 1752 | /* |
| 1753 | * The head is behind the tail in the physical log, |
| 1754 | * so the distance from the head to the tail is just |
| 1755 | * the tail block minus the head block. |
| 1756 | */ |
| 1757 | if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){ |
| 1758 | XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)", |
| 1759 | XFS_ERRLEVEL_LOW, log->l_mp); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1760 | return -EFSCORRUPTED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1761 | } |
| 1762 | tail_distance = tail_block - head_block; |
| 1763 | } |
| 1764 | |
| 1765 | /* |
| 1766 | * If the head is right up against the tail, we can't clear |
| 1767 | * anything. |
| 1768 | */ |
| 1769 | if (tail_distance <= 0) { |
| 1770 | ASSERT(tail_distance == 0); |
| 1771 | return 0; |
| 1772 | } |
| 1773 | |
| 1774 | max_distance = XLOG_TOTAL_REC_SHIFT(log); |
| 1775 | /* |
| 1776 | * Take the smaller of the maximum amount of outstanding I/O |
| 1777 | * we could have and the distance to the tail to clear out. |
| 1778 | * We take the smaller so that we don't overwrite the tail and |
| 1779 | * we don't waste all day writing from the head to the tail |
| 1780 | * for no reason. |
| 1781 | */ |
| 1782 | max_distance = MIN(max_distance, tail_distance); |
| 1783 | |
| 1784 | if ((head_block + max_distance) <= log->l_logBBsize) { |
| 1785 | /* |
| 1786 | * We can stomp all the blocks we need to without |
| 1787 | * wrapping around the end of the log. Just do it |
| 1788 | * in a single write. Use the cycle number of the |
| 1789 | * current cycle minus one so that the log will look like: |
| 1790 | * n ... | n - 1 ... |
| 1791 | */ |
| 1792 | error = xlog_write_log_records(log, (head_cycle - 1), |
| 1793 | head_block, max_distance, tail_cycle, |
| 1794 | tail_block); |
| 1795 | if (error) |
| 1796 | return error; |
| 1797 | } else { |
| 1798 | /* |
| 1799 | * We need to wrap around the end of the physical log in |
| 1800 | * order to clear all the blocks. Do it in two separate |
| 1801 | * I/Os. The first write should be from the head to the |
| 1802 | * end of the physical log, and it should use the current |
| 1803 | * cycle number minus one just like above. |
| 1804 | */ |
| 1805 | distance = log->l_logBBsize - head_block; |
| 1806 | error = xlog_write_log_records(log, (head_cycle - 1), |
| 1807 | head_block, distance, tail_cycle, |
| 1808 | tail_block); |
| 1809 | |
| 1810 | if (error) |
| 1811 | return error; |
| 1812 | |
| 1813 | /* |
| 1814 | * Now write the blocks at the start of the physical log. |
| 1815 | * This writes the remainder of the blocks we want to clear. |
| 1816 | * It uses the current cycle number since we're now on the |
| 1817 | * same cycle as the head so that we get: |
| 1818 | * n ... n ... | n - 1 ... |
| 1819 | * ^^^^^ blocks we're writing |
| 1820 | */ |
| 1821 | distance = max_distance - (log->l_logBBsize - head_block); |
| 1822 | error = xlog_write_log_records(log, head_cycle, 0, distance, |
| 1823 | tail_cycle, tail_block); |
| 1824 | if (error) |
| 1825 | return error; |
| 1826 | } |
| 1827 | |
| 1828 | return 0; |
| 1829 | } |
| 1830 | |
| 1831 | /****************************************************************************** |
| 1832 | * |
| 1833 | * Log recover routines |
| 1834 | * |
| 1835 | ****************************************************************************** |
| 1836 | */ |
| 1837 | |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1838 | /* |
Dave Chinner | a775ad7 | 2013-06-05 12:09:07 +1000 | [diff] [blame] | 1839 | * Sort the log items in the transaction. |
| 1840 | * |
| 1841 | * The ordering constraints are defined by the inode allocation and unlink |
| 1842 | * behaviour. The rules are: |
| 1843 | * |
| 1844 | * 1. Every item is only logged once in a given transaction. Hence it |
| 1845 | * represents the last logged state of the item. Hence ordering is |
| 1846 | * dependent on the order in which operations need to be performed so |
| 1847 | * required initial conditions are always met. |
| 1848 | * |
| 1849 | * 2. Cancelled buffers are recorded in pass 1 in a separate table and |
| 1850 | * there's nothing to replay from them so we can simply cull them |
| 1851 | * from the transaction. However, we can't do that until after we've |
| 1852 | * replayed all the other items because they may be dependent on the |
| 1853 | * cancelled buffer and replaying the cancelled buffer can remove it |
| 1854 | * form the cancelled buffer table. Hence they have tobe done last. |
| 1855 | * |
| 1856 | * 3. Inode allocation buffers must be replayed before inode items that |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 1857 | * read the buffer and replay changes into it. For filesystems using the |
| 1858 | * ICREATE transactions, this means XFS_LI_ICREATE objects need to get |
| 1859 | * treated the same as inode allocation buffers as they create and |
| 1860 | * initialise the buffers directly. |
Dave Chinner | a775ad7 | 2013-06-05 12:09:07 +1000 | [diff] [blame] | 1861 | * |
| 1862 | * 4. Inode unlink buffers must be replayed after inode items are replayed. |
| 1863 | * This ensures that inodes are completely flushed to the inode buffer |
| 1864 | * in a "free" state before we remove the unlinked inode list pointer. |
| 1865 | * |
| 1866 | * Hence the ordering needs to be inode allocation buffers first, inode items |
| 1867 | * second, inode unlink buffers third and cancelled buffers last. |
| 1868 | * |
| 1869 | * But there's a problem with that - we can't tell an inode allocation buffer |
| 1870 | * apart from a regular buffer, so we can't separate them. We can, however, |
| 1871 | * tell an inode unlink buffer from the others, and so we can separate them out |
| 1872 | * from all the other buffers and move them to last. |
| 1873 | * |
| 1874 | * Hence, 4 lists, in order from head to tail: |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 1875 | * - buffer_list for all buffers except cancelled/inode unlink buffers |
| 1876 | * - item_list for all non-buffer items |
| 1877 | * - inode_buffer_list for inode unlink buffers |
| 1878 | * - cancel_list for the cancelled buffers |
| 1879 | * |
| 1880 | * Note that we add objects to the tail of the lists so that first-to-last |
| 1881 | * ordering is preserved within the lists. Adding objects to the head of the |
| 1882 | * list means when we traverse from the head we walk them in last-to-first |
| 1883 | * order. For cancelled buffers and inode unlink buffers this doesn't matter, |
| 1884 | * but for all other items there may be specific ordering that we need to |
| 1885 | * preserve. |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1886 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1887 | STATIC int |
| 1888 | xlog_recover_reorder_trans( |
Mark Tinguely | ad223e6 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 1889 | struct xlog *log, |
| 1890 | struct xlog_recover *trans, |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 1891 | int pass) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1892 | { |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1893 | xlog_recover_item_t *item, *n; |
Mark Tinguely | 2a84108 | 2013-10-02 07:51:12 -0500 | [diff] [blame] | 1894 | int error = 0; |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1895 | LIST_HEAD(sort_list); |
Dave Chinner | a775ad7 | 2013-06-05 12:09:07 +1000 | [diff] [blame] | 1896 | LIST_HEAD(cancel_list); |
| 1897 | LIST_HEAD(buffer_list); |
| 1898 | LIST_HEAD(inode_buffer_list); |
| 1899 | LIST_HEAD(inode_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1900 | |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1901 | list_splice_init(&trans->r_itemq, &sort_list); |
| 1902 | list_for_each_entry_safe(item, n, &sort_list, ri_list) { |
Christoph Hellwig | 4e0d5f9 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 1903 | xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr; |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1904 | |
| 1905 | switch (ITEM_TYPE(item)) { |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 1906 | case XFS_LI_ICREATE: |
| 1907 | list_move_tail(&item->ri_list, &buffer_list); |
| 1908 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1909 | case XFS_LI_BUF: |
Dave Chinner | a775ad7 | 2013-06-05 12:09:07 +1000 | [diff] [blame] | 1910 | if (buf_f->blf_flags & XFS_BLF_CANCEL) { |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 1911 | trace_xfs_log_recover_item_reorder_head(log, |
| 1912 | trans, item, pass); |
Dave Chinner | a775ad7 | 2013-06-05 12:09:07 +1000 | [diff] [blame] | 1913 | list_move(&item->ri_list, &cancel_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1914 | break; |
| 1915 | } |
Dave Chinner | a775ad7 | 2013-06-05 12:09:07 +1000 | [diff] [blame] | 1916 | if (buf_f->blf_flags & XFS_BLF_INODE_BUF) { |
| 1917 | list_move(&item->ri_list, &inode_buffer_list); |
| 1918 | break; |
| 1919 | } |
| 1920 | list_move_tail(&item->ri_list, &buffer_list); |
| 1921 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1922 | case XFS_LI_INODE: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1923 | case XFS_LI_DQUOT: |
| 1924 | case XFS_LI_QUOTAOFF: |
| 1925 | case XFS_LI_EFD: |
| 1926 | case XFS_LI_EFI: |
Darrick J. Wong | 9e88b5d | 2016-08-03 12:09:48 +1000 | [diff] [blame] | 1927 | case XFS_LI_RUI: |
| 1928 | case XFS_LI_RUD: |
Darrick J. Wong | f997ee2 | 2016-10-03 09:11:21 -0700 | [diff] [blame] | 1929 | case XFS_LI_CUI: |
| 1930 | case XFS_LI_CUD: |
Darrick J. Wong | 77d61fe | 2016-10-03 09:11:26 -0700 | [diff] [blame] | 1931 | case XFS_LI_BUI: |
| 1932 | case XFS_LI_BUD: |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 1933 | trace_xfs_log_recover_item_reorder_tail(log, |
| 1934 | trans, item, pass); |
Dave Chinner | a775ad7 | 2013-06-05 12:09:07 +1000 | [diff] [blame] | 1935 | list_move_tail(&item->ri_list, &inode_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1936 | break; |
| 1937 | default: |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 1938 | xfs_warn(log->l_mp, |
| 1939 | "%s: unrecognized type of log operation", |
| 1940 | __func__); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1941 | ASSERT(0); |
Mark Tinguely | 2a84108 | 2013-10-02 07:51:12 -0500 | [diff] [blame] | 1942 | /* |
| 1943 | * return the remaining items back to the transaction |
| 1944 | * item list so they can be freed in caller. |
| 1945 | */ |
| 1946 | if (!list_empty(&sort_list)) |
| 1947 | list_splice_init(&sort_list, &trans->r_itemq); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1948 | error = -EIO; |
Mark Tinguely | 2a84108 | 2013-10-02 07:51:12 -0500 | [diff] [blame] | 1949 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1950 | } |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1951 | } |
Mark Tinguely | 2a84108 | 2013-10-02 07:51:12 -0500 | [diff] [blame] | 1952 | out: |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1953 | ASSERT(list_empty(&sort_list)); |
Dave Chinner | a775ad7 | 2013-06-05 12:09:07 +1000 | [diff] [blame] | 1954 | if (!list_empty(&buffer_list)) |
| 1955 | list_splice(&buffer_list, &trans->r_itemq); |
| 1956 | if (!list_empty(&inode_list)) |
| 1957 | list_splice_tail(&inode_list, &trans->r_itemq); |
| 1958 | if (!list_empty(&inode_buffer_list)) |
| 1959 | list_splice_tail(&inode_buffer_list, &trans->r_itemq); |
| 1960 | if (!list_empty(&cancel_list)) |
| 1961 | list_splice_tail(&cancel_list, &trans->r_itemq); |
Mark Tinguely | 2a84108 | 2013-10-02 07:51:12 -0500 | [diff] [blame] | 1962 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1963 | } |
| 1964 | |
| 1965 | /* |
| 1966 | * Build up the table of buf cancel records so that we don't replay |
| 1967 | * cancelled data in the second pass. For buffer records that are |
| 1968 | * not cancel records, there is nothing to do here so we just return. |
| 1969 | * |
| 1970 | * If we get a cancel record which is already in the table, this indicates |
| 1971 | * that the buffer was cancelled multiple times. In order to ensure |
| 1972 | * that during pass 2 we keep the record in the table until we reach its |
| 1973 | * last occurrence in the log, we keep a reference count in the cancel |
| 1974 | * record in the table to tell us how many times we expect to see this |
| 1975 | * record during the second pass. |
| 1976 | */ |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 1977 | STATIC int |
| 1978 | xlog_recover_buffer_pass1( |
Mark Tinguely | ad223e6 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 1979 | struct xlog *log, |
| 1980 | struct xlog_recover_item *item) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1981 | { |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 1982 | xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr; |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 1983 | struct list_head *bucket; |
| 1984 | struct xfs_buf_cancel *bcp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1985 | |
| 1986 | /* |
| 1987 | * If this isn't a cancel buffer item, then just return. |
| 1988 | */ |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 1989 | if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) { |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 1990 | trace_xfs_log_recover_buf_not_cancel(log, buf_f); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 1991 | return 0; |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 1992 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1993 | |
| 1994 | /* |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 1995 | * Insert an xfs_buf_cancel record into the hash table of them. |
| 1996 | * If there is already an identical record, bump its reference count. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1997 | */ |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 1998 | bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno); |
| 1999 | list_for_each_entry(bcp, bucket, bc_list) { |
| 2000 | if (bcp->bc_blkno == buf_f->blf_blkno && |
| 2001 | bcp->bc_len == buf_f->blf_len) { |
| 2002 | bcp->bc_refcount++; |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 2003 | trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2004 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2005 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2006 | } |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 2007 | |
| 2008 | bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP); |
| 2009 | bcp->bc_blkno = buf_f->blf_blkno; |
| 2010 | bcp->bc_len = buf_f->blf_len; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2011 | bcp->bc_refcount = 1; |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 2012 | list_add_tail(&bcp->bc_list, bucket); |
| 2013 | |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 2014 | trace_xfs_log_recover_buf_cancel_add(log, buf_f); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2015 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2016 | } |
| 2017 | |
| 2018 | /* |
| 2019 | * Check to see whether the buffer being recovered has a corresponding |
Dave Chinner | 84a5b73 | 2013-08-27 08:10:53 +1000 | [diff] [blame] | 2020 | * entry in the buffer cancel record table. If it is, return the cancel |
| 2021 | * buffer structure to the caller. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2022 | */ |
Dave Chinner | 84a5b73 | 2013-08-27 08:10:53 +1000 | [diff] [blame] | 2023 | STATIC struct xfs_buf_cancel * |
| 2024 | xlog_peek_buffer_cancelled( |
Mark Tinguely | ad223e6 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 2025 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2026 | xfs_daddr_t blkno, |
| 2027 | uint len, |
| 2028 | ushort flags) |
| 2029 | { |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 2030 | struct list_head *bucket; |
| 2031 | struct xfs_buf_cancel *bcp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2032 | |
Dave Chinner | 84a5b73 | 2013-08-27 08:10:53 +1000 | [diff] [blame] | 2033 | if (!log->l_buf_cancel_table) { |
| 2034 | /* empty table means no cancelled buffers in the log */ |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 2035 | ASSERT(!(flags & XFS_BLF_CANCEL)); |
Dave Chinner | 84a5b73 | 2013-08-27 08:10:53 +1000 | [diff] [blame] | 2036 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2037 | } |
| 2038 | |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 2039 | bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno); |
| 2040 | list_for_each_entry(bcp, bucket, bc_list) { |
| 2041 | if (bcp->bc_blkno == blkno && bcp->bc_len == len) |
Dave Chinner | 84a5b73 | 2013-08-27 08:10:53 +1000 | [diff] [blame] | 2042 | return bcp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2043 | } |
| 2044 | |
| 2045 | /* |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 2046 | * We didn't find a corresponding entry in the table, so return 0 so |
| 2047 | * that the buffer is NOT cancelled. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2048 | */ |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 2049 | ASSERT(!(flags & XFS_BLF_CANCEL)); |
Dave Chinner | 84a5b73 | 2013-08-27 08:10:53 +1000 | [diff] [blame] | 2050 | return NULL; |
| 2051 | } |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 2052 | |
Dave Chinner | 84a5b73 | 2013-08-27 08:10:53 +1000 | [diff] [blame] | 2053 | /* |
| 2054 | * If the buffer is being cancelled then return 1 so that it will be cancelled, |
| 2055 | * otherwise return 0. If the buffer is actually a buffer cancel item |
| 2056 | * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the |
| 2057 | * table and remove it from the table if this is the last reference. |
| 2058 | * |
| 2059 | * We remove the cancel record from the table when we encounter its last |
| 2060 | * occurrence in the log so that if the same buffer is re-used again after its |
| 2061 | * last cancellation we actually replay the changes made at that point. |
| 2062 | */ |
| 2063 | STATIC int |
| 2064 | xlog_check_buffer_cancelled( |
| 2065 | struct xlog *log, |
| 2066 | xfs_daddr_t blkno, |
| 2067 | uint len, |
| 2068 | ushort flags) |
| 2069 | { |
| 2070 | struct xfs_buf_cancel *bcp; |
| 2071 | |
| 2072 | bcp = xlog_peek_buffer_cancelled(log, blkno, len, flags); |
| 2073 | if (!bcp) |
| 2074 | return 0; |
| 2075 | |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 2076 | /* |
| 2077 | * We've go a match, so return 1 so that the recovery of this buffer |
| 2078 | * is cancelled. If this buffer is actually a buffer cancel log |
| 2079 | * item, then decrement the refcount on the one in the table and |
| 2080 | * remove it if this is the last reference. |
| 2081 | */ |
| 2082 | if (flags & XFS_BLF_CANCEL) { |
| 2083 | if (--bcp->bc_refcount == 0) { |
| 2084 | list_del(&bcp->bc_list); |
| 2085 | kmem_free(bcp); |
| 2086 | } |
| 2087 | } |
| 2088 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2089 | } |
| 2090 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2091 | /* |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2092 | * Perform recovery for a buffer full of inodes. In these buffers, the only |
| 2093 | * data which should be recovered is that which corresponds to the |
| 2094 | * di_next_unlinked pointers in the on disk inode structures. The rest of the |
| 2095 | * data for the inodes is always logged through the inodes themselves rather |
| 2096 | * than the inode buffer and is recovered in xlog_recover_inode_pass2(). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2097 | * |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2098 | * The only time when buffers full of inodes are fully recovered is when the |
| 2099 | * buffer is full of newly allocated inodes. In this case the buffer will |
| 2100 | * not be marked as an inode buffer and so will be sent to |
| 2101 | * xlog_recover_do_reg_buffer() below during recovery. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2102 | */ |
| 2103 | STATIC int |
| 2104 | xlog_recover_do_inode_buffer( |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2105 | struct xfs_mount *mp, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2106 | xlog_recover_item_t *item, |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2107 | struct xfs_buf *bp, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2108 | xfs_buf_log_format_t *buf_f) |
| 2109 | { |
| 2110 | int i; |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2111 | int item_index = 0; |
| 2112 | int bit = 0; |
| 2113 | int nbits = 0; |
| 2114 | int reg_buf_offset = 0; |
| 2115 | int reg_buf_bytes = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2116 | int next_unlinked_offset; |
| 2117 | int inodes_per_buf; |
| 2118 | xfs_agino_t *logged_nextp; |
| 2119 | xfs_agino_t *buffer_nextp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2120 | |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 2121 | trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f); |
Dave Chinner | 9222a9c | 2013-06-12 12:19:06 +1000 | [diff] [blame] | 2122 | |
| 2123 | /* |
| 2124 | * Post recovery validation only works properly on CRC enabled |
| 2125 | * filesystems. |
| 2126 | */ |
| 2127 | if (xfs_sb_version_hascrc(&mp->m_sb)) |
| 2128 | bp->b_ops = &xfs_inode_buf_ops; |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 2129 | |
Dave Chinner | aa0e883 | 2012-04-23 15:58:52 +1000 | [diff] [blame] | 2130 | inodes_per_buf = BBTOB(bp->b_io_length) >> mp->m_sb.sb_inodelog; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2131 | for (i = 0; i < inodes_per_buf; i++) { |
| 2132 | next_unlinked_offset = (i * mp->m_sb.sb_inodesize) + |
| 2133 | offsetof(xfs_dinode_t, di_next_unlinked); |
| 2134 | |
| 2135 | while (next_unlinked_offset >= |
| 2136 | (reg_buf_offset + reg_buf_bytes)) { |
| 2137 | /* |
| 2138 | * The next di_next_unlinked field is beyond |
| 2139 | * the current logged region. Find the next |
| 2140 | * logged region that contains or is beyond |
| 2141 | * the current di_next_unlinked field. |
| 2142 | */ |
| 2143 | bit += nbits; |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2144 | bit = xfs_next_bit(buf_f->blf_data_map, |
| 2145 | buf_f->blf_map_size, bit); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2146 | |
| 2147 | /* |
| 2148 | * If there are no more logged regions in the |
| 2149 | * buffer, then we're done. |
| 2150 | */ |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2151 | if (bit == -1) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2152 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2153 | |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2154 | nbits = xfs_contig_bits(buf_f->blf_data_map, |
| 2155 | buf_f->blf_map_size, bit); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2156 | ASSERT(nbits > 0); |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 2157 | reg_buf_offset = bit << XFS_BLF_SHIFT; |
| 2158 | reg_buf_bytes = nbits << XFS_BLF_SHIFT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2159 | item_index++; |
| 2160 | } |
| 2161 | |
| 2162 | /* |
| 2163 | * If the current logged region starts after the current |
| 2164 | * di_next_unlinked field, then move on to the next |
| 2165 | * di_next_unlinked field. |
| 2166 | */ |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2167 | if (next_unlinked_offset < reg_buf_offset) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2168 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2169 | |
| 2170 | ASSERT(item->ri_buf[item_index].i_addr != NULL); |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 2171 | ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0); |
Dave Chinner | aa0e883 | 2012-04-23 15:58:52 +1000 | [diff] [blame] | 2172 | ASSERT((reg_buf_offset + reg_buf_bytes) <= |
| 2173 | BBTOB(bp->b_io_length)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2174 | |
| 2175 | /* |
| 2176 | * The current logged region contains a copy of the |
| 2177 | * current di_next_unlinked field. Extract its value |
| 2178 | * and copy it to the buffer copy. |
| 2179 | */ |
Christoph Hellwig | 4e0d5f9 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 2180 | logged_nextp = item->ri_buf[item_index].i_addr + |
| 2181 | next_unlinked_offset - reg_buf_offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2182 | if (unlikely(*logged_nextp == 0)) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2183 | xfs_alert(mp, |
| 2184 | "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). " |
| 2185 | "Trying to replay bad (0) inode di_next_unlinked field.", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2186 | item, bp); |
| 2187 | XFS_ERROR_REPORT("xlog_recover_do_inode_buf", |
| 2188 | XFS_ERRLEVEL_LOW, mp); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 2189 | return -EFSCORRUPTED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2190 | } |
| 2191 | |
Christoph Hellwig | 88ee2df | 2015-06-22 09:44:29 +1000 | [diff] [blame] | 2192 | buffer_nextp = xfs_buf_offset(bp, next_unlinked_offset); |
Tim Shimmin | 87c199c | 2006-06-09 14:56:16 +1000 | [diff] [blame] | 2193 | *buffer_nextp = *logged_nextp; |
Dave Chinner | 0a32c26 | 2013-06-05 12:09:08 +1000 | [diff] [blame] | 2194 | |
| 2195 | /* |
| 2196 | * If necessary, recalculate the CRC in the on-disk inode. We |
| 2197 | * have to leave the inode in a consistent state for whoever |
| 2198 | * reads it next.... |
| 2199 | */ |
Christoph Hellwig | 88ee2df | 2015-06-22 09:44:29 +1000 | [diff] [blame] | 2200 | xfs_dinode_calc_crc(mp, |
Dave Chinner | 0a32c26 | 2013-06-05 12:09:08 +1000 | [diff] [blame] | 2201 | xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize)); |
| 2202 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2203 | } |
| 2204 | |
| 2205 | return 0; |
| 2206 | } |
| 2207 | |
| 2208 | /* |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2209 | * V5 filesystems know the age of the buffer on disk being recovered. We can |
| 2210 | * have newer objects on disk than we are replaying, and so for these cases we |
| 2211 | * don't want to replay the current change as that will make the buffer contents |
| 2212 | * temporarily invalid on disk. |
| 2213 | * |
| 2214 | * The magic number might not match the buffer type we are going to recover |
| 2215 | * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags. Hence |
| 2216 | * extract the LSN of the existing object in the buffer based on it's current |
| 2217 | * magic number. If we don't recognise the magic number in the buffer, then |
| 2218 | * return a LSN of -1 so that the caller knows it was an unrecognised block and |
| 2219 | * so can recover the buffer. |
Dave Chinner | 566055d | 2013-09-24 16:01:16 +1000 | [diff] [blame] | 2220 | * |
| 2221 | * Note: we cannot rely solely on magic number matches to determine that the |
| 2222 | * buffer has a valid LSN - we also need to verify that it belongs to this |
| 2223 | * filesystem, so we need to extract the object's LSN and compare it to that |
| 2224 | * which we read from the superblock. If the UUIDs don't match, then we've got a |
| 2225 | * stale metadata block from an old filesystem instance that we need to recover |
| 2226 | * over the top of. |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2227 | */ |
| 2228 | static xfs_lsn_t |
| 2229 | xlog_recover_get_buf_lsn( |
| 2230 | struct xfs_mount *mp, |
| 2231 | struct xfs_buf *bp) |
| 2232 | { |
| 2233 | __uint32_t magic32; |
| 2234 | __uint16_t magic16; |
| 2235 | __uint16_t magicda; |
| 2236 | void *blk = bp->b_addr; |
Dave Chinner | 566055d | 2013-09-24 16:01:16 +1000 | [diff] [blame] | 2237 | uuid_t *uuid; |
| 2238 | xfs_lsn_t lsn = -1; |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2239 | |
| 2240 | /* v4 filesystems always recover immediately */ |
| 2241 | if (!xfs_sb_version_hascrc(&mp->m_sb)) |
| 2242 | goto recover_immediately; |
| 2243 | |
| 2244 | magic32 = be32_to_cpu(*(__be32 *)blk); |
| 2245 | switch (magic32) { |
| 2246 | case XFS_ABTB_CRC_MAGIC: |
| 2247 | case XFS_ABTC_CRC_MAGIC: |
| 2248 | case XFS_ABTB_MAGIC: |
| 2249 | case XFS_ABTC_MAGIC: |
Darrick J. Wong | a650e8f | 2016-08-03 12:17:11 +1000 | [diff] [blame] | 2250 | case XFS_RMAP_CRC_MAGIC: |
Darrick J. Wong | a90c00f | 2016-10-03 09:11:23 -0700 | [diff] [blame] | 2251 | case XFS_REFC_CRC_MAGIC: |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2252 | case XFS_IBT_CRC_MAGIC: |
Dave Chinner | 566055d | 2013-09-24 16:01:16 +1000 | [diff] [blame] | 2253 | case XFS_IBT_MAGIC: { |
| 2254 | struct xfs_btree_block *btb = blk; |
| 2255 | |
| 2256 | lsn = be64_to_cpu(btb->bb_u.s.bb_lsn); |
| 2257 | uuid = &btb->bb_u.s.bb_uuid; |
| 2258 | break; |
| 2259 | } |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2260 | case XFS_BMAP_CRC_MAGIC: |
Dave Chinner | 566055d | 2013-09-24 16:01:16 +1000 | [diff] [blame] | 2261 | case XFS_BMAP_MAGIC: { |
| 2262 | struct xfs_btree_block *btb = blk; |
| 2263 | |
| 2264 | lsn = be64_to_cpu(btb->bb_u.l.bb_lsn); |
| 2265 | uuid = &btb->bb_u.l.bb_uuid; |
| 2266 | break; |
| 2267 | } |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2268 | case XFS_AGF_MAGIC: |
Dave Chinner | 566055d | 2013-09-24 16:01:16 +1000 | [diff] [blame] | 2269 | lsn = be64_to_cpu(((struct xfs_agf *)blk)->agf_lsn); |
| 2270 | uuid = &((struct xfs_agf *)blk)->agf_uuid; |
| 2271 | break; |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2272 | case XFS_AGFL_MAGIC: |
Dave Chinner | 566055d | 2013-09-24 16:01:16 +1000 | [diff] [blame] | 2273 | lsn = be64_to_cpu(((struct xfs_agfl *)blk)->agfl_lsn); |
| 2274 | uuid = &((struct xfs_agfl *)blk)->agfl_uuid; |
| 2275 | break; |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2276 | case XFS_AGI_MAGIC: |
Dave Chinner | 566055d | 2013-09-24 16:01:16 +1000 | [diff] [blame] | 2277 | lsn = be64_to_cpu(((struct xfs_agi *)blk)->agi_lsn); |
| 2278 | uuid = &((struct xfs_agi *)blk)->agi_uuid; |
| 2279 | break; |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2280 | case XFS_SYMLINK_MAGIC: |
Dave Chinner | 566055d | 2013-09-24 16:01:16 +1000 | [diff] [blame] | 2281 | lsn = be64_to_cpu(((struct xfs_dsymlink_hdr *)blk)->sl_lsn); |
| 2282 | uuid = &((struct xfs_dsymlink_hdr *)blk)->sl_uuid; |
| 2283 | break; |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2284 | case XFS_DIR3_BLOCK_MAGIC: |
| 2285 | case XFS_DIR3_DATA_MAGIC: |
| 2286 | case XFS_DIR3_FREE_MAGIC: |
Dave Chinner | 566055d | 2013-09-24 16:01:16 +1000 | [diff] [blame] | 2287 | lsn = be64_to_cpu(((struct xfs_dir3_blk_hdr *)blk)->lsn); |
| 2288 | uuid = &((struct xfs_dir3_blk_hdr *)blk)->uuid; |
| 2289 | break; |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2290 | case XFS_ATTR3_RMT_MAGIC: |
Dave Chinner | e3c32ee | 2015-07-29 11:48:01 +1000 | [diff] [blame] | 2291 | /* |
| 2292 | * Remote attr blocks are written synchronously, rather than |
| 2293 | * being logged. That means they do not contain a valid LSN |
| 2294 | * (i.e. transactionally ordered) in them, and hence any time we |
| 2295 | * see a buffer to replay over the top of a remote attribute |
| 2296 | * block we should simply do so. |
| 2297 | */ |
| 2298 | goto recover_immediately; |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2299 | case XFS_SB_MAGIC: |
Dave Chinner | fcfbe2c | 2015-08-19 10:31:54 +1000 | [diff] [blame] | 2300 | /* |
| 2301 | * superblock uuids are magic. We may or may not have a |
| 2302 | * sb_meta_uuid on disk, but it will be set in the in-core |
| 2303 | * superblock. We set the uuid pointer for verification |
| 2304 | * according to the superblock feature mask to ensure we check |
| 2305 | * the relevant UUID in the superblock. |
| 2306 | */ |
Dave Chinner | 566055d | 2013-09-24 16:01:16 +1000 | [diff] [blame] | 2307 | lsn = be64_to_cpu(((struct xfs_dsb *)blk)->sb_lsn); |
Dave Chinner | fcfbe2c | 2015-08-19 10:31:54 +1000 | [diff] [blame] | 2308 | if (xfs_sb_version_hasmetauuid(&mp->m_sb)) |
| 2309 | uuid = &((struct xfs_dsb *)blk)->sb_meta_uuid; |
| 2310 | else |
| 2311 | uuid = &((struct xfs_dsb *)blk)->sb_uuid; |
Dave Chinner | 566055d | 2013-09-24 16:01:16 +1000 | [diff] [blame] | 2312 | break; |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2313 | default: |
| 2314 | break; |
| 2315 | } |
| 2316 | |
Dave Chinner | 566055d | 2013-09-24 16:01:16 +1000 | [diff] [blame] | 2317 | if (lsn != (xfs_lsn_t)-1) { |
Dave Chinner | fcfbe2c | 2015-08-19 10:31:54 +1000 | [diff] [blame] | 2318 | if (!uuid_equal(&mp->m_sb.sb_meta_uuid, uuid)) |
Dave Chinner | 566055d | 2013-09-24 16:01:16 +1000 | [diff] [blame] | 2319 | goto recover_immediately; |
| 2320 | return lsn; |
| 2321 | } |
| 2322 | |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2323 | magicda = be16_to_cpu(((struct xfs_da_blkinfo *)blk)->magic); |
| 2324 | switch (magicda) { |
| 2325 | case XFS_DIR3_LEAF1_MAGIC: |
| 2326 | case XFS_DIR3_LEAFN_MAGIC: |
| 2327 | case XFS_DA3_NODE_MAGIC: |
Dave Chinner | 566055d | 2013-09-24 16:01:16 +1000 | [diff] [blame] | 2328 | lsn = be64_to_cpu(((struct xfs_da3_blkinfo *)blk)->lsn); |
| 2329 | uuid = &((struct xfs_da3_blkinfo *)blk)->uuid; |
| 2330 | break; |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2331 | default: |
| 2332 | break; |
| 2333 | } |
| 2334 | |
Dave Chinner | 566055d | 2013-09-24 16:01:16 +1000 | [diff] [blame] | 2335 | if (lsn != (xfs_lsn_t)-1) { |
| 2336 | if (!uuid_equal(&mp->m_sb.sb_uuid, uuid)) |
| 2337 | goto recover_immediately; |
| 2338 | return lsn; |
| 2339 | } |
| 2340 | |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2341 | /* |
| 2342 | * We do individual object checks on dquot and inode buffers as they |
| 2343 | * have their own individual LSN records. Also, we could have a stale |
| 2344 | * buffer here, so we have to at least recognise these buffer types. |
| 2345 | * |
| 2346 | * A notd complexity here is inode unlinked list processing - it logs |
| 2347 | * the inode directly in the buffer, but we don't know which inodes have |
| 2348 | * been modified, and there is no global buffer LSN. Hence we need to |
| 2349 | * recover all inode buffer types immediately. This problem will be |
| 2350 | * fixed by logical logging of the unlinked list modifications. |
| 2351 | */ |
| 2352 | magic16 = be16_to_cpu(*(__be16 *)blk); |
| 2353 | switch (magic16) { |
| 2354 | case XFS_DQUOT_MAGIC: |
| 2355 | case XFS_DINODE_MAGIC: |
| 2356 | goto recover_immediately; |
| 2357 | default: |
| 2358 | break; |
| 2359 | } |
| 2360 | |
| 2361 | /* unknown buffer contents, recover immediately */ |
| 2362 | |
| 2363 | recover_immediately: |
| 2364 | return (xfs_lsn_t)-1; |
| 2365 | |
| 2366 | } |
| 2367 | |
| 2368 | /* |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2369 | * Validate the recovered buffer is of the correct type and attach the |
| 2370 | * appropriate buffer operations to them for writeback. Magic numbers are in a |
| 2371 | * few places: |
| 2372 | * the first 16 bits of the buffer (inode buffer, dquot buffer), |
| 2373 | * the first 32 bits of the buffer (most blocks), |
| 2374 | * inside a struct xfs_da_blkinfo at the start of the buffer. |
| 2375 | */ |
| 2376 | static void |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2377 | xlog_recover_validate_buf_type( |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2378 | struct xfs_mount *mp, |
| 2379 | struct xfs_buf *bp, |
Brian Foster | 22db9af | 2016-09-26 08:32:07 +1000 | [diff] [blame] | 2380 | xfs_buf_log_format_t *buf_f, |
| 2381 | xfs_lsn_t current_lsn) |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2382 | { |
| 2383 | struct xfs_da_blkinfo *info = bp->b_addr; |
| 2384 | __uint32_t magic32; |
| 2385 | __uint16_t magic16; |
| 2386 | __uint16_t magicda; |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2387 | char *warnmsg = NULL; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2388 | |
Dave Chinner | 67dc288 | 2014-08-04 12:43:06 +1000 | [diff] [blame] | 2389 | /* |
| 2390 | * We can only do post recovery validation on items on CRC enabled |
| 2391 | * fielsystems as we need to know when the buffer was written to be able |
| 2392 | * to determine if we should have replayed the item. If we replay old |
| 2393 | * metadata over a newer buffer, then it will enter a temporarily |
| 2394 | * inconsistent state resulting in verification failures. Hence for now |
| 2395 | * just avoid the verification stage for non-crc filesystems |
| 2396 | */ |
| 2397 | if (!xfs_sb_version_hascrc(&mp->m_sb)) |
| 2398 | return; |
| 2399 | |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2400 | magic32 = be32_to_cpu(*(__be32 *)bp->b_addr); |
| 2401 | magic16 = be16_to_cpu(*(__be16*)bp->b_addr); |
| 2402 | magicda = be16_to_cpu(info->magic); |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2403 | switch (xfs_blft_from_flags(buf_f)) { |
| 2404 | case XFS_BLFT_BTREE_BUF: |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2405 | switch (magic32) { |
| 2406 | case XFS_ABTB_CRC_MAGIC: |
| 2407 | case XFS_ABTC_CRC_MAGIC: |
| 2408 | case XFS_ABTB_MAGIC: |
| 2409 | case XFS_ABTC_MAGIC: |
| 2410 | bp->b_ops = &xfs_allocbt_buf_ops; |
| 2411 | break; |
| 2412 | case XFS_IBT_CRC_MAGIC: |
Brian Foster | aafc3c2 | 2014-04-24 16:00:52 +1000 | [diff] [blame] | 2413 | case XFS_FIBT_CRC_MAGIC: |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2414 | case XFS_IBT_MAGIC: |
Brian Foster | aafc3c2 | 2014-04-24 16:00:52 +1000 | [diff] [blame] | 2415 | case XFS_FIBT_MAGIC: |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2416 | bp->b_ops = &xfs_inobt_buf_ops; |
| 2417 | break; |
| 2418 | case XFS_BMAP_CRC_MAGIC: |
| 2419 | case XFS_BMAP_MAGIC: |
| 2420 | bp->b_ops = &xfs_bmbt_buf_ops; |
| 2421 | break; |
Darrick J. Wong | a650e8f | 2016-08-03 12:17:11 +1000 | [diff] [blame] | 2422 | case XFS_RMAP_CRC_MAGIC: |
| 2423 | bp->b_ops = &xfs_rmapbt_buf_ops; |
| 2424 | break; |
Darrick J. Wong | a90c00f | 2016-10-03 09:11:23 -0700 | [diff] [blame] | 2425 | case XFS_REFC_CRC_MAGIC: |
| 2426 | bp->b_ops = &xfs_refcountbt_buf_ops; |
| 2427 | break; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2428 | default: |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2429 | warnmsg = "Bad btree block magic!"; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2430 | break; |
| 2431 | } |
| 2432 | break; |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2433 | case XFS_BLFT_AGF_BUF: |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2434 | if (magic32 != XFS_AGF_MAGIC) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2435 | warnmsg = "Bad AGF block magic!"; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2436 | break; |
| 2437 | } |
| 2438 | bp->b_ops = &xfs_agf_buf_ops; |
| 2439 | break; |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2440 | case XFS_BLFT_AGFL_BUF: |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2441 | if (magic32 != XFS_AGFL_MAGIC) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2442 | warnmsg = "Bad AGFL block magic!"; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2443 | break; |
| 2444 | } |
| 2445 | bp->b_ops = &xfs_agfl_buf_ops; |
| 2446 | break; |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2447 | case XFS_BLFT_AGI_BUF: |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2448 | if (magic32 != XFS_AGI_MAGIC) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2449 | warnmsg = "Bad AGI block magic!"; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2450 | break; |
| 2451 | } |
| 2452 | bp->b_ops = &xfs_agi_buf_ops; |
| 2453 | break; |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2454 | case XFS_BLFT_UDQUOT_BUF: |
| 2455 | case XFS_BLFT_PDQUOT_BUF: |
| 2456 | case XFS_BLFT_GDQUOT_BUF: |
Dave Chinner | 123887e | 2013-04-30 21:39:33 +1000 | [diff] [blame] | 2457 | #ifdef CONFIG_XFS_QUOTA |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2458 | if (magic16 != XFS_DQUOT_MAGIC) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2459 | warnmsg = "Bad DQUOT block magic!"; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2460 | break; |
| 2461 | } |
| 2462 | bp->b_ops = &xfs_dquot_buf_ops; |
Dave Chinner | 123887e | 2013-04-30 21:39:33 +1000 | [diff] [blame] | 2463 | #else |
| 2464 | xfs_alert(mp, |
| 2465 | "Trying to recover dquots without QUOTA support built in!"); |
| 2466 | ASSERT(0); |
| 2467 | #endif |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2468 | break; |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2469 | case XFS_BLFT_DINO_BUF: |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2470 | if (magic16 != XFS_DINODE_MAGIC) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2471 | warnmsg = "Bad INODE block magic!"; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2472 | break; |
| 2473 | } |
| 2474 | bp->b_ops = &xfs_inode_buf_ops; |
| 2475 | break; |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2476 | case XFS_BLFT_SYMLINK_BUF: |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2477 | if (magic32 != XFS_SYMLINK_MAGIC) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2478 | warnmsg = "Bad symlink block magic!"; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2479 | break; |
| 2480 | } |
| 2481 | bp->b_ops = &xfs_symlink_buf_ops; |
| 2482 | break; |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2483 | case XFS_BLFT_DIR_BLOCK_BUF: |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2484 | if (magic32 != XFS_DIR2_BLOCK_MAGIC && |
| 2485 | magic32 != XFS_DIR3_BLOCK_MAGIC) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2486 | warnmsg = "Bad dir block magic!"; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2487 | break; |
| 2488 | } |
| 2489 | bp->b_ops = &xfs_dir3_block_buf_ops; |
| 2490 | break; |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2491 | case XFS_BLFT_DIR_DATA_BUF: |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2492 | if (magic32 != XFS_DIR2_DATA_MAGIC && |
| 2493 | magic32 != XFS_DIR3_DATA_MAGIC) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2494 | warnmsg = "Bad dir data magic!"; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2495 | break; |
| 2496 | } |
| 2497 | bp->b_ops = &xfs_dir3_data_buf_ops; |
| 2498 | break; |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2499 | case XFS_BLFT_DIR_FREE_BUF: |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2500 | if (magic32 != XFS_DIR2_FREE_MAGIC && |
| 2501 | magic32 != XFS_DIR3_FREE_MAGIC) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2502 | warnmsg = "Bad dir3 free magic!"; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2503 | break; |
| 2504 | } |
| 2505 | bp->b_ops = &xfs_dir3_free_buf_ops; |
| 2506 | break; |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2507 | case XFS_BLFT_DIR_LEAF1_BUF: |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2508 | if (magicda != XFS_DIR2_LEAF1_MAGIC && |
| 2509 | magicda != XFS_DIR3_LEAF1_MAGIC) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2510 | warnmsg = "Bad dir leaf1 magic!"; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2511 | break; |
| 2512 | } |
| 2513 | bp->b_ops = &xfs_dir3_leaf1_buf_ops; |
| 2514 | break; |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2515 | case XFS_BLFT_DIR_LEAFN_BUF: |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2516 | if (magicda != XFS_DIR2_LEAFN_MAGIC && |
| 2517 | magicda != XFS_DIR3_LEAFN_MAGIC) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2518 | warnmsg = "Bad dir leafn magic!"; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2519 | break; |
| 2520 | } |
| 2521 | bp->b_ops = &xfs_dir3_leafn_buf_ops; |
| 2522 | break; |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2523 | case XFS_BLFT_DA_NODE_BUF: |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2524 | if (magicda != XFS_DA_NODE_MAGIC && |
| 2525 | magicda != XFS_DA3_NODE_MAGIC) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2526 | warnmsg = "Bad da node magic!"; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2527 | break; |
| 2528 | } |
| 2529 | bp->b_ops = &xfs_da3_node_buf_ops; |
| 2530 | break; |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2531 | case XFS_BLFT_ATTR_LEAF_BUF: |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2532 | if (magicda != XFS_ATTR_LEAF_MAGIC && |
| 2533 | magicda != XFS_ATTR3_LEAF_MAGIC) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2534 | warnmsg = "Bad attr leaf magic!"; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2535 | break; |
| 2536 | } |
| 2537 | bp->b_ops = &xfs_attr3_leaf_buf_ops; |
| 2538 | break; |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2539 | case XFS_BLFT_ATTR_RMT_BUF: |
Dave Chinner | cab09a8 | 2013-04-30 21:39:36 +1000 | [diff] [blame] | 2540 | if (magic32 != XFS_ATTR3_RMT_MAGIC) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2541 | warnmsg = "Bad attr remote magic!"; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2542 | break; |
| 2543 | } |
| 2544 | bp->b_ops = &xfs_attr3_rmt_buf_ops; |
| 2545 | break; |
Dave Chinner | 04a1e6c | 2013-04-03 16:11:31 +1100 | [diff] [blame] | 2546 | case XFS_BLFT_SB_BUF: |
| 2547 | if (magic32 != XFS_SB_MAGIC) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2548 | warnmsg = "Bad SB block magic!"; |
Dave Chinner | 04a1e6c | 2013-04-03 16:11:31 +1100 | [diff] [blame] | 2549 | break; |
| 2550 | } |
| 2551 | bp->b_ops = &xfs_sb_buf_ops; |
| 2552 | break; |
Dave Chinner | f67ca6e | 2016-02-09 16:41:31 +1100 | [diff] [blame] | 2553 | #ifdef CONFIG_XFS_RT |
| 2554 | case XFS_BLFT_RTBITMAP_BUF: |
| 2555 | case XFS_BLFT_RTSUMMARY_BUF: |
Dave Chinner | bf85e09 | 2016-02-09 16:41:45 +1100 | [diff] [blame] | 2556 | /* no magic numbers for verification of RT buffers */ |
| 2557 | bp->b_ops = &xfs_rtbuf_ops; |
Dave Chinner | f67ca6e | 2016-02-09 16:41:31 +1100 | [diff] [blame] | 2558 | break; |
| 2559 | #endif /* CONFIG_XFS_RT */ |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2560 | default: |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2561 | xfs_warn(mp, "Unknown buffer type %d!", |
| 2562 | xfs_blft_from_flags(buf_f)); |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2563 | break; |
| 2564 | } |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2565 | |
| 2566 | /* |
Brian Foster | 60a4a22 | 2016-09-26 08:34:27 +1000 | [diff] [blame] | 2567 | * Nothing else to do in the case of a NULL current LSN as this means |
| 2568 | * the buffer is more recent than the change in the log and will be |
| 2569 | * skipped. |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2570 | */ |
Brian Foster | 60a4a22 | 2016-09-26 08:34:27 +1000 | [diff] [blame] | 2571 | if (current_lsn == NULLCOMMITLSN) |
| 2572 | return; |
| 2573 | |
| 2574 | if (warnmsg) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2575 | xfs_warn(mp, warnmsg); |
| 2576 | ASSERT(0); |
| 2577 | } |
Brian Foster | 60a4a22 | 2016-09-26 08:34:27 +1000 | [diff] [blame] | 2578 | |
| 2579 | /* |
| 2580 | * We must update the metadata LSN of the buffer as it is written out to |
| 2581 | * ensure that older transactions never replay over this one and corrupt |
| 2582 | * the buffer. This can occur if log recovery is interrupted at some |
| 2583 | * point after the current transaction completes, at which point a |
| 2584 | * subsequent mount starts recovery from the beginning. |
| 2585 | * |
| 2586 | * Write verifiers update the metadata LSN from log items attached to |
| 2587 | * the buffer. Therefore, initialize a bli purely to carry the LSN to |
| 2588 | * the verifier. We'll clean it up in our ->iodone() callback. |
| 2589 | */ |
| 2590 | if (bp->b_ops) { |
| 2591 | struct xfs_buf_log_item *bip; |
| 2592 | |
| 2593 | ASSERT(!bp->b_iodone || bp->b_iodone == xlog_recover_iodone); |
| 2594 | bp->b_iodone = xlog_recover_iodone; |
| 2595 | xfs_buf_item_init(bp, mp); |
| 2596 | bip = bp->b_fspriv; |
| 2597 | bip->bli_item.li_lsn = current_lsn; |
| 2598 | } |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2599 | } |
| 2600 | |
| 2601 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2602 | * Perform a 'normal' buffer recovery. Each logged region of the |
| 2603 | * buffer should be copied over the corresponding region in the |
| 2604 | * given buffer. The bitmap in the buf log format structure indicates |
| 2605 | * where to place the logged data. |
| 2606 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2607 | STATIC void |
| 2608 | xlog_recover_do_reg_buffer( |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 2609 | struct xfs_mount *mp, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2610 | xlog_recover_item_t *item, |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2611 | struct xfs_buf *bp, |
Brian Foster | 22db9af | 2016-09-26 08:32:07 +1000 | [diff] [blame] | 2612 | xfs_buf_log_format_t *buf_f, |
| 2613 | xfs_lsn_t current_lsn) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2614 | { |
| 2615 | int i; |
| 2616 | int bit; |
| 2617 | int nbits; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2618 | int error; |
| 2619 | |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 2620 | trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f); |
| 2621 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2622 | bit = 0; |
| 2623 | i = 1; /* 0 is the buf format structure */ |
| 2624 | while (1) { |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2625 | bit = xfs_next_bit(buf_f->blf_data_map, |
| 2626 | buf_f->blf_map_size, bit); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2627 | if (bit == -1) |
| 2628 | break; |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2629 | nbits = xfs_contig_bits(buf_f->blf_data_map, |
| 2630 | buf_f->blf_map_size, bit); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2631 | ASSERT(nbits > 0); |
Christoph Hellwig | 4b80916 | 2007-08-16 15:37:36 +1000 | [diff] [blame] | 2632 | ASSERT(item->ri_buf[i].i_addr != NULL); |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 2633 | ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0); |
Dave Chinner | aa0e883 | 2012-04-23 15:58:52 +1000 | [diff] [blame] | 2634 | ASSERT(BBTOB(bp->b_io_length) >= |
| 2635 | ((uint)bit << XFS_BLF_SHIFT) + (nbits << XFS_BLF_SHIFT)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2636 | |
| 2637 | /* |
Dave Chinner | 709da6a | 2013-05-27 16:38:23 +1000 | [diff] [blame] | 2638 | * The dirty regions logged in the buffer, even though |
| 2639 | * contiguous, may span multiple chunks. This is because the |
| 2640 | * dirty region may span a physical page boundary in a buffer |
| 2641 | * and hence be split into two separate vectors for writing into |
| 2642 | * the log. Hence we need to trim nbits back to the length of |
| 2643 | * the current region being copied out of the log. |
| 2644 | */ |
| 2645 | if (item->ri_buf[i].i_len < (nbits << XFS_BLF_SHIFT)) |
| 2646 | nbits = item->ri_buf[i].i_len >> XFS_BLF_SHIFT; |
| 2647 | |
| 2648 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2649 | * Do a sanity check if this is a dquot buffer. Just checking |
| 2650 | * the first dquot in the buffer should do. XXXThis is |
| 2651 | * probably a good thing to do for other buf types also. |
| 2652 | */ |
| 2653 | error = 0; |
Nathan Scott | c8ad20f | 2005-06-21 15:38:48 +1000 | [diff] [blame] | 2654 | if (buf_f->blf_flags & |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 2655 | (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) { |
Christoph Hellwig | 0c5e1ce | 2009-06-08 15:33:21 +0200 | [diff] [blame] | 2656 | if (item->ri_buf[i].i_addr == NULL) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2657 | xfs_alert(mp, |
Christoph Hellwig | 0c5e1ce | 2009-06-08 15:33:21 +0200 | [diff] [blame] | 2658 | "XFS: NULL dquot in %s.", __func__); |
| 2659 | goto next; |
| 2660 | } |
Jan Rekorajski | 8ec6dba | 2009-11-16 11:57:02 +0000 | [diff] [blame] | 2661 | if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2662 | xfs_alert(mp, |
Christoph Hellwig | 0c5e1ce | 2009-06-08 15:33:21 +0200 | [diff] [blame] | 2663 | "XFS: dquot too small (%d) in %s.", |
| 2664 | item->ri_buf[i].i_len, __func__); |
| 2665 | goto next; |
| 2666 | } |
Dave Chinner | 9aede1d | 2013-10-15 09:17:52 +1100 | [diff] [blame] | 2667 | error = xfs_dqcheck(mp, item->ri_buf[i].i_addr, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2668 | -1, 0, XFS_QMOPT_DOWARN, |
| 2669 | "dquot_buf_recover"); |
Christoph Hellwig | 0c5e1ce | 2009-06-08 15:33:21 +0200 | [diff] [blame] | 2670 | if (error) |
| 2671 | goto next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2672 | } |
Christoph Hellwig | 0c5e1ce | 2009-06-08 15:33:21 +0200 | [diff] [blame] | 2673 | |
| 2674 | memcpy(xfs_buf_offset(bp, |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 2675 | (uint)bit << XFS_BLF_SHIFT), /* dest */ |
Christoph Hellwig | 0c5e1ce | 2009-06-08 15:33:21 +0200 | [diff] [blame] | 2676 | item->ri_buf[i].i_addr, /* source */ |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 2677 | nbits<<XFS_BLF_SHIFT); /* length */ |
Christoph Hellwig | 0c5e1ce | 2009-06-08 15:33:21 +0200 | [diff] [blame] | 2678 | next: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2679 | i++; |
| 2680 | bit += nbits; |
| 2681 | } |
| 2682 | |
| 2683 | /* Shouldn't be any more regions */ |
| 2684 | ASSERT(i == item->ri_total); |
Christoph Hellwig | ee1a47a | 2013-04-21 14:53:46 -0500 | [diff] [blame] | 2685 | |
Brian Foster | 22db9af | 2016-09-26 08:32:07 +1000 | [diff] [blame] | 2686 | xlog_recover_validate_buf_type(mp, bp, buf_f, current_lsn); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2687 | } |
| 2688 | |
| 2689 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2690 | * Perform a dquot buffer recovery. |
Zhi Yong Wu | 8ba701e | 2013-08-12 03:15:01 +0000 | [diff] [blame] | 2691 | * Simple algorithm: if we have found a QUOTAOFF log item of the same type |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2692 | * (ie. USR or GRP), then just toss this buffer away; don't recover it. |
| 2693 | * Else, treat it as a regular buffer and do recovery. |
Dave Chinner | ad3714b | 2014-08-04 12:59:31 +1000 | [diff] [blame] | 2694 | * |
| 2695 | * Return false if the buffer was tossed and true if we recovered the buffer to |
| 2696 | * indicate to the caller if the buffer needs writing. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2697 | */ |
Dave Chinner | ad3714b | 2014-08-04 12:59:31 +1000 | [diff] [blame] | 2698 | STATIC bool |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2699 | xlog_recover_do_dquot_buffer( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 2700 | struct xfs_mount *mp, |
| 2701 | struct xlog *log, |
| 2702 | struct xlog_recover_item *item, |
| 2703 | struct xfs_buf *bp, |
| 2704 | struct xfs_buf_log_format *buf_f) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2705 | { |
| 2706 | uint type; |
| 2707 | |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 2708 | trace_xfs_log_recover_buf_dquot_buf(log, buf_f); |
| 2709 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2710 | /* |
| 2711 | * Filesystems are required to send in quota flags at mount time. |
| 2712 | */ |
Dave Chinner | ad3714b | 2014-08-04 12:59:31 +1000 | [diff] [blame] | 2713 | if (!mp->m_qflags) |
| 2714 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2715 | |
| 2716 | type = 0; |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 2717 | if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2718 | type |= XFS_DQ_USER; |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 2719 | if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF) |
Nathan Scott | c8ad20f | 2005-06-21 15:38:48 +1000 | [diff] [blame] | 2720 | type |= XFS_DQ_PROJ; |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 2721 | if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2722 | type |= XFS_DQ_GROUP; |
| 2723 | /* |
| 2724 | * This type of quotas was turned off, so ignore this buffer |
| 2725 | */ |
| 2726 | if (log->l_quotaoffs_flag & type) |
Dave Chinner | ad3714b | 2014-08-04 12:59:31 +1000 | [diff] [blame] | 2727 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2728 | |
Brian Foster | 22db9af | 2016-09-26 08:32:07 +1000 | [diff] [blame] | 2729 | xlog_recover_do_reg_buffer(mp, item, bp, buf_f, NULLCOMMITLSN); |
Dave Chinner | ad3714b | 2014-08-04 12:59:31 +1000 | [diff] [blame] | 2730 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2731 | } |
| 2732 | |
| 2733 | /* |
| 2734 | * This routine replays a modification made to a buffer at runtime. |
| 2735 | * There are actually two types of buffer, regular and inode, which |
| 2736 | * are handled differently. Inode buffers are handled differently |
| 2737 | * in that we only recover a specific set of data from them, namely |
| 2738 | * the inode di_next_unlinked fields. This is because all other inode |
| 2739 | * data is actually logged via inode records and any data we replay |
| 2740 | * here which overlaps that may be stale. |
| 2741 | * |
| 2742 | * When meta-data buffers are freed at run time we log a buffer item |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 2743 | * with the XFS_BLF_CANCEL bit set to indicate that previous copies |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2744 | * of the buffer in the log should not be replayed at recovery time. |
| 2745 | * This is so that if the blocks covered by the buffer are reused for |
| 2746 | * file data before we crash we don't end up replaying old, freed |
| 2747 | * meta-data into a user's file. |
| 2748 | * |
| 2749 | * To handle the cancellation of buffer log items, we make two passes |
| 2750 | * over the log during recovery. During the first we build a table of |
| 2751 | * those buffers which have been cancelled, and during the second we |
| 2752 | * only replay those buffers which do not have corresponding cancel |
Zhi Yong Wu | 34be5ff | 2013-08-07 10:11:07 +0000 | [diff] [blame] | 2753 | * records in the table. See xlog_recover_buffer_pass[1,2] above |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2754 | * for more details on the implementation of the table of cancel records. |
| 2755 | */ |
| 2756 | STATIC int |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2757 | xlog_recover_buffer_pass2( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 2758 | struct xlog *log, |
| 2759 | struct list_head *buffer_list, |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2760 | struct xlog_recover_item *item, |
| 2761 | xfs_lsn_t current_lsn) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2762 | { |
Christoph Hellwig | 4e0d5f9 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 2763 | xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr; |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2764 | xfs_mount_t *mp = log->l_mp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2765 | xfs_buf_t *bp; |
| 2766 | int error; |
Christoph Hellwig | 6ad112b | 2009-11-24 18:02:23 +0000 | [diff] [blame] | 2767 | uint buf_flags; |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2768 | xfs_lsn_t lsn; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2769 | |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2770 | /* |
| 2771 | * In this pass we only want to recover all the buffers which have |
| 2772 | * not been cancelled and are not cancellation buffers themselves. |
| 2773 | */ |
| 2774 | if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno, |
| 2775 | buf_f->blf_len, buf_f->blf_flags)) { |
| 2776 | trace_xfs_log_recover_buf_cancel(log, buf_f); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2777 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2778 | } |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2779 | |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 2780 | trace_xfs_log_recover_buf_recover(log, buf_f); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2781 | |
Dave Chinner | a8acad7 | 2012-04-23 15:58:54 +1000 | [diff] [blame] | 2782 | buf_flags = 0; |
Dave Chinner | 611c994 | 2012-04-23 15:59:07 +1000 | [diff] [blame] | 2783 | if (buf_f->blf_flags & XFS_BLF_INODE_BUF) |
| 2784 | buf_flags |= XBF_UNMAPPED; |
Christoph Hellwig | 6ad112b | 2009-11-24 18:02:23 +0000 | [diff] [blame] | 2785 | |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2786 | bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len, |
Dave Chinner | c3f8fc7 | 2012-11-12 22:54:01 +1100 | [diff] [blame] | 2787 | buf_flags, NULL); |
Chandra Seetharaman | ac4d688 | 2011-08-03 02:18:29 +0000 | [diff] [blame] | 2788 | if (!bp) |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 2789 | return -ENOMEM; |
Chandra Seetharaman | e570280 | 2011-08-03 02:18:34 +0000 | [diff] [blame] | 2790 | error = bp->b_error; |
Chandra Seetharaman | 5a52c2a58 | 2011-07-22 23:39:51 +0000 | [diff] [blame] | 2791 | if (error) { |
Christoph Hellwig | 901796a | 2011-10-10 16:52:49 +0000 | [diff] [blame] | 2792 | xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)"); |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2793 | goto out_release; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2794 | } |
| 2795 | |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2796 | /* |
Dave Chinner | 67dc288 | 2014-08-04 12:43:06 +1000 | [diff] [blame] | 2797 | * Recover the buffer only if we get an LSN from it and it's less than |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2798 | * the lsn of the transaction we are replaying. |
Dave Chinner | 67dc288 | 2014-08-04 12:43:06 +1000 | [diff] [blame] | 2799 | * |
| 2800 | * Note that we have to be extremely careful of readahead here. |
| 2801 | * Readahead does not attach verfiers to the buffers so if we don't |
| 2802 | * actually do any replay after readahead because of the LSN we found |
| 2803 | * in the buffer if more recent than that current transaction then we |
| 2804 | * need to attach the verifier directly. Failure to do so can lead to |
| 2805 | * future recovery actions (e.g. EFI and unlinked list recovery) can |
| 2806 | * operate on the buffers and they won't get the verifier attached. This |
| 2807 | * can lead to blocks on disk having the correct content but a stale |
| 2808 | * CRC. |
| 2809 | * |
| 2810 | * It is safe to assume these clean buffers are currently up to date. |
| 2811 | * If the buffer is dirtied by a later transaction being replayed, then |
| 2812 | * the verifier will be reset to match whatever recover turns that |
| 2813 | * buffer into. |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2814 | */ |
| 2815 | lsn = xlog_recover_get_buf_lsn(mp, bp); |
Dave Chinner | 67dc288 | 2014-08-04 12:43:06 +1000 | [diff] [blame] | 2816 | if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) { |
Brian Foster | 5cd9cee | 2016-09-26 08:34:52 +1000 | [diff] [blame] | 2817 | trace_xfs_log_recover_buf_skip(log, buf_f); |
Brian Foster | 22db9af | 2016-09-26 08:32:07 +1000 | [diff] [blame] | 2818 | xlog_recover_validate_buf_type(mp, bp, buf_f, NULLCOMMITLSN); |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2819 | goto out_release; |
Dave Chinner | 67dc288 | 2014-08-04 12:43:06 +1000 | [diff] [blame] | 2820 | } |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2821 | |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2822 | if (buf_f->blf_flags & XFS_BLF_INODE_BUF) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2823 | error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f); |
Dave Chinner | ad3714b | 2014-08-04 12:59:31 +1000 | [diff] [blame] | 2824 | if (error) |
| 2825 | goto out_release; |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2826 | } else if (buf_f->blf_flags & |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 2827 | (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) { |
Dave Chinner | ad3714b | 2014-08-04 12:59:31 +1000 | [diff] [blame] | 2828 | bool dirty; |
| 2829 | |
| 2830 | dirty = xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f); |
| 2831 | if (!dirty) |
| 2832 | goto out_release; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2833 | } else { |
Brian Foster | 22db9af | 2016-09-26 08:32:07 +1000 | [diff] [blame] | 2834 | xlog_recover_do_reg_buffer(mp, item, bp, buf_f, current_lsn); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2835 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2836 | |
| 2837 | /* |
| 2838 | * Perform delayed write on the buffer. Asynchronous writes will be |
| 2839 | * slower when taking into account all the buffers to be flushed. |
| 2840 | * |
| 2841 | * Also make sure that only inode buffers with good sizes stay in |
| 2842 | * the buffer cache. The kernel moves inodes in buffers of 1 block |
Jie Liu | 0f49efd | 2013-12-13 15:51:48 +1100 | [diff] [blame] | 2843 | * or mp->m_inode_cluster_size bytes, whichever is bigger. The inode |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2844 | * buffers in the log can be a different size if the log was generated |
| 2845 | * by an older kernel using unclustered inode buffers or a newer kernel |
| 2846 | * running with a different inode cluster size. Regardless, if the |
Jie Liu | 0f49efd | 2013-12-13 15:51:48 +1100 | [diff] [blame] | 2847 | * the inode buffer size isn't MAX(blocksize, mp->m_inode_cluster_size) |
| 2848 | * for *our* value of mp->m_inode_cluster_size, then we need to keep |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2849 | * the buffer out of the buffer cache so that the buffer won't |
| 2850 | * overlap with future reads of those inodes. |
| 2851 | */ |
| 2852 | if (XFS_DINODE_MAGIC == |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 2853 | be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) && |
Dave Chinner | aa0e883 | 2012-04-23 15:58:52 +1000 | [diff] [blame] | 2854 | (BBTOB(bp->b_io_length) != MAX(log->l_mp->m_sb.sb_blocksize, |
Jie Liu | 0f49efd | 2013-12-13 15:51:48 +1100 | [diff] [blame] | 2855 | (__uint32_t)log->l_mp->m_inode_cluster_size))) { |
Christoph Hellwig | c867cb6 | 2011-10-10 16:52:46 +0000 | [diff] [blame] | 2856 | xfs_buf_stale(bp); |
Christoph Hellwig | c2b006c | 2011-08-23 08:28:07 +0000 | [diff] [blame] | 2857 | error = xfs_bwrite(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2858 | } else { |
Dave Chinner | ebad861 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 2859 | ASSERT(bp->b_target->bt_mount == mp); |
Christoph Hellwig | cb669ca | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 2860 | bp->b_iodone = xlog_recover_iodone; |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 2861 | xfs_buf_delwri_queue(bp, buffer_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2862 | } |
| 2863 | |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2864 | out_release: |
Christoph Hellwig | c2b006c | 2011-08-23 08:28:07 +0000 | [diff] [blame] | 2865 | xfs_buf_relse(bp); |
| 2866 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2867 | } |
| 2868 | |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 2869 | /* |
| 2870 | * Inode fork owner changes |
| 2871 | * |
| 2872 | * If we have been told that we have to reparent the inode fork, it's because an |
| 2873 | * extent swap operation on a CRC enabled filesystem has been done and we are |
| 2874 | * replaying it. We need to walk the BMBT of the appropriate fork and change the |
| 2875 | * owners of it. |
| 2876 | * |
| 2877 | * The complexity here is that we don't have an inode context to work with, so |
| 2878 | * after we've replayed the inode we need to instantiate one. This is where the |
| 2879 | * fun begins. |
| 2880 | * |
| 2881 | * We are in the middle of log recovery, so we can't run transactions. That |
| 2882 | * means we cannot use cache coherent inode instantiation via xfs_iget(), as |
| 2883 | * that will result in the corresponding iput() running the inode through |
| 2884 | * xfs_inactive(). If we've just replayed an inode core that changes the link |
| 2885 | * count to zero (i.e. it's been unlinked), then xfs_inactive() will run |
| 2886 | * transactions (bad!). |
| 2887 | * |
| 2888 | * So, to avoid this, we instantiate an inode directly from the inode core we've |
| 2889 | * just recovered. We have the buffer still locked, and all we really need to |
| 2890 | * instantiate is the inode core and the forks being modified. We can do this |
| 2891 | * manually, then run the inode btree owner change, and then tear down the |
| 2892 | * xfs_inode without having to run any transactions at all. |
| 2893 | * |
| 2894 | * Also, because we don't have a transaction context available here but need to |
| 2895 | * gather all the buffers we modify for writeback so we pass the buffer_list |
| 2896 | * instead for the operation to use. |
| 2897 | */ |
| 2898 | |
| 2899 | STATIC int |
| 2900 | xfs_recover_inode_owner_change( |
| 2901 | struct xfs_mount *mp, |
| 2902 | struct xfs_dinode *dip, |
| 2903 | struct xfs_inode_log_format *in_f, |
| 2904 | struct list_head *buffer_list) |
| 2905 | { |
| 2906 | struct xfs_inode *ip; |
| 2907 | int error; |
| 2908 | |
| 2909 | ASSERT(in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER)); |
| 2910 | |
| 2911 | ip = xfs_inode_alloc(mp, in_f->ilf_ino); |
| 2912 | if (!ip) |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 2913 | return -ENOMEM; |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 2914 | |
| 2915 | /* instantiate the inode */ |
Dave Chinner | 3987848 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 2916 | xfs_inode_from_disk(ip, dip); |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 2917 | ASSERT(ip->i_d.di_version >= 3); |
| 2918 | |
| 2919 | error = xfs_iformat_fork(ip, dip); |
| 2920 | if (error) |
| 2921 | goto out_free_ip; |
| 2922 | |
| 2923 | |
| 2924 | if (in_f->ilf_fields & XFS_ILOG_DOWNER) { |
| 2925 | ASSERT(in_f->ilf_fields & XFS_ILOG_DBROOT); |
| 2926 | error = xfs_bmbt_change_owner(NULL, ip, XFS_DATA_FORK, |
| 2927 | ip->i_ino, buffer_list); |
| 2928 | if (error) |
| 2929 | goto out_free_ip; |
| 2930 | } |
| 2931 | |
| 2932 | if (in_f->ilf_fields & XFS_ILOG_AOWNER) { |
| 2933 | ASSERT(in_f->ilf_fields & XFS_ILOG_ABROOT); |
| 2934 | error = xfs_bmbt_change_owner(NULL, ip, XFS_ATTR_FORK, |
| 2935 | ip->i_ino, buffer_list); |
| 2936 | if (error) |
| 2937 | goto out_free_ip; |
| 2938 | } |
| 2939 | |
| 2940 | out_free_ip: |
| 2941 | xfs_inode_free(ip); |
| 2942 | return error; |
| 2943 | } |
| 2944 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2945 | STATIC int |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2946 | xlog_recover_inode_pass2( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 2947 | struct xlog *log, |
| 2948 | struct list_head *buffer_list, |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2949 | struct xlog_recover_item *item, |
| 2950 | xfs_lsn_t current_lsn) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2951 | { |
| 2952 | xfs_inode_log_format_t *in_f; |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2953 | xfs_mount_t *mp = log->l_mp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2954 | xfs_buf_t *bp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2955 | xfs_dinode_t *dip; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2956 | int len; |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 2957 | char *src; |
| 2958 | char *dest; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2959 | int error; |
| 2960 | int attr_index; |
| 2961 | uint fields; |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 2962 | struct xfs_log_dinode *ldip; |
Christoph Hellwig | 93848a9 | 2013-04-03 16:11:17 +1100 | [diff] [blame] | 2963 | uint isize; |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2964 | int need_free = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2965 | |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2966 | if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) { |
Christoph Hellwig | 4e0d5f9 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 2967 | in_f = item->ri_buf[0].i_addr; |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2968 | } else { |
Christoph Hellwig | 4e0d5f9 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 2969 | in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP); |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2970 | need_free = 1; |
| 2971 | error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f); |
| 2972 | if (error) |
| 2973 | goto error; |
| 2974 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2975 | |
| 2976 | /* |
| 2977 | * Inode buffers can be freed, look out for it, |
| 2978 | * and do not replay the inode. |
| 2979 | */ |
Christoph Hellwig | a194189 | 2008-11-28 14:23:40 +1100 | [diff] [blame] | 2980 | if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno, |
| 2981 | in_f->ilf_len, 0)) { |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2982 | error = 0; |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 2983 | trace_xfs_log_recover_inode_cancel(log, in_f); |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2984 | goto error; |
| 2985 | } |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 2986 | trace_xfs_log_recover_inode_recover(log, in_f); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2987 | |
Dave Chinner | c3f8fc7 | 2012-11-12 22:54:01 +1100 | [diff] [blame] | 2988 | bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len, 0, |
Christoph Hellwig | 93848a9 | 2013-04-03 16:11:17 +1100 | [diff] [blame] | 2989 | &xfs_inode_buf_ops); |
Chandra Seetharaman | ac4d688 | 2011-08-03 02:18:29 +0000 | [diff] [blame] | 2990 | if (!bp) { |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 2991 | error = -ENOMEM; |
Chandra Seetharaman | ac4d688 | 2011-08-03 02:18:29 +0000 | [diff] [blame] | 2992 | goto error; |
| 2993 | } |
Chandra Seetharaman | e570280 | 2011-08-03 02:18:34 +0000 | [diff] [blame] | 2994 | error = bp->b_error; |
Chandra Seetharaman | 5a52c2a58 | 2011-07-22 23:39:51 +0000 | [diff] [blame] | 2995 | if (error) { |
Christoph Hellwig | 901796a | 2011-10-10 16:52:49 +0000 | [diff] [blame] | 2996 | xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)"); |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 2997 | goto out_release; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2998 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2999 | ASSERT(in_f->ilf_fields & XFS_ILOG_CORE); |
Christoph Hellwig | 88ee2df | 2015-06-22 09:44:29 +1000 | [diff] [blame] | 3000 | dip = xfs_buf_offset(bp, in_f->ilf_boffset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3001 | |
| 3002 | /* |
| 3003 | * Make sure the place we're flushing out to really looks |
| 3004 | * like an inode! |
| 3005 | */ |
Christoph Hellwig | 69ef921 | 2011-07-08 14:36:05 +0200 | [diff] [blame] | 3006 | if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 3007 | xfs_alert(mp, |
| 3008 | "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld", |
| 3009 | __func__, dip, bp, in_f->ilf_ino); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3010 | XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3011 | XFS_ERRLEVEL_LOW, mp); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3012 | error = -EFSCORRUPTED; |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 3013 | goto out_release; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3014 | } |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3015 | ldip = item->ri_buf[1].i_addr; |
| 3016 | if (unlikely(ldip->di_magic != XFS_DINODE_MAGIC)) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 3017 | xfs_alert(mp, |
| 3018 | "%s: Bad inode log record, rec ptr 0x%p, ino %Ld", |
| 3019 | __func__, item, in_f->ilf_ino); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3020 | XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3021 | XFS_ERRLEVEL_LOW, mp); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3022 | error = -EFSCORRUPTED; |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 3023 | goto out_release; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3024 | } |
| 3025 | |
Dave Chinner | e60896d | 2013-07-24 15:47:30 +1000 | [diff] [blame] | 3026 | /* |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 3027 | * If the inode has an LSN in it, recover the inode only if it's less |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 3028 | * than the lsn of the transaction we are replaying. Note: we still |
| 3029 | * need to replay an owner change even though the inode is more recent |
| 3030 | * than the transaction as there is no guarantee that all the btree |
| 3031 | * blocks are more recent than this transaction, too. |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 3032 | */ |
| 3033 | if (dip->di_version >= 3) { |
| 3034 | xfs_lsn_t lsn = be64_to_cpu(dip->di_lsn); |
| 3035 | |
| 3036 | if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) { |
| 3037 | trace_xfs_log_recover_inode_skip(log, in_f); |
| 3038 | error = 0; |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 3039 | goto out_owner_change; |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 3040 | } |
| 3041 | } |
| 3042 | |
| 3043 | /* |
Dave Chinner | e60896d | 2013-07-24 15:47:30 +1000 | [diff] [blame] | 3044 | * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes |
| 3045 | * are transactional and if ordering is necessary we can determine that |
| 3046 | * more accurately by the LSN field in the V3 inode core. Don't trust |
| 3047 | * the inode versions we might be changing them here - use the |
| 3048 | * superblock flag to determine whether we need to look at di_flushiter |
| 3049 | * to skip replay when the on disk inode is newer than the log one |
| 3050 | */ |
| 3051 | if (!xfs_sb_version_hascrc(&mp->m_sb) && |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3052 | ldip->di_flushiter < be16_to_cpu(dip->di_flushiter)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3053 | /* |
| 3054 | * Deal with the wrap case, DI_MAX_FLUSH is less |
| 3055 | * than smaller numbers |
| 3056 | */ |
Christoph Hellwig | 81591fe | 2008-11-28 14:23:39 +1100 | [diff] [blame] | 3057 | if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH && |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3058 | ldip->di_flushiter < (DI_MAX_FLUSH >> 1)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3059 | /* do nothing */ |
| 3060 | } else { |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 3061 | trace_xfs_log_recover_inode_skip(log, in_f); |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 3062 | error = 0; |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 3063 | goto out_release; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3064 | } |
| 3065 | } |
Dave Chinner | e60896d | 2013-07-24 15:47:30 +1000 | [diff] [blame] | 3066 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3067 | /* Take the opportunity to reset the flush iteration count */ |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3068 | ldip->di_flushiter = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3069 | |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3070 | if (unlikely(S_ISREG(ldip->di_mode))) { |
| 3071 | if ((ldip->di_format != XFS_DINODE_FMT_EXTENTS) && |
| 3072 | (ldip->di_format != XFS_DINODE_FMT_BTREE)) { |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3073 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)", |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3074 | XFS_ERRLEVEL_LOW, mp, ldip); |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 3075 | xfs_alert(mp, |
| 3076 | "%s: Bad regular inode log record, rec ptr 0x%p, " |
| 3077 | "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld", |
| 3078 | __func__, item, dip, bp, in_f->ilf_ino); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3079 | error = -EFSCORRUPTED; |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 3080 | goto out_release; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3081 | } |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3082 | } else if (unlikely(S_ISDIR(ldip->di_mode))) { |
| 3083 | if ((ldip->di_format != XFS_DINODE_FMT_EXTENTS) && |
| 3084 | (ldip->di_format != XFS_DINODE_FMT_BTREE) && |
| 3085 | (ldip->di_format != XFS_DINODE_FMT_LOCAL)) { |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3086 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)", |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3087 | XFS_ERRLEVEL_LOW, mp, ldip); |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 3088 | xfs_alert(mp, |
| 3089 | "%s: Bad dir inode log record, rec ptr 0x%p, " |
| 3090 | "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld", |
| 3091 | __func__, item, dip, bp, in_f->ilf_ino); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3092 | error = -EFSCORRUPTED; |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 3093 | goto out_release; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3094 | } |
| 3095 | } |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3096 | if (unlikely(ldip->di_nextents + ldip->di_anextents > ldip->di_nblocks)){ |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3097 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)", |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3098 | XFS_ERRLEVEL_LOW, mp, ldip); |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 3099 | xfs_alert(mp, |
| 3100 | "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, " |
| 3101 | "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld", |
| 3102 | __func__, item, dip, bp, in_f->ilf_ino, |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3103 | ldip->di_nextents + ldip->di_anextents, |
| 3104 | ldip->di_nblocks); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3105 | error = -EFSCORRUPTED; |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 3106 | goto out_release; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3107 | } |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3108 | if (unlikely(ldip->di_forkoff > mp->m_sb.sb_inodesize)) { |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3109 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)", |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3110 | XFS_ERRLEVEL_LOW, mp, ldip); |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 3111 | xfs_alert(mp, |
| 3112 | "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, " |
| 3113 | "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__, |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3114 | item, dip, bp, in_f->ilf_ino, ldip->di_forkoff); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3115 | error = -EFSCORRUPTED; |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 3116 | goto out_release; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3117 | } |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3118 | isize = xfs_log_dinode_size(ldip->di_version); |
Christoph Hellwig | 93848a9 | 2013-04-03 16:11:17 +1100 | [diff] [blame] | 3119 | if (unlikely(item->ri_buf[1].i_len > isize)) { |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3120 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)", |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3121 | XFS_ERRLEVEL_LOW, mp, ldip); |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 3122 | xfs_alert(mp, |
| 3123 | "%s: Bad inode log record length %d, rec ptr 0x%p", |
| 3124 | __func__, item->ri_buf[1].i_len, item); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3125 | error = -EFSCORRUPTED; |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 3126 | goto out_release; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3127 | } |
| 3128 | |
Dave Chinner | 3987848 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3129 | /* recover the log dinode inode into the on disk inode */ |
| 3130 | xfs_log_dinode_to_disk(ldip, dip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3131 | |
| 3132 | /* the rest is in on-disk format */ |
Christoph Hellwig | 93848a9 | 2013-04-03 16:11:17 +1100 | [diff] [blame] | 3133 | if (item->ri_buf[1].i_len > isize) { |
| 3134 | memcpy((char *)dip + isize, |
| 3135 | item->ri_buf[1].i_addr + isize, |
| 3136 | item->ri_buf[1].i_len - isize); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3137 | } |
| 3138 | |
| 3139 | fields = in_f->ilf_fields; |
| 3140 | switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) { |
| 3141 | case XFS_ILOG_DEV: |
Christoph Hellwig | 81591fe | 2008-11-28 14:23:39 +1100 | [diff] [blame] | 3142 | xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3143 | break; |
| 3144 | case XFS_ILOG_UUID: |
Christoph Hellwig | 81591fe | 2008-11-28 14:23:39 +1100 | [diff] [blame] | 3145 | memcpy(XFS_DFORK_DPTR(dip), |
| 3146 | &in_f->ilf_u.ilfu_uuid, |
| 3147 | sizeof(uuid_t)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3148 | break; |
| 3149 | } |
| 3150 | |
| 3151 | if (in_f->ilf_size == 2) |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 3152 | goto out_owner_change; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3153 | len = item->ri_buf[2].i_len; |
| 3154 | src = item->ri_buf[2].i_addr; |
| 3155 | ASSERT(in_f->ilf_size <= 4); |
| 3156 | ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK)); |
| 3157 | ASSERT(!(fields & XFS_ILOG_DFORK) || |
| 3158 | (len == in_f->ilf_dsize)); |
| 3159 | |
| 3160 | switch (fields & XFS_ILOG_DFORK) { |
| 3161 | case XFS_ILOG_DDATA: |
| 3162 | case XFS_ILOG_DEXT: |
Christoph Hellwig | 81591fe | 2008-11-28 14:23:39 +1100 | [diff] [blame] | 3163 | memcpy(XFS_DFORK_DPTR(dip), src, len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3164 | break; |
| 3165 | |
| 3166 | case XFS_ILOG_DBROOT: |
Christoph Hellwig | 7cc95a8 | 2008-10-30 17:14:34 +1100 | [diff] [blame] | 3167 | xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len, |
Christoph Hellwig | 81591fe | 2008-11-28 14:23:39 +1100 | [diff] [blame] | 3168 | (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3169 | XFS_DFORK_DSIZE(dip, mp)); |
| 3170 | break; |
| 3171 | |
| 3172 | default: |
| 3173 | /* |
| 3174 | * There are no data fork flags set. |
| 3175 | */ |
| 3176 | ASSERT((fields & XFS_ILOG_DFORK) == 0); |
| 3177 | break; |
| 3178 | } |
| 3179 | |
| 3180 | /* |
| 3181 | * If we logged any attribute data, recover it. There may or |
| 3182 | * may not have been any other non-core data logged in this |
| 3183 | * transaction. |
| 3184 | */ |
| 3185 | if (in_f->ilf_fields & XFS_ILOG_AFORK) { |
| 3186 | if (in_f->ilf_fields & XFS_ILOG_DFORK) { |
| 3187 | attr_index = 3; |
| 3188 | } else { |
| 3189 | attr_index = 2; |
| 3190 | } |
| 3191 | len = item->ri_buf[attr_index].i_len; |
| 3192 | src = item->ri_buf[attr_index].i_addr; |
| 3193 | ASSERT(len == in_f->ilf_asize); |
| 3194 | |
| 3195 | switch (in_f->ilf_fields & XFS_ILOG_AFORK) { |
| 3196 | case XFS_ILOG_ADATA: |
| 3197 | case XFS_ILOG_AEXT: |
| 3198 | dest = XFS_DFORK_APTR(dip); |
| 3199 | ASSERT(len <= XFS_DFORK_ASIZE(dip, mp)); |
| 3200 | memcpy(dest, src, len); |
| 3201 | break; |
| 3202 | |
| 3203 | case XFS_ILOG_ABROOT: |
| 3204 | dest = XFS_DFORK_APTR(dip); |
Christoph Hellwig | 7cc95a8 | 2008-10-30 17:14:34 +1100 | [diff] [blame] | 3205 | xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, |
| 3206 | len, (xfs_bmdr_block_t*)dest, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3207 | XFS_DFORK_ASIZE(dip, mp)); |
| 3208 | break; |
| 3209 | |
| 3210 | default: |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 3211 | xfs_warn(log->l_mp, "%s: Invalid flag", __func__); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3212 | ASSERT(0); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3213 | error = -EIO; |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 3214 | goto out_release; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3215 | } |
| 3216 | } |
| 3217 | |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 3218 | out_owner_change: |
| 3219 | if (in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER)) |
| 3220 | error = xfs_recover_inode_owner_change(mp, dip, in_f, |
| 3221 | buffer_list); |
Christoph Hellwig | 93848a9 | 2013-04-03 16:11:17 +1100 | [diff] [blame] | 3222 | /* re-generate the checksum. */ |
| 3223 | xfs_dinode_calc_crc(log->l_mp, dip); |
| 3224 | |
Dave Chinner | ebad861 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 3225 | ASSERT(bp->b_target->bt_mount == mp); |
Christoph Hellwig | cb669ca | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 3226 | bp->b_iodone = xlog_recover_iodone; |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 3227 | xfs_buf_delwri_queue(bp, buffer_list); |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 3228 | |
| 3229 | out_release: |
Christoph Hellwig | 61551f1 | 2011-08-23 08:28:06 +0000 | [diff] [blame] | 3230 | xfs_buf_relse(bp); |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 3231 | error: |
| 3232 | if (need_free) |
Denys Vlasenko | f0e2d93 | 2008-05-19 16:31:57 +1000 | [diff] [blame] | 3233 | kmem_free(in_f); |
Eric Sandeen | b474c7a | 2014-06-22 15:04:54 +1000 | [diff] [blame] | 3234 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3235 | } |
| 3236 | |
| 3237 | /* |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 3238 | * Recover QUOTAOFF records. We simply make a note of it in the xlog |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3239 | * structure, so that we know not to do any dquot item or dquot buffer recovery, |
| 3240 | * of that type. |
| 3241 | */ |
| 3242 | STATIC int |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3243 | xlog_recover_quotaoff_pass1( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 3244 | struct xlog *log, |
| 3245 | struct xlog_recover_item *item) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3246 | { |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3247 | xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3248 | ASSERT(qoff_f); |
| 3249 | |
| 3250 | /* |
| 3251 | * The logitem format's flag tells us if this was user quotaoff, |
Nathan Scott | 77a7cce | 2006-01-11 15:35:57 +1100 | [diff] [blame] | 3252 | * group/project quotaoff or both. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3253 | */ |
| 3254 | if (qoff_f->qf_flags & XFS_UQUOTA_ACCT) |
| 3255 | log->l_quotaoffs_flag |= XFS_DQ_USER; |
Nathan Scott | 77a7cce | 2006-01-11 15:35:57 +1100 | [diff] [blame] | 3256 | if (qoff_f->qf_flags & XFS_PQUOTA_ACCT) |
| 3257 | log->l_quotaoffs_flag |= XFS_DQ_PROJ; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3258 | if (qoff_f->qf_flags & XFS_GQUOTA_ACCT) |
| 3259 | log->l_quotaoffs_flag |= XFS_DQ_GROUP; |
| 3260 | |
Eric Sandeen | d99831f | 2014-06-22 15:03:54 +1000 | [diff] [blame] | 3261 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3262 | } |
| 3263 | |
| 3264 | /* |
| 3265 | * Recover a dquot record |
| 3266 | */ |
| 3267 | STATIC int |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3268 | xlog_recover_dquot_pass2( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 3269 | struct xlog *log, |
| 3270 | struct list_head *buffer_list, |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 3271 | struct xlog_recover_item *item, |
| 3272 | xfs_lsn_t current_lsn) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3273 | { |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3274 | xfs_mount_t *mp = log->l_mp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3275 | xfs_buf_t *bp; |
| 3276 | struct xfs_disk_dquot *ddq, *recddq; |
| 3277 | int error; |
| 3278 | xfs_dq_logformat_t *dq_f; |
| 3279 | uint type; |
| 3280 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3281 | |
| 3282 | /* |
| 3283 | * Filesystems are required to send in quota flags at mount time. |
| 3284 | */ |
| 3285 | if (mp->m_qflags == 0) |
Eric Sandeen | d99831f | 2014-06-22 15:03:54 +1000 | [diff] [blame] | 3286 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3287 | |
Christoph Hellwig | 4e0d5f9 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 3288 | recddq = item->ri_buf[1].i_addr; |
| 3289 | if (recddq == NULL) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 3290 | xfs_alert(log->l_mp, "NULL dquot in %s.", __func__); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3291 | return -EIO; |
Christoph Hellwig | 0c5e1ce | 2009-06-08 15:33:21 +0200 | [diff] [blame] | 3292 | } |
Jan Rekorajski | 8ec6dba | 2009-11-16 11:57:02 +0000 | [diff] [blame] | 3293 | if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 3294 | xfs_alert(log->l_mp, "dquot too small (%d) in %s.", |
Christoph Hellwig | 0c5e1ce | 2009-06-08 15:33:21 +0200 | [diff] [blame] | 3295 | item->ri_buf[1].i_len, __func__); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3296 | return -EIO; |
Christoph Hellwig | 0c5e1ce | 2009-06-08 15:33:21 +0200 | [diff] [blame] | 3297 | } |
| 3298 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3299 | /* |
| 3300 | * This type of quotas was turned off, so ignore this record. |
| 3301 | */ |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 3302 | type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3303 | ASSERT(type); |
| 3304 | if (log->l_quotaoffs_flag & type) |
Eric Sandeen | d99831f | 2014-06-22 15:03:54 +1000 | [diff] [blame] | 3305 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3306 | |
| 3307 | /* |
| 3308 | * At this point we know that quota was _not_ turned off. |
| 3309 | * Since the mount flags are not indicating to us otherwise, this |
| 3310 | * must mean that quota is on, and the dquot needs to be replayed. |
| 3311 | * Remember that we may not have fully recovered the superblock yet, |
| 3312 | * so we can't do the usual trick of looking at the SB quota bits. |
| 3313 | * |
| 3314 | * The other possibility, of course, is that the quota subsystem was |
| 3315 | * removed since the last mount - ENOSYS. |
| 3316 | */ |
Christoph Hellwig | 4e0d5f9 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 3317 | dq_f = item->ri_buf[0].i_addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3318 | ASSERT(dq_f); |
Dave Chinner | 9aede1d | 2013-10-15 09:17:52 +1100 | [diff] [blame] | 3319 | error = xfs_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN, |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 3320 | "xlog_recover_dquot_pass2 (log copy)"); |
| 3321 | if (error) |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3322 | return -EIO; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3323 | ASSERT(dq_f->qlf_len == 1); |
| 3324 | |
Dave Chinner | ad3714b | 2014-08-04 12:59:31 +1000 | [diff] [blame] | 3325 | /* |
| 3326 | * At this point we are assuming that the dquots have been allocated |
| 3327 | * and hence the buffer has valid dquots stamped in it. It should, |
| 3328 | * therefore, pass verifier validation. If the dquot is bad, then the |
| 3329 | * we'll return an error here, so we don't need to specifically check |
| 3330 | * the dquot in the buffer after the verifier has run. |
| 3331 | */ |
Dave Chinner | 7ca790a | 2012-04-23 15:58:55 +1000 | [diff] [blame] | 3332 | error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dq_f->qlf_blkno, |
Dave Chinner | c3f8fc7 | 2012-11-12 22:54:01 +1100 | [diff] [blame] | 3333 | XFS_FSB_TO_BB(mp, dq_f->qlf_len), 0, &bp, |
Dave Chinner | ad3714b | 2014-08-04 12:59:31 +1000 | [diff] [blame] | 3334 | &xfs_dquot_buf_ops); |
Dave Chinner | 7ca790a | 2012-04-23 15:58:55 +1000 | [diff] [blame] | 3335 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3336 | return error; |
Dave Chinner | 7ca790a | 2012-04-23 15:58:55 +1000 | [diff] [blame] | 3337 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3338 | ASSERT(bp); |
Christoph Hellwig | 88ee2df | 2015-06-22 09:44:29 +1000 | [diff] [blame] | 3339 | ddq = xfs_buf_offset(bp, dq_f->qlf_boffset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3340 | |
| 3341 | /* |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 3342 | * If the dquot has an LSN in it, recover the dquot only if it's less |
| 3343 | * than the lsn of the transaction we are replaying. |
| 3344 | */ |
| 3345 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
| 3346 | struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddq; |
| 3347 | xfs_lsn_t lsn = be64_to_cpu(dqb->dd_lsn); |
| 3348 | |
| 3349 | if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) { |
| 3350 | goto out_release; |
| 3351 | } |
| 3352 | } |
| 3353 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3354 | memcpy(ddq, recddq, item->ri_buf[1].i_len); |
Dave Chinner | 6fcdc59 | 2013-06-03 15:28:46 +1000 | [diff] [blame] | 3355 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
| 3356 | xfs_update_cksum((char *)ddq, sizeof(struct xfs_dqblk), |
| 3357 | XFS_DQUOT_CRC_OFF); |
| 3358 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3359 | |
| 3360 | ASSERT(dq_f->qlf_size == 2); |
Dave Chinner | ebad861 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 3361 | ASSERT(bp->b_target->bt_mount == mp); |
Christoph Hellwig | cb669ca | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 3362 | bp->b_iodone = xlog_recover_iodone; |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 3363 | xfs_buf_delwri_queue(bp, buffer_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3364 | |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 3365 | out_release: |
| 3366 | xfs_buf_relse(bp); |
| 3367 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3368 | } |
| 3369 | |
| 3370 | /* |
| 3371 | * This routine is called to create an in-core extent free intent |
| 3372 | * item from the efi format structure which was logged on disk. |
| 3373 | * It allocates an in-core efi, copies the extents from the format |
| 3374 | * structure into it, and adds the efi to the AIL with the given |
| 3375 | * LSN. |
| 3376 | */ |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 3377 | STATIC int |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3378 | xlog_recover_efi_pass2( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 3379 | struct xlog *log, |
| 3380 | struct xlog_recover_item *item, |
| 3381 | xfs_lsn_t lsn) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3382 | { |
Brian Foster | e32a1d1 | 2015-08-19 09:52:21 +1000 | [diff] [blame] | 3383 | int error; |
| 3384 | struct xfs_mount *mp = log->l_mp; |
| 3385 | struct xfs_efi_log_item *efip; |
| 3386 | struct xfs_efi_log_format *efi_formatp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3387 | |
Christoph Hellwig | 4e0d5f9 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 3388 | efi_formatp = item->ri_buf[0].i_addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3389 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3390 | efip = xfs_efi_init(mp, efi_formatp->efi_nextents); |
Brian Foster | e32a1d1 | 2015-08-19 09:52:21 +1000 | [diff] [blame] | 3391 | error = xfs_efi_copy_format(&item->ri_buf[0], &efip->efi_format); |
| 3392 | if (error) { |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 3393 | xfs_efi_item_free(efip); |
| 3394 | return error; |
| 3395 | } |
Dave Chinner | b199c8a | 2010-12-20 11:59:49 +1100 | [diff] [blame] | 3396 | atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3397 | |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 3398 | spin_lock(&log->l_ailp->xa_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3399 | /* |
Brian Foster | e32a1d1 | 2015-08-19 09:52:21 +1000 | [diff] [blame] | 3400 | * The EFI has two references. One for the EFD and one for EFI to ensure |
| 3401 | * it makes it into the AIL. Insert the EFI into the AIL directly and |
| 3402 | * drop the EFI reference. Note that xfs_trans_ail_update() drops the |
| 3403 | * AIL lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3404 | */ |
Dave Chinner | e605994 | 2010-12-20 12:34:26 +1100 | [diff] [blame] | 3405 | xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn); |
Brian Foster | e32a1d1 | 2015-08-19 09:52:21 +1000 | [diff] [blame] | 3406 | xfs_efi_release(efip); |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 3407 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3408 | } |
| 3409 | |
| 3410 | |
| 3411 | /* |
Brian Foster | e32a1d1 | 2015-08-19 09:52:21 +1000 | [diff] [blame] | 3412 | * This routine is called when an EFD format structure is found in a committed |
| 3413 | * transaction in the log. Its purpose is to cancel the corresponding EFI if it |
| 3414 | * was still in the log. To do this it searches the AIL for the EFI with an id |
| 3415 | * equal to that in the EFD format structure. If we find it we drop the EFD |
| 3416 | * reference, which removes the EFI from the AIL and frees it. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3417 | */ |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3418 | STATIC int |
| 3419 | xlog_recover_efd_pass2( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 3420 | struct xlog *log, |
| 3421 | struct xlog_recover_item *item) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3422 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3423 | xfs_efd_log_format_t *efd_formatp; |
| 3424 | xfs_efi_log_item_t *efip = NULL; |
| 3425 | xfs_log_item_t *lip; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3426 | __uint64_t efi_id; |
David Chinner | 27d8d5f | 2008-10-30 17:38:39 +1100 | [diff] [blame] | 3427 | struct xfs_ail_cursor cur; |
David Chinner | 783a2f6 | 2008-10-30 17:39:58 +1100 | [diff] [blame] | 3428 | struct xfs_ail *ailp = log->l_ailp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3429 | |
Christoph Hellwig | 4e0d5f9 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 3430 | efd_formatp = item->ri_buf[0].i_addr; |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 3431 | ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) + |
| 3432 | ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) || |
| 3433 | (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) + |
| 3434 | ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t))))); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3435 | efi_id = efd_formatp->efd_efi_id; |
| 3436 | |
| 3437 | /* |
Brian Foster | e32a1d1 | 2015-08-19 09:52:21 +1000 | [diff] [blame] | 3438 | * Search for the EFI with the id in the EFD format structure in the |
| 3439 | * AIL. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3440 | */ |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 3441 | spin_lock(&ailp->xa_lock); |
| 3442 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3443 | while (lip != NULL) { |
| 3444 | if (lip->li_type == XFS_LI_EFI) { |
| 3445 | efip = (xfs_efi_log_item_t *)lip; |
| 3446 | if (efip->efi_format.efi_id == efi_id) { |
| 3447 | /* |
Brian Foster | e32a1d1 | 2015-08-19 09:52:21 +1000 | [diff] [blame] | 3448 | * Drop the EFD reference to the EFI. This |
| 3449 | * removes the EFI from the AIL and frees it. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3450 | */ |
Brian Foster | e32a1d1 | 2015-08-19 09:52:21 +1000 | [diff] [blame] | 3451 | spin_unlock(&ailp->xa_lock); |
| 3452 | xfs_efi_release(efip); |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 3453 | spin_lock(&ailp->xa_lock); |
David Chinner | 27d8d5f | 2008-10-30 17:38:39 +1100 | [diff] [blame] | 3454 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3455 | } |
| 3456 | } |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 3457 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3458 | } |
Brian Foster | e32a1d1 | 2015-08-19 09:52:21 +1000 | [diff] [blame] | 3459 | |
Eric Sandeen | e4a1e29 | 2014-04-14 19:06:05 +1000 | [diff] [blame] | 3460 | xfs_trans_ail_cursor_done(&cur); |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 3461 | spin_unlock(&ailp->xa_lock); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3462 | |
| 3463 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3464 | } |
| 3465 | |
| 3466 | /* |
Darrick J. Wong | 9e88b5d | 2016-08-03 12:09:48 +1000 | [diff] [blame] | 3467 | * This routine is called to create an in-core extent rmap update |
| 3468 | * item from the rui format structure which was logged on disk. |
| 3469 | * It allocates an in-core rui, copies the extents from the format |
| 3470 | * structure into it, and adds the rui to the AIL with the given |
| 3471 | * LSN. |
| 3472 | */ |
| 3473 | STATIC int |
| 3474 | xlog_recover_rui_pass2( |
| 3475 | struct xlog *log, |
| 3476 | struct xlog_recover_item *item, |
| 3477 | xfs_lsn_t lsn) |
| 3478 | { |
| 3479 | int error; |
| 3480 | struct xfs_mount *mp = log->l_mp; |
| 3481 | struct xfs_rui_log_item *ruip; |
| 3482 | struct xfs_rui_log_format *rui_formatp; |
| 3483 | |
| 3484 | rui_formatp = item->ri_buf[0].i_addr; |
| 3485 | |
| 3486 | ruip = xfs_rui_init(mp, rui_formatp->rui_nextents); |
| 3487 | error = xfs_rui_copy_format(&item->ri_buf[0], &ruip->rui_format); |
| 3488 | if (error) { |
| 3489 | xfs_rui_item_free(ruip); |
| 3490 | return error; |
| 3491 | } |
| 3492 | atomic_set(&ruip->rui_next_extent, rui_formatp->rui_nextents); |
| 3493 | |
| 3494 | spin_lock(&log->l_ailp->xa_lock); |
| 3495 | /* |
| 3496 | * The RUI has two references. One for the RUD and one for RUI to ensure |
| 3497 | * it makes it into the AIL. Insert the RUI into the AIL directly and |
| 3498 | * drop the RUI reference. Note that xfs_trans_ail_update() drops the |
| 3499 | * AIL lock. |
| 3500 | */ |
| 3501 | xfs_trans_ail_update(log->l_ailp, &ruip->rui_item, lsn); |
| 3502 | xfs_rui_release(ruip); |
| 3503 | return 0; |
| 3504 | } |
| 3505 | |
| 3506 | |
| 3507 | /* |
| 3508 | * This routine is called when an RUD format structure is found in a committed |
| 3509 | * transaction in the log. Its purpose is to cancel the corresponding RUI if it |
| 3510 | * was still in the log. To do this it searches the AIL for the RUI with an id |
| 3511 | * equal to that in the RUD format structure. If we find it we drop the RUD |
| 3512 | * reference, which removes the RUI from the AIL and frees it. |
| 3513 | */ |
| 3514 | STATIC int |
| 3515 | xlog_recover_rud_pass2( |
| 3516 | struct xlog *log, |
| 3517 | struct xlog_recover_item *item) |
| 3518 | { |
| 3519 | struct xfs_rud_log_format *rud_formatp; |
| 3520 | struct xfs_rui_log_item *ruip = NULL; |
| 3521 | struct xfs_log_item *lip; |
| 3522 | __uint64_t rui_id; |
| 3523 | struct xfs_ail_cursor cur; |
| 3524 | struct xfs_ail *ailp = log->l_ailp; |
| 3525 | |
| 3526 | rud_formatp = item->ri_buf[0].i_addr; |
Darrick J. Wong | 722e251 | 2016-08-03 12:28:43 +1000 | [diff] [blame] | 3527 | ASSERT(item->ri_buf[0].i_len == sizeof(struct xfs_rud_log_format)); |
Darrick J. Wong | 9e88b5d | 2016-08-03 12:09:48 +1000 | [diff] [blame] | 3528 | rui_id = rud_formatp->rud_rui_id; |
| 3529 | |
| 3530 | /* |
| 3531 | * Search for the RUI with the id in the RUD format structure in the |
| 3532 | * AIL. |
| 3533 | */ |
| 3534 | spin_lock(&ailp->xa_lock); |
| 3535 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); |
| 3536 | while (lip != NULL) { |
| 3537 | if (lip->li_type == XFS_LI_RUI) { |
| 3538 | ruip = (struct xfs_rui_log_item *)lip; |
| 3539 | if (ruip->rui_format.rui_id == rui_id) { |
| 3540 | /* |
| 3541 | * Drop the RUD reference to the RUI. This |
| 3542 | * removes the RUI from the AIL and frees it. |
| 3543 | */ |
| 3544 | spin_unlock(&ailp->xa_lock); |
| 3545 | xfs_rui_release(ruip); |
| 3546 | spin_lock(&ailp->xa_lock); |
| 3547 | break; |
| 3548 | } |
| 3549 | } |
| 3550 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
| 3551 | } |
| 3552 | |
| 3553 | xfs_trans_ail_cursor_done(&cur); |
| 3554 | spin_unlock(&ailp->xa_lock); |
| 3555 | |
| 3556 | return 0; |
| 3557 | } |
| 3558 | |
| 3559 | /* |
Darrick J. Wong | f997ee2 | 2016-10-03 09:11:21 -0700 | [diff] [blame] | 3560 | * Copy an CUI format buffer from the given buf, and into the destination |
| 3561 | * CUI format structure. The CUI/CUD items were designed not to need any |
| 3562 | * special alignment handling. |
| 3563 | */ |
| 3564 | static int |
| 3565 | xfs_cui_copy_format( |
| 3566 | struct xfs_log_iovec *buf, |
| 3567 | struct xfs_cui_log_format *dst_cui_fmt) |
| 3568 | { |
| 3569 | struct xfs_cui_log_format *src_cui_fmt; |
| 3570 | uint len; |
| 3571 | |
| 3572 | src_cui_fmt = buf->i_addr; |
| 3573 | len = xfs_cui_log_format_sizeof(src_cui_fmt->cui_nextents); |
| 3574 | |
| 3575 | if (buf->i_len == len) { |
| 3576 | memcpy(dst_cui_fmt, src_cui_fmt, len); |
| 3577 | return 0; |
| 3578 | } |
| 3579 | return -EFSCORRUPTED; |
| 3580 | } |
| 3581 | |
| 3582 | /* |
| 3583 | * This routine is called to create an in-core extent refcount update |
| 3584 | * item from the cui format structure which was logged on disk. |
| 3585 | * It allocates an in-core cui, copies the extents from the format |
| 3586 | * structure into it, and adds the cui to the AIL with the given |
| 3587 | * LSN. |
| 3588 | */ |
| 3589 | STATIC int |
| 3590 | xlog_recover_cui_pass2( |
| 3591 | struct xlog *log, |
| 3592 | struct xlog_recover_item *item, |
| 3593 | xfs_lsn_t lsn) |
| 3594 | { |
| 3595 | int error; |
| 3596 | struct xfs_mount *mp = log->l_mp; |
| 3597 | struct xfs_cui_log_item *cuip; |
| 3598 | struct xfs_cui_log_format *cui_formatp; |
| 3599 | |
| 3600 | cui_formatp = item->ri_buf[0].i_addr; |
| 3601 | |
| 3602 | cuip = xfs_cui_init(mp, cui_formatp->cui_nextents); |
| 3603 | error = xfs_cui_copy_format(&item->ri_buf[0], &cuip->cui_format); |
| 3604 | if (error) { |
| 3605 | xfs_cui_item_free(cuip); |
| 3606 | return error; |
| 3607 | } |
| 3608 | atomic_set(&cuip->cui_next_extent, cui_formatp->cui_nextents); |
| 3609 | |
| 3610 | spin_lock(&log->l_ailp->xa_lock); |
| 3611 | /* |
| 3612 | * The CUI has two references. One for the CUD and one for CUI to ensure |
| 3613 | * it makes it into the AIL. Insert the CUI into the AIL directly and |
| 3614 | * drop the CUI reference. Note that xfs_trans_ail_update() drops the |
| 3615 | * AIL lock. |
| 3616 | */ |
| 3617 | xfs_trans_ail_update(log->l_ailp, &cuip->cui_item, lsn); |
| 3618 | xfs_cui_release(cuip); |
| 3619 | return 0; |
| 3620 | } |
| 3621 | |
| 3622 | |
| 3623 | /* |
| 3624 | * This routine is called when an CUD format structure is found in a committed |
| 3625 | * transaction in the log. Its purpose is to cancel the corresponding CUI if it |
| 3626 | * was still in the log. To do this it searches the AIL for the CUI with an id |
| 3627 | * equal to that in the CUD format structure. If we find it we drop the CUD |
| 3628 | * reference, which removes the CUI from the AIL and frees it. |
| 3629 | */ |
| 3630 | STATIC int |
| 3631 | xlog_recover_cud_pass2( |
| 3632 | struct xlog *log, |
| 3633 | struct xlog_recover_item *item) |
| 3634 | { |
| 3635 | struct xfs_cud_log_format *cud_formatp; |
| 3636 | struct xfs_cui_log_item *cuip = NULL; |
| 3637 | struct xfs_log_item *lip; |
| 3638 | __uint64_t cui_id; |
| 3639 | struct xfs_ail_cursor cur; |
| 3640 | struct xfs_ail *ailp = log->l_ailp; |
| 3641 | |
| 3642 | cud_formatp = item->ri_buf[0].i_addr; |
| 3643 | if (item->ri_buf[0].i_len != sizeof(struct xfs_cud_log_format)) |
| 3644 | return -EFSCORRUPTED; |
| 3645 | cui_id = cud_formatp->cud_cui_id; |
| 3646 | |
| 3647 | /* |
| 3648 | * Search for the CUI with the id in the CUD format structure in the |
| 3649 | * AIL. |
| 3650 | */ |
| 3651 | spin_lock(&ailp->xa_lock); |
| 3652 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); |
| 3653 | while (lip != NULL) { |
| 3654 | if (lip->li_type == XFS_LI_CUI) { |
| 3655 | cuip = (struct xfs_cui_log_item *)lip; |
| 3656 | if (cuip->cui_format.cui_id == cui_id) { |
| 3657 | /* |
| 3658 | * Drop the CUD reference to the CUI. This |
| 3659 | * removes the CUI from the AIL and frees it. |
| 3660 | */ |
| 3661 | spin_unlock(&ailp->xa_lock); |
| 3662 | xfs_cui_release(cuip); |
| 3663 | spin_lock(&ailp->xa_lock); |
| 3664 | break; |
| 3665 | } |
| 3666 | } |
| 3667 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
| 3668 | } |
| 3669 | |
| 3670 | xfs_trans_ail_cursor_done(&cur); |
| 3671 | spin_unlock(&ailp->xa_lock); |
| 3672 | |
| 3673 | return 0; |
| 3674 | } |
| 3675 | |
| 3676 | /* |
Darrick J. Wong | 77d61fe | 2016-10-03 09:11:26 -0700 | [diff] [blame] | 3677 | * Copy an BUI format buffer from the given buf, and into the destination |
| 3678 | * BUI format structure. The BUI/BUD items were designed not to need any |
| 3679 | * special alignment handling. |
| 3680 | */ |
| 3681 | static int |
| 3682 | xfs_bui_copy_format( |
| 3683 | struct xfs_log_iovec *buf, |
| 3684 | struct xfs_bui_log_format *dst_bui_fmt) |
| 3685 | { |
| 3686 | struct xfs_bui_log_format *src_bui_fmt; |
| 3687 | uint len; |
| 3688 | |
| 3689 | src_bui_fmt = buf->i_addr; |
| 3690 | len = xfs_bui_log_format_sizeof(src_bui_fmt->bui_nextents); |
| 3691 | |
| 3692 | if (buf->i_len == len) { |
| 3693 | memcpy(dst_bui_fmt, src_bui_fmt, len); |
| 3694 | return 0; |
| 3695 | } |
| 3696 | return -EFSCORRUPTED; |
| 3697 | } |
| 3698 | |
| 3699 | /* |
| 3700 | * This routine is called to create an in-core extent bmap update |
| 3701 | * item from the bui format structure which was logged on disk. |
| 3702 | * It allocates an in-core bui, copies the extents from the format |
| 3703 | * structure into it, and adds the bui to the AIL with the given |
| 3704 | * LSN. |
| 3705 | */ |
| 3706 | STATIC int |
| 3707 | xlog_recover_bui_pass2( |
| 3708 | struct xlog *log, |
| 3709 | struct xlog_recover_item *item, |
| 3710 | xfs_lsn_t lsn) |
| 3711 | { |
| 3712 | int error; |
| 3713 | struct xfs_mount *mp = log->l_mp; |
| 3714 | struct xfs_bui_log_item *buip; |
| 3715 | struct xfs_bui_log_format *bui_formatp; |
| 3716 | |
| 3717 | bui_formatp = item->ri_buf[0].i_addr; |
| 3718 | |
| 3719 | if (bui_formatp->bui_nextents != XFS_BUI_MAX_FAST_EXTENTS) |
| 3720 | return -EFSCORRUPTED; |
| 3721 | buip = xfs_bui_init(mp); |
| 3722 | error = xfs_bui_copy_format(&item->ri_buf[0], &buip->bui_format); |
| 3723 | if (error) { |
| 3724 | xfs_bui_item_free(buip); |
| 3725 | return error; |
| 3726 | } |
| 3727 | atomic_set(&buip->bui_next_extent, bui_formatp->bui_nextents); |
| 3728 | |
| 3729 | spin_lock(&log->l_ailp->xa_lock); |
| 3730 | /* |
| 3731 | * The RUI has two references. One for the RUD and one for RUI to ensure |
| 3732 | * it makes it into the AIL. Insert the RUI into the AIL directly and |
| 3733 | * drop the RUI reference. Note that xfs_trans_ail_update() drops the |
| 3734 | * AIL lock. |
| 3735 | */ |
| 3736 | xfs_trans_ail_update(log->l_ailp, &buip->bui_item, lsn); |
| 3737 | xfs_bui_release(buip); |
| 3738 | return 0; |
| 3739 | } |
| 3740 | |
| 3741 | |
| 3742 | /* |
| 3743 | * This routine is called when an BUD format structure is found in a committed |
| 3744 | * transaction in the log. Its purpose is to cancel the corresponding BUI if it |
| 3745 | * was still in the log. To do this it searches the AIL for the BUI with an id |
| 3746 | * equal to that in the BUD format structure. If we find it we drop the BUD |
| 3747 | * reference, which removes the BUI from the AIL and frees it. |
| 3748 | */ |
| 3749 | STATIC int |
| 3750 | xlog_recover_bud_pass2( |
| 3751 | struct xlog *log, |
| 3752 | struct xlog_recover_item *item) |
| 3753 | { |
| 3754 | struct xfs_bud_log_format *bud_formatp; |
| 3755 | struct xfs_bui_log_item *buip = NULL; |
| 3756 | struct xfs_log_item *lip; |
| 3757 | __uint64_t bui_id; |
| 3758 | struct xfs_ail_cursor cur; |
| 3759 | struct xfs_ail *ailp = log->l_ailp; |
| 3760 | |
| 3761 | bud_formatp = item->ri_buf[0].i_addr; |
| 3762 | if (item->ri_buf[0].i_len != sizeof(struct xfs_bud_log_format)) |
| 3763 | return -EFSCORRUPTED; |
| 3764 | bui_id = bud_formatp->bud_bui_id; |
| 3765 | |
| 3766 | /* |
| 3767 | * Search for the BUI with the id in the BUD format structure in the |
| 3768 | * AIL. |
| 3769 | */ |
| 3770 | spin_lock(&ailp->xa_lock); |
| 3771 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); |
| 3772 | while (lip != NULL) { |
| 3773 | if (lip->li_type == XFS_LI_BUI) { |
| 3774 | buip = (struct xfs_bui_log_item *)lip; |
| 3775 | if (buip->bui_format.bui_id == bui_id) { |
| 3776 | /* |
| 3777 | * Drop the BUD reference to the BUI. This |
| 3778 | * removes the BUI from the AIL and frees it. |
| 3779 | */ |
| 3780 | spin_unlock(&ailp->xa_lock); |
| 3781 | xfs_bui_release(buip); |
| 3782 | spin_lock(&ailp->xa_lock); |
| 3783 | break; |
| 3784 | } |
| 3785 | } |
| 3786 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
| 3787 | } |
| 3788 | |
| 3789 | xfs_trans_ail_cursor_done(&cur); |
| 3790 | spin_unlock(&ailp->xa_lock); |
| 3791 | |
| 3792 | return 0; |
| 3793 | } |
| 3794 | |
| 3795 | /* |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3796 | * This routine is called when an inode create format structure is found in a |
| 3797 | * committed transaction in the log. It's purpose is to initialise the inodes |
| 3798 | * being allocated on disk. This requires us to get inode cluster buffers that |
| 3799 | * match the range to be intialised, stamped with inode templates and written |
| 3800 | * by delayed write so that subsequent modifications will hit the cached buffer |
| 3801 | * and only need writing out at the end of recovery. |
| 3802 | */ |
| 3803 | STATIC int |
| 3804 | xlog_recover_do_icreate_pass2( |
| 3805 | struct xlog *log, |
| 3806 | struct list_head *buffer_list, |
| 3807 | xlog_recover_item_t *item) |
| 3808 | { |
| 3809 | struct xfs_mount *mp = log->l_mp; |
| 3810 | struct xfs_icreate_log *icl; |
| 3811 | xfs_agnumber_t agno; |
| 3812 | xfs_agblock_t agbno; |
| 3813 | unsigned int count; |
| 3814 | unsigned int isize; |
| 3815 | xfs_agblock_t length; |
Brian Foster | fc0d165 | 2015-08-19 09:59:38 +1000 | [diff] [blame] | 3816 | int blks_per_cluster; |
| 3817 | int bb_per_cluster; |
| 3818 | int cancel_count; |
| 3819 | int nbufs; |
| 3820 | int i; |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3821 | |
| 3822 | icl = (struct xfs_icreate_log *)item->ri_buf[0].i_addr; |
| 3823 | if (icl->icl_type != XFS_LI_ICREATE) { |
| 3824 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad type"); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3825 | return -EINVAL; |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3826 | } |
| 3827 | |
| 3828 | if (icl->icl_size != 1) { |
| 3829 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad icl size"); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3830 | return -EINVAL; |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3831 | } |
| 3832 | |
| 3833 | agno = be32_to_cpu(icl->icl_ag); |
| 3834 | if (agno >= mp->m_sb.sb_agcount) { |
| 3835 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agno"); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3836 | return -EINVAL; |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3837 | } |
| 3838 | agbno = be32_to_cpu(icl->icl_agbno); |
| 3839 | if (!agbno || agbno == NULLAGBLOCK || agbno >= mp->m_sb.sb_agblocks) { |
| 3840 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agbno"); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3841 | return -EINVAL; |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3842 | } |
| 3843 | isize = be32_to_cpu(icl->icl_isize); |
| 3844 | if (isize != mp->m_sb.sb_inodesize) { |
| 3845 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad isize"); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3846 | return -EINVAL; |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3847 | } |
| 3848 | count = be32_to_cpu(icl->icl_count); |
| 3849 | if (!count) { |
| 3850 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count"); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3851 | return -EINVAL; |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3852 | } |
| 3853 | length = be32_to_cpu(icl->icl_length); |
| 3854 | if (!length || length >= mp->m_sb.sb_agblocks) { |
| 3855 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad length"); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3856 | return -EINVAL; |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3857 | } |
| 3858 | |
Brian Foster | 7f43c90 | 2015-05-29 09:06:30 +1000 | [diff] [blame] | 3859 | /* |
| 3860 | * The inode chunk is either full or sparse and we only support |
| 3861 | * m_ialloc_min_blks sized sparse allocations at this time. |
| 3862 | */ |
| 3863 | if (length != mp->m_ialloc_blks && |
| 3864 | length != mp->m_ialloc_min_blks) { |
| 3865 | xfs_warn(log->l_mp, |
| 3866 | "%s: unsupported chunk length", __FUNCTION__); |
| 3867 | return -EINVAL; |
| 3868 | } |
| 3869 | |
| 3870 | /* verify inode count is consistent with extent length */ |
| 3871 | if ((count >> mp->m_sb.sb_inopblog) != length) { |
| 3872 | xfs_warn(log->l_mp, |
| 3873 | "%s: inconsistent inode count and chunk length", |
| 3874 | __FUNCTION__); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3875 | return -EINVAL; |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3876 | } |
| 3877 | |
| 3878 | /* |
Brian Foster | fc0d165 | 2015-08-19 09:59:38 +1000 | [diff] [blame] | 3879 | * The icreate transaction can cover multiple cluster buffers and these |
| 3880 | * buffers could have been freed and reused. Check the individual |
| 3881 | * buffers for cancellation so we don't overwrite anything written after |
| 3882 | * a cancellation. |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3883 | */ |
Brian Foster | fc0d165 | 2015-08-19 09:59:38 +1000 | [diff] [blame] | 3884 | blks_per_cluster = xfs_icluster_size_fsb(mp); |
| 3885 | bb_per_cluster = XFS_FSB_TO_BB(mp, blks_per_cluster); |
| 3886 | nbufs = length / blks_per_cluster; |
| 3887 | for (i = 0, cancel_count = 0; i < nbufs; i++) { |
| 3888 | xfs_daddr_t daddr; |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3889 | |
Brian Foster | fc0d165 | 2015-08-19 09:59:38 +1000 | [diff] [blame] | 3890 | daddr = XFS_AGB_TO_DADDR(mp, agno, |
| 3891 | agbno + i * blks_per_cluster); |
| 3892 | if (xlog_check_buffer_cancelled(log, daddr, bb_per_cluster, 0)) |
| 3893 | cancel_count++; |
| 3894 | } |
| 3895 | |
| 3896 | /* |
| 3897 | * We currently only use icreate for a single allocation at a time. This |
| 3898 | * means we should expect either all or none of the buffers to be |
| 3899 | * cancelled. Be conservative and skip replay if at least one buffer is |
| 3900 | * cancelled, but warn the user that something is awry if the buffers |
| 3901 | * are not consistent. |
| 3902 | * |
| 3903 | * XXX: This must be refined to only skip cancelled clusters once we use |
| 3904 | * icreate for multiple chunk allocations. |
| 3905 | */ |
| 3906 | ASSERT(!cancel_count || cancel_count == nbufs); |
| 3907 | if (cancel_count) { |
| 3908 | if (cancel_count != nbufs) |
| 3909 | xfs_warn(mp, |
| 3910 | "WARNING: partial inode chunk cancellation, skipped icreate."); |
Brian Foster | 78d57e4 | 2015-08-19 09:58:48 +1000 | [diff] [blame] | 3911 | trace_xfs_log_recover_icreate_cancel(log, icl); |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3912 | return 0; |
Brian Foster | 78d57e4 | 2015-08-19 09:58:48 +1000 | [diff] [blame] | 3913 | } |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3914 | |
Brian Foster | 78d57e4 | 2015-08-19 09:58:48 +1000 | [diff] [blame] | 3915 | trace_xfs_log_recover_icreate_recover(log, icl); |
Brian Foster | fc0d165 | 2015-08-19 09:59:38 +1000 | [diff] [blame] | 3916 | return xfs_ialloc_inode_init(mp, NULL, buffer_list, count, agno, agbno, |
| 3917 | length, be32_to_cpu(icl->icl_gen)); |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3918 | } |
| 3919 | |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 3920 | STATIC void |
| 3921 | xlog_recover_buffer_ra_pass2( |
| 3922 | struct xlog *log, |
| 3923 | struct xlog_recover_item *item) |
| 3924 | { |
| 3925 | struct xfs_buf_log_format *buf_f = item->ri_buf[0].i_addr; |
| 3926 | struct xfs_mount *mp = log->l_mp; |
| 3927 | |
Dave Chinner | 84a5b73 | 2013-08-27 08:10:53 +1000 | [diff] [blame] | 3928 | if (xlog_peek_buffer_cancelled(log, buf_f->blf_blkno, |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 3929 | buf_f->blf_len, buf_f->blf_flags)) { |
| 3930 | return; |
| 3931 | } |
| 3932 | |
| 3933 | xfs_buf_readahead(mp->m_ddev_targp, buf_f->blf_blkno, |
| 3934 | buf_f->blf_len, NULL); |
| 3935 | } |
| 3936 | |
| 3937 | STATIC void |
| 3938 | xlog_recover_inode_ra_pass2( |
| 3939 | struct xlog *log, |
| 3940 | struct xlog_recover_item *item) |
| 3941 | { |
| 3942 | struct xfs_inode_log_format ilf_buf; |
| 3943 | struct xfs_inode_log_format *ilfp; |
| 3944 | struct xfs_mount *mp = log->l_mp; |
| 3945 | int error; |
| 3946 | |
| 3947 | if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) { |
| 3948 | ilfp = item->ri_buf[0].i_addr; |
| 3949 | } else { |
| 3950 | ilfp = &ilf_buf; |
| 3951 | memset(ilfp, 0, sizeof(*ilfp)); |
| 3952 | error = xfs_inode_item_format_convert(&item->ri_buf[0], ilfp); |
| 3953 | if (error) |
| 3954 | return; |
| 3955 | } |
| 3956 | |
Dave Chinner | 84a5b73 | 2013-08-27 08:10:53 +1000 | [diff] [blame] | 3957 | if (xlog_peek_buffer_cancelled(log, ilfp->ilf_blkno, ilfp->ilf_len, 0)) |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 3958 | return; |
| 3959 | |
| 3960 | xfs_buf_readahead(mp->m_ddev_targp, ilfp->ilf_blkno, |
Dave Chinner | d891400 | 2013-08-27 11:39:37 +1000 | [diff] [blame] | 3961 | ilfp->ilf_len, &xfs_inode_buf_ra_ops); |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 3962 | } |
| 3963 | |
| 3964 | STATIC void |
| 3965 | xlog_recover_dquot_ra_pass2( |
| 3966 | struct xlog *log, |
| 3967 | struct xlog_recover_item *item) |
| 3968 | { |
| 3969 | struct xfs_mount *mp = log->l_mp; |
| 3970 | struct xfs_disk_dquot *recddq; |
| 3971 | struct xfs_dq_logformat *dq_f; |
| 3972 | uint type; |
Dave Chinner | 7d6a13f | 2016-01-12 07:04:01 +1100 | [diff] [blame] | 3973 | int len; |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 3974 | |
| 3975 | |
| 3976 | if (mp->m_qflags == 0) |
| 3977 | return; |
| 3978 | |
| 3979 | recddq = item->ri_buf[1].i_addr; |
| 3980 | if (recddq == NULL) |
| 3981 | return; |
| 3982 | if (item->ri_buf[1].i_len < sizeof(struct xfs_disk_dquot)) |
| 3983 | return; |
| 3984 | |
| 3985 | type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP); |
| 3986 | ASSERT(type); |
| 3987 | if (log->l_quotaoffs_flag & type) |
| 3988 | return; |
| 3989 | |
| 3990 | dq_f = item->ri_buf[0].i_addr; |
| 3991 | ASSERT(dq_f); |
| 3992 | ASSERT(dq_f->qlf_len == 1); |
| 3993 | |
Dave Chinner | 7d6a13f | 2016-01-12 07:04:01 +1100 | [diff] [blame] | 3994 | len = XFS_FSB_TO_BB(mp, dq_f->qlf_len); |
| 3995 | if (xlog_peek_buffer_cancelled(log, dq_f->qlf_blkno, len, 0)) |
| 3996 | return; |
| 3997 | |
| 3998 | xfs_buf_readahead(mp->m_ddev_targp, dq_f->qlf_blkno, len, |
| 3999 | &xfs_dquot_buf_ra_ops); |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 4000 | } |
| 4001 | |
| 4002 | STATIC void |
| 4003 | xlog_recover_ra_pass2( |
| 4004 | struct xlog *log, |
| 4005 | struct xlog_recover_item *item) |
| 4006 | { |
| 4007 | switch (ITEM_TYPE(item)) { |
| 4008 | case XFS_LI_BUF: |
| 4009 | xlog_recover_buffer_ra_pass2(log, item); |
| 4010 | break; |
| 4011 | case XFS_LI_INODE: |
| 4012 | xlog_recover_inode_ra_pass2(log, item); |
| 4013 | break; |
| 4014 | case XFS_LI_DQUOT: |
| 4015 | xlog_recover_dquot_ra_pass2(log, item); |
| 4016 | break; |
| 4017 | case XFS_LI_EFI: |
| 4018 | case XFS_LI_EFD: |
| 4019 | case XFS_LI_QUOTAOFF: |
Darrick J. Wong | 9e88b5d | 2016-08-03 12:09:48 +1000 | [diff] [blame] | 4020 | case XFS_LI_RUI: |
| 4021 | case XFS_LI_RUD: |
Darrick J. Wong | f997ee2 | 2016-10-03 09:11:21 -0700 | [diff] [blame] | 4022 | case XFS_LI_CUI: |
| 4023 | case XFS_LI_CUD: |
Darrick J. Wong | 77d61fe | 2016-10-03 09:11:26 -0700 | [diff] [blame] | 4024 | case XFS_LI_BUI: |
| 4025 | case XFS_LI_BUD: |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 4026 | default: |
| 4027 | break; |
| 4028 | } |
| 4029 | } |
| 4030 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4031 | STATIC int |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 4032 | xlog_recover_commit_pass1( |
Mark Tinguely | ad223e6 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 4033 | struct xlog *log, |
| 4034 | struct xlog_recover *trans, |
| 4035 | struct xlog_recover_item *item) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4036 | { |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 4037 | trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1); |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 4038 | |
| 4039 | switch (ITEM_TYPE(item)) { |
| 4040 | case XFS_LI_BUF: |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 4041 | return xlog_recover_buffer_pass1(log, item); |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 4042 | case XFS_LI_QUOTAOFF: |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 4043 | return xlog_recover_quotaoff_pass1(log, item); |
| 4044 | case XFS_LI_INODE: |
| 4045 | case XFS_LI_EFI: |
| 4046 | case XFS_LI_EFD: |
| 4047 | case XFS_LI_DQUOT: |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 4048 | case XFS_LI_ICREATE: |
Darrick J. Wong | 9e88b5d | 2016-08-03 12:09:48 +1000 | [diff] [blame] | 4049 | case XFS_LI_RUI: |
| 4050 | case XFS_LI_RUD: |
Darrick J. Wong | f997ee2 | 2016-10-03 09:11:21 -0700 | [diff] [blame] | 4051 | case XFS_LI_CUI: |
| 4052 | case XFS_LI_CUD: |
Darrick J. Wong | 77d61fe | 2016-10-03 09:11:26 -0700 | [diff] [blame] | 4053 | case XFS_LI_BUI: |
| 4054 | case XFS_LI_BUD: |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 4055 | /* nothing to do in pass 1 */ |
| 4056 | return 0; |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 4057 | default: |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 4058 | xfs_warn(log->l_mp, "%s: invalid item type (%d)", |
| 4059 | __func__, ITEM_TYPE(item)); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 4060 | ASSERT(0); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 4061 | return -EIO; |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 4062 | } |
| 4063 | } |
| 4064 | |
| 4065 | STATIC int |
| 4066 | xlog_recover_commit_pass2( |
Mark Tinguely | ad223e6 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 4067 | struct xlog *log, |
| 4068 | struct xlog_recover *trans, |
| 4069 | struct list_head *buffer_list, |
| 4070 | struct xlog_recover_item *item) |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 4071 | { |
| 4072 | trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2); |
| 4073 | |
| 4074 | switch (ITEM_TYPE(item)) { |
| 4075 | case XFS_LI_BUF: |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 4076 | return xlog_recover_buffer_pass2(log, buffer_list, item, |
| 4077 | trans->r_lsn); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 4078 | case XFS_LI_INODE: |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 4079 | return xlog_recover_inode_pass2(log, buffer_list, item, |
| 4080 | trans->r_lsn); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 4081 | case XFS_LI_EFI: |
| 4082 | return xlog_recover_efi_pass2(log, item, trans->r_lsn); |
| 4083 | case XFS_LI_EFD: |
| 4084 | return xlog_recover_efd_pass2(log, item); |
Darrick J. Wong | 9e88b5d | 2016-08-03 12:09:48 +1000 | [diff] [blame] | 4085 | case XFS_LI_RUI: |
| 4086 | return xlog_recover_rui_pass2(log, item, trans->r_lsn); |
| 4087 | case XFS_LI_RUD: |
| 4088 | return xlog_recover_rud_pass2(log, item); |
Darrick J. Wong | f997ee2 | 2016-10-03 09:11:21 -0700 | [diff] [blame] | 4089 | case XFS_LI_CUI: |
| 4090 | return xlog_recover_cui_pass2(log, item, trans->r_lsn); |
| 4091 | case XFS_LI_CUD: |
| 4092 | return xlog_recover_cud_pass2(log, item); |
Darrick J. Wong | 77d61fe | 2016-10-03 09:11:26 -0700 | [diff] [blame] | 4093 | case XFS_LI_BUI: |
| 4094 | return xlog_recover_bui_pass2(log, item, trans->r_lsn); |
| 4095 | case XFS_LI_BUD: |
| 4096 | return xlog_recover_bud_pass2(log, item); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 4097 | case XFS_LI_DQUOT: |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 4098 | return xlog_recover_dquot_pass2(log, buffer_list, item, |
| 4099 | trans->r_lsn); |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 4100 | case XFS_LI_ICREATE: |
| 4101 | return xlog_recover_do_icreate_pass2(log, buffer_list, item); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 4102 | case XFS_LI_QUOTAOFF: |
| 4103 | /* nothing to do in pass2 */ |
| 4104 | return 0; |
| 4105 | default: |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 4106 | xfs_warn(log->l_mp, "%s: invalid item type (%d)", |
| 4107 | __func__, ITEM_TYPE(item)); |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 4108 | ASSERT(0); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 4109 | return -EIO; |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 4110 | } |
| 4111 | } |
| 4112 | |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 4113 | STATIC int |
| 4114 | xlog_recover_items_pass2( |
| 4115 | struct xlog *log, |
| 4116 | struct xlog_recover *trans, |
| 4117 | struct list_head *buffer_list, |
| 4118 | struct list_head *item_list) |
| 4119 | { |
| 4120 | struct xlog_recover_item *item; |
| 4121 | int error = 0; |
| 4122 | |
| 4123 | list_for_each_entry(item, item_list, ri_list) { |
| 4124 | error = xlog_recover_commit_pass2(log, trans, |
| 4125 | buffer_list, item); |
| 4126 | if (error) |
| 4127 | return error; |
| 4128 | } |
| 4129 | |
| 4130 | return error; |
| 4131 | } |
| 4132 | |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 4133 | /* |
| 4134 | * Perform the transaction. |
| 4135 | * |
| 4136 | * If the transaction modifies a buffer or inode, do it now. Otherwise, |
| 4137 | * EFIs and EFDs get queued up by adding entries into the AIL for them. |
| 4138 | */ |
| 4139 | STATIC int |
| 4140 | xlog_recover_commit_trans( |
Mark Tinguely | ad223e6 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 4141 | struct xlog *log, |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 4142 | struct xlog_recover *trans, |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 4143 | int pass, |
| 4144 | struct list_head *buffer_list) |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 4145 | { |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 4146 | int error = 0; |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 4147 | int items_queued = 0; |
| 4148 | struct xlog_recover_item *item; |
| 4149 | struct xlog_recover_item *next; |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 4150 | LIST_HEAD (ra_list); |
| 4151 | LIST_HEAD (done_list); |
| 4152 | |
| 4153 | #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4154 | |
Brian Foster | 171192c | 2017-09-17 14:06:33 -0700 | [diff] [blame] | 4155 | hlist_del_init(&trans->r_list); |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 4156 | |
| 4157 | error = xlog_recover_reorder_trans(log, trans, pass); |
| 4158 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4159 | return error; |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 4160 | |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 4161 | list_for_each_entry_safe(item, next, &trans->r_itemq, ri_list) { |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 4162 | switch (pass) { |
| 4163 | case XLOG_RECOVER_PASS1: |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 4164 | error = xlog_recover_commit_pass1(log, trans, item); |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 4165 | break; |
| 4166 | case XLOG_RECOVER_PASS2: |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 4167 | xlog_recover_ra_pass2(log, item); |
| 4168 | list_move_tail(&item->ri_list, &ra_list); |
| 4169 | items_queued++; |
| 4170 | if (items_queued >= XLOG_RECOVER_COMMIT_QUEUE_MAX) { |
| 4171 | error = xlog_recover_items_pass2(log, trans, |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 4172 | buffer_list, &ra_list); |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 4173 | list_splice_tail_init(&ra_list, &done_list); |
| 4174 | items_queued = 0; |
| 4175 | } |
| 4176 | |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 4177 | break; |
| 4178 | default: |
| 4179 | ASSERT(0); |
| 4180 | } |
| 4181 | |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 4182 | if (error) |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 4183 | goto out; |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 4184 | } |
| 4185 | |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 4186 | out: |
| 4187 | if (!list_empty(&ra_list)) { |
| 4188 | if (!error) |
| 4189 | error = xlog_recover_items_pass2(log, trans, |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 4190 | buffer_list, &ra_list); |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 4191 | list_splice_tail_init(&ra_list, &done_list); |
| 4192 | } |
| 4193 | |
| 4194 | if (!list_empty(&done_list)) |
| 4195 | list_splice_init(&done_list, &trans->r_itemq); |
| 4196 | |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 4197 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4198 | } |
| 4199 | |
Dave Chinner | 7656066 | 2014-09-29 09:45:42 +1000 | [diff] [blame] | 4200 | STATIC void |
| 4201 | xlog_recover_add_item( |
| 4202 | struct list_head *head) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4203 | { |
Dave Chinner | 7656066 | 2014-09-29 09:45:42 +1000 | [diff] [blame] | 4204 | xlog_recover_item_t *item; |
| 4205 | |
| 4206 | item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP); |
| 4207 | INIT_LIST_HEAD(&item->ri_list); |
| 4208 | list_add_tail(&item->ri_list, head); |
| 4209 | } |
| 4210 | |
| 4211 | STATIC int |
| 4212 | xlog_recover_add_to_cont_trans( |
| 4213 | struct xlog *log, |
| 4214 | struct xlog_recover *trans, |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 4215 | char *dp, |
Dave Chinner | 7656066 | 2014-09-29 09:45:42 +1000 | [diff] [blame] | 4216 | int len) |
| 4217 | { |
| 4218 | xlog_recover_item_t *item; |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 4219 | char *ptr, *old_ptr; |
Dave Chinner | 7656066 | 2014-09-29 09:45:42 +1000 | [diff] [blame] | 4220 | int old_len; |
| 4221 | |
Brian Foster | 89cebc84 | 2015-07-29 11:51:10 +1000 | [diff] [blame] | 4222 | /* |
| 4223 | * If the transaction is empty, the header was split across this and the |
| 4224 | * previous record. Copy the rest of the header. |
| 4225 | */ |
Dave Chinner | 7656066 | 2014-09-29 09:45:42 +1000 | [diff] [blame] | 4226 | if (list_empty(&trans->r_itemq)) { |
Brian Foster | 848ccfc | 2015-11-10 10:10:33 +1100 | [diff] [blame] | 4227 | ASSERT(len <= sizeof(struct xfs_trans_header)); |
Brian Foster | 89cebc84 | 2015-07-29 11:51:10 +1000 | [diff] [blame] | 4228 | if (len > sizeof(struct xfs_trans_header)) { |
| 4229 | xfs_warn(log->l_mp, "%s: bad header length", __func__); |
| 4230 | return -EIO; |
| 4231 | } |
| 4232 | |
Dave Chinner | 7656066 | 2014-09-29 09:45:42 +1000 | [diff] [blame] | 4233 | xlog_recover_add_item(&trans->r_itemq); |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 4234 | ptr = (char *)&trans->r_theader + |
Brian Foster | 89cebc84 | 2015-07-29 11:51:10 +1000 | [diff] [blame] | 4235 | sizeof(struct xfs_trans_header) - len; |
Dave Chinner | 7656066 | 2014-09-29 09:45:42 +1000 | [diff] [blame] | 4236 | memcpy(ptr, dp, len); |
| 4237 | return 0; |
| 4238 | } |
Brian Foster | 89cebc84 | 2015-07-29 11:51:10 +1000 | [diff] [blame] | 4239 | |
Dave Chinner | 7656066 | 2014-09-29 09:45:42 +1000 | [diff] [blame] | 4240 | /* take the tail entry */ |
| 4241 | item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list); |
| 4242 | |
| 4243 | old_ptr = item->ri_buf[item->ri_cnt-1].i_addr; |
| 4244 | old_len = item->ri_buf[item->ri_cnt-1].i_len; |
| 4245 | |
Christoph Hellwig | 664b60f | 2016-04-06 09:47:01 +1000 | [diff] [blame] | 4246 | ptr = kmem_realloc(old_ptr, len + old_len, KM_SLEEP); |
Dave Chinner | 7656066 | 2014-09-29 09:45:42 +1000 | [diff] [blame] | 4247 | memcpy(&ptr[old_len], dp, len); |
| 4248 | item->ri_buf[item->ri_cnt-1].i_len += len; |
| 4249 | item->ri_buf[item->ri_cnt-1].i_addr = ptr; |
| 4250 | trace_xfs_log_recover_item_add_cont(log, trans, item, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4251 | return 0; |
| 4252 | } |
| 4253 | |
| 4254 | /* |
Dave Chinner | 7656066 | 2014-09-29 09:45:42 +1000 | [diff] [blame] | 4255 | * The next region to add is the start of a new region. It could be |
| 4256 | * a whole region or it could be the first part of a new region. Because |
| 4257 | * of this, the assumption here is that the type and size fields of all |
| 4258 | * format structures fit into the first 32 bits of the structure. |
| 4259 | * |
| 4260 | * This works because all regions must be 32 bit aligned. Therefore, we |
| 4261 | * either have both fields or we have neither field. In the case we have |
| 4262 | * neither field, the data part of the region is zero length. We only have |
| 4263 | * a log_op_header and can throw away the header since a new one will appear |
| 4264 | * later. If we have at least 4 bytes, then we can determine how many regions |
| 4265 | * will appear in the current log item. |
| 4266 | */ |
| 4267 | STATIC int |
| 4268 | xlog_recover_add_to_trans( |
| 4269 | struct xlog *log, |
| 4270 | struct xlog_recover *trans, |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 4271 | char *dp, |
Dave Chinner | 7656066 | 2014-09-29 09:45:42 +1000 | [diff] [blame] | 4272 | int len) |
| 4273 | { |
| 4274 | xfs_inode_log_format_t *in_f; /* any will do */ |
| 4275 | xlog_recover_item_t *item; |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 4276 | char *ptr; |
Dave Chinner | 7656066 | 2014-09-29 09:45:42 +1000 | [diff] [blame] | 4277 | |
| 4278 | if (!len) |
| 4279 | return 0; |
| 4280 | if (list_empty(&trans->r_itemq)) { |
| 4281 | /* we need to catch log corruptions here */ |
| 4282 | if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) { |
| 4283 | xfs_warn(log->l_mp, "%s: bad header magic number", |
| 4284 | __func__); |
| 4285 | ASSERT(0); |
| 4286 | return -EIO; |
| 4287 | } |
Brian Foster | 89cebc84 | 2015-07-29 11:51:10 +1000 | [diff] [blame] | 4288 | |
| 4289 | if (len > sizeof(struct xfs_trans_header)) { |
| 4290 | xfs_warn(log->l_mp, "%s: bad header length", __func__); |
| 4291 | ASSERT(0); |
| 4292 | return -EIO; |
| 4293 | } |
| 4294 | |
| 4295 | /* |
| 4296 | * The transaction header can be arbitrarily split across op |
| 4297 | * records. If we don't have the whole thing here, copy what we |
| 4298 | * do have and handle the rest in the next record. |
| 4299 | */ |
| 4300 | if (len == sizeof(struct xfs_trans_header)) |
Dave Chinner | 7656066 | 2014-09-29 09:45:42 +1000 | [diff] [blame] | 4301 | xlog_recover_add_item(&trans->r_itemq); |
| 4302 | memcpy(&trans->r_theader, dp, len); |
| 4303 | return 0; |
| 4304 | } |
| 4305 | |
| 4306 | ptr = kmem_alloc(len, KM_SLEEP); |
| 4307 | memcpy(ptr, dp, len); |
| 4308 | in_f = (xfs_inode_log_format_t *)ptr; |
| 4309 | |
| 4310 | /* take the tail entry */ |
| 4311 | item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list); |
| 4312 | if (item->ri_total != 0 && |
| 4313 | item->ri_total == item->ri_cnt) { |
| 4314 | /* tail item is in use, get a new one */ |
| 4315 | xlog_recover_add_item(&trans->r_itemq); |
| 4316 | item = list_entry(trans->r_itemq.prev, |
| 4317 | xlog_recover_item_t, ri_list); |
| 4318 | } |
| 4319 | |
| 4320 | if (item->ri_total == 0) { /* first region to be added */ |
| 4321 | if (in_f->ilf_size == 0 || |
| 4322 | in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) { |
| 4323 | xfs_warn(log->l_mp, |
| 4324 | "bad number of regions (%d) in inode log format", |
| 4325 | in_f->ilf_size); |
| 4326 | ASSERT(0); |
| 4327 | kmem_free(ptr); |
| 4328 | return -EIO; |
| 4329 | } |
| 4330 | |
| 4331 | item->ri_total = in_f->ilf_size; |
| 4332 | item->ri_buf = |
| 4333 | kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t), |
| 4334 | KM_SLEEP); |
| 4335 | } |
| 4336 | ASSERT(item->ri_total > item->ri_cnt); |
| 4337 | /* Description region is ri_buf[0] */ |
| 4338 | item->ri_buf[item->ri_cnt].i_addr = ptr; |
| 4339 | item->ri_buf[item->ri_cnt].i_len = len; |
| 4340 | item->ri_cnt++; |
| 4341 | trace_xfs_log_recover_item_add(log, trans, item, 0); |
| 4342 | return 0; |
| 4343 | } |
Dave Chinner | b818cca | 2014-09-29 09:45:54 +1000 | [diff] [blame] | 4344 | |
Dave Chinner | 7656066 | 2014-09-29 09:45:42 +1000 | [diff] [blame] | 4345 | /* |
| 4346 | * Free up any resources allocated by the transaction |
| 4347 | * |
| 4348 | * Remember that EFIs, EFDs, and IUNLINKs are handled later. |
| 4349 | */ |
| 4350 | STATIC void |
| 4351 | xlog_recover_free_trans( |
| 4352 | struct xlog_recover *trans) |
| 4353 | { |
| 4354 | xlog_recover_item_t *item, *n; |
| 4355 | int i; |
| 4356 | |
Brian Foster | 171192c | 2017-09-17 14:06:33 -0700 | [diff] [blame] | 4357 | hlist_del_init(&trans->r_list); |
| 4358 | |
Dave Chinner | 7656066 | 2014-09-29 09:45:42 +1000 | [diff] [blame] | 4359 | list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) { |
| 4360 | /* Free the regions in the item. */ |
| 4361 | list_del(&item->ri_list); |
| 4362 | for (i = 0; i < item->ri_cnt; i++) |
| 4363 | kmem_free(item->ri_buf[i].i_addr); |
| 4364 | /* Free the item itself */ |
| 4365 | kmem_free(item->ri_buf); |
| 4366 | kmem_free(item); |
| 4367 | } |
| 4368 | /* Free the transaction recover structure */ |
| 4369 | kmem_free(trans); |
| 4370 | } |
| 4371 | |
Dave Chinner | e9131e5 | 2014-09-29 09:45:18 +1000 | [diff] [blame] | 4372 | /* |
| 4373 | * On error or completion, trans is freed. |
| 4374 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4375 | STATIC int |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4376 | xlog_recovery_process_trans( |
| 4377 | struct xlog *log, |
| 4378 | struct xlog_recover *trans, |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 4379 | char *dp, |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4380 | unsigned int len, |
| 4381 | unsigned int flags, |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 4382 | int pass, |
| 4383 | struct list_head *buffer_list) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4384 | { |
Dave Chinner | e9131e5 | 2014-09-29 09:45:18 +1000 | [diff] [blame] | 4385 | int error = 0; |
| 4386 | bool freeit = false; |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4387 | |
| 4388 | /* mask off ophdr transaction container flags */ |
| 4389 | flags &= ~XLOG_END_TRANS; |
| 4390 | if (flags & XLOG_WAS_CONT_TRANS) |
| 4391 | flags &= ~XLOG_CONTINUE_TRANS; |
| 4392 | |
Dave Chinner | 88b863d | 2014-09-29 09:45:32 +1000 | [diff] [blame] | 4393 | /* |
| 4394 | * Callees must not free the trans structure. We'll decide if we need to |
| 4395 | * free it or not based on the operation being done and it's result. |
| 4396 | */ |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4397 | switch (flags) { |
| 4398 | /* expected flag values */ |
| 4399 | case 0: |
| 4400 | case XLOG_CONTINUE_TRANS: |
| 4401 | error = xlog_recover_add_to_trans(log, trans, dp, len); |
| 4402 | break; |
| 4403 | case XLOG_WAS_CONT_TRANS: |
| 4404 | error = xlog_recover_add_to_cont_trans(log, trans, dp, len); |
| 4405 | break; |
| 4406 | case XLOG_COMMIT_TRANS: |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 4407 | error = xlog_recover_commit_trans(log, trans, pass, |
| 4408 | buffer_list); |
Dave Chinner | 88b863d | 2014-09-29 09:45:32 +1000 | [diff] [blame] | 4409 | /* success or fail, we are now done with this transaction. */ |
| 4410 | freeit = true; |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4411 | break; |
| 4412 | |
| 4413 | /* unexpected flag values */ |
| 4414 | case XLOG_UNMOUNT_TRANS: |
Dave Chinner | e9131e5 | 2014-09-29 09:45:18 +1000 | [diff] [blame] | 4415 | /* just skip trans */ |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4416 | xfs_warn(log->l_mp, "%s: Unmount LR", __func__); |
Dave Chinner | e9131e5 | 2014-09-29 09:45:18 +1000 | [diff] [blame] | 4417 | freeit = true; |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4418 | break; |
| 4419 | case XLOG_START_TRANS: |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4420 | default: |
| 4421 | xfs_warn(log->l_mp, "%s: bad flag 0x%x", __func__, flags); |
| 4422 | ASSERT(0); |
Dave Chinner | e9131e5 | 2014-09-29 09:45:18 +1000 | [diff] [blame] | 4423 | error = -EIO; |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4424 | break; |
| 4425 | } |
Dave Chinner | e9131e5 | 2014-09-29 09:45:18 +1000 | [diff] [blame] | 4426 | if (error || freeit) |
| 4427 | xlog_recover_free_trans(trans); |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4428 | return error; |
| 4429 | } |
| 4430 | |
Dave Chinner | b818cca | 2014-09-29 09:45:54 +1000 | [diff] [blame] | 4431 | /* |
| 4432 | * Lookup the transaction recovery structure associated with the ID in the |
| 4433 | * current ophdr. If the transaction doesn't exist and the start flag is set in |
| 4434 | * the ophdr, then allocate a new transaction for future ID matches to find. |
| 4435 | * Either way, return what we found during the lookup - an existing transaction |
| 4436 | * or nothing. |
| 4437 | */ |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4438 | STATIC struct xlog_recover * |
| 4439 | xlog_recover_ophdr_to_trans( |
| 4440 | struct hlist_head rhash[], |
| 4441 | struct xlog_rec_header *rhead, |
| 4442 | struct xlog_op_header *ohead) |
| 4443 | { |
| 4444 | struct xlog_recover *trans; |
| 4445 | xlog_tid_t tid; |
| 4446 | struct hlist_head *rhp; |
| 4447 | |
| 4448 | tid = be32_to_cpu(ohead->oh_tid); |
| 4449 | rhp = &rhash[XLOG_RHASH(tid)]; |
Dave Chinner | b818cca | 2014-09-29 09:45:54 +1000 | [diff] [blame] | 4450 | hlist_for_each_entry(trans, rhp, r_list) { |
| 4451 | if (trans->r_log_tid == tid) |
| 4452 | return trans; |
| 4453 | } |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4454 | |
| 4455 | /* |
Dave Chinner | b818cca | 2014-09-29 09:45:54 +1000 | [diff] [blame] | 4456 | * skip over non-start transaction headers - we could be |
| 4457 | * processing slack space before the next transaction starts |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4458 | */ |
Dave Chinner | b818cca | 2014-09-29 09:45:54 +1000 | [diff] [blame] | 4459 | if (!(ohead->oh_flags & XLOG_START_TRANS)) |
| 4460 | return NULL; |
| 4461 | |
| 4462 | ASSERT(be32_to_cpu(ohead->oh_len) == 0); |
| 4463 | |
| 4464 | /* |
| 4465 | * This is a new transaction so allocate a new recovery container to |
| 4466 | * hold the recovery ops that will follow. |
| 4467 | */ |
| 4468 | trans = kmem_zalloc(sizeof(struct xlog_recover), KM_SLEEP); |
| 4469 | trans->r_log_tid = tid; |
| 4470 | trans->r_lsn = be64_to_cpu(rhead->h_lsn); |
| 4471 | INIT_LIST_HEAD(&trans->r_itemq); |
| 4472 | INIT_HLIST_NODE(&trans->r_list); |
| 4473 | hlist_add_head(&trans->r_list, rhp); |
| 4474 | |
| 4475 | /* |
| 4476 | * Nothing more to do for this ophdr. Items to be added to this new |
| 4477 | * transaction will be in subsequent ophdr containers. |
| 4478 | */ |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4479 | return NULL; |
| 4480 | } |
| 4481 | |
| 4482 | STATIC int |
| 4483 | xlog_recover_process_ophdr( |
| 4484 | struct xlog *log, |
| 4485 | struct hlist_head rhash[], |
| 4486 | struct xlog_rec_header *rhead, |
| 4487 | struct xlog_op_header *ohead, |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 4488 | char *dp, |
| 4489 | char *end, |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 4490 | int pass, |
| 4491 | struct list_head *buffer_list) |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4492 | { |
| 4493 | struct xlog_recover *trans; |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4494 | unsigned int len; |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 4495 | int error; |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4496 | |
| 4497 | /* Do we understand who wrote this op? */ |
| 4498 | if (ohead->oh_clientid != XFS_TRANSACTION && |
| 4499 | ohead->oh_clientid != XFS_LOG) { |
| 4500 | xfs_warn(log->l_mp, "%s: bad clientid 0x%x", |
| 4501 | __func__, ohead->oh_clientid); |
| 4502 | ASSERT(0); |
| 4503 | return -EIO; |
| 4504 | } |
| 4505 | |
| 4506 | /* |
| 4507 | * Check the ophdr contains all the data it is supposed to contain. |
| 4508 | */ |
| 4509 | len = be32_to_cpu(ohead->oh_len); |
| 4510 | if (dp + len > end) { |
| 4511 | xfs_warn(log->l_mp, "%s: bad length 0x%x", __func__, len); |
| 4512 | WARN_ON(1); |
| 4513 | return -EIO; |
| 4514 | } |
| 4515 | |
| 4516 | trans = xlog_recover_ophdr_to_trans(rhash, rhead, ohead); |
| 4517 | if (!trans) { |
| 4518 | /* nothing to do, so skip over this ophdr */ |
| 4519 | return 0; |
| 4520 | } |
| 4521 | |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 4522 | /* |
| 4523 | * The recovered buffer queue is drained only once we know that all |
| 4524 | * recovery items for the current LSN have been processed. This is |
| 4525 | * required because: |
| 4526 | * |
| 4527 | * - Buffer write submission updates the metadata LSN of the buffer. |
| 4528 | * - Log recovery skips items with a metadata LSN >= the current LSN of |
| 4529 | * the recovery item. |
| 4530 | * - Separate recovery items against the same metadata buffer can share |
| 4531 | * a current LSN. I.e., consider that the LSN of a recovery item is |
| 4532 | * defined as the starting LSN of the first record in which its |
| 4533 | * transaction appears, that a record can hold multiple transactions, |
| 4534 | * and/or that a transaction can span multiple records. |
| 4535 | * |
| 4536 | * In other words, we are allowed to submit a buffer from log recovery |
| 4537 | * once per current LSN. Otherwise, we may incorrectly skip recovery |
| 4538 | * items and cause corruption. |
| 4539 | * |
| 4540 | * We don't know up front whether buffers are updated multiple times per |
| 4541 | * LSN. Therefore, track the current LSN of each commit log record as it |
| 4542 | * is processed and drain the queue when it changes. Use commit records |
| 4543 | * because they are ordered correctly by the logging code. |
| 4544 | */ |
| 4545 | if (log->l_recovery_lsn != trans->r_lsn && |
| 4546 | ohead->oh_flags & XLOG_COMMIT_TRANS) { |
| 4547 | error = xfs_buf_delwri_submit(buffer_list); |
| 4548 | if (error) |
| 4549 | return error; |
| 4550 | log->l_recovery_lsn = trans->r_lsn; |
| 4551 | } |
| 4552 | |
Dave Chinner | e9131e5 | 2014-09-29 09:45:18 +1000 | [diff] [blame] | 4553 | return xlog_recovery_process_trans(log, trans, dp, len, |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 4554 | ohead->oh_flags, pass, buffer_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4555 | } |
| 4556 | |
| 4557 | /* |
| 4558 | * There are two valid states of the r_state field. 0 indicates that the |
| 4559 | * transaction structure is in a normal state. We have either seen the |
| 4560 | * start of the transaction or the last operation we added was not a partial |
| 4561 | * operation. If the last operation we added to the transaction was a |
| 4562 | * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS. |
| 4563 | * |
| 4564 | * NOTE: skip LRs with 0 data length. |
| 4565 | */ |
| 4566 | STATIC int |
| 4567 | xlog_recover_process_data( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 4568 | struct xlog *log, |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 4569 | struct hlist_head rhash[], |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 4570 | struct xlog_rec_header *rhead, |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 4571 | char *dp, |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 4572 | int pass, |
| 4573 | struct list_head *buffer_list) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4574 | { |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4575 | struct xlog_op_header *ohead; |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 4576 | char *end; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4577 | int num_logops; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4578 | int error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4579 | |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4580 | end = dp + be32_to_cpu(rhead->h_len); |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 4581 | num_logops = be32_to_cpu(rhead->h_num_logops); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4582 | |
| 4583 | /* check the log format matches our own - else we can't recover */ |
| 4584 | if (xlog_header_check_recover(log->l_mp, rhead)) |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 4585 | return -EIO; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4586 | |
Brian Foster | 5cd9cee | 2016-09-26 08:34:52 +1000 | [diff] [blame] | 4587 | trace_xfs_log_recover_record(log, rhead, pass); |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4588 | while ((dp < end) && num_logops) { |
| 4589 | |
| 4590 | ohead = (struct xlog_op_header *)dp; |
| 4591 | dp += sizeof(*ohead); |
| 4592 | ASSERT(dp <= end); |
| 4593 | |
| 4594 | /* errors will abort recovery */ |
| 4595 | error = xlog_recover_process_ophdr(log, rhash, rhead, ohead, |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 4596 | dp, end, pass, buffer_list); |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4597 | if (error) |
| 4598 | return error; |
| 4599 | |
Christoph Hellwig | 67fcb7b | 2007-10-12 10:58:59 +1000 | [diff] [blame] | 4600 | dp += be32_to_cpu(ohead->oh_len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4601 | num_logops--; |
| 4602 | } |
| 4603 | return 0; |
| 4604 | } |
| 4605 | |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4606 | /* Recover the EFI if necessary. */ |
David Chinner | 3c1e2bb | 2008-04-10 12:21:11 +1000 | [diff] [blame] | 4607 | STATIC int |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4608 | xlog_recover_process_efi( |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4609 | struct xfs_mount *mp, |
| 4610 | struct xfs_ail *ailp, |
| 4611 | struct xfs_log_item *lip) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4612 | { |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4613 | struct xfs_efi_log_item *efip; |
| 4614 | int error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4615 | |
| 4616 | /* |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4617 | * Skip EFIs that we've already processed. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4618 | */ |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4619 | efip = container_of(lip, struct xfs_efi_log_item, efi_item); |
| 4620 | if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) |
| 4621 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4622 | |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4623 | spin_unlock(&ailp->xa_lock); |
| 4624 | error = xfs_efi_recover(mp, efip); |
| 4625 | spin_lock(&ailp->xa_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4626 | |
David Chinner | fc6149d | 2008-04-10 12:21:53 +1000 | [diff] [blame] | 4627 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4628 | } |
| 4629 | |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4630 | /* Release the EFI since we're cancelling everything. */ |
| 4631 | STATIC void |
| 4632 | xlog_recover_cancel_efi( |
| 4633 | struct xfs_mount *mp, |
| 4634 | struct xfs_ail *ailp, |
| 4635 | struct xfs_log_item *lip) |
| 4636 | { |
| 4637 | struct xfs_efi_log_item *efip; |
| 4638 | |
| 4639 | efip = container_of(lip, struct xfs_efi_log_item, efi_item); |
| 4640 | |
| 4641 | spin_unlock(&ailp->xa_lock); |
| 4642 | xfs_efi_release(efip); |
| 4643 | spin_lock(&ailp->xa_lock); |
| 4644 | } |
| 4645 | |
Darrick J. Wong | 9e88b5d | 2016-08-03 12:09:48 +1000 | [diff] [blame] | 4646 | /* Recover the RUI if necessary. */ |
| 4647 | STATIC int |
| 4648 | xlog_recover_process_rui( |
| 4649 | struct xfs_mount *mp, |
| 4650 | struct xfs_ail *ailp, |
| 4651 | struct xfs_log_item *lip) |
| 4652 | { |
| 4653 | struct xfs_rui_log_item *ruip; |
| 4654 | int error; |
| 4655 | |
| 4656 | /* |
| 4657 | * Skip RUIs that we've already processed. |
| 4658 | */ |
| 4659 | ruip = container_of(lip, struct xfs_rui_log_item, rui_item); |
| 4660 | if (test_bit(XFS_RUI_RECOVERED, &ruip->rui_flags)) |
| 4661 | return 0; |
| 4662 | |
| 4663 | spin_unlock(&ailp->xa_lock); |
| 4664 | error = xfs_rui_recover(mp, ruip); |
| 4665 | spin_lock(&ailp->xa_lock); |
| 4666 | |
| 4667 | return error; |
| 4668 | } |
| 4669 | |
| 4670 | /* Release the RUI since we're cancelling everything. */ |
| 4671 | STATIC void |
| 4672 | xlog_recover_cancel_rui( |
| 4673 | struct xfs_mount *mp, |
| 4674 | struct xfs_ail *ailp, |
| 4675 | struct xfs_log_item *lip) |
| 4676 | { |
| 4677 | struct xfs_rui_log_item *ruip; |
| 4678 | |
| 4679 | ruip = container_of(lip, struct xfs_rui_log_item, rui_item); |
| 4680 | |
| 4681 | spin_unlock(&ailp->xa_lock); |
| 4682 | xfs_rui_release(ruip); |
| 4683 | spin_lock(&ailp->xa_lock); |
| 4684 | } |
| 4685 | |
Darrick J. Wong | f997ee2 | 2016-10-03 09:11:21 -0700 | [diff] [blame] | 4686 | /* Recover the CUI if necessary. */ |
| 4687 | STATIC int |
| 4688 | xlog_recover_process_cui( |
| 4689 | struct xfs_mount *mp, |
| 4690 | struct xfs_ail *ailp, |
| 4691 | struct xfs_log_item *lip) |
| 4692 | { |
| 4693 | struct xfs_cui_log_item *cuip; |
| 4694 | int error; |
| 4695 | |
| 4696 | /* |
| 4697 | * Skip CUIs that we've already processed. |
| 4698 | */ |
| 4699 | cuip = container_of(lip, struct xfs_cui_log_item, cui_item); |
| 4700 | if (test_bit(XFS_CUI_RECOVERED, &cuip->cui_flags)) |
| 4701 | return 0; |
| 4702 | |
| 4703 | spin_unlock(&ailp->xa_lock); |
| 4704 | error = xfs_cui_recover(mp, cuip); |
| 4705 | spin_lock(&ailp->xa_lock); |
| 4706 | |
| 4707 | return error; |
| 4708 | } |
| 4709 | |
| 4710 | /* Release the CUI since we're cancelling everything. */ |
| 4711 | STATIC void |
| 4712 | xlog_recover_cancel_cui( |
| 4713 | struct xfs_mount *mp, |
| 4714 | struct xfs_ail *ailp, |
| 4715 | struct xfs_log_item *lip) |
| 4716 | { |
| 4717 | struct xfs_cui_log_item *cuip; |
| 4718 | |
| 4719 | cuip = container_of(lip, struct xfs_cui_log_item, cui_item); |
| 4720 | |
| 4721 | spin_unlock(&ailp->xa_lock); |
| 4722 | xfs_cui_release(cuip); |
| 4723 | spin_lock(&ailp->xa_lock); |
| 4724 | } |
| 4725 | |
Darrick J. Wong | 77d61fe | 2016-10-03 09:11:26 -0700 | [diff] [blame] | 4726 | /* Recover the BUI if necessary. */ |
| 4727 | STATIC int |
| 4728 | xlog_recover_process_bui( |
| 4729 | struct xfs_mount *mp, |
| 4730 | struct xfs_ail *ailp, |
| 4731 | struct xfs_log_item *lip) |
| 4732 | { |
| 4733 | struct xfs_bui_log_item *buip; |
| 4734 | int error; |
| 4735 | |
| 4736 | /* |
| 4737 | * Skip BUIs that we've already processed. |
| 4738 | */ |
| 4739 | buip = container_of(lip, struct xfs_bui_log_item, bui_item); |
| 4740 | if (test_bit(XFS_BUI_RECOVERED, &buip->bui_flags)) |
| 4741 | return 0; |
| 4742 | |
| 4743 | spin_unlock(&ailp->xa_lock); |
| 4744 | error = xfs_bui_recover(mp, buip); |
| 4745 | spin_lock(&ailp->xa_lock); |
| 4746 | |
| 4747 | return error; |
| 4748 | } |
| 4749 | |
| 4750 | /* Release the BUI since we're cancelling everything. */ |
| 4751 | STATIC void |
| 4752 | xlog_recover_cancel_bui( |
| 4753 | struct xfs_mount *mp, |
| 4754 | struct xfs_ail *ailp, |
| 4755 | struct xfs_log_item *lip) |
| 4756 | { |
| 4757 | struct xfs_bui_log_item *buip; |
| 4758 | |
| 4759 | buip = container_of(lip, struct xfs_bui_log_item, bui_item); |
| 4760 | |
| 4761 | spin_unlock(&ailp->xa_lock); |
| 4762 | xfs_bui_release(buip); |
| 4763 | spin_lock(&ailp->xa_lock); |
| 4764 | } |
| 4765 | |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4766 | /* Is this log item a deferred action intent? */ |
| 4767 | static inline bool xlog_item_is_intent(struct xfs_log_item *lip) |
| 4768 | { |
| 4769 | switch (lip->li_type) { |
| 4770 | case XFS_LI_EFI: |
Darrick J. Wong | 9e88b5d | 2016-08-03 12:09:48 +1000 | [diff] [blame] | 4771 | case XFS_LI_RUI: |
Darrick J. Wong | f997ee2 | 2016-10-03 09:11:21 -0700 | [diff] [blame] | 4772 | case XFS_LI_CUI: |
Darrick J. Wong | 77d61fe | 2016-10-03 09:11:26 -0700 | [diff] [blame] | 4773 | case XFS_LI_BUI: |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4774 | return true; |
| 4775 | default: |
| 4776 | return false; |
| 4777 | } |
| 4778 | } |
| 4779 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4780 | /* |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4781 | * When this is called, all of the log intent items which did not have |
| 4782 | * corresponding log done items should be in the AIL. What we do now |
| 4783 | * is update the data structures associated with each one. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4784 | * |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4785 | * Since we process the log intent items in normal transactions, they |
| 4786 | * will be removed at some point after the commit. This prevents us |
| 4787 | * from just walking down the list processing each one. We'll use a |
| 4788 | * flag in the intent item to skip those that we've already processed |
| 4789 | * and use the AIL iteration mechanism's generation count to try to |
| 4790 | * speed this up at least a bit. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4791 | * |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4792 | * When we start, we know that the intents are the only things in the |
| 4793 | * AIL. As we process them, however, other items are added to the |
| 4794 | * AIL. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4795 | */ |
David Chinner | 3c1e2bb | 2008-04-10 12:21:11 +1000 | [diff] [blame] | 4796 | STATIC int |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4797 | xlog_recover_process_intents( |
Brian Foster | f0b2efa | 2015-08-19 09:58:36 +1000 | [diff] [blame] | 4798 | struct xlog *log) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4799 | { |
Brian Foster | f0b2efa | 2015-08-19 09:58:36 +1000 | [diff] [blame] | 4800 | struct xfs_log_item *lip; |
David Chinner | 3c1e2bb | 2008-04-10 12:21:11 +1000 | [diff] [blame] | 4801 | int error = 0; |
David Chinner | 27d8d5f | 2008-10-30 17:38:39 +1100 | [diff] [blame] | 4802 | struct xfs_ail_cursor cur; |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 4803 | struct xfs_ail *ailp; |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4804 | xfs_lsn_t last_lsn; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4805 | |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 4806 | ailp = log->l_ailp; |
| 4807 | spin_lock(&ailp->xa_lock); |
| 4808 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4809 | last_lsn = xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4810 | while (lip != NULL) { |
| 4811 | /* |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4812 | * We're done when we see something other than an intent. |
| 4813 | * There should be no intents left in the AIL now. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4814 | */ |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4815 | if (!xlog_item_is_intent(lip)) { |
David Chinner | 27d8d5f | 2008-10-30 17:38:39 +1100 | [diff] [blame] | 4816 | #ifdef DEBUG |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 4817 | for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur)) |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4818 | ASSERT(!xlog_item_is_intent(lip)); |
David Chinner | 27d8d5f | 2008-10-30 17:38:39 +1100 | [diff] [blame] | 4819 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4820 | break; |
| 4821 | } |
| 4822 | |
| 4823 | /* |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4824 | * We should never see a redo item with a LSN higher than |
| 4825 | * the last transaction we found in the log at the start |
| 4826 | * of recovery. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4827 | */ |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4828 | ASSERT(XFS_LSN_CMP(last_lsn, lip->li_lsn) >= 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4829 | |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4830 | switch (lip->li_type) { |
| 4831 | case XFS_LI_EFI: |
| 4832 | error = xlog_recover_process_efi(log->l_mp, ailp, lip); |
| 4833 | break; |
Darrick J. Wong | 9e88b5d | 2016-08-03 12:09:48 +1000 | [diff] [blame] | 4834 | case XFS_LI_RUI: |
| 4835 | error = xlog_recover_process_rui(log->l_mp, ailp, lip); |
| 4836 | break; |
Darrick J. Wong | f997ee2 | 2016-10-03 09:11:21 -0700 | [diff] [blame] | 4837 | case XFS_LI_CUI: |
| 4838 | error = xlog_recover_process_cui(log->l_mp, ailp, lip); |
| 4839 | break; |
Darrick J. Wong | 77d61fe | 2016-10-03 09:11:26 -0700 | [diff] [blame] | 4840 | case XFS_LI_BUI: |
| 4841 | error = xlog_recover_process_bui(log->l_mp, ailp, lip); |
| 4842 | break; |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4843 | } |
David Chinner | 27d8d5f | 2008-10-30 17:38:39 +1100 | [diff] [blame] | 4844 | if (error) |
| 4845 | goto out; |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 4846 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4847 | } |
David Chinner | 27d8d5f | 2008-10-30 17:38:39 +1100 | [diff] [blame] | 4848 | out: |
Eric Sandeen | e4a1e29 | 2014-04-14 19:06:05 +1000 | [diff] [blame] | 4849 | xfs_trans_ail_cursor_done(&cur); |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 4850 | spin_unlock(&ailp->xa_lock); |
David Chinner | 3c1e2bb | 2008-04-10 12:21:11 +1000 | [diff] [blame] | 4851 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4852 | } |
| 4853 | |
| 4854 | /* |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4855 | * A cancel occurs when the mount has failed and we're bailing out. |
| 4856 | * Release all pending log intent items so they don't pin the AIL. |
Brian Foster | f0b2efa | 2015-08-19 09:58:36 +1000 | [diff] [blame] | 4857 | */ |
| 4858 | STATIC int |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4859 | xlog_recover_cancel_intents( |
Brian Foster | f0b2efa | 2015-08-19 09:58:36 +1000 | [diff] [blame] | 4860 | struct xlog *log) |
| 4861 | { |
| 4862 | struct xfs_log_item *lip; |
Brian Foster | f0b2efa | 2015-08-19 09:58:36 +1000 | [diff] [blame] | 4863 | int error = 0; |
| 4864 | struct xfs_ail_cursor cur; |
| 4865 | struct xfs_ail *ailp; |
| 4866 | |
| 4867 | ailp = log->l_ailp; |
| 4868 | spin_lock(&ailp->xa_lock); |
| 4869 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); |
| 4870 | while (lip != NULL) { |
| 4871 | /* |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4872 | * We're done when we see something other than an intent. |
| 4873 | * There should be no intents left in the AIL now. |
Brian Foster | f0b2efa | 2015-08-19 09:58:36 +1000 | [diff] [blame] | 4874 | */ |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4875 | if (!xlog_item_is_intent(lip)) { |
Brian Foster | f0b2efa | 2015-08-19 09:58:36 +1000 | [diff] [blame] | 4876 | #ifdef DEBUG |
| 4877 | for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur)) |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4878 | ASSERT(!xlog_item_is_intent(lip)); |
Brian Foster | f0b2efa | 2015-08-19 09:58:36 +1000 | [diff] [blame] | 4879 | #endif |
| 4880 | break; |
| 4881 | } |
| 4882 | |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4883 | switch (lip->li_type) { |
| 4884 | case XFS_LI_EFI: |
| 4885 | xlog_recover_cancel_efi(log->l_mp, ailp, lip); |
| 4886 | break; |
Darrick J. Wong | 9e88b5d | 2016-08-03 12:09:48 +1000 | [diff] [blame] | 4887 | case XFS_LI_RUI: |
| 4888 | xlog_recover_cancel_rui(log->l_mp, ailp, lip); |
| 4889 | break; |
Darrick J. Wong | f997ee2 | 2016-10-03 09:11:21 -0700 | [diff] [blame] | 4890 | case XFS_LI_CUI: |
| 4891 | xlog_recover_cancel_cui(log->l_mp, ailp, lip); |
| 4892 | break; |
Darrick J. Wong | 77d61fe | 2016-10-03 09:11:26 -0700 | [diff] [blame] | 4893 | case XFS_LI_BUI: |
| 4894 | xlog_recover_cancel_bui(log->l_mp, ailp, lip); |
| 4895 | break; |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4896 | } |
Brian Foster | f0b2efa | 2015-08-19 09:58:36 +1000 | [diff] [blame] | 4897 | |
| 4898 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
| 4899 | } |
| 4900 | |
| 4901 | xfs_trans_ail_cursor_done(&cur); |
| 4902 | spin_unlock(&ailp->xa_lock); |
| 4903 | return error; |
| 4904 | } |
| 4905 | |
| 4906 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4907 | * This routine performs a transaction to null out a bad inode pointer |
| 4908 | * in an agi unlinked inode hash bucket. |
| 4909 | */ |
| 4910 | STATIC void |
| 4911 | xlog_recover_clear_agi_bucket( |
| 4912 | xfs_mount_t *mp, |
| 4913 | xfs_agnumber_t agno, |
| 4914 | int bucket) |
| 4915 | { |
| 4916 | xfs_trans_t *tp; |
| 4917 | xfs_agi_t *agi; |
| 4918 | xfs_buf_t *agibp; |
| 4919 | int offset; |
| 4920 | int error; |
| 4921 | |
Christoph Hellwig | 253f491 | 2016-04-06 09:19:55 +1000 | [diff] [blame] | 4922 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_clearagi, 0, 0, 0, &tp); |
David Chinner | e5720ee | 2008-04-10 12:21:18 +1000 | [diff] [blame] | 4923 | if (error) |
Christoph Hellwig | 253f491 | 2016-04-06 09:19:55 +1000 | [diff] [blame] | 4924 | goto out_error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4925 | |
Christoph Hellwig | 5e1be0f | 2008-11-28 14:23:37 +1100 | [diff] [blame] | 4926 | error = xfs_read_agi(mp, tp, agno, &agibp); |
| 4927 | if (error) |
David Chinner | e5720ee | 2008-04-10 12:21:18 +1000 | [diff] [blame] | 4928 | goto out_abort; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4929 | |
Christoph Hellwig | 5e1be0f | 2008-11-28 14:23:37 +1100 | [diff] [blame] | 4930 | agi = XFS_BUF_TO_AGI(agibp); |
Christoph Hellwig | 16259e7 | 2005-11-02 15:11:25 +1100 | [diff] [blame] | 4931 | agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4932 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
| 4933 | (sizeof(xfs_agino_t) * bucket); |
| 4934 | xfs_trans_log_buf(tp, agibp, offset, |
| 4935 | (offset + sizeof(xfs_agino_t) - 1)); |
| 4936 | |
Christoph Hellwig | 7039331 | 2015-06-04 13:48:08 +1000 | [diff] [blame] | 4937 | error = xfs_trans_commit(tp); |
David Chinner | e5720ee | 2008-04-10 12:21:18 +1000 | [diff] [blame] | 4938 | if (error) |
| 4939 | goto out_error; |
| 4940 | return; |
| 4941 | |
| 4942 | out_abort: |
Christoph Hellwig | 4906e21 | 2015-06-04 13:47:56 +1000 | [diff] [blame] | 4943 | xfs_trans_cancel(tp); |
David Chinner | e5720ee | 2008-04-10 12:21:18 +1000 | [diff] [blame] | 4944 | out_error: |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 4945 | xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno); |
David Chinner | e5720ee | 2008-04-10 12:21:18 +1000 | [diff] [blame] | 4946 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4947 | } |
| 4948 | |
Christoph Hellwig | 23fac50 | 2008-11-28 14:23:40 +1100 | [diff] [blame] | 4949 | STATIC xfs_agino_t |
| 4950 | xlog_recover_process_one_iunlink( |
| 4951 | struct xfs_mount *mp, |
| 4952 | xfs_agnumber_t agno, |
| 4953 | xfs_agino_t agino, |
| 4954 | int bucket) |
| 4955 | { |
| 4956 | struct xfs_buf *ibp; |
| 4957 | struct xfs_dinode *dip; |
| 4958 | struct xfs_inode *ip; |
| 4959 | xfs_ino_t ino; |
| 4960 | int error; |
| 4961 | |
| 4962 | ino = XFS_AGINO_TO_INO(mp, agno, agino); |
Dave Chinner | 7b6259e | 2010-06-24 11:35:17 +1000 | [diff] [blame] | 4963 | error = xfs_iget(mp, NULL, ino, 0, 0, &ip); |
Christoph Hellwig | 23fac50 | 2008-11-28 14:23:40 +1100 | [diff] [blame] | 4964 | if (error) |
| 4965 | goto fail; |
| 4966 | |
| 4967 | /* |
| 4968 | * Get the on disk inode to find the next inode in the bucket. |
| 4969 | */ |
Christoph Hellwig | 475ee41 | 2012-07-03 12:21:22 -0400 | [diff] [blame] | 4970 | error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &ibp, 0, 0); |
Christoph Hellwig | 23fac50 | 2008-11-28 14:23:40 +1100 | [diff] [blame] | 4971 | if (error) |
Christoph Hellwig | 0e44667 | 2008-11-28 14:23:42 +1100 | [diff] [blame] | 4972 | goto fail_iput; |
Christoph Hellwig | 23fac50 | 2008-11-28 14:23:40 +1100 | [diff] [blame] | 4973 | |
Darrick J. Wong | 17c12bc | 2016-10-03 09:11:29 -0700 | [diff] [blame] | 4974 | xfs_iflags_clear(ip, XFS_IRECOVERY); |
Dave Chinner | 54d7b5c | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 4975 | ASSERT(VFS_I(ip)->i_nlink == 0); |
Dave Chinner | c19b3b05 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 4976 | ASSERT(VFS_I(ip)->i_mode != 0); |
Christoph Hellwig | 23fac50 | 2008-11-28 14:23:40 +1100 | [diff] [blame] | 4977 | |
| 4978 | /* setup for the next pass */ |
| 4979 | agino = be32_to_cpu(dip->di_next_unlinked); |
| 4980 | xfs_buf_relse(ibp); |
| 4981 | |
| 4982 | /* |
| 4983 | * Prevent any DMAPI event from being sent when the reference on |
| 4984 | * the inode is dropped. |
| 4985 | */ |
| 4986 | ip->i_d.di_dmevmask = 0; |
| 4987 | |
Christoph Hellwig | 0e44667 | 2008-11-28 14:23:42 +1100 | [diff] [blame] | 4988 | IRELE(ip); |
Christoph Hellwig | 23fac50 | 2008-11-28 14:23:40 +1100 | [diff] [blame] | 4989 | return agino; |
| 4990 | |
Christoph Hellwig | 0e44667 | 2008-11-28 14:23:42 +1100 | [diff] [blame] | 4991 | fail_iput: |
| 4992 | IRELE(ip); |
Christoph Hellwig | 23fac50 | 2008-11-28 14:23:40 +1100 | [diff] [blame] | 4993 | fail: |
| 4994 | /* |
| 4995 | * We can't read in the inode this bucket points to, or this inode |
| 4996 | * is messed up. Just ditch this bucket of inodes. We will lose |
| 4997 | * some inodes and space, but at least we won't hang. |
| 4998 | * |
| 4999 | * Call xlog_recover_clear_agi_bucket() to perform a transaction to |
| 5000 | * clear the inode pointer in the bucket. |
| 5001 | */ |
| 5002 | xlog_recover_clear_agi_bucket(mp, agno, bucket); |
| 5003 | return NULLAGINO; |
| 5004 | } |
| 5005 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5006 | /* |
| 5007 | * xlog_iunlink_recover |
| 5008 | * |
| 5009 | * This is called during recovery to process any inodes which |
| 5010 | * we unlinked but not freed when the system crashed. These |
| 5011 | * inodes will be on the lists in the AGI blocks. What we do |
| 5012 | * here is scan all the AGIs and fully truncate and free any |
| 5013 | * inodes found on the lists. Each inode is removed from the |
| 5014 | * lists when it has been fully truncated and is freed. The |
| 5015 | * freeing of the inode and its removal from the list must be |
| 5016 | * atomic. |
| 5017 | */ |
Eric Sandeen | d96f8f8 | 2009-07-02 00:09:33 -0500 | [diff] [blame] | 5018 | STATIC void |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5019 | xlog_recover_process_iunlinks( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 5020 | struct xlog *log) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5021 | { |
| 5022 | xfs_mount_t *mp; |
| 5023 | xfs_agnumber_t agno; |
| 5024 | xfs_agi_t *agi; |
| 5025 | xfs_buf_t *agibp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5026 | xfs_agino_t agino; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5027 | int bucket; |
| 5028 | int error; |
| 5029 | uint mp_dmevmask; |
| 5030 | |
| 5031 | mp = log->l_mp; |
| 5032 | |
| 5033 | /* |
| 5034 | * Prevent any DMAPI event from being sent while in this function. |
| 5035 | */ |
| 5036 | mp_dmevmask = mp->m_dmevmask; |
| 5037 | mp->m_dmevmask = 0; |
| 5038 | |
| 5039 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { |
| 5040 | /* |
| 5041 | * Find the agi for this ag. |
| 5042 | */ |
Christoph Hellwig | 5e1be0f | 2008-11-28 14:23:37 +1100 | [diff] [blame] | 5043 | error = xfs_read_agi(mp, NULL, agno, &agibp); |
| 5044 | if (error) { |
| 5045 | /* |
| 5046 | * AGI is b0rked. Don't process it. |
| 5047 | * |
| 5048 | * We should probably mark the filesystem as corrupt |
| 5049 | * after we've recovered all the ag's we can.... |
| 5050 | */ |
| 5051 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5052 | } |
Jan Kara | d97d32e | 2012-03-15 09:34:02 +0000 | [diff] [blame] | 5053 | /* |
| 5054 | * Unlock the buffer so that it can be acquired in the normal |
| 5055 | * course of the transaction to truncate and free each inode. |
| 5056 | * Because we are not racing with anyone else here for the AGI |
| 5057 | * buffer, we don't even need to hold it locked to read the |
| 5058 | * initial unlinked bucket entries out of the buffer. We keep |
| 5059 | * buffer reference though, so that it stays pinned in memory |
| 5060 | * while we need the buffer. |
| 5061 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5062 | agi = XFS_BUF_TO_AGI(agibp); |
Jan Kara | d97d32e | 2012-03-15 09:34:02 +0000 | [diff] [blame] | 5063 | xfs_buf_unlock(agibp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5064 | |
| 5065 | for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) { |
Christoph Hellwig | 16259e7 | 2005-11-02 15:11:25 +1100 | [diff] [blame] | 5066 | agino = be32_to_cpu(agi->agi_unlinked[bucket]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5067 | while (agino != NULLAGINO) { |
Christoph Hellwig | 23fac50 | 2008-11-28 14:23:40 +1100 | [diff] [blame] | 5068 | agino = xlog_recover_process_one_iunlink(mp, |
| 5069 | agno, agino, bucket); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5070 | } |
| 5071 | } |
Jan Kara | d97d32e | 2012-03-15 09:34:02 +0000 | [diff] [blame] | 5072 | xfs_buf_rele(agibp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5073 | } |
| 5074 | |
| 5075 | mp->m_dmevmask = mp_dmevmask; |
| 5076 | } |
| 5077 | |
Christoph Hellwig | 0e446be | 2012-11-12 22:54:24 +1100 | [diff] [blame] | 5078 | STATIC int |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5079 | xlog_unpack_data( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 5080 | struct xlog_rec_header *rhead, |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 5081 | char *dp, |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 5082 | struct xlog *log) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5083 | { |
| 5084 | int i, j, k; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5085 | |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 5086 | for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) && |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5087 | i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) { |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 5088 | *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5089 | dp += BBSIZE; |
| 5090 | } |
| 5091 | |
Eric Sandeen | 6211870 | 2008-03-06 13:44:28 +1100 | [diff] [blame] | 5092 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
Christoph Hellwig | b28708d | 2008-11-28 14:23:38 +1100 | [diff] [blame] | 5093 | xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead; |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 5094 | for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5095 | j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); |
| 5096 | k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 5097 | *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5098 | dp += BBSIZE; |
| 5099 | } |
| 5100 | } |
Christoph Hellwig | 0e446be | 2012-11-12 22:54:24 +1100 | [diff] [blame] | 5101 | |
| 5102 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5103 | } |
| 5104 | |
Brian Foster | 9d94901 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5105 | /* |
Brian Foster | b94fb2d | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5106 | * CRC check, unpack and process a log record. |
Brian Foster | 9d94901 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5107 | */ |
| 5108 | STATIC int |
| 5109 | xlog_recover_process( |
| 5110 | struct xlog *log, |
| 5111 | struct hlist_head rhash[], |
| 5112 | struct xlog_rec_header *rhead, |
| 5113 | char *dp, |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 5114 | int pass, |
| 5115 | struct list_head *buffer_list) |
Brian Foster | 9d94901 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5116 | { |
| 5117 | int error; |
Brian Foster | b94fb2d | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5118 | __le32 crc; |
| 5119 | |
Brian Foster | b94fb2d | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5120 | crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len)); |
Brian Foster | 6528250 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5121 | |
| 5122 | /* |
| 5123 | * Nothing else to do if this is a CRC verification pass. Just return |
| 5124 | * if this a record with a non-zero crc. Unfortunately, mkfs always |
| 5125 | * sets h_crc to 0 so we must consider this valid even on v5 supers. |
| 5126 | * Otherwise, return EFSBADCRC on failure so the callers up the stack |
| 5127 | * know precisely what failed. |
| 5128 | */ |
| 5129 | if (pass == XLOG_RECOVER_CRCPASS) { |
Darrick J. Wong | 8e0bd49 | 2016-02-08 11:03:58 +1100 | [diff] [blame] | 5130 | if (rhead->h_crc && crc != rhead->h_crc) |
Brian Foster | 6528250 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5131 | return -EFSBADCRC; |
| 5132 | return 0; |
| 5133 | } |
| 5134 | |
| 5135 | /* |
| 5136 | * We're in the normal recovery path. Issue a warning if and only if the |
| 5137 | * CRC in the header is non-zero. This is an advisory warning and the |
| 5138 | * zero CRC check prevents warnings from being emitted when upgrading |
| 5139 | * the kernel from one that does not add CRCs by default. |
| 5140 | */ |
Darrick J. Wong | 8e0bd49 | 2016-02-08 11:03:58 +1100 | [diff] [blame] | 5141 | if (crc != rhead->h_crc) { |
Brian Foster | b94fb2d | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5142 | if (rhead->h_crc || xfs_sb_version_hascrc(&log->l_mp->m_sb)) { |
| 5143 | xfs_alert(log->l_mp, |
| 5144 | "log record CRC mismatch: found 0x%x, expected 0x%x.", |
| 5145 | le32_to_cpu(rhead->h_crc), |
| 5146 | le32_to_cpu(crc)); |
| 5147 | xfs_hex_dump(dp, 32); |
| 5148 | } |
| 5149 | |
| 5150 | /* |
| 5151 | * If the filesystem is CRC enabled, this mismatch becomes a |
| 5152 | * fatal log corruption failure. |
| 5153 | */ |
| 5154 | if (xfs_sb_version_hascrc(&log->l_mp->m_sb)) |
| 5155 | return -EFSCORRUPTED; |
| 5156 | } |
Brian Foster | 9d94901 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5157 | |
| 5158 | error = xlog_unpack_data(rhead, dp, log); |
| 5159 | if (error) |
| 5160 | return error; |
| 5161 | |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 5162 | return xlog_recover_process_data(log, rhash, rhead, dp, pass, |
| 5163 | buffer_list); |
Brian Foster | 9d94901 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5164 | } |
| 5165 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5166 | STATIC int |
| 5167 | xlog_valid_rec_header( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 5168 | struct xlog *log, |
| 5169 | struct xlog_rec_header *rhead, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5170 | xfs_daddr_t blkno) |
| 5171 | { |
| 5172 | int hlen; |
| 5173 | |
Christoph Hellwig | 69ef921 | 2011-07-08 14:36:05 +0200 | [diff] [blame] | 5174 | if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5175 | XFS_ERROR_REPORT("xlog_valid_rec_header(1)", |
| 5176 | XFS_ERRLEVEL_LOW, log->l_mp); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 5177 | return -EFSCORRUPTED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5178 | } |
| 5179 | if (unlikely( |
| 5180 | (!rhead->h_version || |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 5181 | (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 5182 | xfs_warn(log->l_mp, "%s: unrecognised log version (%d).", |
Harvey Harrison | 34a622b | 2008-04-10 12:19:21 +1000 | [diff] [blame] | 5183 | __func__, be32_to_cpu(rhead->h_version)); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 5184 | return -EIO; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5185 | } |
| 5186 | |
| 5187 | /* LR body must have data or it wouldn't have been written */ |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 5188 | hlen = be32_to_cpu(rhead->h_len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5189 | if (unlikely( hlen <= 0 || hlen > INT_MAX )) { |
| 5190 | XFS_ERROR_REPORT("xlog_valid_rec_header(2)", |
| 5191 | XFS_ERRLEVEL_LOW, log->l_mp); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 5192 | return -EFSCORRUPTED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5193 | } |
| 5194 | if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) { |
| 5195 | XFS_ERROR_REPORT("xlog_valid_rec_header(3)", |
| 5196 | XFS_ERRLEVEL_LOW, log->l_mp); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 5197 | return -EFSCORRUPTED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5198 | } |
| 5199 | return 0; |
| 5200 | } |
| 5201 | |
| 5202 | /* |
| 5203 | * Read the log from tail to head and process the log records found. |
| 5204 | * Handle the two cases where the tail and head are in the same cycle |
| 5205 | * and where the active portion of the log wraps around the end of |
| 5206 | * the physical log separately. The pass parameter is passed through |
| 5207 | * to the routines called to process the data and is not looked at |
| 5208 | * here. |
| 5209 | */ |
| 5210 | STATIC int |
| 5211 | xlog_do_recovery_pass( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 5212 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5213 | xfs_daddr_t head_blk, |
| 5214 | xfs_daddr_t tail_blk, |
Brian Foster | d7f3769 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5215 | int pass, |
| 5216 | xfs_daddr_t *first_bad) /* out: first bad log rec */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5217 | { |
| 5218 | xlog_rec_header_t *rhead; |
| 5219 | xfs_daddr_t blk_no; |
Brian Foster | d7f3769 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5220 | xfs_daddr_t rhead_blk; |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 5221 | char *offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5222 | xfs_buf_t *hbp, *dbp; |
Brian Foster | a70f9fe | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5223 | int error = 0, h_size, h_len; |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 5224 | int error2 = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5225 | int bblks, split_bblks; |
| 5226 | int hblks, split_hblks, wrapped_hblks; |
Brian Foster | 171192c | 2017-09-17 14:06:33 -0700 | [diff] [blame] | 5227 | int i; |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 5228 | struct hlist_head rhash[XLOG_RHASH_SIZE]; |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 5229 | LIST_HEAD (buffer_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5230 | |
| 5231 | ASSERT(head_blk != tail_blk); |
Brian Foster | d7f3769 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5232 | rhead_blk = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5233 | |
Brian Foster | 171192c | 2017-09-17 14:06:33 -0700 | [diff] [blame] | 5234 | for (i = 0; i < XLOG_RHASH_SIZE; i++) |
| 5235 | INIT_HLIST_HEAD(&rhash[i]); |
| 5236 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5237 | /* |
| 5238 | * Read the header of the tail block and get the iclog buffer size from |
| 5239 | * h_size. Use this to tell how many sectors make up the log header. |
| 5240 | */ |
Eric Sandeen | 6211870 | 2008-03-06 13:44:28 +1100 | [diff] [blame] | 5241 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5242 | /* |
| 5243 | * When using variable length iclogs, read first sector of |
| 5244 | * iclog header and extract the header size from it. Get a |
| 5245 | * new hbp that is the correct size. |
| 5246 | */ |
| 5247 | hbp = xlog_get_bp(log, 1); |
| 5248 | if (!hbp) |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 5249 | return -ENOMEM; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 5250 | |
| 5251 | error = xlog_bread(log, tail_blk, 1, hbp, &offset); |
| 5252 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5253 | goto bread_err1; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 5254 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5255 | rhead = (xlog_rec_header_t *)offset; |
| 5256 | error = xlog_valid_rec_header(log, rhead, tail_blk); |
| 5257 | if (error) |
| 5258 | goto bread_err1; |
Brian Foster | a70f9fe | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5259 | |
| 5260 | /* |
| 5261 | * xfsprogs has a bug where record length is based on lsunit but |
| 5262 | * h_size (iclog size) is hardcoded to 32k. Now that we |
| 5263 | * unconditionally CRC verify the unmount record, this means the |
| 5264 | * log buffer can be too small for the record and cause an |
| 5265 | * overrun. |
| 5266 | * |
| 5267 | * Detect this condition here. Use lsunit for the buffer size as |
| 5268 | * long as this looks like the mkfs case. Otherwise, return an |
| 5269 | * error to avoid a buffer overrun. |
| 5270 | */ |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 5271 | h_size = be32_to_cpu(rhead->h_size); |
Brian Foster | a70f9fe | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5272 | h_len = be32_to_cpu(rhead->h_len); |
| 5273 | if (h_len > h_size) { |
| 5274 | if (h_len <= log->l_mp->m_logbsize && |
| 5275 | be32_to_cpu(rhead->h_num_logops) == 1) { |
| 5276 | xfs_warn(log->l_mp, |
| 5277 | "invalid iclog size (%d bytes), using lsunit (%d bytes)", |
| 5278 | h_size, log->l_mp->m_logbsize); |
| 5279 | h_size = log->l_mp->m_logbsize; |
| 5280 | } else |
| 5281 | return -EFSCORRUPTED; |
| 5282 | } |
| 5283 | |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 5284 | if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) && |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5285 | (h_size > XLOG_HEADER_CYCLE_SIZE)) { |
| 5286 | hblks = h_size / XLOG_HEADER_CYCLE_SIZE; |
| 5287 | if (h_size % XLOG_HEADER_CYCLE_SIZE) |
| 5288 | hblks++; |
| 5289 | xlog_put_bp(hbp); |
| 5290 | hbp = xlog_get_bp(log, hblks); |
| 5291 | } else { |
| 5292 | hblks = 1; |
| 5293 | } |
| 5294 | } else { |
Alex Elder | 69ce58f | 2010-04-20 17:09:59 +1000 | [diff] [blame] | 5295 | ASSERT(log->l_sectBBsize == 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5296 | hblks = 1; |
| 5297 | hbp = xlog_get_bp(log, 1); |
| 5298 | h_size = XLOG_BIG_RECORD_BSIZE; |
| 5299 | } |
| 5300 | |
| 5301 | if (!hbp) |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 5302 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5303 | dbp = xlog_get_bp(log, BTOBB(h_size)); |
| 5304 | if (!dbp) { |
| 5305 | xlog_put_bp(hbp); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 5306 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5307 | } |
| 5308 | |
| 5309 | memset(rhash, 0, sizeof(rhash)); |
Brian Foster | d7f3769 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5310 | blk_no = rhead_blk = tail_blk; |
Eric Sandeen | 970fd3f | 2014-09-09 11:57:29 +1000 | [diff] [blame] | 5311 | if (tail_blk > head_blk) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5312 | /* |
| 5313 | * Perform recovery around the end of the physical log. |
| 5314 | * When the head is not on the same cycle number as the tail, |
Eric Sandeen | 970fd3f | 2014-09-09 11:57:29 +1000 | [diff] [blame] | 5315 | * we can't do a sequential recovery. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5316 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5317 | while (blk_no < log->l_logBBsize) { |
| 5318 | /* |
| 5319 | * Check for header wrapping around physical end-of-log |
| 5320 | */ |
Chandra Seetharaman | 6292604 | 2011-07-22 23:40:15 +0000 | [diff] [blame] | 5321 | offset = hbp->b_addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5322 | split_hblks = 0; |
| 5323 | wrapped_hblks = 0; |
| 5324 | if (blk_no + hblks <= log->l_logBBsize) { |
| 5325 | /* Read header in one read */ |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 5326 | error = xlog_bread(log, blk_no, hblks, hbp, |
| 5327 | &offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5328 | if (error) |
| 5329 | goto bread_err2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5330 | } else { |
| 5331 | /* This LR is split across physical log end */ |
| 5332 | if (blk_no != log->l_logBBsize) { |
| 5333 | /* some data before physical log end */ |
| 5334 | ASSERT(blk_no <= INT_MAX); |
| 5335 | split_hblks = log->l_logBBsize - (int)blk_no; |
| 5336 | ASSERT(split_hblks > 0); |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 5337 | error = xlog_bread(log, blk_no, |
| 5338 | split_hblks, hbp, |
| 5339 | &offset); |
| 5340 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5341 | goto bread_err2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5342 | } |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 5343 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5344 | /* |
| 5345 | * Note: this black magic still works with |
| 5346 | * large sector sizes (non-512) only because: |
| 5347 | * - we increased the buffer size originally |
| 5348 | * by 1 sector giving us enough extra space |
| 5349 | * for the second read; |
| 5350 | * - the log start is guaranteed to be sector |
| 5351 | * aligned; |
| 5352 | * - we read the log end (LR header start) |
| 5353 | * _first_, then the log start (LR header end) |
| 5354 | * - order is important. |
| 5355 | */ |
David Chinner | 234f56a | 2008-04-10 12:24:24 +1000 | [diff] [blame] | 5356 | wrapped_hblks = hblks - split_hblks; |
Dave Chinner | 4439647 | 2011-04-21 09:34:27 +0000 | [diff] [blame] | 5357 | error = xlog_bread_offset(log, 0, |
| 5358 | wrapped_hblks, hbp, |
| 5359 | offset + BBTOB(split_hblks)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5360 | if (error) |
| 5361 | goto bread_err2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5362 | } |
| 5363 | rhead = (xlog_rec_header_t *)offset; |
| 5364 | error = xlog_valid_rec_header(log, rhead, |
| 5365 | split_hblks ? blk_no : 0); |
| 5366 | if (error) |
| 5367 | goto bread_err2; |
| 5368 | |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 5369 | bblks = (int)BTOBB(be32_to_cpu(rhead->h_len)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5370 | blk_no += hblks; |
| 5371 | |
| 5372 | /* Read in data for log record */ |
| 5373 | if (blk_no + bblks <= log->l_logBBsize) { |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 5374 | error = xlog_bread(log, blk_no, bblks, dbp, |
| 5375 | &offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5376 | if (error) |
| 5377 | goto bread_err2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5378 | } else { |
| 5379 | /* This log record is split across the |
| 5380 | * physical end of log */ |
Chandra Seetharaman | 6292604 | 2011-07-22 23:40:15 +0000 | [diff] [blame] | 5381 | offset = dbp->b_addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5382 | split_bblks = 0; |
| 5383 | if (blk_no != log->l_logBBsize) { |
| 5384 | /* some data is before the physical |
| 5385 | * end of log */ |
| 5386 | ASSERT(!wrapped_hblks); |
| 5387 | ASSERT(blk_no <= INT_MAX); |
| 5388 | split_bblks = |
| 5389 | log->l_logBBsize - (int)blk_no; |
| 5390 | ASSERT(split_bblks > 0); |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 5391 | error = xlog_bread(log, blk_no, |
| 5392 | split_bblks, dbp, |
| 5393 | &offset); |
| 5394 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5395 | goto bread_err2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5396 | } |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 5397 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5398 | /* |
| 5399 | * Note: this black magic still works with |
| 5400 | * large sector sizes (non-512) only because: |
| 5401 | * - we increased the buffer size originally |
| 5402 | * by 1 sector giving us enough extra space |
| 5403 | * for the second read; |
| 5404 | * - the log start is guaranteed to be sector |
| 5405 | * aligned; |
| 5406 | * - we read the log end (LR header start) |
| 5407 | * _first_, then the log start (LR header end) |
| 5408 | * - order is important. |
| 5409 | */ |
Dave Chinner | 4439647 | 2011-04-21 09:34:27 +0000 | [diff] [blame] | 5410 | error = xlog_bread_offset(log, 0, |
Dave Chinner | 009507b | 2012-11-02 11:38:44 +1100 | [diff] [blame] | 5411 | bblks - split_bblks, dbp, |
Dave Chinner | 4439647 | 2011-04-21 09:34:27 +0000 | [diff] [blame] | 5412 | offset + BBTOB(split_bblks)); |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 5413 | if (error) |
| 5414 | goto bread_err2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5415 | } |
Christoph Hellwig | 0e446be | 2012-11-12 22:54:24 +1100 | [diff] [blame] | 5416 | |
Brian Foster | 9d94901 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5417 | error = xlog_recover_process(log, rhash, rhead, offset, |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 5418 | pass, &buffer_list); |
Christoph Hellwig | 0e446be | 2012-11-12 22:54:24 +1100 | [diff] [blame] | 5419 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5420 | goto bread_err2; |
Brian Foster | d7f3769 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5421 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5422 | blk_no += bblks; |
Brian Foster | d7f3769 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5423 | rhead_blk = blk_no; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5424 | } |
| 5425 | |
| 5426 | ASSERT(blk_no >= log->l_logBBsize); |
| 5427 | blk_no -= log->l_logBBsize; |
Brian Foster | d7f3769 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5428 | rhead_blk = blk_no; |
Eric Sandeen | 970fd3f | 2014-09-09 11:57:29 +1000 | [diff] [blame] | 5429 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5430 | |
Eric Sandeen | 970fd3f | 2014-09-09 11:57:29 +1000 | [diff] [blame] | 5431 | /* read first part of physical log */ |
| 5432 | while (blk_no < head_blk) { |
| 5433 | error = xlog_bread(log, blk_no, hblks, hbp, &offset); |
| 5434 | if (error) |
| 5435 | goto bread_err2; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 5436 | |
Eric Sandeen | 970fd3f | 2014-09-09 11:57:29 +1000 | [diff] [blame] | 5437 | rhead = (xlog_rec_header_t *)offset; |
| 5438 | error = xlog_valid_rec_header(log, rhead, blk_no); |
| 5439 | if (error) |
| 5440 | goto bread_err2; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 5441 | |
Eric Sandeen | 970fd3f | 2014-09-09 11:57:29 +1000 | [diff] [blame] | 5442 | /* blocks in data section */ |
| 5443 | bblks = (int)BTOBB(be32_to_cpu(rhead->h_len)); |
| 5444 | error = xlog_bread(log, blk_no+hblks, bblks, dbp, |
| 5445 | &offset); |
| 5446 | if (error) |
| 5447 | goto bread_err2; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 5448 | |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 5449 | error = xlog_recover_process(log, rhash, rhead, offset, pass, |
| 5450 | &buffer_list); |
Eric Sandeen | 970fd3f | 2014-09-09 11:57:29 +1000 | [diff] [blame] | 5451 | if (error) |
| 5452 | goto bread_err2; |
Brian Foster | d7f3769 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5453 | |
Eric Sandeen | 970fd3f | 2014-09-09 11:57:29 +1000 | [diff] [blame] | 5454 | blk_no += bblks + hblks; |
Brian Foster | d7f3769 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5455 | rhead_blk = blk_no; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5456 | } |
| 5457 | |
| 5458 | bread_err2: |
| 5459 | xlog_put_bp(dbp); |
| 5460 | bread_err1: |
| 5461 | xlog_put_bp(hbp); |
Brian Foster | d7f3769 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5462 | |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 5463 | /* |
| 5464 | * Submit buffers that have been added from the last record processed, |
| 5465 | * regardless of error status. |
| 5466 | */ |
| 5467 | if (!list_empty(&buffer_list)) |
| 5468 | error2 = xfs_buf_delwri_submit(&buffer_list); |
| 5469 | |
Brian Foster | d7f3769 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5470 | if (error && first_bad) |
| 5471 | *first_bad = rhead_blk; |
| 5472 | |
Brian Foster | 171192c | 2017-09-17 14:06:33 -0700 | [diff] [blame] | 5473 | /* |
| 5474 | * Transactions are freed at commit time but transactions without commit |
| 5475 | * records on disk are never committed. Free any that may be left in the |
| 5476 | * hash table. |
| 5477 | */ |
| 5478 | for (i = 0; i < XLOG_RHASH_SIZE; i++) { |
| 5479 | struct hlist_node *tmp; |
| 5480 | struct xlog_recover *trans; |
| 5481 | |
| 5482 | hlist_for_each_entry_safe(trans, tmp, &rhash[i], r_list) |
| 5483 | xlog_recover_free_trans(trans); |
| 5484 | } |
| 5485 | |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 5486 | return error ? error : error2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5487 | } |
| 5488 | |
| 5489 | /* |
| 5490 | * Do the recovery of the log. We actually do this in two phases. |
| 5491 | * The two passes are necessary in order to implement the function |
| 5492 | * of cancelling a record written into the log. The first pass |
| 5493 | * determines those things which have been cancelled, and the |
| 5494 | * second pass replays log items normally except for those which |
| 5495 | * have been cancelled. The handling of the replay and cancellations |
| 5496 | * takes place in the log item type specific routines. |
| 5497 | * |
| 5498 | * The table of items which have cancel records in the log is allocated |
| 5499 | * and freed at this level, since only here do we know when all of |
| 5500 | * the log recovery has been completed. |
| 5501 | */ |
| 5502 | STATIC int |
| 5503 | xlog_do_log_recovery( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 5504 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5505 | xfs_daddr_t head_blk, |
| 5506 | xfs_daddr_t tail_blk) |
| 5507 | { |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 5508 | int error, i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5509 | |
| 5510 | ASSERT(head_blk != tail_blk); |
| 5511 | |
| 5512 | /* |
| 5513 | * First do a pass to find all of the cancelled buf log items. |
| 5514 | * Store them in the buf_cancel_table for use in the second pass. |
| 5515 | */ |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 5516 | log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE * |
| 5517 | sizeof(struct list_head), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5518 | KM_SLEEP); |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 5519 | for (i = 0; i < XLOG_BC_TABLE_SIZE; i++) |
| 5520 | INIT_LIST_HEAD(&log->l_buf_cancel_table[i]); |
| 5521 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5522 | error = xlog_do_recovery_pass(log, head_blk, tail_blk, |
Brian Foster | d7f3769 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5523 | XLOG_RECOVER_PASS1, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5524 | if (error != 0) { |
Denys Vlasenko | f0e2d93 | 2008-05-19 16:31:57 +1000 | [diff] [blame] | 5525 | kmem_free(log->l_buf_cancel_table); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5526 | log->l_buf_cancel_table = NULL; |
| 5527 | return error; |
| 5528 | } |
| 5529 | /* |
| 5530 | * Then do a second pass to actually recover the items in the log. |
| 5531 | * When it is complete free the table of buf cancel items. |
| 5532 | */ |
| 5533 | error = xlog_do_recovery_pass(log, head_blk, tail_blk, |
Brian Foster | d7f3769 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5534 | XLOG_RECOVER_PASS2, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5535 | #ifdef DEBUG |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 5536 | if (!error) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5537 | int i; |
| 5538 | |
| 5539 | for (i = 0; i < XLOG_BC_TABLE_SIZE; i++) |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 5540 | ASSERT(list_empty(&log->l_buf_cancel_table[i])); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5541 | } |
| 5542 | #endif /* DEBUG */ |
| 5543 | |
Denys Vlasenko | f0e2d93 | 2008-05-19 16:31:57 +1000 | [diff] [blame] | 5544 | kmem_free(log->l_buf_cancel_table); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5545 | log->l_buf_cancel_table = NULL; |
| 5546 | |
| 5547 | return error; |
| 5548 | } |
| 5549 | |
| 5550 | /* |
| 5551 | * Do the actual recovery |
| 5552 | */ |
| 5553 | STATIC int |
| 5554 | xlog_do_recover( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 5555 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5556 | xfs_daddr_t head_blk, |
| 5557 | xfs_daddr_t tail_blk) |
| 5558 | { |
Dave Chinner | a798011 | 2016-03-07 08:39:36 +1100 | [diff] [blame] | 5559 | struct xfs_mount *mp = log->l_mp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5560 | int error; |
| 5561 | xfs_buf_t *bp; |
| 5562 | xfs_sb_t *sbp; |
| 5563 | |
| 5564 | /* |
| 5565 | * First replay the images in the log. |
| 5566 | */ |
| 5567 | error = xlog_do_log_recovery(log, head_blk, tail_blk); |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 5568 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5569 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5570 | |
| 5571 | /* |
| 5572 | * If IO errors happened during recovery, bail out. |
| 5573 | */ |
Dave Chinner | a798011 | 2016-03-07 08:39:36 +1100 | [diff] [blame] | 5574 | if (XFS_FORCED_SHUTDOWN(mp)) { |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 5575 | return -EIO; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5576 | } |
| 5577 | |
| 5578 | /* |
| 5579 | * We now update the tail_lsn since much of the recovery has completed |
| 5580 | * and there may be space available to use. If there were no extent |
| 5581 | * or iunlinks, we can free up the entire log and set the tail_lsn to |
| 5582 | * be the last_sync_lsn. This was set in xlog_find_tail to be the |
| 5583 | * lsn of the last known good LR on disk. If there are extent frees |
| 5584 | * or iunlinks they will have some entries in the AIL; so we look at |
| 5585 | * the AIL to determine how to set the tail_lsn. |
| 5586 | */ |
Dave Chinner | a798011 | 2016-03-07 08:39:36 +1100 | [diff] [blame] | 5587 | xlog_assign_tail_lsn(mp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5588 | |
| 5589 | /* |
| 5590 | * Now that we've finished replaying all buffer and inode |
Dave Chinner | 9802182 | 2012-11-12 22:54:03 +1100 | [diff] [blame] | 5591 | * updates, re-read in the superblock and reverify it. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5592 | */ |
Dave Chinner | a798011 | 2016-03-07 08:39:36 +1100 | [diff] [blame] | 5593 | bp = xfs_getsb(mp, 0); |
Dave Chinner | 1157b32c | 2016-02-10 15:01:11 +1100 | [diff] [blame] | 5594 | bp->b_flags &= ~(XBF_DONE | XBF_ASYNC); |
Dave Chinner | b68c082 | 2016-02-10 15:01:11 +1100 | [diff] [blame] | 5595 | ASSERT(!(bp->b_flags & XBF_WRITE)); |
Dave Chinner | 0cac682 | 2016-02-10 15:01:11 +1100 | [diff] [blame] | 5596 | bp->b_flags |= XBF_READ; |
Dave Chinner | 1813dd6 | 2012-11-14 17:54:40 +1100 | [diff] [blame] | 5597 | bp->b_ops = &xfs_sb_buf_ops; |
Christoph Hellwig | 83a0adc | 2013-12-17 00:03:52 -0800 | [diff] [blame] | 5598 | |
Dave Chinner | 595bff7 | 2014-10-02 09:05:14 +1000 | [diff] [blame] | 5599 | error = xfs_buf_submit_wait(bp); |
David Chinner | d64e31a | 2008-04-10 12:22:17 +1000 | [diff] [blame] | 5600 | if (error) { |
Dave Chinner | a798011 | 2016-03-07 08:39:36 +1100 | [diff] [blame] | 5601 | if (!XFS_FORCED_SHUTDOWN(mp)) { |
Dave Chinner | 595bff7 | 2014-10-02 09:05:14 +1000 | [diff] [blame] | 5602 | xfs_buf_ioerror_alert(bp, __func__); |
| 5603 | ASSERT(0); |
| 5604 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5605 | xfs_buf_relse(bp); |
| 5606 | return error; |
| 5607 | } |
| 5608 | |
| 5609 | /* Convert superblock from on-disk format */ |
Dave Chinner | a798011 | 2016-03-07 08:39:36 +1100 | [diff] [blame] | 5610 | sbp = &mp->m_sb; |
Dave Chinner | 9802182 | 2012-11-12 22:54:03 +1100 | [diff] [blame] | 5611 | xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5612 | xfs_buf_relse(bp); |
| 5613 | |
Dave Chinner | a798011 | 2016-03-07 08:39:36 +1100 | [diff] [blame] | 5614 | /* re-initialise in-core superblock and geometry structures */ |
| 5615 | xfs_reinit_percpu_counters(mp); |
| 5616 | error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi); |
| 5617 | if (error) { |
| 5618 | xfs_warn(mp, "Failed post-recovery per-ag init: %d", error); |
| 5619 | return error; |
| 5620 | } |
Darrick J. Wong | 5254885 | 2016-08-03 11:38:24 +1000 | [diff] [blame] | 5621 | mp->m_alloc_set_aside = xfs_alloc_set_aside(mp); |
Lachlan McIlroy | 5478eea | 2007-02-10 18:36:29 +1100 | [diff] [blame] | 5622 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5623 | xlog_recover_check_summary(log); |
| 5624 | |
| 5625 | /* Normal transactions can now occur */ |
| 5626 | log->l_flags &= ~XLOG_ACTIVE_RECOVERY; |
| 5627 | return 0; |
| 5628 | } |
| 5629 | |
| 5630 | /* |
| 5631 | * Perform recovery and re-initialize some log variables in xlog_find_tail. |
| 5632 | * |
| 5633 | * Return error or zero. |
| 5634 | */ |
| 5635 | int |
| 5636 | xlog_recover( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 5637 | struct xlog *log) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5638 | { |
| 5639 | xfs_daddr_t head_blk, tail_blk; |
| 5640 | int error; |
| 5641 | |
| 5642 | /* find the tail of the log */ |
Brian Foster | a45086e | 2015-10-12 15:59:25 +1100 | [diff] [blame] | 5643 | error = xlog_find_tail(log, &head_blk, &tail_blk); |
| 5644 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5645 | return error; |
| 5646 | |
Brian Foster | a45086e | 2015-10-12 15:59:25 +1100 | [diff] [blame] | 5647 | /* |
| 5648 | * The superblock was read before the log was available and thus the LSN |
| 5649 | * could not be verified. Check the superblock LSN against the current |
| 5650 | * LSN now that it's known. |
| 5651 | */ |
| 5652 | if (xfs_sb_version_hascrc(&log->l_mp->m_sb) && |
| 5653 | !xfs_log_check_lsn(log->l_mp, log->l_mp->m_sb.sb_lsn)) |
| 5654 | return -EINVAL; |
| 5655 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5656 | if (tail_blk != head_blk) { |
| 5657 | /* There used to be a comment here: |
| 5658 | * |
| 5659 | * disallow recovery on read-only mounts. note -- mount |
| 5660 | * checks for ENOSPC and turns it into an intelligent |
| 5661 | * error message. |
| 5662 | * ...but this is no longer true. Now, unless you specify |
| 5663 | * NORECOVERY (in which case this function would never be |
| 5664 | * called), we just go ahead and recover. We do this all |
| 5665 | * under the vfs layer, so we can get away with it unless |
| 5666 | * the device itself is read-only, in which case we fail. |
| 5667 | */ |
Utako Kusaka | 3a02ee1 | 2007-05-08 13:50:06 +1000 | [diff] [blame] | 5668 | if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5669 | return error; |
| 5670 | } |
| 5671 | |
Dave Chinner | e721f50 | 2013-04-03 16:11:32 +1100 | [diff] [blame] | 5672 | /* |
| 5673 | * Version 5 superblock log feature mask validation. We know the |
| 5674 | * log is dirty so check if there are any unknown log features |
| 5675 | * in what we need to recover. If there are unknown features |
| 5676 | * (e.g. unsupported transactions, then simply reject the |
| 5677 | * attempt at recovery before touching anything. |
| 5678 | */ |
| 5679 | if (XFS_SB_VERSION_NUM(&log->l_mp->m_sb) == XFS_SB_VERSION_5 && |
| 5680 | xfs_sb_has_incompat_log_feature(&log->l_mp->m_sb, |
| 5681 | XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) { |
| 5682 | xfs_warn(log->l_mp, |
Joe Perches | f41febd | 2015-07-29 11:52:04 +1000 | [diff] [blame] | 5683 | "Superblock has unknown incompatible log features (0x%x) enabled.", |
Dave Chinner | e721f50 | 2013-04-03 16:11:32 +1100 | [diff] [blame] | 5684 | (log->l_mp->m_sb.sb_features_log_incompat & |
| 5685 | XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)); |
Joe Perches | f41febd | 2015-07-29 11:52:04 +1000 | [diff] [blame] | 5686 | xfs_warn(log->l_mp, |
| 5687 | "The log can not be fully and/or safely recovered by this kernel."); |
| 5688 | xfs_warn(log->l_mp, |
| 5689 | "Please recover the log on a kernel that supports the unknown features."); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 5690 | return -EINVAL; |
Dave Chinner | e721f50 | 2013-04-03 16:11:32 +1100 | [diff] [blame] | 5691 | } |
| 5692 | |
Brian Foster | 2e22717 | 2014-09-09 11:56:13 +1000 | [diff] [blame] | 5693 | /* |
| 5694 | * Delay log recovery if the debug hook is set. This is debug |
| 5695 | * instrumention to coordinate simulation of I/O failures with |
| 5696 | * log recovery. |
| 5697 | */ |
| 5698 | if (xfs_globals.log_recovery_delay) { |
| 5699 | xfs_notice(log->l_mp, |
| 5700 | "Delaying log recovery for %d seconds.", |
| 5701 | xfs_globals.log_recovery_delay); |
| 5702 | msleep(xfs_globals.log_recovery_delay * 1000); |
| 5703 | } |
| 5704 | |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 5705 | xfs_notice(log->l_mp, "Starting recovery (logdev: %s)", |
| 5706 | log->l_mp->m_logname ? log->l_mp->m_logname |
| 5707 | : "internal"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5708 | |
| 5709 | error = xlog_do_recover(log, head_blk, tail_blk); |
| 5710 | log->l_flags |= XLOG_RECOVERY_NEEDED; |
| 5711 | } |
| 5712 | return error; |
| 5713 | } |
| 5714 | |
| 5715 | /* |
| 5716 | * In the first part of recovery we replay inodes and buffers and build |
| 5717 | * up the list of extent free items which need to be processed. Here |
| 5718 | * we process the extent free items and clean up the on disk unlinked |
| 5719 | * inode lists. This is separated from the first part of recovery so |
| 5720 | * that the root and real-time bitmap inodes can be read in from disk in |
| 5721 | * between the two stages. This is necessary so that we can free space |
| 5722 | * in the real-time portion of the file system. |
| 5723 | */ |
| 5724 | int |
| 5725 | xlog_recover_finish( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 5726 | struct xlog *log) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5727 | { |
| 5728 | /* |
| 5729 | * Now we're ready to do the transactions needed for the |
| 5730 | * rest of recovery. Start with completing all the extent |
| 5731 | * free intent records and then process the unlinked inode |
| 5732 | * lists. At this point, we essentially run in normal mode |
| 5733 | * except that we're still performing recovery actions |
| 5734 | * rather than accepting new requests. |
| 5735 | */ |
| 5736 | if (log->l_flags & XLOG_RECOVERY_NEEDED) { |
David Chinner | 3c1e2bb | 2008-04-10 12:21:11 +1000 | [diff] [blame] | 5737 | int error; |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 5738 | error = xlog_recover_process_intents(log); |
David Chinner | 3c1e2bb | 2008-04-10 12:21:11 +1000 | [diff] [blame] | 5739 | if (error) { |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 5740 | xfs_alert(log->l_mp, "Failed to recover intents"); |
David Chinner | 3c1e2bb | 2008-04-10 12:21:11 +1000 | [diff] [blame] | 5741 | return error; |
| 5742 | } |
Darrick J. Wong | 9e88b5d | 2016-08-03 12:09:48 +1000 | [diff] [blame] | 5743 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5744 | /* |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 5745 | * Sync the log to get all the intents out of the AIL. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5746 | * This isn't absolutely necessary, but it helps in |
| 5747 | * case the unlink transactions would have problems |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 5748 | * pushing the intents out of the way. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5749 | */ |
Christoph Hellwig | a14a348 | 2010-01-19 09:56:46 +0000 | [diff] [blame] | 5750 | xfs_log_force(log->l_mp, XFS_LOG_SYNC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5751 | |
Christoph Hellwig | 4249023 | 2008-08-13 16:49:32 +1000 | [diff] [blame] | 5752 | xlog_recover_process_iunlinks(log); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5753 | |
| 5754 | xlog_recover_check_summary(log); |
| 5755 | |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 5756 | xfs_notice(log->l_mp, "Ending recovery (logdev: %s)", |
| 5757 | log->l_mp->m_logname ? log->l_mp->m_logname |
| 5758 | : "internal"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5759 | log->l_flags &= ~XLOG_RECOVERY_NEEDED; |
| 5760 | } else { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 5761 | xfs_info(log->l_mp, "Ending clean mount"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5762 | } |
| 5763 | return 0; |
| 5764 | } |
| 5765 | |
Brian Foster | f0b2efa | 2015-08-19 09:58:36 +1000 | [diff] [blame] | 5766 | int |
| 5767 | xlog_recover_cancel( |
| 5768 | struct xlog *log) |
| 5769 | { |
| 5770 | int error = 0; |
| 5771 | |
| 5772 | if (log->l_flags & XLOG_RECOVERY_NEEDED) |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 5773 | error = xlog_recover_cancel_intents(log); |
Brian Foster | f0b2efa | 2015-08-19 09:58:36 +1000 | [diff] [blame] | 5774 | |
| 5775 | return error; |
| 5776 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5777 | |
| 5778 | #if defined(DEBUG) |
| 5779 | /* |
| 5780 | * Read all of the agf and agi counters and check that they |
| 5781 | * are consistent with the superblock counters. |
| 5782 | */ |
| 5783 | void |
| 5784 | xlog_recover_check_summary( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 5785 | struct xlog *log) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5786 | { |
| 5787 | xfs_mount_t *mp; |
| 5788 | xfs_agf_t *agfp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5789 | xfs_buf_t *agfbp; |
| 5790 | xfs_buf_t *agibp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5791 | xfs_agnumber_t agno; |
| 5792 | __uint64_t freeblks; |
| 5793 | __uint64_t itotal; |
| 5794 | __uint64_t ifree; |
Christoph Hellwig | 5e1be0f | 2008-11-28 14:23:37 +1100 | [diff] [blame] | 5795 | int error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5796 | |
| 5797 | mp = log->l_mp; |
| 5798 | |
| 5799 | freeblks = 0LL; |
| 5800 | itotal = 0LL; |
| 5801 | ifree = 0LL; |
| 5802 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { |
From: Christoph Hellwig | 4805621 | 2008-11-28 14:23:38 +1100 | [diff] [blame] | 5803 | error = xfs_read_agf(mp, NULL, agno, 0, &agfbp); |
| 5804 | if (error) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 5805 | xfs_alert(mp, "%s agf read failed agno %d error %d", |
| 5806 | __func__, agno, error); |
From: Christoph Hellwig | 4805621 | 2008-11-28 14:23:38 +1100 | [diff] [blame] | 5807 | } else { |
| 5808 | agfp = XFS_BUF_TO_AGF(agfbp); |
| 5809 | freeblks += be32_to_cpu(agfp->agf_freeblks) + |
| 5810 | be32_to_cpu(agfp->agf_flcount); |
| 5811 | xfs_buf_relse(agfbp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5812 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5813 | |
Christoph Hellwig | 5e1be0f | 2008-11-28 14:23:37 +1100 | [diff] [blame] | 5814 | error = xfs_read_agi(mp, NULL, agno, &agibp); |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 5815 | if (error) { |
| 5816 | xfs_alert(mp, "%s agi read failed agno %d error %d", |
| 5817 | __func__, agno, error); |
| 5818 | } else { |
Christoph Hellwig | 5e1be0f | 2008-11-28 14:23:37 +1100 | [diff] [blame] | 5819 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5820 | |
Christoph Hellwig | 5e1be0f | 2008-11-28 14:23:37 +1100 | [diff] [blame] | 5821 | itotal += be32_to_cpu(agi->agi_count); |
| 5822 | ifree += be32_to_cpu(agi->agi_freecount); |
| 5823 | xfs_buf_relse(agibp); |
| 5824 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5825 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5826 | } |
| 5827 | #endif /* DEBUG */ |