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" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 48 | |
Dave Chinner | fc06c6d | 2013-08-12 20:49:22 +1000 | [diff] [blame] | 49 | #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1) |
| 50 | |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 51 | STATIC int |
| 52 | xlog_find_zeroed( |
| 53 | struct xlog *, |
| 54 | xfs_daddr_t *); |
| 55 | STATIC int |
| 56 | xlog_clear_stale_blocks( |
| 57 | struct xlog *, |
| 58 | xfs_lsn_t); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 59 | #if defined(DEBUG) |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 60 | STATIC void |
| 61 | xlog_recover_check_summary( |
| 62 | struct xlog *); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 63 | #else |
| 64 | #define xlog_recover_check_summary(log) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 65 | #endif |
Brian Foster | 7088c41 | 2016-01-05 07:40:16 +1100 | [diff] [blame] | 66 | STATIC int |
| 67 | xlog_do_recovery_pass( |
| 68 | struct xlog *, xfs_daddr_t, xfs_daddr_t, int, xfs_daddr_t *); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 69 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 70 | /* |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 71 | * This structure is used during recovery to record the buf log items which |
| 72 | * have been canceled and should not be replayed. |
| 73 | */ |
| 74 | struct xfs_buf_cancel { |
| 75 | xfs_daddr_t bc_blkno; |
| 76 | uint bc_len; |
| 77 | int bc_refcount; |
| 78 | struct list_head bc_list; |
| 79 | }; |
| 80 | |
| 81 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 82 | * Sector aligned buffer routines for buffer create/read/write/access |
| 83 | */ |
| 84 | |
Alex Elder | ff30a62 | 2010-04-13 15:22:58 +1000 | [diff] [blame] | 85 | /* |
| 86 | * Verify the given count of basic blocks is valid number of blocks |
| 87 | * to specify for an operation involving the given XFS log buffer. |
| 88 | * Returns nonzero if the count is valid, 0 otherwise. |
| 89 | */ |
| 90 | |
| 91 | static inline int |
| 92 | xlog_buf_bbcount_valid( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 93 | struct xlog *log, |
Alex Elder | ff30a62 | 2010-04-13 15:22:58 +1000 | [diff] [blame] | 94 | int bbcount) |
| 95 | { |
| 96 | return bbcount > 0 && bbcount <= log->l_logBBsize; |
| 97 | } |
| 98 | |
Alex Elder | 36adecf | 2010-04-13 15:21:13 +1000 | [diff] [blame] | 99 | /* |
| 100 | * Allocate a buffer to hold log data. The buffer needs to be able |
| 101 | * to map to a range of nbblks basic blocks at any valid (basic |
| 102 | * block) offset within the log. |
| 103 | */ |
Eric Sandeen | 5d77c0d | 2009-11-19 15:52:00 +0000 | [diff] [blame] | 104 | STATIC xfs_buf_t * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 105 | xlog_get_bp( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 106 | struct xlog *log, |
Dave Chinner | 3228149 | 2009-01-22 15:37:47 +1100 | [diff] [blame] | 107 | int nbblks) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 108 | { |
Christoph Hellwig | c8da0fa | 2011-07-08 14:36:25 +0200 | [diff] [blame] | 109 | struct xfs_buf *bp; |
| 110 | |
Alex Elder | ff30a62 | 2010-04-13 15:22:58 +1000 | [diff] [blame] | 111 | if (!xlog_buf_bbcount_valid(log, nbblks)) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 112 | xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer", |
Alex Elder | ff30a62 | 2010-04-13 15:22:58 +1000 | [diff] [blame] | 113 | nbblks); |
| 114 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp); |
Dave Chinner | 3228149 | 2009-01-22 15:37:47 +1100 | [diff] [blame] | 115 | return NULL; |
| 116 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 117 | |
Alex Elder | 36adecf | 2010-04-13 15:21:13 +1000 | [diff] [blame] | 118 | /* |
| 119 | * We do log I/O in units of log sectors (a power-of-2 |
| 120 | * multiple of the basic block size), so we round up the |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 121 | * requested size to accommodate the basic blocks required |
Alex Elder | 36adecf | 2010-04-13 15:21:13 +1000 | [diff] [blame] | 122 | * for complete log sectors. |
| 123 | * |
| 124 | * In addition, the buffer may be used for a non-sector- |
| 125 | * aligned block offset, in which case an I/O of the |
| 126 | * requested size could extend beyond the end of the |
| 127 | * buffer. If the requested size is only 1 basic block it |
| 128 | * will never straddle a sector boundary, so this won't be |
| 129 | * an issue. Nor will this be a problem if the log I/O is |
| 130 | * done in basic blocks (sector size 1). But otherwise we |
| 131 | * extend the buffer by one extra log sector to ensure |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 132 | * there's space to accommodate this possibility. |
Alex Elder | 36adecf | 2010-04-13 15:21:13 +1000 | [diff] [blame] | 133 | */ |
Alex Elder | 69ce58f | 2010-04-20 17:09:59 +1000 | [diff] [blame] | 134 | if (nbblks > 1 && log->l_sectBBsize > 1) |
| 135 | nbblks += log->l_sectBBsize; |
| 136 | nbblks = round_up(nbblks, log->l_sectBBsize); |
Alex Elder | 36adecf | 2010-04-13 15:21:13 +1000 | [diff] [blame] | 137 | |
Dave Chinner | e70b73f | 2012-04-23 15:58:49 +1000 | [diff] [blame] | 138 | 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] | 139 | if (bp) |
| 140 | xfs_buf_unlock(bp); |
| 141 | return bp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 142 | } |
| 143 | |
Eric Sandeen | 5d77c0d | 2009-11-19 15:52:00 +0000 | [diff] [blame] | 144 | STATIC void |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 145 | xlog_put_bp( |
| 146 | xfs_buf_t *bp) |
| 147 | { |
| 148 | xfs_buf_free(bp); |
| 149 | } |
| 150 | |
Alex Elder | 48389ef | 2010-04-20 17:10:21 +1000 | [diff] [blame] | 151 | /* |
| 152 | * Return the address of the start of the given block number's data |
| 153 | * in a log buffer. The buffer covers a log sector-aligned region. |
| 154 | */ |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 155 | STATIC char * |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 156 | xlog_align( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 157 | struct xlog *log, |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 158 | xfs_daddr_t blk_no, |
| 159 | int nbblks, |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 160 | struct xfs_buf *bp) |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 161 | { |
Christoph Hellwig | fdc07f4 | 2010-05-10 17:28:14 +0000 | [diff] [blame] | 162 | 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] | 163 | |
Dave Chinner | 4e94b71 | 2012-04-23 15:58:51 +1000 | [diff] [blame] | 164 | ASSERT(offset + nbblks <= bp->b_length); |
Chandra Seetharaman | 6292604 | 2011-07-22 23:40:15 +0000 | [diff] [blame] | 165 | return bp->b_addr + BBTOB(offset); |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 166 | } |
| 167 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 168 | |
| 169 | /* |
| 170 | * nbblks should be uint, but oh well. Just want to catch that 32-bit length. |
| 171 | */ |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 172 | STATIC int |
| 173 | xlog_bread_noalign( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 174 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 175 | xfs_daddr_t blk_no, |
| 176 | int nbblks, |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 177 | struct xfs_buf *bp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 178 | { |
| 179 | int error; |
| 180 | |
Alex Elder | ff30a62 | 2010-04-13 15:22:58 +1000 | [diff] [blame] | 181 | if (!xlog_buf_bbcount_valid(log, nbblks)) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 182 | xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer", |
Alex Elder | ff30a62 | 2010-04-13 15:22:58 +1000 | [diff] [blame] | 183 | nbblks); |
| 184 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 185 | return -EFSCORRUPTED; |
Dave Chinner | 3228149 | 2009-01-22 15:37:47 +1100 | [diff] [blame] | 186 | } |
| 187 | |
Alex Elder | 69ce58f | 2010-04-20 17:09:59 +1000 | [diff] [blame] | 188 | blk_no = round_down(blk_no, log->l_sectBBsize); |
| 189 | nbblks = round_up(nbblks, log->l_sectBBsize); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 190 | |
| 191 | ASSERT(nbblks > 0); |
Dave Chinner | 4e94b71 | 2012-04-23 15:58:51 +1000 | [diff] [blame] | 192 | ASSERT(nbblks <= bp->b_length); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 193 | |
| 194 | XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no); |
Dave Chinner | 0cac682 | 2016-02-10 15:01:11 +1100 | [diff] [blame] | 195 | bp->b_flags |= XBF_READ; |
Dave Chinner | aa0e883 | 2012-04-23 15:58:52 +1000 | [diff] [blame] | 196 | bp->b_io_length = nbblks; |
Dave Chinner | 0e95f19 | 2012-04-23 15:58:46 +1000 | [diff] [blame] | 197 | bp->b_error = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 198 | |
Dave Chinner | 595bff7 | 2014-10-02 09:05:14 +1000 | [diff] [blame] | 199 | error = xfs_buf_submit_wait(bp); |
| 200 | if (error && !XFS_FORCED_SHUTDOWN(log->l_mp)) |
Christoph Hellwig | 901796a | 2011-10-10 16:52:49 +0000 | [diff] [blame] | 201 | xfs_buf_ioerror_alert(bp, __func__); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 202 | return error; |
| 203 | } |
| 204 | |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 205 | STATIC int |
| 206 | xlog_bread( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 207 | struct xlog *log, |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 208 | xfs_daddr_t blk_no, |
| 209 | int nbblks, |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 210 | struct xfs_buf *bp, |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 211 | char **offset) |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 212 | { |
| 213 | int error; |
| 214 | |
| 215 | error = xlog_bread_noalign(log, blk_no, nbblks, bp); |
| 216 | if (error) |
| 217 | return error; |
| 218 | |
| 219 | *offset = xlog_align(log, blk_no, nbblks, bp); |
| 220 | return 0; |
| 221 | } |
| 222 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 223 | /* |
Dave Chinner | 4439647 | 2011-04-21 09:34:27 +0000 | [diff] [blame] | 224 | * Read at an offset into the buffer. Returns with the buffer in it's original |
| 225 | * state regardless of the result of the read. |
| 226 | */ |
| 227 | STATIC int |
| 228 | xlog_bread_offset( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 229 | struct xlog *log, |
Dave Chinner | 4439647 | 2011-04-21 09:34:27 +0000 | [diff] [blame] | 230 | xfs_daddr_t blk_no, /* block to read from */ |
| 231 | int nbblks, /* blocks to read */ |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 232 | struct xfs_buf *bp, |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 233 | char *offset) |
Dave Chinner | 4439647 | 2011-04-21 09:34:27 +0000 | [diff] [blame] | 234 | { |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 235 | char *orig_offset = bp->b_addr; |
Dave Chinner | 4e94b71 | 2012-04-23 15:58:51 +1000 | [diff] [blame] | 236 | int orig_len = BBTOB(bp->b_length); |
Dave Chinner | 4439647 | 2011-04-21 09:34:27 +0000 | [diff] [blame] | 237 | int error, error2; |
| 238 | |
Chandra Seetharaman | 02fe03d | 2011-07-22 23:40:22 +0000 | [diff] [blame] | 239 | error = xfs_buf_associate_memory(bp, offset, BBTOB(nbblks)); |
Dave Chinner | 4439647 | 2011-04-21 09:34:27 +0000 | [diff] [blame] | 240 | if (error) |
| 241 | return error; |
| 242 | |
| 243 | error = xlog_bread_noalign(log, blk_no, nbblks, bp); |
| 244 | |
| 245 | /* must reset buffer pointer even on error */ |
Chandra Seetharaman | 02fe03d | 2011-07-22 23:40:22 +0000 | [diff] [blame] | 246 | error2 = xfs_buf_associate_memory(bp, orig_offset, orig_len); |
Dave Chinner | 4439647 | 2011-04-21 09:34:27 +0000 | [diff] [blame] | 247 | if (error) |
| 248 | return error; |
| 249 | return error2; |
| 250 | } |
| 251 | |
| 252 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 253 | * Write out the buffer at the given block for the given number of blocks. |
| 254 | * The buffer is kept locked across the write and is returned locked. |
| 255 | * This can only be used for synchronous log writes. |
| 256 | */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 257 | STATIC int |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 258 | xlog_bwrite( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 259 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 260 | xfs_daddr_t blk_no, |
| 261 | int nbblks, |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 262 | struct xfs_buf *bp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 263 | { |
| 264 | int error; |
| 265 | |
Alex Elder | ff30a62 | 2010-04-13 15:22:58 +1000 | [diff] [blame] | 266 | if (!xlog_buf_bbcount_valid(log, nbblks)) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 267 | xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer", |
Alex Elder | ff30a62 | 2010-04-13 15:22:58 +1000 | [diff] [blame] | 268 | nbblks); |
| 269 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 270 | return -EFSCORRUPTED; |
Dave Chinner | 3228149 | 2009-01-22 15:37:47 +1100 | [diff] [blame] | 271 | } |
| 272 | |
Alex Elder | 69ce58f | 2010-04-20 17:09:59 +1000 | [diff] [blame] | 273 | blk_no = round_down(blk_no, log->l_sectBBsize); |
| 274 | nbblks = round_up(nbblks, log->l_sectBBsize); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 275 | |
| 276 | ASSERT(nbblks > 0); |
Dave Chinner | 4e94b71 | 2012-04-23 15:58:51 +1000 | [diff] [blame] | 277 | ASSERT(nbblks <= bp->b_length); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 278 | |
| 279 | XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no); |
Chandra Seetharaman | 72790aa | 2011-07-22 23:40:04 +0000 | [diff] [blame] | 280 | xfs_buf_hold(bp); |
Christoph Hellwig | 0c842ad | 2011-07-08 14:36:19 +0200 | [diff] [blame] | 281 | xfs_buf_lock(bp); |
Dave Chinner | aa0e883 | 2012-04-23 15:58:52 +1000 | [diff] [blame] | 282 | bp->b_io_length = nbblks; |
Dave Chinner | 0e95f19 | 2012-04-23 15:58:46 +1000 | [diff] [blame] | 283 | bp->b_error = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 284 | |
Christoph Hellwig | c2b006c | 2011-08-23 08:28:07 +0000 | [diff] [blame] | 285 | error = xfs_bwrite(bp); |
Christoph Hellwig | 901796a | 2011-10-10 16:52:49 +0000 | [diff] [blame] | 286 | if (error) |
| 287 | xfs_buf_ioerror_alert(bp, __func__); |
Christoph Hellwig | c2b006c | 2011-08-23 08:28:07 +0000 | [diff] [blame] | 288 | xfs_buf_relse(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 289 | return error; |
| 290 | } |
| 291 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 292 | #ifdef DEBUG |
| 293 | /* |
| 294 | * dump debug superblock and log record information |
| 295 | */ |
| 296 | STATIC void |
| 297 | xlog_header_check_dump( |
| 298 | xfs_mount_t *mp, |
| 299 | xlog_rec_header_t *head) |
| 300 | { |
Eric Sandeen | 08e96e1 | 2013-10-11 20:59:05 -0500 | [diff] [blame] | 301 | xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d", |
Joe Perches | 03daa57 | 2009-12-14 18:01:10 -0800 | [diff] [blame] | 302 | __func__, &mp->m_sb.sb_uuid, XLOG_FMT); |
Eric Sandeen | 08e96e1 | 2013-10-11 20:59:05 -0500 | [diff] [blame] | 303 | xfs_debug(mp, " log : uuid = %pU, fmt = %d", |
Joe Perches | 03daa57 | 2009-12-14 18:01:10 -0800 | [diff] [blame] | 304 | &head->h_fs_uuid, be32_to_cpu(head->h_fmt)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 305 | } |
| 306 | #else |
| 307 | #define xlog_header_check_dump(mp, head) |
| 308 | #endif |
| 309 | |
| 310 | /* |
| 311 | * check log record header for recovery |
| 312 | */ |
| 313 | STATIC int |
| 314 | xlog_header_check_recover( |
| 315 | xfs_mount_t *mp, |
| 316 | xlog_rec_header_t *head) |
| 317 | { |
Christoph Hellwig | 69ef921 | 2011-07-08 14:36:05 +0200 | [diff] [blame] | 318 | ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 319 | |
| 320 | /* |
| 321 | * IRIX doesn't write the h_fmt field and leaves it zeroed |
| 322 | * (XLOG_FMT_UNKNOWN). This stops us from trying to recover |
| 323 | * a dirty log created in IRIX. |
| 324 | */ |
Christoph Hellwig | 69ef921 | 2011-07-08 14:36:05 +0200 | [diff] [blame] | 325 | if (unlikely(head->h_fmt != cpu_to_be32(XLOG_FMT))) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 326 | xfs_warn(mp, |
| 327 | "dirty log written in incompatible format - can't recover"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 328 | xlog_header_check_dump(mp, head); |
| 329 | XFS_ERROR_REPORT("xlog_header_check_recover(1)", |
| 330 | XFS_ERRLEVEL_HIGH, mp); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 331 | return -EFSCORRUPTED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 332 | } 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] | 333 | xfs_warn(mp, |
| 334 | "dirty log entry has mismatched uuid - can't recover"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 335 | xlog_header_check_dump(mp, head); |
| 336 | XFS_ERROR_REPORT("xlog_header_check_recover(2)", |
| 337 | XFS_ERRLEVEL_HIGH, mp); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 338 | return -EFSCORRUPTED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 339 | } |
| 340 | return 0; |
| 341 | } |
| 342 | |
| 343 | /* |
| 344 | * read the head block of the log and check the header |
| 345 | */ |
| 346 | STATIC int |
| 347 | xlog_header_check_mount( |
| 348 | xfs_mount_t *mp, |
| 349 | xlog_rec_header_t *head) |
| 350 | { |
Christoph Hellwig | 69ef921 | 2011-07-08 14:36:05 +0200 | [diff] [blame] | 351 | ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 352 | |
| 353 | if (uuid_is_nil(&head->h_fs_uuid)) { |
| 354 | /* |
| 355 | * IRIX doesn't write the h_fs_uuid or h_fmt fields. If |
| 356 | * h_fs_uuid is nil, we assume this log was last mounted |
| 357 | * by IRIX and continue. |
| 358 | */ |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 359 | xfs_warn(mp, "nil uuid in log - IRIX style log"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 360 | } 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] | 361 | xfs_warn(mp, "log has mismatched uuid - can't recover"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 362 | xlog_header_check_dump(mp, head); |
| 363 | XFS_ERROR_REPORT("xlog_header_check_mount", |
| 364 | XFS_ERRLEVEL_HIGH, mp); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 365 | return -EFSCORRUPTED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 366 | } |
| 367 | return 0; |
| 368 | } |
| 369 | |
| 370 | STATIC void |
| 371 | xlog_recover_iodone( |
| 372 | struct xfs_buf *bp) |
| 373 | { |
Chandra Seetharaman | 5a52c2a58 | 2011-07-22 23:39:51 +0000 | [diff] [blame] | 374 | if (bp->b_error) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 375 | /* |
| 376 | * We're not going to bother about retrying |
| 377 | * this during recovery. One strike! |
| 378 | */ |
Dave Chinner | 595bff7 | 2014-10-02 09:05:14 +1000 | [diff] [blame] | 379 | if (!XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) { |
| 380 | xfs_buf_ioerror_alert(bp, __func__); |
| 381 | xfs_force_shutdown(bp->b_target->bt_mount, |
| 382 | SHUTDOWN_META_IO_ERROR); |
| 383 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 384 | } |
Brian Foster | 60a4a22 | 2016-09-26 08:34:27 +1000 | [diff] [blame^] | 385 | |
| 386 | /* |
| 387 | * On v5 supers, a bli could be attached to update the metadata LSN. |
| 388 | * Clean it up. |
| 389 | */ |
| 390 | if (bp->b_fspriv) |
| 391 | xfs_buf_item_relse(bp); |
| 392 | ASSERT(bp->b_fspriv == NULL); |
| 393 | |
Christoph Hellwig | cb669ca | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 394 | bp->b_iodone = NULL; |
Dave Chinner | e8aaba9 | 2014-10-02 09:04:22 +1000 | [diff] [blame] | 395 | xfs_buf_ioend(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 396 | } |
| 397 | |
| 398 | /* |
| 399 | * This routine finds (to an approximation) the first block in the physical |
| 400 | * log which contains the given cycle. It uses a binary search algorithm. |
| 401 | * Note that the algorithm can not be perfect because the disk will not |
| 402 | * necessarily be perfect. |
| 403 | */ |
David Chinner | a8272ce | 2007-11-23 16:28:09 +1100 | [diff] [blame] | 404 | STATIC int |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 405 | xlog_find_cycle_start( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 406 | struct xlog *log, |
| 407 | struct xfs_buf *bp, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 408 | xfs_daddr_t first_blk, |
| 409 | xfs_daddr_t *last_blk, |
| 410 | uint cycle) |
| 411 | { |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 412 | char *offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 413 | xfs_daddr_t mid_blk; |
Alex Elder | e3bb2e3 | 2010-04-15 18:17:30 +0000 | [diff] [blame] | 414 | xfs_daddr_t end_blk; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 415 | uint mid_cycle; |
| 416 | int error; |
| 417 | |
Alex Elder | e3bb2e3 | 2010-04-15 18:17:30 +0000 | [diff] [blame] | 418 | end_blk = *last_blk; |
| 419 | mid_blk = BLK_AVG(first_blk, end_blk); |
| 420 | while (mid_blk != first_blk && mid_blk != end_blk) { |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 421 | error = xlog_bread(log, mid_blk, 1, bp, &offset); |
| 422 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 423 | return error; |
Christoph Hellwig | 03bea6f | 2007-10-12 10:58:05 +1000 | [diff] [blame] | 424 | mid_cycle = xlog_get_cycle(offset); |
Alex Elder | e3bb2e3 | 2010-04-15 18:17:30 +0000 | [diff] [blame] | 425 | if (mid_cycle == cycle) |
| 426 | end_blk = mid_blk; /* last_half_cycle == mid_cycle */ |
| 427 | else |
| 428 | first_blk = mid_blk; /* first_half_cycle == mid_cycle */ |
| 429 | mid_blk = BLK_AVG(first_blk, end_blk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 430 | } |
Alex Elder | e3bb2e3 | 2010-04-15 18:17:30 +0000 | [diff] [blame] | 431 | ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) || |
| 432 | (mid_blk == end_blk && mid_blk-1 == first_blk)); |
| 433 | |
| 434 | *last_blk = end_blk; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 435 | |
| 436 | return 0; |
| 437 | } |
| 438 | |
| 439 | /* |
Alex Elder | 3f943d8 | 2010-04-15 18:17:34 +0000 | [diff] [blame] | 440 | * Check that a range of blocks does not contain stop_on_cycle_no. |
| 441 | * Fill in *new_blk with the block offset where such a block is |
| 442 | * found, or with -1 (an invalid block number) if there is no such |
| 443 | * block in the range. The scan needs to occur from front to back |
| 444 | * and the pointer into the region must be updated since a later |
| 445 | * routine will need to perform another test. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 446 | */ |
| 447 | STATIC int |
| 448 | xlog_find_verify_cycle( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 449 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 450 | xfs_daddr_t start_blk, |
| 451 | int nbblks, |
| 452 | uint stop_on_cycle_no, |
| 453 | xfs_daddr_t *new_blk) |
| 454 | { |
| 455 | xfs_daddr_t i, j; |
| 456 | uint cycle; |
| 457 | xfs_buf_t *bp; |
| 458 | xfs_daddr_t bufblks; |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 459 | char *buf = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 460 | int error = 0; |
| 461 | |
Alex Elder | 6881a22 | 2010-04-13 15:22:29 +1000 | [diff] [blame] | 462 | /* |
| 463 | * Greedily allocate a buffer big enough to handle the full |
| 464 | * range of basic blocks we'll be examining. If that fails, |
| 465 | * try a smaller size. We need to be able to read at least |
| 466 | * a log sector, or we're out of luck. |
| 467 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 468 | bufblks = 1 << ffs(nbblks); |
Dave Chinner | 81158e0 | 2012-04-27 19:45:22 +1000 | [diff] [blame] | 469 | while (bufblks > log->l_logBBsize) |
| 470 | bufblks >>= 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 471 | while (!(bp = xlog_get_bp(log, bufblks))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 472 | bufblks >>= 1; |
Alex Elder | 69ce58f | 2010-04-20 17:09:59 +1000 | [diff] [blame] | 473 | if (bufblks < log->l_sectBBsize) |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 474 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 475 | } |
| 476 | |
| 477 | for (i = start_blk; i < start_blk + nbblks; i += bufblks) { |
| 478 | int bcount; |
| 479 | |
| 480 | bcount = min(bufblks, (start_blk + nbblks - i)); |
| 481 | |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 482 | error = xlog_bread(log, i, bcount, bp, &buf); |
| 483 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 484 | goto out; |
| 485 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 486 | for (j = 0; j < bcount; j++) { |
Christoph Hellwig | 03bea6f | 2007-10-12 10:58:05 +1000 | [diff] [blame] | 487 | cycle = xlog_get_cycle(buf); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 488 | if (cycle == stop_on_cycle_no) { |
| 489 | *new_blk = i+j; |
| 490 | goto out; |
| 491 | } |
| 492 | |
| 493 | buf += BBSIZE; |
| 494 | } |
| 495 | } |
| 496 | |
| 497 | *new_blk = -1; |
| 498 | |
| 499 | out: |
| 500 | xlog_put_bp(bp); |
| 501 | return error; |
| 502 | } |
| 503 | |
| 504 | /* |
| 505 | * Potentially backup over partial log record write. |
| 506 | * |
| 507 | * In the typical case, last_blk is the number of the block directly after |
| 508 | * a good log record. Therefore, we subtract one to get the block number |
| 509 | * of the last block in the given buffer. extra_bblks contains the number |
| 510 | * of blocks we would have read on a previous read. This happens when the |
| 511 | * last log record is split over the end of the physical log. |
| 512 | * |
| 513 | * extra_bblks is the number of blocks potentially verified on a previous |
| 514 | * call to this routine. |
| 515 | */ |
| 516 | STATIC int |
| 517 | xlog_find_verify_log_record( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 518 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 519 | xfs_daddr_t start_blk, |
| 520 | xfs_daddr_t *last_blk, |
| 521 | int extra_bblks) |
| 522 | { |
| 523 | xfs_daddr_t i; |
| 524 | xfs_buf_t *bp; |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 525 | char *offset = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 526 | xlog_rec_header_t *head = NULL; |
| 527 | int error = 0; |
| 528 | int smallmem = 0; |
| 529 | int num_blks = *last_blk - start_blk; |
| 530 | int xhdrs; |
| 531 | |
| 532 | ASSERT(start_blk != 0 || *last_blk != start_blk); |
| 533 | |
| 534 | if (!(bp = xlog_get_bp(log, num_blks))) { |
| 535 | if (!(bp = xlog_get_bp(log, 1))) |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 536 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 537 | smallmem = 1; |
| 538 | } else { |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 539 | error = xlog_bread(log, start_blk, num_blks, bp, &offset); |
| 540 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 541 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 542 | offset += ((num_blks - 1) << BBSHIFT); |
| 543 | } |
| 544 | |
| 545 | for (i = (*last_blk) - 1; i >= 0; i--) { |
| 546 | if (i < start_blk) { |
| 547 | /* valid log record not found */ |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 548 | xfs_warn(log->l_mp, |
| 549 | "Log inconsistent (didn't find previous header)"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 550 | ASSERT(0); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 551 | error = -EIO; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 552 | goto out; |
| 553 | } |
| 554 | |
| 555 | if (smallmem) { |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 556 | error = xlog_bread(log, i, 1, bp, &offset); |
| 557 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 558 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 559 | } |
| 560 | |
| 561 | head = (xlog_rec_header_t *)offset; |
| 562 | |
Christoph Hellwig | 69ef921 | 2011-07-08 14:36:05 +0200 | [diff] [blame] | 563 | if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 564 | break; |
| 565 | |
| 566 | if (!smallmem) |
| 567 | offset -= BBSIZE; |
| 568 | } |
| 569 | |
| 570 | /* |
| 571 | * We hit the beginning of the physical log & still no header. Return |
| 572 | * to caller. If caller can handle a return of -1, then this routine |
| 573 | * will be called again for the end of the physical log. |
| 574 | */ |
| 575 | if (i == -1) { |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 576 | error = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 577 | goto out; |
| 578 | } |
| 579 | |
| 580 | /* |
| 581 | * We have the final block of the good log (the first block |
| 582 | * of the log record _before_ the head. So we check the uuid. |
| 583 | */ |
| 584 | if ((error = xlog_header_check_mount(log->l_mp, head))) |
| 585 | goto out; |
| 586 | |
| 587 | /* |
| 588 | * We may have found a log record header before we expected one. |
| 589 | * last_blk will be the 1st block # with a given cycle #. We may end |
| 590 | * up reading an entire log record. In this case, we don't want to |
| 591 | * reset last_blk. Only when last_blk points in the middle of a log |
| 592 | * record do we update last_blk. |
| 593 | */ |
Eric Sandeen | 6211870 | 2008-03-06 13:44:28 +1100 | [diff] [blame] | 594 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 595 | uint h_size = be32_to_cpu(head->h_size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 596 | |
| 597 | xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE; |
| 598 | if (h_size % XLOG_HEADER_CYCLE_SIZE) |
| 599 | xhdrs++; |
| 600 | } else { |
| 601 | xhdrs = 1; |
| 602 | } |
| 603 | |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 604 | if (*last_blk - i + extra_bblks != |
| 605 | BTOBB(be32_to_cpu(head->h_len)) + xhdrs) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 606 | *last_blk = i; |
| 607 | |
| 608 | out: |
| 609 | xlog_put_bp(bp); |
| 610 | return error; |
| 611 | } |
| 612 | |
| 613 | /* |
| 614 | * 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] | 615 | * could go. This means that incomplete LR writes at the end are |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 616 | * eliminated when calculating the head. We aren't guaranteed that previous |
| 617 | * LR have complete transactions. We only know that a cycle number of |
| 618 | * current cycle number -1 won't be present in the log if we start writing |
| 619 | * from our current block number. |
| 620 | * |
| 621 | * last_blk contains the block number of the first block with a given |
| 622 | * cycle number. |
| 623 | * |
| 624 | * Return: zero if normal, non-zero if error. |
| 625 | */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 626 | STATIC int |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 627 | xlog_find_head( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 628 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 629 | xfs_daddr_t *return_head_blk) |
| 630 | { |
| 631 | xfs_buf_t *bp; |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 632 | char *offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 633 | xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk; |
| 634 | int num_scan_bblks; |
| 635 | uint first_half_cycle, last_half_cycle; |
| 636 | uint stop_on_cycle; |
| 637 | int error, log_bbnum = log->l_logBBsize; |
| 638 | |
| 639 | /* Is the end of the log device zeroed? */ |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 640 | error = xlog_find_zeroed(log, &first_blk); |
| 641 | if (error < 0) { |
| 642 | xfs_warn(log->l_mp, "empty log check failed"); |
| 643 | return error; |
| 644 | } |
| 645 | if (error == 1) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 646 | *return_head_blk = first_blk; |
| 647 | |
| 648 | /* Is the whole lot zeroed? */ |
| 649 | if (!first_blk) { |
| 650 | /* Linux XFS shouldn't generate totally zeroed logs - |
| 651 | * mkfs etc write a dummy unmount record to a fresh |
| 652 | * log so we can store the uuid in there |
| 653 | */ |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 654 | xfs_warn(log->l_mp, "totally zeroed log"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 655 | } |
| 656 | |
| 657 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 658 | } |
| 659 | |
| 660 | first_blk = 0; /* get cycle # of 1st block */ |
| 661 | bp = xlog_get_bp(log, 1); |
| 662 | if (!bp) |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 663 | return -ENOMEM; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 664 | |
| 665 | error = xlog_bread(log, 0, 1, bp, &offset); |
| 666 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 667 | goto bp_err; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 668 | |
Christoph Hellwig | 03bea6f | 2007-10-12 10:58:05 +1000 | [diff] [blame] | 669 | first_half_cycle = xlog_get_cycle(offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 670 | |
| 671 | last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */ |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 672 | error = xlog_bread(log, last_blk, 1, bp, &offset); |
| 673 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 674 | goto bp_err; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 675 | |
Christoph Hellwig | 03bea6f | 2007-10-12 10:58:05 +1000 | [diff] [blame] | 676 | last_half_cycle = xlog_get_cycle(offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 677 | ASSERT(last_half_cycle != 0); |
| 678 | |
| 679 | /* |
| 680 | * If the 1st half cycle number is equal to the last half cycle number, |
| 681 | * then the entire log is stamped with the same cycle number. In this |
| 682 | * case, head_blk can't be set to zero (which makes sense). The below |
| 683 | * math doesn't work out properly with head_blk equal to zero. Instead, |
| 684 | * we set it to log_bbnum which is an invalid block number, but this |
| 685 | * value makes the math correct. If head_blk doesn't changed through |
| 686 | * all the tests below, *head_blk is set to zero at the very end rather |
| 687 | * than log_bbnum. In a sense, log_bbnum and zero are the same block |
| 688 | * in a circular file. |
| 689 | */ |
| 690 | if (first_half_cycle == last_half_cycle) { |
| 691 | /* |
| 692 | * In this case we believe that the entire log should have |
| 693 | * cycle number last_half_cycle. We need to scan backwards |
| 694 | * from the end verifying that there are no holes still |
| 695 | * containing last_half_cycle - 1. If we find such a hole, |
| 696 | * then the start of that hole will be the new head. The |
| 697 | * simple case looks like |
| 698 | * x | x ... | x - 1 | x |
| 699 | * Another case that fits this picture would be |
| 700 | * x | x + 1 | x ... | x |
Nathan Scott | c41564b | 2006-03-29 08:55:14 +1000 | [diff] [blame] | 701 | * 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] | 702 | * log, as one of the latest writes at the beginning was |
| 703 | * incomplete. |
| 704 | * One more case is |
| 705 | * x | x + 1 | x ... | x - 1 | x |
| 706 | * This is really the combination of the above two cases, and |
| 707 | * the head has to end up at the start of the x-1 hole at the |
| 708 | * end of the log. |
| 709 | * |
| 710 | * In the 256k log case, we will read from the beginning to the |
| 711 | * end of the log and search for cycle numbers equal to x-1. |
| 712 | * We don't worry about the x+1 blocks that we encounter, |
| 713 | * because we know that they cannot be the head since the log |
| 714 | * started with x. |
| 715 | */ |
| 716 | head_blk = log_bbnum; |
| 717 | stop_on_cycle = last_half_cycle - 1; |
| 718 | } else { |
| 719 | /* |
| 720 | * In this case we want to find the first block with cycle |
| 721 | * number matching last_half_cycle. We expect the log to be |
| 722 | * some variation on |
Alex Elder | 3f943d8 | 2010-04-15 18:17:34 +0000 | [diff] [blame] | 723 | * x + 1 ... | x ... | x |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 724 | * The first block with cycle number x (last_half_cycle) will |
| 725 | * be where the new head belongs. First we do a binary search |
| 726 | * for the first occurrence of last_half_cycle. The binary |
| 727 | * search may not be totally accurate, so then we scan back |
| 728 | * from there looking for occurrences of last_half_cycle before |
| 729 | * us. If that backwards scan wraps around the beginning of |
| 730 | * the log, then we look for occurrences of last_half_cycle - 1 |
| 731 | * at the end of the log. The cases we're looking for look |
| 732 | * like |
Alex Elder | 3f943d8 | 2010-04-15 18:17:34 +0000 | [diff] [blame] | 733 | * v binary search stopped here |
| 734 | * x + 1 ... | x | x + 1 | x ... | x |
| 735 | * ^ but we want to locate this spot |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 736 | * or |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 737 | * <---------> less than scan distance |
Alex Elder | 3f943d8 | 2010-04-15 18:17:34 +0000 | [diff] [blame] | 738 | * x + 1 ... | x ... | x - 1 | x |
| 739 | * ^ we want to locate this spot |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 740 | */ |
| 741 | stop_on_cycle = last_half_cycle; |
| 742 | if ((error = xlog_find_cycle_start(log, bp, first_blk, |
| 743 | &head_blk, last_half_cycle))) |
| 744 | goto bp_err; |
| 745 | } |
| 746 | |
| 747 | /* |
| 748 | * Now validate the answer. Scan back some number of maximum possible |
| 749 | * blocks and make sure each one has the expected cycle number. The |
| 750 | * maximum is determined by the total possible amount of buffering |
| 751 | * in the in-core log. The following number can be made tighter if |
| 752 | * we actually look at the block size of the filesystem. |
| 753 | */ |
| 754 | num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log); |
| 755 | if (head_blk >= num_scan_bblks) { |
| 756 | /* |
| 757 | * We are guaranteed that the entire check can be performed |
| 758 | * in one buffer. |
| 759 | */ |
| 760 | start_blk = head_blk - num_scan_bblks; |
| 761 | if ((error = xlog_find_verify_cycle(log, |
| 762 | start_blk, num_scan_bblks, |
| 763 | stop_on_cycle, &new_blk))) |
| 764 | goto bp_err; |
| 765 | if (new_blk != -1) |
| 766 | head_blk = new_blk; |
| 767 | } else { /* need to read 2 parts of log */ |
| 768 | /* |
| 769 | * We are going to scan backwards in the log in two parts. |
| 770 | * First we scan the physical end of the log. In this part |
| 771 | * of the log, we are looking for blocks with cycle number |
| 772 | * last_half_cycle - 1. |
| 773 | * If we find one, then we know that the log starts there, as |
| 774 | * we've found a hole that didn't get written in going around |
| 775 | * the end of the physical log. The simple case for this is |
| 776 | * x + 1 ... | x ... | x - 1 | x |
| 777 | * <---------> less than scan distance |
| 778 | * If all of the blocks at the end of the log have cycle number |
| 779 | * last_half_cycle, then we check the blocks at the start of |
| 780 | * the log looking for occurrences of last_half_cycle. If we |
| 781 | * find one, then our current estimate for the location of the |
| 782 | * first occurrence of last_half_cycle is wrong and we move |
| 783 | * back to the hole we've found. This case looks like |
| 784 | * x + 1 ... | x | x + 1 | x ... |
| 785 | * ^ binary search stopped here |
| 786 | * Another case we need to handle that only occurs in 256k |
| 787 | * logs is |
| 788 | * x + 1 ... | x ... | x+1 | x ... |
| 789 | * ^ binary search stops here |
| 790 | * In a 256k log, the scan at the end of the log will see the |
| 791 | * x + 1 blocks. We need to skip past those since that is |
| 792 | * certainly not the head of the log. By searching for |
| 793 | * last_half_cycle-1 we accomplish that. |
| 794 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 795 | ASSERT(head_blk <= INT_MAX && |
Alex Elder | 3f943d8 | 2010-04-15 18:17:34 +0000 | [diff] [blame] | 796 | (xfs_daddr_t) num_scan_bblks >= head_blk); |
| 797 | start_blk = log_bbnum - (num_scan_bblks - head_blk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 798 | if ((error = xlog_find_verify_cycle(log, start_blk, |
| 799 | num_scan_bblks - (int)head_blk, |
| 800 | (stop_on_cycle - 1), &new_blk))) |
| 801 | goto bp_err; |
| 802 | if (new_blk != -1) { |
| 803 | head_blk = new_blk; |
Alex Elder | 9db127e | 2010-04-15 18:17:26 +0000 | [diff] [blame] | 804 | goto validate_head; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 805 | } |
| 806 | |
| 807 | /* |
| 808 | * Scan beginning of log now. The last part of the physical |
| 809 | * log is good. This scan needs to verify that it doesn't find |
| 810 | * the last_half_cycle. |
| 811 | */ |
| 812 | start_blk = 0; |
| 813 | ASSERT(head_blk <= INT_MAX); |
| 814 | if ((error = xlog_find_verify_cycle(log, |
| 815 | start_blk, (int)head_blk, |
| 816 | stop_on_cycle, &new_blk))) |
| 817 | goto bp_err; |
| 818 | if (new_blk != -1) |
| 819 | head_blk = new_blk; |
| 820 | } |
| 821 | |
Alex Elder | 9db127e | 2010-04-15 18:17:26 +0000 | [diff] [blame] | 822 | validate_head: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 823 | /* |
| 824 | * Now we need to make sure head_blk is not pointing to a block in |
| 825 | * the middle of a log record. |
| 826 | */ |
| 827 | num_scan_bblks = XLOG_REC_SHIFT(log); |
| 828 | if (head_blk >= num_scan_bblks) { |
| 829 | start_blk = head_blk - num_scan_bblks; /* don't read head_blk */ |
| 830 | |
| 831 | /* start ptr at last block ptr before head_blk */ |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 832 | error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0); |
| 833 | if (error == 1) |
| 834 | error = -EIO; |
| 835 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 836 | goto bp_err; |
| 837 | } else { |
| 838 | start_blk = 0; |
| 839 | ASSERT(head_blk <= INT_MAX); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 840 | error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0); |
| 841 | if (error < 0) |
| 842 | goto bp_err; |
| 843 | if (error == 1) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 844 | /* We hit the beginning of the log during our search */ |
Alex Elder | 3f943d8 | 2010-04-15 18:17:34 +0000 | [diff] [blame] | 845 | start_blk = log_bbnum - (num_scan_bblks - head_blk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 846 | new_blk = log_bbnum; |
| 847 | ASSERT(start_blk <= INT_MAX && |
| 848 | (xfs_daddr_t) log_bbnum-start_blk >= 0); |
| 849 | ASSERT(head_blk <= INT_MAX); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 850 | error = xlog_find_verify_log_record(log, start_blk, |
| 851 | &new_blk, (int)head_blk); |
| 852 | if (error == 1) |
| 853 | error = -EIO; |
| 854 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 855 | goto bp_err; |
| 856 | if (new_blk != log_bbnum) |
| 857 | head_blk = new_blk; |
| 858 | } else if (error) |
| 859 | goto bp_err; |
| 860 | } |
| 861 | |
| 862 | xlog_put_bp(bp); |
| 863 | if (head_blk == log_bbnum) |
| 864 | *return_head_blk = 0; |
| 865 | else |
| 866 | *return_head_blk = head_blk; |
| 867 | /* |
| 868 | * When returning here, we have a good block number. Bad block |
| 869 | * means that during a previous crash, we didn't have a clean break |
| 870 | * from cycle number N to cycle number N-1. In this case, we need |
| 871 | * to find the first block with cycle number N-1. |
| 872 | */ |
| 873 | return 0; |
| 874 | |
| 875 | bp_err: |
| 876 | xlog_put_bp(bp); |
| 877 | |
| 878 | if (error) |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 879 | xfs_warn(log->l_mp, "failed to find log head"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 880 | return error; |
| 881 | } |
| 882 | |
| 883 | /* |
Brian Foster | eed6b46 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 884 | * Seek backwards in the log for log record headers. |
| 885 | * |
| 886 | * Given a starting log block, walk backwards until we find the provided number |
| 887 | * of records or hit the provided tail block. The return value is the number of |
| 888 | * records encountered or a negative error code. The log block and buffer |
| 889 | * pointer of the last record seen are returned in rblk and rhead respectively. |
| 890 | */ |
| 891 | STATIC int |
| 892 | xlog_rseek_logrec_hdr( |
| 893 | struct xlog *log, |
| 894 | xfs_daddr_t head_blk, |
| 895 | xfs_daddr_t tail_blk, |
| 896 | int count, |
| 897 | struct xfs_buf *bp, |
| 898 | xfs_daddr_t *rblk, |
| 899 | struct xlog_rec_header **rhead, |
| 900 | bool *wrapped) |
| 901 | { |
| 902 | int i; |
| 903 | int error; |
| 904 | int found = 0; |
| 905 | char *offset = NULL; |
| 906 | xfs_daddr_t end_blk; |
| 907 | |
| 908 | *wrapped = false; |
| 909 | |
| 910 | /* |
| 911 | * Walk backwards from the head block until we hit the tail or the first |
| 912 | * block in the log. |
| 913 | */ |
| 914 | end_blk = head_blk > tail_blk ? tail_blk : 0; |
| 915 | for (i = (int) head_blk - 1; i >= end_blk; i--) { |
| 916 | error = xlog_bread(log, i, 1, bp, &offset); |
| 917 | if (error) |
| 918 | goto out_error; |
| 919 | |
| 920 | if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) { |
| 921 | *rblk = i; |
| 922 | *rhead = (struct xlog_rec_header *) offset; |
| 923 | if (++found == count) |
| 924 | break; |
| 925 | } |
| 926 | } |
| 927 | |
| 928 | /* |
| 929 | * If we haven't hit the tail block or the log record header count, |
| 930 | * start looking again from the end of the physical log. Note that |
| 931 | * callers can pass head == tail if the tail is not yet known. |
| 932 | */ |
| 933 | if (tail_blk >= head_blk && found != count) { |
| 934 | for (i = log->l_logBBsize - 1; i >= (int) tail_blk; i--) { |
| 935 | error = xlog_bread(log, i, 1, bp, &offset); |
| 936 | if (error) |
| 937 | goto out_error; |
| 938 | |
| 939 | if (*(__be32 *)offset == |
| 940 | cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) { |
| 941 | *wrapped = true; |
| 942 | *rblk = i; |
| 943 | *rhead = (struct xlog_rec_header *) offset; |
| 944 | if (++found == count) |
| 945 | break; |
| 946 | } |
| 947 | } |
| 948 | } |
| 949 | |
| 950 | return found; |
| 951 | |
| 952 | out_error: |
| 953 | return error; |
| 954 | } |
| 955 | |
| 956 | /* |
Brian Foster | 7088c41 | 2016-01-05 07:40:16 +1100 | [diff] [blame] | 957 | * Seek forward in the log for log record headers. |
| 958 | * |
| 959 | * Given head and tail blocks, walk forward from the tail block until we find |
| 960 | * the provided number of records or hit the head block. The return value is the |
| 961 | * number of records encountered or a negative error code. The log block and |
| 962 | * buffer pointer of the last record seen are returned in rblk and rhead |
| 963 | * respectively. |
| 964 | */ |
| 965 | STATIC int |
| 966 | xlog_seek_logrec_hdr( |
| 967 | struct xlog *log, |
| 968 | xfs_daddr_t head_blk, |
| 969 | xfs_daddr_t tail_blk, |
| 970 | int count, |
| 971 | struct xfs_buf *bp, |
| 972 | xfs_daddr_t *rblk, |
| 973 | struct xlog_rec_header **rhead, |
| 974 | bool *wrapped) |
| 975 | { |
| 976 | int i; |
| 977 | int error; |
| 978 | int found = 0; |
| 979 | char *offset = NULL; |
| 980 | xfs_daddr_t end_blk; |
| 981 | |
| 982 | *wrapped = false; |
| 983 | |
| 984 | /* |
| 985 | * Walk forward from the tail block until we hit the head or the last |
| 986 | * block in the log. |
| 987 | */ |
| 988 | end_blk = head_blk > tail_blk ? head_blk : log->l_logBBsize - 1; |
| 989 | for (i = (int) tail_blk; i <= end_blk; i++) { |
| 990 | error = xlog_bread(log, i, 1, bp, &offset); |
| 991 | if (error) |
| 992 | goto out_error; |
| 993 | |
| 994 | if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) { |
| 995 | *rblk = i; |
| 996 | *rhead = (struct xlog_rec_header *) offset; |
| 997 | if (++found == count) |
| 998 | break; |
| 999 | } |
| 1000 | } |
| 1001 | |
| 1002 | /* |
| 1003 | * If we haven't hit the head block or the log record header count, |
| 1004 | * start looking again from the start of the physical log. |
| 1005 | */ |
| 1006 | if (tail_blk > head_blk && found != count) { |
| 1007 | for (i = 0; i < (int) head_blk; i++) { |
| 1008 | error = xlog_bread(log, i, 1, bp, &offset); |
| 1009 | if (error) |
| 1010 | goto out_error; |
| 1011 | |
| 1012 | if (*(__be32 *)offset == |
| 1013 | cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) { |
| 1014 | *wrapped = true; |
| 1015 | *rblk = i; |
| 1016 | *rhead = (struct xlog_rec_header *) offset; |
| 1017 | if (++found == count) |
| 1018 | break; |
| 1019 | } |
| 1020 | } |
| 1021 | } |
| 1022 | |
| 1023 | return found; |
| 1024 | |
| 1025 | out_error: |
| 1026 | return error; |
| 1027 | } |
| 1028 | |
| 1029 | /* |
| 1030 | * Check the log tail for torn writes. This is required when torn writes are |
| 1031 | * detected at the head and the head had to be walked back to a previous record. |
| 1032 | * The tail of the previous record must now be verified to ensure the torn |
| 1033 | * writes didn't corrupt the previous tail. |
| 1034 | * |
| 1035 | * Return an error if CRC verification fails as recovery cannot proceed. |
| 1036 | */ |
| 1037 | STATIC int |
| 1038 | xlog_verify_tail( |
| 1039 | struct xlog *log, |
| 1040 | xfs_daddr_t head_blk, |
| 1041 | xfs_daddr_t tail_blk) |
| 1042 | { |
| 1043 | struct xlog_rec_header *thead; |
| 1044 | struct xfs_buf *bp; |
| 1045 | xfs_daddr_t first_bad; |
| 1046 | int count; |
| 1047 | int error = 0; |
| 1048 | bool wrapped; |
| 1049 | xfs_daddr_t tmp_head; |
| 1050 | |
| 1051 | bp = xlog_get_bp(log, 1); |
| 1052 | if (!bp) |
| 1053 | return -ENOMEM; |
| 1054 | |
| 1055 | /* |
| 1056 | * Seek XLOG_MAX_ICLOGS + 1 records past the current tail record to get |
| 1057 | * a temporary head block that points after the last possible |
| 1058 | * concurrently written record of the tail. |
| 1059 | */ |
| 1060 | count = xlog_seek_logrec_hdr(log, head_blk, tail_blk, |
| 1061 | XLOG_MAX_ICLOGS + 1, bp, &tmp_head, &thead, |
| 1062 | &wrapped); |
| 1063 | if (count < 0) { |
| 1064 | error = count; |
| 1065 | goto out; |
| 1066 | } |
| 1067 | |
| 1068 | /* |
| 1069 | * If the call above didn't find XLOG_MAX_ICLOGS + 1 records, we ran |
| 1070 | * into the actual log head. tmp_head points to the start of the record |
| 1071 | * so update it to the actual head block. |
| 1072 | */ |
| 1073 | if (count < XLOG_MAX_ICLOGS + 1) |
| 1074 | tmp_head = head_blk; |
| 1075 | |
| 1076 | /* |
| 1077 | * We now have a tail and temporary head block that covers at least |
| 1078 | * XLOG_MAX_ICLOGS records from the tail. We need to verify that these |
| 1079 | * records were completely written. Run a CRC verification pass from |
| 1080 | * tail to head and return the result. |
| 1081 | */ |
| 1082 | error = xlog_do_recovery_pass(log, tmp_head, tail_blk, |
| 1083 | XLOG_RECOVER_CRCPASS, &first_bad); |
| 1084 | |
| 1085 | out: |
| 1086 | xlog_put_bp(bp); |
| 1087 | return error; |
| 1088 | } |
| 1089 | |
| 1090 | /* |
| 1091 | * Detect and trim torn writes from the head of the log. |
| 1092 | * |
| 1093 | * Storage without sector atomicity guarantees can result in torn writes in the |
| 1094 | * log in the event of a crash. Our only means to detect this scenario is via |
| 1095 | * CRC verification. While we can't always be certain that CRC verification |
| 1096 | * failure is due to a torn write vs. an unrelated corruption, we do know that |
| 1097 | * only a certain number (XLOG_MAX_ICLOGS) of log records can be written out at |
| 1098 | * one time. Therefore, CRC verify up to XLOG_MAX_ICLOGS records at the head of |
| 1099 | * the log and treat failures in this range as torn writes as a matter of |
| 1100 | * policy. In the event of CRC failure, the head is walked back to the last good |
| 1101 | * record in the log and the tail is updated from that record and verified. |
| 1102 | */ |
| 1103 | STATIC int |
| 1104 | xlog_verify_head( |
| 1105 | struct xlog *log, |
| 1106 | xfs_daddr_t *head_blk, /* in/out: unverified head */ |
| 1107 | xfs_daddr_t *tail_blk, /* out: tail block */ |
| 1108 | struct xfs_buf *bp, |
| 1109 | xfs_daddr_t *rhead_blk, /* start blk of last record */ |
| 1110 | struct xlog_rec_header **rhead, /* ptr to last record */ |
| 1111 | bool *wrapped) /* last rec. wraps phys. log */ |
| 1112 | { |
| 1113 | struct xlog_rec_header *tmp_rhead; |
| 1114 | struct xfs_buf *tmp_bp; |
| 1115 | xfs_daddr_t first_bad; |
| 1116 | xfs_daddr_t tmp_rhead_blk; |
| 1117 | int found; |
| 1118 | int error; |
| 1119 | bool tmp_wrapped; |
| 1120 | |
| 1121 | /* |
Brian Foster | 82ff6cc | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1122 | * Check the head of the log for torn writes. Search backwards from the |
| 1123 | * head until we hit the tail or the maximum number of log record I/Os |
| 1124 | * that could have been in flight at one time. Use a temporary buffer so |
| 1125 | * we don't trash the rhead/bp pointers from the caller. |
Brian Foster | 7088c41 | 2016-01-05 07:40:16 +1100 | [diff] [blame] | 1126 | */ |
| 1127 | tmp_bp = xlog_get_bp(log, 1); |
| 1128 | if (!tmp_bp) |
| 1129 | return -ENOMEM; |
| 1130 | error = xlog_rseek_logrec_hdr(log, *head_blk, *tail_blk, |
| 1131 | XLOG_MAX_ICLOGS, tmp_bp, &tmp_rhead_blk, |
| 1132 | &tmp_rhead, &tmp_wrapped); |
| 1133 | xlog_put_bp(tmp_bp); |
| 1134 | if (error < 0) |
| 1135 | return error; |
| 1136 | |
| 1137 | /* |
| 1138 | * Now run a CRC verification pass over the records starting at the |
| 1139 | * block found above to the current head. If a CRC failure occurs, the |
| 1140 | * log block of the first bad record is saved in first_bad. |
| 1141 | */ |
| 1142 | error = xlog_do_recovery_pass(log, *head_blk, tmp_rhead_blk, |
| 1143 | XLOG_RECOVER_CRCPASS, &first_bad); |
| 1144 | if (error == -EFSBADCRC) { |
| 1145 | /* |
| 1146 | * We've hit a potential torn write. Reset the error and warn |
| 1147 | * about it. |
| 1148 | */ |
| 1149 | error = 0; |
| 1150 | xfs_warn(log->l_mp, |
| 1151 | "Torn write (CRC failure) detected at log block 0x%llx. Truncating head block from 0x%llx.", |
| 1152 | first_bad, *head_blk); |
| 1153 | |
| 1154 | /* |
| 1155 | * Get the header block and buffer pointer for the last good |
| 1156 | * record before the bad record. |
| 1157 | * |
| 1158 | * Note that xlog_find_tail() clears the blocks at the new head |
| 1159 | * (i.e., the records with invalid CRC) if the cycle number |
| 1160 | * matches the the current cycle. |
| 1161 | */ |
| 1162 | found = xlog_rseek_logrec_hdr(log, first_bad, *tail_blk, 1, bp, |
| 1163 | rhead_blk, rhead, wrapped); |
| 1164 | if (found < 0) |
| 1165 | return found; |
| 1166 | if (found == 0) /* XXX: right thing to do here? */ |
| 1167 | return -EIO; |
| 1168 | |
| 1169 | /* |
| 1170 | * Reset the head block to the starting block of the first bad |
| 1171 | * log record and set the tail block based on the last good |
| 1172 | * record. |
| 1173 | * |
| 1174 | * Bail out if the updated head/tail match as this indicates |
| 1175 | * possible corruption outside of the acceptable |
| 1176 | * (XLOG_MAX_ICLOGS) range. This is a job for xfs_repair... |
| 1177 | */ |
| 1178 | *head_blk = first_bad; |
| 1179 | *tail_blk = BLOCK_LSN(be64_to_cpu((*rhead)->h_tail_lsn)); |
| 1180 | if (*head_blk == *tail_blk) { |
| 1181 | ASSERT(0); |
| 1182 | return 0; |
| 1183 | } |
| 1184 | |
| 1185 | /* |
| 1186 | * Now verify the tail based on the updated head. This is |
| 1187 | * required because the torn writes trimmed from the head could |
| 1188 | * have been written over the tail of a previous record. Return |
| 1189 | * any errors since recovery cannot proceed if the tail is |
| 1190 | * corrupt. |
| 1191 | * |
| 1192 | * XXX: This leaves a gap in truly robust protection from torn |
| 1193 | * writes in the log. If the head is behind the tail, the tail |
| 1194 | * pushes forward to create some space and then a crash occurs |
| 1195 | * causing the writes into the previous record's tail region to |
| 1196 | * tear, log recovery isn't able to recover. |
| 1197 | * |
| 1198 | * How likely is this to occur? If possible, can we do something |
| 1199 | * more intelligent here? Is it safe to push the tail forward if |
| 1200 | * we can determine that the tail is within the range of the |
| 1201 | * torn write (e.g., the kernel can only overwrite the tail if |
| 1202 | * it has actually been pushed forward)? Alternatively, could we |
| 1203 | * somehow prevent this condition at runtime? |
| 1204 | */ |
| 1205 | error = xlog_verify_tail(log, *head_blk, *tail_blk); |
| 1206 | } |
| 1207 | |
| 1208 | return error; |
| 1209 | } |
| 1210 | |
| 1211 | /* |
Brian Foster | 65b99a0 | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1212 | * Check whether the head of the log points to an unmount record. In other |
| 1213 | * words, determine whether the log is clean. If so, update the in-core state |
| 1214 | * appropriately. |
| 1215 | */ |
| 1216 | static int |
| 1217 | xlog_check_unmount_rec( |
| 1218 | struct xlog *log, |
| 1219 | xfs_daddr_t *head_blk, |
| 1220 | xfs_daddr_t *tail_blk, |
| 1221 | struct xlog_rec_header *rhead, |
| 1222 | xfs_daddr_t rhead_blk, |
| 1223 | struct xfs_buf *bp, |
| 1224 | bool *clean) |
| 1225 | { |
| 1226 | struct xlog_op_header *op_head; |
| 1227 | xfs_daddr_t umount_data_blk; |
| 1228 | xfs_daddr_t after_umount_blk; |
| 1229 | int hblks; |
| 1230 | int error; |
| 1231 | char *offset; |
| 1232 | |
| 1233 | *clean = false; |
| 1234 | |
| 1235 | /* |
| 1236 | * Look for unmount record. If we find it, then we know there was a |
| 1237 | * clean unmount. Since 'i' could be the last block in the physical |
| 1238 | * log, we convert to a log block before comparing to the head_blk. |
| 1239 | * |
| 1240 | * Save the current tail lsn to use to pass to xlog_clear_stale_blocks() |
| 1241 | * below. We won't want to clear the unmount record if there is one, so |
| 1242 | * we pass the lsn of the unmount record rather than the block after it. |
| 1243 | */ |
| 1244 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
| 1245 | int h_size = be32_to_cpu(rhead->h_size); |
| 1246 | int h_version = be32_to_cpu(rhead->h_version); |
| 1247 | |
| 1248 | if ((h_version & XLOG_VERSION_2) && |
| 1249 | (h_size > XLOG_HEADER_CYCLE_SIZE)) { |
| 1250 | hblks = h_size / XLOG_HEADER_CYCLE_SIZE; |
| 1251 | if (h_size % XLOG_HEADER_CYCLE_SIZE) |
| 1252 | hblks++; |
| 1253 | } else { |
| 1254 | hblks = 1; |
| 1255 | } |
| 1256 | } else { |
| 1257 | hblks = 1; |
| 1258 | } |
| 1259 | after_umount_blk = rhead_blk + hblks + BTOBB(be32_to_cpu(rhead->h_len)); |
| 1260 | after_umount_blk = do_mod(after_umount_blk, log->l_logBBsize); |
| 1261 | if (*head_blk == after_umount_blk && |
| 1262 | be32_to_cpu(rhead->h_num_logops) == 1) { |
| 1263 | umount_data_blk = rhead_blk + hblks; |
| 1264 | umount_data_blk = do_mod(umount_data_blk, log->l_logBBsize); |
| 1265 | error = xlog_bread(log, umount_data_blk, 1, bp, &offset); |
| 1266 | if (error) |
| 1267 | return error; |
| 1268 | |
| 1269 | op_head = (struct xlog_op_header *)offset; |
| 1270 | if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) { |
| 1271 | /* |
| 1272 | * Set tail and last sync so that newly written log |
| 1273 | * records will point recovery to after the current |
| 1274 | * unmount record. |
| 1275 | */ |
| 1276 | xlog_assign_atomic_lsn(&log->l_tail_lsn, |
| 1277 | log->l_curr_cycle, after_umount_blk); |
| 1278 | xlog_assign_atomic_lsn(&log->l_last_sync_lsn, |
| 1279 | log->l_curr_cycle, after_umount_blk); |
| 1280 | *tail_blk = after_umount_blk; |
| 1281 | |
| 1282 | *clean = true; |
| 1283 | } |
| 1284 | } |
| 1285 | |
| 1286 | return 0; |
| 1287 | } |
| 1288 | |
Brian Foster | 717bc0e | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1289 | static void |
| 1290 | xlog_set_state( |
| 1291 | struct xlog *log, |
| 1292 | xfs_daddr_t head_blk, |
| 1293 | struct xlog_rec_header *rhead, |
| 1294 | xfs_daddr_t rhead_blk, |
| 1295 | bool bump_cycle) |
| 1296 | { |
| 1297 | /* |
| 1298 | * Reset log values according to the state of the log when we |
| 1299 | * crashed. In the case where head_blk == 0, we bump curr_cycle |
| 1300 | * one because the next write starts a new cycle rather than |
| 1301 | * continuing the cycle of the last good log record. At this |
| 1302 | * point we have guaranteed that all partial log records have been |
| 1303 | * accounted for. Therefore, we know that the last good log record |
| 1304 | * written was complete and ended exactly on the end boundary |
| 1305 | * of the physical log. |
| 1306 | */ |
| 1307 | log->l_prev_block = rhead_blk; |
| 1308 | log->l_curr_block = (int)head_blk; |
| 1309 | log->l_curr_cycle = be32_to_cpu(rhead->h_cycle); |
| 1310 | if (bump_cycle) |
| 1311 | log->l_curr_cycle++; |
| 1312 | atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn)); |
| 1313 | atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn)); |
| 1314 | xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle, |
| 1315 | BBTOB(log->l_curr_block)); |
| 1316 | xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle, |
| 1317 | BBTOB(log->l_curr_block)); |
| 1318 | } |
| 1319 | |
Brian Foster | 65b99a0 | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1320 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1321 | * Find the sync block number or the tail of the log. |
| 1322 | * |
| 1323 | * This will be the block number of the last record to have its |
| 1324 | * associated buffers synced to disk. Every log record header has |
| 1325 | * a sync lsn embedded in it. LSNs hold block numbers, so it is easy |
| 1326 | * to get a sync block number. The only concern is to figure out which |
| 1327 | * log record header to believe. |
| 1328 | * |
| 1329 | * The following algorithm uses the log record header with the largest |
| 1330 | * lsn. The entire log record does not need to be valid. We only care |
| 1331 | * that the header is valid. |
| 1332 | * |
| 1333 | * We could speed up search by using current head_blk buffer, but it is not |
| 1334 | * available. |
| 1335 | */ |
Eric Sandeen | 5d77c0d | 2009-11-19 15:52:00 +0000 | [diff] [blame] | 1336 | STATIC int |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1337 | xlog_find_tail( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 1338 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1339 | xfs_daddr_t *head_blk, |
Eric Sandeen | 65be605 | 2006-01-11 15:34:19 +1100 | [diff] [blame] | 1340 | xfs_daddr_t *tail_blk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1341 | { |
| 1342 | xlog_rec_header_t *rhead; |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 1343 | char *offset = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1344 | xfs_buf_t *bp; |
Brian Foster | 7088c41 | 2016-01-05 07:40:16 +1100 | [diff] [blame] | 1345 | int error; |
Brian Foster | 7088c41 | 2016-01-05 07:40:16 +1100 | [diff] [blame] | 1346 | xfs_daddr_t rhead_blk; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1347 | xfs_lsn_t tail_lsn; |
Brian Foster | eed6b46 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 1348 | bool wrapped = false; |
Brian Foster | 65b99a0 | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1349 | bool clean = false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1350 | |
| 1351 | /* |
| 1352 | * Find previous log record |
| 1353 | */ |
| 1354 | if ((error = xlog_find_head(log, head_blk))) |
| 1355 | return error; |
Brian Foster | 82ff6cc | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1356 | ASSERT(*head_blk < INT_MAX); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1357 | |
| 1358 | bp = xlog_get_bp(log, 1); |
| 1359 | if (!bp) |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1360 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1361 | if (*head_blk == 0) { /* special case */ |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1362 | error = xlog_bread(log, 0, 1, bp, &offset); |
| 1363 | if (error) |
Alex Elder | 9db127e | 2010-04-15 18:17:26 +0000 | [diff] [blame] | 1364 | goto done; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1365 | |
Christoph Hellwig | 03bea6f | 2007-10-12 10:58:05 +1000 | [diff] [blame] | 1366 | if (xlog_get_cycle(offset) == 0) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1367 | *tail_blk = 0; |
| 1368 | /* leave all other log inited values alone */ |
Alex Elder | 9db127e | 2010-04-15 18:17:26 +0000 | [diff] [blame] | 1369 | goto done; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1370 | } |
| 1371 | } |
| 1372 | |
| 1373 | /* |
Brian Foster | 82ff6cc | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1374 | * Search backwards through the log looking for the log record header |
| 1375 | * block. This wraps all the way back around to the head so something is |
| 1376 | * seriously wrong if we can't find it. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1377 | */ |
Brian Foster | 82ff6cc | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1378 | error = xlog_rseek_logrec_hdr(log, *head_blk, *head_blk, 1, bp, |
| 1379 | &rhead_blk, &rhead, &wrapped); |
| 1380 | if (error < 0) |
| 1381 | return error; |
| 1382 | if (!error) { |
| 1383 | xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__); |
| 1384 | return -EIO; |
| 1385 | } |
| 1386 | *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn)); |
| 1387 | |
| 1388 | /* |
Brian Foster | 717bc0e | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1389 | * Set the log state based on the current head record. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1390 | */ |
Brian Foster | 717bc0e | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1391 | xlog_set_state(log, *head_blk, rhead, rhead_blk, wrapped); |
Brian Foster | 65b99a0 | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1392 | tail_lsn = atomic64_read(&log->l_tail_lsn); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1393 | |
| 1394 | /* |
Brian Foster | 65b99a0 | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1395 | * Look for an unmount record at the head of the log. This sets the log |
| 1396 | * state to determine whether recovery is necessary. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1397 | */ |
Brian Foster | 65b99a0 | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1398 | error = xlog_check_unmount_rec(log, head_blk, tail_blk, rhead, |
| 1399 | rhead_blk, bp, &clean); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1400 | if (error) |
| 1401 | goto done; |
| 1402 | |
| 1403 | /* |
Brian Foster | 7f6aff3a | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1404 | * Verify the log head if the log is not clean (e.g., we have anything |
| 1405 | * but an unmount record at the head). This uses CRC verification to |
| 1406 | * detect and trim torn writes. If discovered, CRC failures are |
| 1407 | * considered torn writes and the log head is trimmed accordingly. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1408 | * |
Brian Foster | 7f6aff3a | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1409 | * Note that we can only run CRC verification when the log is dirty |
| 1410 | * because there's no guarantee that the log data behind an unmount |
| 1411 | * record is compatible with the current architecture. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1412 | */ |
Brian Foster | 7f6aff3a | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1413 | if (!clean) { |
| 1414 | xfs_daddr_t orig_head = *head_blk; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1415 | |
Brian Foster | 7f6aff3a | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1416 | error = xlog_verify_head(log, head_blk, tail_blk, bp, |
| 1417 | &rhead_blk, &rhead, &wrapped); |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1418 | if (error) |
Alex Elder | 9db127e | 2010-04-15 18:17:26 +0000 | [diff] [blame] | 1419 | goto done; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1420 | |
Brian Foster | 7f6aff3a | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1421 | /* update in-core state again if the head changed */ |
| 1422 | if (*head_blk != orig_head) { |
| 1423 | xlog_set_state(log, *head_blk, rhead, rhead_blk, |
| 1424 | wrapped); |
| 1425 | tail_lsn = atomic64_read(&log->l_tail_lsn); |
| 1426 | error = xlog_check_unmount_rec(log, head_blk, tail_blk, |
| 1427 | rhead, rhead_blk, bp, |
| 1428 | &clean); |
| 1429 | if (error) |
| 1430 | goto done; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1431 | } |
| 1432 | } |
| 1433 | |
| 1434 | /* |
Brian Foster | 65b99a0 | 2016-03-07 08:22:22 +1100 | [diff] [blame] | 1435 | * Note that the unmount was clean. If the unmount was not clean, we |
| 1436 | * need to know this to rebuild the superblock counters from the perag |
| 1437 | * headers if we have a filesystem using non-persistent counters. |
| 1438 | */ |
| 1439 | if (clean) |
| 1440 | log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1441 | |
| 1442 | /* |
| 1443 | * Make sure that there are no blocks in front of the head |
| 1444 | * with the same cycle number as the head. This can happen |
| 1445 | * because we allow multiple outstanding log writes concurrently, |
| 1446 | * and the later writes might make it out before earlier ones. |
| 1447 | * |
| 1448 | * We use the lsn from before modifying it so that we'll never |
| 1449 | * overwrite the unmount record after a clean unmount. |
| 1450 | * |
| 1451 | * Do this only if we are going to recover the filesystem |
| 1452 | * |
| 1453 | * NOTE: This used to say "if (!readonly)" |
| 1454 | * However on Linux, we can & do recover a read-only filesystem. |
| 1455 | * We only skip recovery if NORECOVERY is specified on mount, |
| 1456 | * in which case we would not be here. |
| 1457 | * |
| 1458 | * But... if the -device- itself is readonly, just skip this. |
| 1459 | * We can't recover this device anyway, so it won't matter. |
| 1460 | */ |
Alex Elder | 9db127e | 2010-04-15 18:17:26 +0000 | [diff] [blame] | 1461 | if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1462 | error = xlog_clear_stale_blocks(log, tail_lsn); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1463 | |
Alex Elder | 9db127e | 2010-04-15 18:17:26 +0000 | [diff] [blame] | 1464 | done: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1465 | xlog_put_bp(bp); |
| 1466 | |
| 1467 | if (error) |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 1468 | xfs_warn(log->l_mp, "failed to locate log tail"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1469 | return error; |
| 1470 | } |
| 1471 | |
| 1472 | /* |
| 1473 | * Is the log zeroed at all? |
| 1474 | * |
| 1475 | * The last binary search should be changed to perform an X block read |
| 1476 | * once X becomes small enough. You can then search linearly through |
| 1477 | * the X blocks. This will cut down on the number of reads we need to do. |
| 1478 | * |
| 1479 | * If the log is partially zeroed, this routine will pass back the blkno |
| 1480 | * of the first block with cycle number 0. It won't have a complete LR |
| 1481 | * preceding it. |
| 1482 | * |
| 1483 | * Return: |
| 1484 | * 0 => the log is completely written to |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1485 | * 1 => use *blk_no as the first block of the log |
| 1486 | * <0 => error has occurred |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1487 | */ |
David Chinner | a8272ce | 2007-11-23 16:28:09 +1100 | [diff] [blame] | 1488 | STATIC int |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1489 | xlog_find_zeroed( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 1490 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1491 | xfs_daddr_t *blk_no) |
| 1492 | { |
| 1493 | xfs_buf_t *bp; |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 1494 | char *offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1495 | uint first_cycle, last_cycle; |
| 1496 | xfs_daddr_t new_blk, last_blk, start_blk; |
| 1497 | xfs_daddr_t num_scan_bblks; |
| 1498 | int error, log_bbnum = log->l_logBBsize; |
| 1499 | |
Nathan Scott | 6fdf8cc | 2006-06-28 10:13:52 +1000 | [diff] [blame] | 1500 | *blk_no = 0; |
| 1501 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1502 | /* check totally zeroed log */ |
| 1503 | bp = xlog_get_bp(log, 1); |
| 1504 | if (!bp) |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1505 | return -ENOMEM; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1506 | error = xlog_bread(log, 0, 1, bp, &offset); |
| 1507 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1508 | goto bp_err; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1509 | |
Christoph Hellwig | 03bea6f | 2007-10-12 10:58:05 +1000 | [diff] [blame] | 1510 | first_cycle = xlog_get_cycle(offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1511 | if (first_cycle == 0) { /* completely zeroed log */ |
| 1512 | *blk_no = 0; |
| 1513 | xlog_put_bp(bp); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1514 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1515 | } |
| 1516 | |
| 1517 | /* check partially zeroed log */ |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1518 | error = xlog_bread(log, log_bbnum-1, 1, bp, &offset); |
| 1519 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1520 | goto bp_err; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1521 | |
Christoph Hellwig | 03bea6f | 2007-10-12 10:58:05 +1000 | [diff] [blame] | 1522 | last_cycle = xlog_get_cycle(offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1523 | if (last_cycle != 0) { /* log completely written to */ |
| 1524 | xlog_put_bp(bp); |
| 1525 | return 0; |
| 1526 | } else if (first_cycle != 1) { |
| 1527 | /* |
| 1528 | * If the cycle of the last block is zero, the cycle of |
| 1529 | * the first block must be 1. If it's not, maybe we're |
| 1530 | * not looking at a log... Bail out. |
| 1531 | */ |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 1532 | xfs_warn(log->l_mp, |
| 1533 | "Log inconsistent or not a log (last==0, first!=1)"); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1534 | error = -EINVAL; |
Eric Sandeen | 5d0a654 | 2013-07-31 20:32:30 -0500 | [diff] [blame] | 1535 | goto bp_err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1536 | } |
| 1537 | |
| 1538 | /* we have a partially zeroed log */ |
| 1539 | last_blk = log_bbnum-1; |
| 1540 | if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0))) |
| 1541 | goto bp_err; |
| 1542 | |
| 1543 | /* |
| 1544 | * Validate the answer. Because there is no way to guarantee that |
| 1545 | * the entire log is made up of log records which are the same size, |
| 1546 | * we scan over the defined maximum blocks. At this point, the maximum |
| 1547 | * is not chosen to mean anything special. XXXmiken |
| 1548 | */ |
| 1549 | num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log); |
| 1550 | ASSERT(num_scan_bblks <= INT_MAX); |
| 1551 | |
| 1552 | if (last_blk < num_scan_bblks) |
| 1553 | num_scan_bblks = last_blk; |
| 1554 | start_blk = last_blk - num_scan_bblks; |
| 1555 | |
| 1556 | /* |
| 1557 | * We search for any instances of cycle number 0 that occur before |
| 1558 | * our current estimate of the head. What we're trying to detect is |
| 1559 | * 1 ... | 0 | 1 | 0... |
| 1560 | * ^ binary search ends here |
| 1561 | */ |
| 1562 | if ((error = xlog_find_verify_cycle(log, start_blk, |
| 1563 | (int)num_scan_bblks, 0, &new_blk))) |
| 1564 | goto bp_err; |
| 1565 | if (new_blk != -1) |
| 1566 | last_blk = new_blk; |
| 1567 | |
| 1568 | /* |
| 1569 | * Potentially backup over partial log record write. We don't need |
| 1570 | * to search the end of the log because we know it is zero. |
| 1571 | */ |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1572 | error = xlog_find_verify_log_record(log, start_blk, &last_blk, 0); |
| 1573 | if (error == 1) |
| 1574 | error = -EIO; |
| 1575 | if (error) |
| 1576 | goto bp_err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1577 | |
| 1578 | *blk_no = last_blk; |
| 1579 | bp_err: |
| 1580 | xlog_put_bp(bp); |
| 1581 | if (error) |
| 1582 | return error; |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1583 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1584 | } |
| 1585 | |
| 1586 | /* |
| 1587 | * These are simple subroutines used by xlog_clear_stale_blocks() below |
| 1588 | * to initialize a buffer full of empty log record headers and write |
| 1589 | * them into the log. |
| 1590 | */ |
| 1591 | STATIC void |
| 1592 | xlog_add_record( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 1593 | struct xlog *log, |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 1594 | char *buf, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1595 | int cycle, |
| 1596 | int block, |
| 1597 | int tail_cycle, |
| 1598 | int tail_block) |
| 1599 | { |
| 1600 | xlog_rec_header_t *recp = (xlog_rec_header_t *)buf; |
| 1601 | |
| 1602 | memset(buf, 0, BBSIZE); |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 1603 | recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM); |
| 1604 | recp->h_cycle = cpu_to_be32(cycle); |
| 1605 | recp->h_version = cpu_to_be32( |
Eric Sandeen | 6211870 | 2008-03-06 13:44:28 +1100 | [diff] [blame] | 1606 | xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1); |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 1607 | recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block)); |
| 1608 | recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block)); |
| 1609 | recp->h_fmt = cpu_to_be32(XLOG_FMT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1610 | memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t)); |
| 1611 | } |
| 1612 | |
| 1613 | STATIC int |
| 1614 | xlog_write_log_records( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 1615 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1616 | int cycle, |
| 1617 | int start_block, |
| 1618 | int blocks, |
| 1619 | int tail_cycle, |
| 1620 | int tail_block) |
| 1621 | { |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 1622 | char *offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1623 | xfs_buf_t *bp; |
| 1624 | int balign, ealign; |
Alex Elder | 69ce58f | 2010-04-20 17:09:59 +1000 | [diff] [blame] | 1625 | int sectbb = log->l_sectBBsize; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1626 | int end_block = start_block + blocks; |
| 1627 | int bufblks; |
| 1628 | int error = 0; |
| 1629 | int i, j = 0; |
| 1630 | |
Alex Elder | 6881a22 | 2010-04-13 15:22:29 +1000 | [diff] [blame] | 1631 | /* |
| 1632 | * Greedily allocate a buffer big enough to handle the full |
| 1633 | * range of basic blocks to be written. If that fails, try |
| 1634 | * a smaller size. We need to be able to write at least a |
| 1635 | * log sector, or we're out of luck. |
| 1636 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1637 | bufblks = 1 << ffs(blocks); |
Dave Chinner | 81158e0 | 2012-04-27 19:45:22 +1000 | [diff] [blame] | 1638 | while (bufblks > log->l_logBBsize) |
| 1639 | bufblks >>= 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1640 | while (!(bp = xlog_get_bp(log, bufblks))) { |
| 1641 | bufblks >>= 1; |
Alex Elder | 69ce58f | 2010-04-20 17:09:59 +1000 | [diff] [blame] | 1642 | if (bufblks < sectbb) |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1643 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1644 | } |
| 1645 | |
| 1646 | /* We may need to do a read at the start to fill in part of |
| 1647 | * the buffer in the starting sector not covered by the first |
| 1648 | * write below. |
| 1649 | */ |
Alex Elder | 5c17f53 | 2010-04-13 15:22:48 +1000 | [diff] [blame] | 1650 | balign = round_down(start_block, sectbb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1651 | if (balign != start_block) { |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1652 | error = xlog_bread_noalign(log, start_block, 1, bp); |
| 1653 | if (error) |
| 1654 | goto out_put_bp; |
| 1655 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1656 | j = start_block - balign; |
| 1657 | } |
| 1658 | |
| 1659 | for (i = start_block; i < end_block; i += bufblks) { |
| 1660 | int bcount, endcount; |
| 1661 | |
| 1662 | bcount = min(bufblks, end_block - start_block); |
| 1663 | endcount = bcount - j; |
| 1664 | |
| 1665 | /* We may need to do a read at the end to fill in part of |
| 1666 | * the buffer in the final sector not covered by the write. |
| 1667 | * If this is the same sector as the above read, skip it. |
| 1668 | */ |
Alex Elder | 5c17f53 | 2010-04-13 15:22:48 +1000 | [diff] [blame] | 1669 | ealign = round_down(end_block, sectbb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1670 | if (j == 0 && (start_block + endcount > ealign)) { |
Chandra Seetharaman | 6292604 | 2011-07-22 23:40:15 +0000 | [diff] [blame] | 1671 | offset = bp->b_addr + BBTOB(ealign - start_block); |
Dave Chinner | 4439647 | 2011-04-21 09:34:27 +0000 | [diff] [blame] | 1672 | error = xlog_bread_offset(log, ealign, sectbb, |
| 1673 | bp, offset); |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1674 | if (error) |
| 1675 | break; |
| 1676 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1677 | } |
| 1678 | |
| 1679 | offset = xlog_align(log, start_block, endcount, bp); |
| 1680 | for (; j < endcount; j++) { |
| 1681 | xlog_add_record(log, offset, cycle, i+j, |
| 1682 | tail_cycle, tail_block); |
| 1683 | offset += BBSIZE; |
| 1684 | } |
| 1685 | error = xlog_bwrite(log, start_block, endcount, bp); |
| 1686 | if (error) |
| 1687 | break; |
| 1688 | start_block += endcount; |
| 1689 | j = 0; |
| 1690 | } |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1691 | |
| 1692 | out_put_bp: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1693 | xlog_put_bp(bp); |
| 1694 | return error; |
| 1695 | } |
| 1696 | |
| 1697 | /* |
| 1698 | * This routine is called to blow away any incomplete log writes out |
| 1699 | * in front of the log head. We do this so that we won't become confused |
| 1700 | * if we come up, write only a little bit more, and then crash again. |
| 1701 | * If we leave the partial log records out there, this situation could |
| 1702 | * cause us to think those partial writes are valid blocks since they |
| 1703 | * have the current cycle number. We get rid of them by overwriting them |
| 1704 | * with empty log records with the old cycle number rather than the |
| 1705 | * current one. |
| 1706 | * |
| 1707 | * The tail lsn is passed in rather than taken from |
| 1708 | * the log so that we will not write over the unmount record after a |
| 1709 | * clean unmount in a 512 block log. Doing so would leave the log without |
| 1710 | * any valid log records in it until a new one was written. If we crashed |
| 1711 | * during that time we would not be able to recover. |
| 1712 | */ |
| 1713 | STATIC int |
| 1714 | xlog_clear_stale_blocks( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 1715 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1716 | xfs_lsn_t tail_lsn) |
| 1717 | { |
| 1718 | int tail_cycle, head_cycle; |
| 1719 | int tail_block, head_block; |
| 1720 | int tail_distance, max_distance; |
| 1721 | int distance; |
| 1722 | int error; |
| 1723 | |
| 1724 | tail_cycle = CYCLE_LSN(tail_lsn); |
| 1725 | tail_block = BLOCK_LSN(tail_lsn); |
| 1726 | head_cycle = log->l_curr_cycle; |
| 1727 | head_block = log->l_curr_block; |
| 1728 | |
| 1729 | /* |
| 1730 | * Figure out the distance between the new head of the log |
| 1731 | * and the tail. We want to write over any blocks beyond the |
| 1732 | * head that we may have written just before the crash, but |
| 1733 | * we don't want to overwrite the tail of the log. |
| 1734 | */ |
| 1735 | if (head_cycle == tail_cycle) { |
| 1736 | /* |
| 1737 | * The tail is behind the head in the physical log, |
| 1738 | * so the distance from the head to the tail is the |
| 1739 | * distance from the head to the end of the log plus |
| 1740 | * the distance from the beginning of the log to the |
| 1741 | * tail. |
| 1742 | */ |
| 1743 | if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) { |
| 1744 | XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)", |
| 1745 | XFS_ERRLEVEL_LOW, log->l_mp); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1746 | return -EFSCORRUPTED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1747 | } |
| 1748 | tail_distance = tail_block + (log->l_logBBsize - head_block); |
| 1749 | } else { |
| 1750 | /* |
| 1751 | * The head is behind the tail in the physical log, |
| 1752 | * so the distance from the head to the tail is just |
| 1753 | * the tail block minus the head block. |
| 1754 | */ |
| 1755 | if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){ |
| 1756 | XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)", |
| 1757 | XFS_ERRLEVEL_LOW, log->l_mp); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1758 | return -EFSCORRUPTED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1759 | } |
| 1760 | tail_distance = tail_block - head_block; |
| 1761 | } |
| 1762 | |
| 1763 | /* |
| 1764 | * If the head is right up against the tail, we can't clear |
| 1765 | * anything. |
| 1766 | */ |
| 1767 | if (tail_distance <= 0) { |
| 1768 | ASSERT(tail_distance == 0); |
| 1769 | return 0; |
| 1770 | } |
| 1771 | |
| 1772 | max_distance = XLOG_TOTAL_REC_SHIFT(log); |
| 1773 | /* |
| 1774 | * Take the smaller of the maximum amount of outstanding I/O |
| 1775 | * we could have and the distance to the tail to clear out. |
| 1776 | * We take the smaller so that we don't overwrite the tail and |
| 1777 | * we don't waste all day writing from the head to the tail |
| 1778 | * for no reason. |
| 1779 | */ |
| 1780 | max_distance = MIN(max_distance, tail_distance); |
| 1781 | |
| 1782 | if ((head_block + max_distance) <= log->l_logBBsize) { |
| 1783 | /* |
| 1784 | * We can stomp all the blocks we need to without |
| 1785 | * wrapping around the end of the log. Just do it |
| 1786 | * in a single write. Use the cycle number of the |
| 1787 | * current cycle minus one so that the log will look like: |
| 1788 | * n ... | n - 1 ... |
| 1789 | */ |
| 1790 | error = xlog_write_log_records(log, (head_cycle - 1), |
| 1791 | head_block, max_distance, tail_cycle, |
| 1792 | tail_block); |
| 1793 | if (error) |
| 1794 | return error; |
| 1795 | } else { |
| 1796 | /* |
| 1797 | * We need to wrap around the end of the physical log in |
| 1798 | * order to clear all the blocks. Do it in two separate |
| 1799 | * I/Os. The first write should be from the head to the |
| 1800 | * end of the physical log, and it should use the current |
| 1801 | * cycle number minus one just like above. |
| 1802 | */ |
| 1803 | distance = log->l_logBBsize - head_block; |
| 1804 | error = xlog_write_log_records(log, (head_cycle - 1), |
| 1805 | head_block, distance, tail_cycle, |
| 1806 | tail_block); |
| 1807 | |
| 1808 | if (error) |
| 1809 | return error; |
| 1810 | |
| 1811 | /* |
| 1812 | * Now write the blocks at the start of the physical log. |
| 1813 | * This writes the remainder of the blocks we want to clear. |
| 1814 | * It uses the current cycle number since we're now on the |
| 1815 | * same cycle as the head so that we get: |
| 1816 | * n ... n ... | n - 1 ... |
| 1817 | * ^^^^^ blocks we're writing |
| 1818 | */ |
| 1819 | distance = max_distance - (log->l_logBBsize - head_block); |
| 1820 | error = xlog_write_log_records(log, head_cycle, 0, distance, |
| 1821 | tail_cycle, tail_block); |
| 1822 | if (error) |
| 1823 | return error; |
| 1824 | } |
| 1825 | |
| 1826 | return 0; |
| 1827 | } |
| 1828 | |
| 1829 | /****************************************************************************** |
| 1830 | * |
| 1831 | * Log recover routines |
| 1832 | * |
| 1833 | ****************************************************************************** |
| 1834 | */ |
| 1835 | |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1836 | /* |
Dave Chinner | a775ad7 | 2013-06-05 12:09:07 +1000 | [diff] [blame] | 1837 | * Sort the log items in the transaction. |
| 1838 | * |
| 1839 | * The ordering constraints are defined by the inode allocation and unlink |
| 1840 | * behaviour. The rules are: |
| 1841 | * |
| 1842 | * 1. Every item is only logged once in a given transaction. Hence it |
| 1843 | * represents the last logged state of the item. Hence ordering is |
| 1844 | * dependent on the order in which operations need to be performed so |
| 1845 | * required initial conditions are always met. |
| 1846 | * |
| 1847 | * 2. Cancelled buffers are recorded in pass 1 in a separate table and |
| 1848 | * there's nothing to replay from them so we can simply cull them |
| 1849 | * from the transaction. However, we can't do that until after we've |
| 1850 | * replayed all the other items because they may be dependent on the |
| 1851 | * cancelled buffer and replaying the cancelled buffer can remove it |
| 1852 | * form the cancelled buffer table. Hence they have tobe done last. |
| 1853 | * |
| 1854 | * 3. Inode allocation buffers must be replayed before inode items that |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 1855 | * read the buffer and replay changes into it. For filesystems using the |
| 1856 | * ICREATE transactions, this means XFS_LI_ICREATE objects need to get |
| 1857 | * treated the same as inode allocation buffers as they create and |
| 1858 | * initialise the buffers directly. |
Dave Chinner | a775ad7 | 2013-06-05 12:09:07 +1000 | [diff] [blame] | 1859 | * |
| 1860 | * 4. Inode unlink buffers must be replayed after inode items are replayed. |
| 1861 | * This ensures that inodes are completely flushed to the inode buffer |
| 1862 | * in a "free" state before we remove the unlinked inode list pointer. |
| 1863 | * |
| 1864 | * Hence the ordering needs to be inode allocation buffers first, inode items |
| 1865 | * second, inode unlink buffers third and cancelled buffers last. |
| 1866 | * |
| 1867 | * But there's a problem with that - we can't tell an inode allocation buffer |
| 1868 | * apart from a regular buffer, so we can't separate them. We can, however, |
| 1869 | * tell an inode unlink buffer from the others, and so we can separate them out |
| 1870 | * from all the other buffers and move them to last. |
| 1871 | * |
| 1872 | * Hence, 4 lists, in order from head to tail: |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 1873 | * - buffer_list for all buffers except cancelled/inode unlink buffers |
| 1874 | * - item_list for all non-buffer items |
| 1875 | * - inode_buffer_list for inode unlink buffers |
| 1876 | * - cancel_list for the cancelled buffers |
| 1877 | * |
| 1878 | * Note that we add objects to the tail of the lists so that first-to-last |
| 1879 | * ordering is preserved within the lists. Adding objects to the head of the |
| 1880 | * list means when we traverse from the head we walk them in last-to-first |
| 1881 | * order. For cancelled buffers and inode unlink buffers this doesn't matter, |
| 1882 | * but for all other items there may be specific ordering that we need to |
| 1883 | * preserve. |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1884 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1885 | STATIC int |
| 1886 | xlog_recover_reorder_trans( |
Mark Tinguely | ad223e6 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 1887 | struct xlog *log, |
| 1888 | struct xlog_recover *trans, |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 1889 | int pass) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1890 | { |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1891 | xlog_recover_item_t *item, *n; |
Mark Tinguely | 2a84108 | 2013-10-02 07:51:12 -0500 | [diff] [blame] | 1892 | int error = 0; |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1893 | LIST_HEAD(sort_list); |
Dave Chinner | a775ad7 | 2013-06-05 12:09:07 +1000 | [diff] [blame] | 1894 | LIST_HEAD(cancel_list); |
| 1895 | LIST_HEAD(buffer_list); |
| 1896 | LIST_HEAD(inode_buffer_list); |
| 1897 | LIST_HEAD(inode_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1898 | |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1899 | list_splice_init(&trans->r_itemq, &sort_list); |
| 1900 | list_for_each_entry_safe(item, n, &sort_list, ri_list) { |
Christoph Hellwig | 4e0d5f9 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 1901 | 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] | 1902 | |
| 1903 | switch (ITEM_TYPE(item)) { |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 1904 | case XFS_LI_ICREATE: |
| 1905 | list_move_tail(&item->ri_list, &buffer_list); |
| 1906 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1907 | case XFS_LI_BUF: |
Dave Chinner | a775ad7 | 2013-06-05 12:09:07 +1000 | [diff] [blame] | 1908 | if (buf_f->blf_flags & XFS_BLF_CANCEL) { |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 1909 | trace_xfs_log_recover_item_reorder_head(log, |
| 1910 | trans, item, pass); |
Dave Chinner | a775ad7 | 2013-06-05 12:09:07 +1000 | [diff] [blame] | 1911 | list_move(&item->ri_list, &cancel_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1912 | break; |
| 1913 | } |
Dave Chinner | a775ad7 | 2013-06-05 12:09:07 +1000 | [diff] [blame] | 1914 | if (buf_f->blf_flags & XFS_BLF_INODE_BUF) { |
| 1915 | list_move(&item->ri_list, &inode_buffer_list); |
| 1916 | break; |
| 1917 | } |
| 1918 | list_move_tail(&item->ri_list, &buffer_list); |
| 1919 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1920 | case XFS_LI_INODE: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1921 | case XFS_LI_DQUOT: |
| 1922 | case XFS_LI_QUOTAOFF: |
| 1923 | case XFS_LI_EFD: |
| 1924 | case XFS_LI_EFI: |
Darrick J. Wong | 9e88b5d | 2016-08-03 12:09:48 +1000 | [diff] [blame] | 1925 | case XFS_LI_RUI: |
| 1926 | case XFS_LI_RUD: |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 1927 | trace_xfs_log_recover_item_reorder_tail(log, |
| 1928 | trans, item, pass); |
Dave Chinner | a775ad7 | 2013-06-05 12:09:07 +1000 | [diff] [blame] | 1929 | list_move_tail(&item->ri_list, &inode_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1930 | break; |
| 1931 | default: |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 1932 | xfs_warn(log->l_mp, |
| 1933 | "%s: unrecognized type of log operation", |
| 1934 | __func__); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1935 | ASSERT(0); |
Mark Tinguely | 2a84108 | 2013-10-02 07:51:12 -0500 | [diff] [blame] | 1936 | /* |
| 1937 | * return the remaining items back to the transaction |
| 1938 | * item list so they can be freed in caller. |
| 1939 | */ |
| 1940 | if (!list_empty(&sort_list)) |
| 1941 | list_splice_init(&sort_list, &trans->r_itemq); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1942 | error = -EIO; |
Mark Tinguely | 2a84108 | 2013-10-02 07:51:12 -0500 | [diff] [blame] | 1943 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1944 | } |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1945 | } |
Mark Tinguely | 2a84108 | 2013-10-02 07:51:12 -0500 | [diff] [blame] | 1946 | out: |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1947 | ASSERT(list_empty(&sort_list)); |
Dave Chinner | a775ad7 | 2013-06-05 12:09:07 +1000 | [diff] [blame] | 1948 | if (!list_empty(&buffer_list)) |
| 1949 | list_splice(&buffer_list, &trans->r_itemq); |
| 1950 | if (!list_empty(&inode_list)) |
| 1951 | list_splice_tail(&inode_list, &trans->r_itemq); |
| 1952 | if (!list_empty(&inode_buffer_list)) |
| 1953 | list_splice_tail(&inode_buffer_list, &trans->r_itemq); |
| 1954 | if (!list_empty(&cancel_list)) |
| 1955 | list_splice_tail(&cancel_list, &trans->r_itemq); |
Mark Tinguely | 2a84108 | 2013-10-02 07:51:12 -0500 | [diff] [blame] | 1956 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1957 | } |
| 1958 | |
| 1959 | /* |
| 1960 | * Build up the table of buf cancel records so that we don't replay |
| 1961 | * cancelled data in the second pass. For buffer records that are |
| 1962 | * not cancel records, there is nothing to do here so we just return. |
| 1963 | * |
| 1964 | * If we get a cancel record which is already in the table, this indicates |
| 1965 | * that the buffer was cancelled multiple times. In order to ensure |
| 1966 | * that during pass 2 we keep the record in the table until we reach its |
| 1967 | * last occurrence in the log, we keep a reference count in the cancel |
| 1968 | * record in the table to tell us how many times we expect to see this |
| 1969 | * record during the second pass. |
| 1970 | */ |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 1971 | STATIC int |
| 1972 | xlog_recover_buffer_pass1( |
Mark Tinguely | ad223e6 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 1973 | struct xlog *log, |
| 1974 | struct xlog_recover_item *item) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1975 | { |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 1976 | 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] | 1977 | struct list_head *bucket; |
| 1978 | struct xfs_buf_cancel *bcp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1979 | |
| 1980 | /* |
| 1981 | * If this isn't a cancel buffer item, then just return. |
| 1982 | */ |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 1983 | if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) { |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 1984 | trace_xfs_log_recover_buf_not_cancel(log, buf_f); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 1985 | return 0; |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 1986 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1987 | |
| 1988 | /* |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 1989 | * Insert an xfs_buf_cancel record into the hash table of them. |
| 1990 | * If there is already an identical record, bump its reference count. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1991 | */ |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 1992 | bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno); |
| 1993 | list_for_each_entry(bcp, bucket, bc_list) { |
| 1994 | if (bcp->bc_blkno == buf_f->blf_blkno && |
| 1995 | bcp->bc_len == buf_f->blf_len) { |
| 1996 | bcp->bc_refcount++; |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 1997 | trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 1998 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1999 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2000 | } |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 2001 | |
| 2002 | bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP); |
| 2003 | bcp->bc_blkno = buf_f->blf_blkno; |
| 2004 | bcp->bc_len = buf_f->blf_len; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2005 | bcp->bc_refcount = 1; |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 2006 | list_add_tail(&bcp->bc_list, bucket); |
| 2007 | |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 2008 | trace_xfs_log_recover_buf_cancel_add(log, buf_f); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2009 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2010 | } |
| 2011 | |
| 2012 | /* |
| 2013 | * Check to see whether the buffer being recovered has a corresponding |
Dave Chinner | 84a5b73 | 2013-08-27 08:10:53 +1000 | [diff] [blame] | 2014 | * entry in the buffer cancel record table. If it is, return the cancel |
| 2015 | * buffer structure to the caller. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2016 | */ |
Dave Chinner | 84a5b73 | 2013-08-27 08:10:53 +1000 | [diff] [blame] | 2017 | STATIC struct xfs_buf_cancel * |
| 2018 | xlog_peek_buffer_cancelled( |
Mark Tinguely | ad223e6 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 2019 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2020 | xfs_daddr_t blkno, |
| 2021 | uint len, |
| 2022 | ushort flags) |
| 2023 | { |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 2024 | struct list_head *bucket; |
| 2025 | struct xfs_buf_cancel *bcp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2026 | |
Dave Chinner | 84a5b73 | 2013-08-27 08:10:53 +1000 | [diff] [blame] | 2027 | if (!log->l_buf_cancel_table) { |
| 2028 | /* empty table means no cancelled buffers in the log */ |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 2029 | ASSERT(!(flags & XFS_BLF_CANCEL)); |
Dave Chinner | 84a5b73 | 2013-08-27 08:10:53 +1000 | [diff] [blame] | 2030 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2031 | } |
| 2032 | |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 2033 | bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno); |
| 2034 | list_for_each_entry(bcp, bucket, bc_list) { |
| 2035 | if (bcp->bc_blkno == blkno && bcp->bc_len == len) |
Dave Chinner | 84a5b73 | 2013-08-27 08:10:53 +1000 | [diff] [blame] | 2036 | return bcp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2037 | } |
| 2038 | |
| 2039 | /* |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 2040 | * We didn't find a corresponding entry in the table, so return 0 so |
| 2041 | * that the buffer is NOT cancelled. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2042 | */ |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 2043 | ASSERT(!(flags & XFS_BLF_CANCEL)); |
Dave Chinner | 84a5b73 | 2013-08-27 08:10:53 +1000 | [diff] [blame] | 2044 | return NULL; |
| 2045 | } |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 2046 | |
Dave Chinner | 84a5b73 | 2013-08-27 08:10:53 +1000 | [diff] [blame] | 2047 | /* |
| 2048 | * If the buffer is being cancelled then return 1 so that it will be cancelled, |
| 2049 | * otherwise return 0. If the buffer is actually a buffer cancel item |
| 2050 | * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the |
| 2051 | * table and remove it from the table if this is the last reference. |
| 2052 | * |
| 2053 | * We remove the cancel record from the table when we encounter its last |
| 2054 | * occurrence in the log so that if the same buffer is re-used again after its |
| 2055 | * last cancellation we actually replay the changes made at that point. |
| 2056 | */ |
| 2057 | STATIC int |
| 2058 | xlog_check_buffer_cancelled( |
| 2059 | struct xlog *log, |
| 2060 | xfs_daddr_t blkno, |
| 2061 | uint len, |
| 2062 | ushort flags) |
| 2063 | { |
| 2064 | struct xfs_buf_cancel *bcp; |
| 2065 | |
| 2066 | bcp = xlog_peek_buffer_cancelled(log, blkno, len, flags); |
| 2067 | if (!bcp) |
| 2068 | return 0; |
| 2069 | |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 2070 | /* |
| 2071 | * We've go a match, so return 1 so that the recovery of this buffer |
| 2072 | * is cancelled. If this buffer is actually a buffer cancel log |
| 2073 | * item, then decrement the refcount on the one in the table and |
| 2074 | * remove it if this is the last reference. |
| 2075 | */ |
| 2076 | if (flags & XFS_BLF_CANCEL) { |
| 2077 | if (--bcp->bc_refcount == 0) { |
| 2078 | list_del(&bcp->bc_list); |
| 2079 | kmem_free(bcp); |
| 2080 | } |
| 2081 | } |
| 2082 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2083 | } |
| 2084 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2085 | /* |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2086 | * Perform recovery for a buffer full of inodes. In these buffers, the only |
| 2087 | * data which should be recovered is that which corresponds to the |
| 2088 | * di_next_unlinked pointers in the on disk inode structures. The rest of the |
| 2089 | * data for the inodes is always logged through the inodes themselves rather |
| 2090 | * than the inode buffer and is recovered in xlog_recover_inode_pass2(). |
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 | * The only time when buffers full of inodes are fully recovered is when the |
| 2093 | * buffer is full of newly allocated inodes. In this case the buffer will |
| 2094 | * not be marked as an inode buffer and so will be sent to |
| 2095 | * xlog_recover_do_reg_buffer() below during recovery. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2096 | */ |
| 2097 | STATIC int |
| 2098 | xlog_recover_do_inode_buffer( |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2099 | struct xfs_mount *mp, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2100 | xlog_recover_item_t *item, |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2101 | struct xfs_buf *bp, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2102 | xfs_buf_log_format_t *buf_f) |
| 2103 | { |
| 2104 | int i; |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2105 | int item_index = 0; |
| 2106 | int bit = 0; |
| 2107 | int nbits = 0; |
| 2108 | int reg_buf_offset = 0; |
| 2109 | int reg_buf_bytes = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2110 | int next_unlinked_offset; |
| 2111 | int inodes_per_buf; |
| 2112 | xfs_agino_t *logged_nextp; |
| 2113 | xfs_agino_t *buffer_nextp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2114 | |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 2115 | trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f); |
Dave Chinner | 9222a9c | 2013-06-12 12:19:06 +1000 | [diff] [blame] | 2116 | |
| 2117 | /* |
| 2118 | * Post recovery validation only works properly on CRC enabled |
| 2119 | * filesystems. |
| 2120 | */ |
| 2121 | if (xfs_sb_version_hascrc(&mp->m_sb)) |
| 2122 | bp->b_ops = &xfs_inode_buf_ops; |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 2123 | |
Dave Chinner | aa0e883 | 2012-04-23 15:58:52 +1000 | [diff] [blame] | 2124 | 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] | 2125 | for (i = 0; i < inodes_per_buf; i++) { |
| 2126 | next_unlinked_offset = (i * mp->m_sb.sb_inodesize) + |
| 2127 | offsetof(xfs_dinode_t, di_next_unlinked); |
| 2128 | |
| 2129 | while (next_unlinked_offset >= |
| 2130 | (reg_buf_offset + reg_buf_bytes)) { |
| 2131 | /* |
| 2132 | * The next di_next_unlinked field is beyond |
| 2133 | * the current logged region. Find the next |
| 2134 | * logged region that contains or is beyond |
| 2135 | * the current di_next_unlinked field. |
| 2136 | */ |
| 2137 | bit += nbits; |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2138 | bit = xfs_next_bit(buf_f->blf_data_map, |
| 2139 | buf_f->blf_map_size, bit); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2140 | |
| 2141 | /* |
| 2142 | * If there are no more logged regions in the |
| 2143 | * buffer, then we're done. |
| 2144 | */ |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2145 | if (bit == -1) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2146 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2147 | |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2148 | nbits = xfs_contig_bits(buf_f->blf_data_map, |
| 2149 | buf_f->blf_map_size, bit); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2150 | ASSERT(nbits > 0); |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 2151 | reg_buf_offset = bit << XFS_BLF_SHIFT; |
| 2152 | reg_buf_bytes = nbits << XFS_BLF_SHIFT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2153 | item_index++; |
| 2154 | } |
| 2155 | |
| 2156 | /* |
| 2157 | * If the current logged region starts after the current |
| 2158 | * di_next_unlinked field, then move on to the next |
| 2159 | * di_next_unlinked field. |
| 2160 | */ |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2161 | if (next_unlinked_offset < reg_buf_offset) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2162 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2163 | |
| 2164 | ASSERT(item->ri_buf[item_index].i_addr != NULL); |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 2165 | ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0); |
Dave Chinner | aa0e883 | 2012-04-23 15:58:52 +1000 | [diff] [blame] | 2166 | ASSERT((reg_buf_offset + reg_buf_bytes) <= |
| 2167 | BBTOB(bp->b_io_length)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2168 | |
| 2169 | /* |
| 2170 | * The current logged region contains a copy of the |
| 2171 | * current di_next_unlinked field. Extract its value |
| 2172 | * and copy it to the buffer copy. |
| 2173 | */ |
Christoph Hellwig | 4e0d5f9 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 2174 | logged_nextp = item->ri_buf[item_index].i_addr + |
| 2175 | next_unlinked_offset - reg_buf_offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2176 | if (unlikely(*logged_nextp == 0)) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2177 | xfs_alert(mp, |
| 2178 | "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). " |
| 2179 | "Trying to replay bad (0) inode di_next_unlinked field.", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2180 | item, bp); |
| 2181 | XFS_ERROR_REPORT("xlog_recover_do_inode_buf", |
| 2182 | XFS_ERRLEVEL_LOW, mp); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 2183 | return -EFSCORRUPTED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2184 | } |
| 2185 | |
Christoph Hellwig | 88ee2df | 2015-06-22 09:44:29 +1000 | [diff] [blame] | 2186 | buffer_nextp = xfs_buf_offset(bp, next_unlinked_offset); |
Tim Shimmin | 87c199c | 2006-06-09 14:56:16 +1000 | [diff] [blame] | 2187 | *buffer_nextp = *logged_nextp; |
Dave Chinner | 0a32c26 | 2013-06-05 12:09:08 +1000 | [diff] [blame] | 2188 | |
| 2189 | /* |
| 2190 | * If necessary, recalculate the CRC in the on-disk inode. We |
| 2191 | * have to leave the inode in a consistent state for whoever |
| 2192 | * reads it next.... |
| 2193 | */ |
Christoph Hellwig | 88ee2df | 2015-06-22 09:44:29 +1000 | [diff] [blame] | 2194 | xfs_dinode_calc_crc(mp, |
Dave Chinner | 0a32c26 | 2013-06-05 12:09:08 +1000 | [diff] [blame] | 2195 | xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize)); |
| 2196 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2197 | } |
| 2198 | |
| 2199 | return 0; |
| 2200 | } |
| 2201 | |
| 2202 | /* |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2203 | * V5 filesystems know the age of the buffer on disk being recovered. We can |
| 2204 | * have newer objects on disk than we are replaying, and so for these cases we |
| 2205 | * don't want to replay the current change as that will make the buffer contents |
| 2206 | * temporarily invalid on disk. |
| 2207 | * |
| 2208 | * The magic number might not match the buffer type we are going to recover |
| 2209 | * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags. Hence |
| 2210 | * extract the LSN of the existing object in the buffer based on it's current |
| 2211 | * magic number. If we don't recognise the magic number in the buffer, then |
| 2212 | * return a LSN of -1 so that the caller knows it was an unrecognised block and |
| 2213 | * so can recover the buffer. |
Dave Chinner | 566055d | 2013-09-24 16:01:16 +1000 | [diff] [blame] | 2214 | * |
| 2215 | * Note: we cannot rely solely on magic number matches to determine that the |
| 2216 | * buffer has a valid LSN - we also need to verify that it belongs to this |
| 2217 | * filesystem, so we need to extract the object's LSN and compare it to that |
| 2218 | * which we read from the superblock. If the UUIDs don't match, then we've got a |
| 2219 | * stale metadata block from an old filesystem instance that we need to recover |
| 2220 | * over the top of. |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2221 | */ |
| 2222 | static xfs_lsn_t |
| 2223 | xlog_recover_get_buf_lsn( |
| 2224 | struct xfs_mount *mp, |
| 2225 | struct xfs_buf *bp) |
| 2226 | { |
| 2227 | __uint32_t magic32; |
| 2228 | __uint16_t magic16; |
| 2229 | __uint16_t magicda; |
| 2230 | void *blk = bp->b_addr; |
Dave Chinner | 566055d | 2013-09-24 16:01:16 +1000 | [diff] [blame] | 2231 | uuid_t *uuid; |
| 2232 | xfs_lsn_t lsn = -1; |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2233 | |
| 2234 | /* v4 filesystems always recover immediately */ |
| 2235 | if (!xfs_sb_version_hascrc(&mp->m_sb)) |
| 2236 | goto recover_immediately; |
| 2237 | |
| 2238 | magic32 = be32_to_cpu(*(__be32 *)blk); |
| 2239 | switch (magic32) { |
| 2240 | case XFS_ABTB_CRC_MAGIC: |
| 2241 | case XFS_ABTC_CRC_MAGIC: |
| 2242 | case XFS_ABTB_MAGIC: |
| 2243 | case XFS_ABTC_MAGIC: |
Darrick J. Wong | a650e8f | 2016-08-03 12:17:11 +1000 | [diff] [blame] | 2244 | case XFS_RMAP_CRC_MAGIC: |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2245 | case XFS_IBT_CRC_MAGIC: |
Dave Chinner | 566055d | 2013-09-24 16:01:16 +1000 | [diff] [blame] | 2246 | case XFS_IBT_MAGIC: { |
| 2247 | struct xfs_btree_block *btb = blk; |
| 2248 | |
| 2249 | lsn = be64_to_cpu(btb->bb_u.s.bb_lsn); |
| 2250 | uuid = &btb->bb_u.s.bb_uuid; |
| 2251 | break; |
| 2252 | } |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2253 | case XFS_BMAP_CRC_MAGIC: |
Dave Chinner | 566055d | 2013-09-24 16:01:16 +1000 | [diff] [blame] | 2254 | case XFS_BMAP_MAGIC: { |
| 2255 | struct xfs_btree_block *btb = blk; |
| 2256 | |
| 2257 | lsn = be64_to_cpu(btb->bb_u.l.bb_lsn); |
| 2258 | uuid = &btb->bb_u.l.bb_uuid; |
| 2259 | break; |
| 2260 | } |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2261 | case XFS_AGF_MAGIC: |
Dave Chinner | 566055d | 2013-09-24 16:01:16 +1000 | [diff] [blame] | 2262 | lsn = be64_to_cpu(((struct xfs_agf *)blk)->agf_lsn); |
| 2263 | uuid = &((struct xfs_agf *)blk)->agf_uuid; |
| 2264 | break; |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2265 | case XFS_AGFL_MAGIC: |
Dave Chinner | 566055d | 2013-09-24 16:01:16 +1000 | [diff] [blame] | 2266 | lsn = be64_to_cpu(((struct xfs_agfl *)blk)->agfl_lsn); |
| 2267 | uuid = &((struct xfs_agfl *)blk)->agfl_uuid; |
| 2268 | break; |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2269 | case XFS_AGI_MAGIC: |
Dave Chinner | 566055d | 2013-09-24 16:01:16 +1000 | [diff] [blame] | 2270 | lsn = be64_to_cpu(((struct xfs_agi *)blk)->agi_lsn); |
| 2271 | uuid = &((struct xfs_agi *)blk)->agi_uuid; |
| 2272 | break; |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2273 | case XFS_SYMLINK_MAGIC: |
Dave Chinner | 566055d | 2013-09-24 16:01:16 +1000 | [diff] [blame] | 2274 | lsn = be64_to_cpu(((struct xfs_dsymlink_hdr *)blk)->sl_lsn); |
| 2275 | uuid = &((struct xfs_dsymlink_hdr *)blk)->sl_uuid; |
| 2276 | break; |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2277 | case XFS_DIR3_BLOCK_MAGIC: |
| 2278 | case XFS_DIR3_DATA_MAGIC: |
| 2279 | case XFS_DIR3_FREE_MAGIC: |
Dave Chinner | 566055d | 2013-09-24 16:01:16 +1000 | [diff] [blame] | 2280 | lsn = be64_to_cpu(((struct xfs_dir3_blk_hdr *)blk)->lsn); |
| 2281 | uuid = &((struct xfs_dir3_blk_hdr *)blk)->uuid; |
| 2282 | break; |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2283 | case XFS_ATTR3_RMT_MAGIC: |
Dave Chinner | e3c32ee | 2015-07-29 11:48:01 +1000 | [diff] [blame] | 2284 | /* |
| 2285 | * Remote attr blocks are written synchronously, rather than |
| 2286 | * being logged. That means they do not contain a valid LSN |
| 2287 | * (i.e. transactionally ordered) in them, and hence any time we |
| 2288 | * see a buffer to replay over the top of a remote attribute |
| 2289 | * block we should simply do so. |
| 2290 | */ |
| 2291 | goto recover_immediately; |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2292 | case XFS_SB_MAGIC: |
Dave Chinner | fcfbe2c | 2015-08-19 10:31:54 +1000 | [diff] [blame] | 2293 | /* |
| 2294 | * superblock uuids are magic. We may or may not have a |
| 2295 | * sb_meta_uuid on disk, but it will be set in the in-core |
| 2296 | * superblock. We set the uuid pointer for verification |
| 2297 | * according to the superblock feature mask to ensure we check |
| 2298 | * the relevant UUID in the superblock. |
| 2299 | */ |
Dave Chinner | 566055d | 2013-09-24 16:01:16 +1000 | [diff] [blame] | 2300 | lsn = be64_to_cpu(((struct xfs_dsb *)blk)->sb_lsn); |
Dave Chinner | fcfbe2c | 2015-08-19 10:31:54 +1000 | [diff] [blame] | 2301 | if (xfs_sb_version_hasmetauuid(&mp->m_sb)) |
| 2302 | uuid = &((struct xfs_dsb *)blk)->sb_meta_uuid; |
| 2303 | else |
| 2304 | uuid = &((struct xfs_dsb *)blk)->sb_uuid; |
Dave Chinner | 566055d | 2013-09-24 16:01:16 +1000 | [diff] [blame] | 2305 | break; |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2306 | default: |
| 2307 | break; |
| 2308 | } |
| 2309 | |
Dave Chinner | 566055d | 2013-09-24 16:01:16 +1000 | [diff] [blame] | 2310 | if (lsn != (xfs_lsn_t)-1) { |
Dave Chinner | fcfbe2c | 2015-08-19 10:31:54 +1000 | [diff] [blame] | 2311 | if (!uuid_equal(&mp->m_sb.sb_meta_uuid, uuid)) |
Dave Chinner | 566055d | 2013-09-24 16:01:16 +1000 | [diff] [blame] | 2312 | goto recover_immediately; |
| 2313 | return lsn; |
| 2314 | } |
| 2315 | |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2316 | magicda = be16_to_cpu(((struct xfs_da_blkinfo *)blk)->magic); |
| 2317 | switch (magicda) { |
| 2318 | case XFS_DIR3_LEAF1_MAGIC: |
| 2319 | case XFS_DIR3_LEAFN_MAGIC: |
| 2320 | case XFS_DA3_NODE_MAGIC: |
Dave Chinner | 566055d | 2013-09-24 16:01:16 +1000 | [diff] [blame] | 2321 | lsn = be64_to_cpu(((struct xfs_da3_blkinfo *)blk)->lsn); |
| 2322 | uuid = &((struct xfs_da3_blkinfo *)blk)->uuid; |
| 2323 | break; |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2324 | default: |
| 2325 | break; |
| 2326 | } |
| 2327 | |
Dave Chinner | 566055d | 2013-09-24 16:01:16 +1000 | [diff] [blame] | 2328 | if (lsn != (xfs_lsn_t)-1) { |
| 2329 | if (!uuid_equal(&mp->m_sb.sb_uuid, uuid)) |
| 2330 | goto recover_immediately; |
| 2331 | return lsn; |
| 2332 | } |
| 2333 | |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2334 | /* |
| 2335 | * We do individual object checks on dquot and inode buffers as they |
| 2336 | * have their own individual LSN records. Also, we could have a stale |
| 2337 | * buffer here, so we have to at least recognise these buffer types. |
| 2338 | * |
| 2339 | * A notd complexity here is inode unlinked list processing - it logs |
| 2340 | * the inode directly in the buffer, but we don't know which inodes have |
| 2341 | * been modified, and there is no global buffer LSN. Hence we need to |
| 2342 | * recover all inode buffer types immediately. This problem will be |
| 2343 | * fixed by logical logging of the unlinked list modifications. |
| 2344 | */ |
| 2345 | magic16 = be16_to_cpu(*(__be16 *)blk); |
| 2346 | switch (magic16) { |
| 2347 | case XFS_DQUOT_MAGIC: |
| 2348 | case XFS_DINODE_MAGIC: |
| 2349 | goto recover_immediately; |
| 2350 | default: |
| 2351 | break; |
| 2352 | } |
| 2353 | |
| 2354 | /* unknown buffer contents, recover immediately */ |
| 2355 | |
| 2356 | recover_immediately: |
| 2357 | return (xfs_lsn_t)-1; |
| 2358 | |
| 2359 | } |
| 2360 | |
| 2361 | /* |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2362 | * Validate the recovered buffer is of the correct type and attach the |
| 2363 | * appropriate buffer operations to them for writeback. Magic numbers are in a |
| 2364 | * few places: |
| 2365 | * the first 16 bits of the buffer (inode buffer, dquot buffer), |
| 2366 | * the first 32 bits of the buffer (most blocks), |
| 2367 | * inside a struct xfs_da_blkinfo at the start of the buffer. |
| 2368 | */ |
| 2369 | static void |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2370 | xlog_recover_validate_buf_type( |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2371 | struct xfs_mount *mp, |
| 2372 | struct xfs_buf *bp, |
Brian Foster | 22db9af | 2016-09-26 08:32:07 +1000 | [diff] [blame] | 2373 | xfs_buf_log_format_t *buf_f, |
| 2374 | xfs_lsn_t current_lsn) |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2375 | { |
| 2376 | struct xfs_da_blkinfo *info = bp->b_addr; |
| 2377 | __uint32_t magic32; |
| 2378 | __uint16_t magic16; |
| 2379 | __uint16_t magicda; |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2380 | char *warnmsg = NULL; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2381 | |
Dave Chinner | 67dc288 | 2014-08-04 12:43:06 +1000 | [diff] [blame] | 2382 | /* |
| 2383 | * We can only do post recovery validation on items on CRC enabled |
| 2384 | * fielsystems as we need to know when the buffer was written to be able |
| 2385 | * to determine if we should have replayed the item. If we replay old |
| 2386 | * metadata over a newer buffer, then it will enter a temporarily |
| 2387 | * inconsistent state resulting in verification failures. Hence for now |
| 2388 | * just avoid the verification stage for non-crc filesystems |
| 2389 | */ |
| 2390 | if (!xfs_sb_version_hascrc(&mp->m_sb)) |
| 2391 | return; |
| 2392 | |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2393 | magic32 = be32_to_cpu(*(__be32 *)bp->b_addr); |
| 2394 | magic16 = be16_to_cpu(*(__be16*)bp->b_addr); |
| 2395 | magicda = be16_to_cpu(info->magic); |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2396 | switch (xfs_blft_from_flags(buf_f)) { |
| 2397 | case XFS_BLFT_BTREE_BUF: |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2398 | switch (magic32) { |
| 2399 | case XFS_ABTB_CRC_MAGIC: |
| 2400 | case XFS_ABTC_CRC_MAGIC: |
| 2401 | case XFS_ABTB_MAGIC: |
| 2402 | case XFS_ABTC_MAGIC: |
| 2403 | bp->b_ops = &xfs_allocbt_buf_ops; |
| 2404 | break; |
| 2405 | case XFS_IBT_CRC_MAGIC: |
Brian Foster | aafc3c2 | 2014-04-24 16:00:52 +1000 | [diff] [blame] | 2406 | case XFS_FIBT_CRC_MAGIC: |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2407 | case XFS_IBT_MAGIC: |
Brian Foster | aafc3c2 | 2014-04-24 16:00:52 +1000 | [diff] [blame] | 2408 | case XFS_FIBT_MAGIC: |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2409 | bp->b_ops = &xfs_inobt_buf_ops; |
| 2410 | break; |
| 2411 | case XFS_BMAP_CRC_MAGIC: |
| 2412 | case XFS_BMAP_MAGIC: |
| 2413 | bp->b_ops = &xfs_bmbt_buf_ops; |
| 2414 | break; |
Darrick J. Wong | a650e8f | 2016-08-03 12:17:11 +1000 | [diff] [blame] | 2415 | case XFS_RMAP_CRC_MAGIC: |
| 2416 | bp->b_ops = &xfs_rmapbt_buf_ops; |
| 2417 | break; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2418 | default: |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2419 | warnmsg = "Bad btree block magic!"; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2420 | break; |
| 2421 | } |
| 2422 | break; |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2423 | case XFS_BLFT_AGF_BUF: |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2424 | if (magic32 != XFS_AGF_MAGIC) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2425 | warnmsg = "Bad AGF block magic!"; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2426 | break; |
| 2427 | } |
| 2428 | bp->b_ops = &xfs_agf_buf_ops; |
| 2429 | break; |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2430 | case XFS_BLFT_AGFL_BUF: |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2431 | if (magic32 != XFS_AGFL_MAGIC) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2432 | warnmsg = "Bad AGFL block magic!"; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2433 | break; |
| 2434 | } |
| 2435 | bp->b_ops = &xfs_agfl_buf_ops; |
| 2436 | break; |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2437 | case XFS_BLFT_AGI_BUF: |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2438 | if (magic32 != XFS_AGI_MAGIC) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2439 | warnmsg = "Bad AGI block magic!"; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2440 | break; |
| 2441 | } |
| 2442 | bp->b_ops = &xfs_agi_buf_ops; |
| 2443 | break; |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2444 | case XFS_BLFT_UDQUOT_BUF: |
| 2445 | case XFS_BLFT_PDQUOT_BUF: |
| 2446 | case XFS_BLFT_GDQUOT_BUF: |
Dave Chinner | 123887e | 2013-04-30 21:39:33 +1000 | [diff] [blame] | 2447 | #ifdef CONFIG_XFS_QUOTA |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2448 | if (magic16 != XFS_DQUOT_MAGIC) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2449 | warnmsg = "Bad DQUOT block magic!"; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2450 | break; |
| 2451 | } |
| 2452 | bp->b_ops = &xfs_dquot_buf_ops; |
Dave Chinner | 123887e | 2013-04-30 21:39:33 +1000 | [diff] [blame] | 2453 | #else |
| 2454 | xfs_alert(mp, |
| 2455 | "Trying to recover dquots without QUOTA support built in!"); |
| 2456 | ASSERT(0); |
| 2457 | #endif |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2458 | break; |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2459 | case XFS_BLFT_DINO_BUF: |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2460 | if (magic16 != XFS_DINODE_MAGIC) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2461 | warnmsg = "Bad INODE block magic!"; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2462 | break; |
| 2463 | } |
| 2464 | bp->b_ops = &xfs_inode_buf_ops; |
| 2465 | break; |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2466 | case XFS_BLFT_SYMLINK_BUF: |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2467 | if (magic32 != XFS_SYMLINK_MAGIC) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2468 | warnmsg = "Bad symlink block magic!"; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2469 | break; |
| 2470 | } |
| 2471 | bp->b_ops = &xfs_symlink_buf_ops; |
| 2472 | break; |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2473 | case XFS_BLFT_DIR_BLOCK_BUF: |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2474 | if (magic32 != XFS_DIR2_BLOCK_MAGIC && |
| 2475 | magic32 != XFS_DIR3_BLOCK_MAGIC) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2476 | warnmsg = "Bad dir block magic!"; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2477 | break; |
| 2478 | } |
| 2479 | bp->b_ops = &xfs_dir3_block_buf_ops; |
| 2480 | break; |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2481 | case XFS_BLFT_DIR_DATA_BUF: |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2482 | if (magic32 != XFS_DIR2_DATA_MAGIC && |
| 2483 | magic32 != XFS_DIR3_DATA_MAGIC) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2484 | warnmsg = "Bad dir data magic!"; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2485 | break; |
| 2486 | } |
| 2487 | bp->b_ops = &xfs_dir3_data_buf_ops; |
| 2488 | break; |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2489 | case XFS_BLFT_DIR_FREE_BUF: |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2490 | if (magic32 != XFS_DIR2_FREE_MAGIC && |
| 2491 | magic32 != XFS_DIR3_FREE_MAGIC) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2492 | warnmsg = "Bad dir3 free magic!"; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2493 | break; |
| 2494 | } |
| 2495 | bp->b_ops = &xfs_dir3_free_buf_ops; |
| 2496 | break; |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2497 | case XFS_BLFT_DIR_LEAF1_BUF: |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2498 | if (magicda != XFS_DIR2_LEAF1_MAGIC && |
| 2499 | magicda != XFS_DIR3_LEAF1_MAGIC) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2500 | warnmsg = "Bad dir leaf1 magic!"; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2501 | break; |
| 2502 | } |
| 2503 | bp->b_ops = &xfs_dir3_leaf1_buf_ops; |
| 2504 | break; |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2505 | case XFS_BLFT_DIR_LEAFN_BUF: |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2506 | if (magicda != XFS_DIR2_LEAFN_MAGIC && |
| 2507 | magicda != XFS_DIR3_LEAFN_MAGIC) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2508 | warnmsg = "Bad dir leafn magic!"; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2509 | break; |
| 2510 | } |
| 2511 | bp->b_ops = &xfs_dir3_leafn_buf_ops; |
| 2512 | break; |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2513 | case XFS_BLFT_DA_NODE_BUF: |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2514 | if (magicda != XFS_DA_NODE_MAGIC && |
| 2515 | magicda != XFS_DA3_NODE_MAGIC) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2516 | warnmsg = "Bad da node magic!"; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2517 | break; |
| 2518 | } |
| 2519 | bp->b_ops = &xfs_da3_node_buf_ops; |
| 2520 | break; |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2521 | case XFS_BLFT_ATTR_LEAF_BUF: |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2522 | if (magicda != XFS_ATTR_LEAF_MAGIC && |
| 2523 | magicda != XFS_ATTR3_LEAF_MAGIC) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2524 | warnmsg = "Bad attr leaf magic!"; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2525 | break; |
| 2526 | } |
| 2527 | bp->b_ops = &xfs_attr3_leaf_buf_ops; |
| 2528 | break; |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2529 | case XFS_BLFT_ATTR_RMT_BUF: |
Dave Chinner | cab09a8 | 2013-04-30 21:39:36 +1000 | [diff] [blame] | 2530 | if (magic32 != XFS_ATTR3_RMT_MAGIC) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2531 | warnmsg = "Bad attr remote magic!"; |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2532 | break; |
| 2533 | } |
| 2534 | bp->b_ops = &xfs_attr3_rmt_buf_ops; |
| 2535 | break; |
Dave Chinner | 04a1e6c | 2013-04-03 16:11:31 +1100 | [diff] [blame] | 2536 | case XFS_BLFT_SB_BUF: |
| 2537 | if (magic32 != XFS_SB_MAGIC) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2538 | warnmsg = "Bad SB block magic!"; |
Dave Chinner | 04a1e6c | 2013-04-03 16:11:31 +1100 | [diff] [blame] | 2539 | break; |
| 2540 | } |
| 2541 | bp->b_ops = &xfs_sb_buf_ops; |
| 2542 | break; |
Dave Chinner | f67ca6e | 2016-02-09 16:41:31 +1100 | [diff] [blame] | 2543 | #ifdef CONFIG_XFS_RT |
| 2544 | case XFS_BLFT_RTBITMAP_BUF: |
| 2545 | case XFS_BLFT_RTSUMMARY_BUF: |
Dave Chinner | bf85e09 | 2016-02-09 16:41:45 +1100 | [diff] [blame] | 2546 | /* no magic numbers for verification of RT buffers */ |
| 2547 | bp->b_ops = &xfs_rtbuf_ops; |
Dave Chinner | f67ca6e | 2016-02-09 16:41:31 +1100 | [diff] [blame] | 2548 | break; |
| 2549 | #endif /* CONFIG_XFS_RT */ |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2550 | default: |
Dave Chinner | 61fe135 | 2013-04-03 16:11:30 +1100 | [diff] [blame] | 2551 | xfs_warn(mp, "Unknown buffer type %d!", |
| 2552 | xfs_blft_from_flags(buf_f)); |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2553 | break; |
| 2554 | } |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2555 | |
| 2556 | /* |
Brian Foster | 60a4a22 | 2016-09-26 08:34:27 +1000 | [diff] [blame^] | 2557 | * Nothing else to do in the case of a NULL current LSN as this means |
| 2558 | * the buffer is more recent than the change in the log and will be |
| 2559 | * skipped. |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2560 | */ |
Brian Foster | 60a4a22 | 2016-09-26 08:34:27 +1000 | [diff] [blame^] | 2561 | if (current_lsn == NULLCOMMITLSN) |
| 2562 | return; |
| 2563 | |
| 2564 | if (warnmsg) { |
Brian Foster | 040c52c | 2016-09-26 08:32:50 +1000 | [diff] [blame] | 2565 | xfs_warn(mp, warnmsg); |
| 2566 | ASSERT(0); |
| 2567 | } |
Brian Foster | 60a4a22 | 2016-09-26 08:34:27 +1000 | [diff] [blame^] | 2568 | |
| 2569 | /* |
| 2570 | * We must update the metadata LSN of the buffer as it is written out to |
| 2571 | * ensure that older transactions never replay over this one and corrupt |
| 2572 | * the buffer. This can occur if log recovery is interrupted at some |
| 2573 | * point after the current transaction completes, at which point a |
| 2574 | * subsequent mount starts recovery from the beginning. |
| 2575 | * |
| 2576 | * Write verifiers update the metadata LSN from log items attached to |
| 2577 | * the buffer. Therefore, initialize a bli purely to carry the LSN to |
| 2578 | * the verifier. We'll clean it up in our ->iodone() callback. |
| 2579 | */ |
| 2580 | if (bp->b_ops) { |
| 2581 | struct xfs_buf_log_item *bip; |
| 2582 | |
| 2583 | ASSERT(!bp->b_iodone || bp->b_iodone == xlog_recover_iodone); |
| 2584 | bp->b_iodone = xlog_recover_iodone; |
| 2585 | xfs_buf_item_init(bp, mp); |
| 2586 | bip = bp->b_fspriv; |
| 2587 | bip->bli_item.li_lsn = current_lsn; |
| 2588 | } |
Dave Chinner | d75afeb | 2013-04-03 16:11:29 +1100 | [diff] [blame] | 2589 | } |
| 2590 | |
| 2591 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2592 | * Perform a 'normal' buffer recovery. Each logged region of the |
| 2593 | * buffer should be copied over the corresponding region in the |
| 2594 | * given buffer. The bitmap in the buf log format structure indicates |
| 2595 | * where to place the logged data. |
| 2596 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2597 | STATIC void |
| 2598 | xlog_recover_do_reg_buffer( |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 2599 | struct xfs_mount *mp, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2600 | xlog_recover_item_t *item, |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2601 | struct xfs_buf *bp, |
Brian Foster | 22db9af | 2016-09-26 08:32:07 +1000 | [diff] [blame] | 2602 | xfs_buf_log_format_t *buf_f, |
| 2603 | xfs_lsn_t current_lsn) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2604 | { |
| 2605 | int i; |
| 2606 | int bit; |
| 2607 | int nbits; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2608 | int error; |
| 2609 | |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 2610 | trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f); |
| 2611 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2612 | bit = 0; |
| 2613 | i = 1; /* 0 is the buf format structure */ |
| 2614 | while (1) { |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2615 | bit = xfs_next_bit(buf_f->blf_data_map, |
| 2616 | buf_f->blf_map_size, bit); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2617 | if (bit == -1) |
| 2618 | break; |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2619 | nbits = xfs_contig_bits(buf_f->blf_data_map, |
| 2620 | buf_f->blf_map_size, bit); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2621 | ASSERT(nbits > 0); |
Christoph Hellwig | 4b80916 | 2007-08-16 15:37:36 +1000 | [diff] [blame] | 2622 | ASSERT(item->ri_buf[i].i_addr != NULL); |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 2623 | ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0); |
Dave Chinner | aa0e883 | 2012-04-23 15:58:52 +1000 | [diff] [blame] | 2624 | ASSERT(BBTOB(bp->b_io_length) >= |
| 2625 | ((uint)bit << XFS_BLF_SHIFT) + (nbits << XFS_BLF_SHIFT)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2626 | |
| 2627 | /* |
Dave Chinner | 709da6a | 2013-05-27 16:38:23 +1000 | [diff] [blame] | 2628 | * The dirty regions logged in the buffer, even though |
| 2629 | * contiguous, may span multiple chunks. This is because the |
| 2630 | * dirty region may span a physical page boundary in a buffer |
| 2631 | * and hence be split into two separate vectors for writing into |
| 2632 | * the log. Hence we need to trim nbits back to the length of |
| 2633 | * the current region being copied out of the log. |
| 2634 | */ |
| 2635 | if (item->ri_buf[i].i_len < (nbits << XFS_BLF_SHIFT)) |
| 2636 | nbits = item->ri_buf[i].i_len >> XFS_BLF_SHIFT; |
| 2637 | |
| 2638 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2639 | * Do a sanity check if this is a dquot buffer. Just checking |
| 2640 | * the first dquot in the buffer should do. XXXThis is |
| 2641 | * probably a good thing to do for other buf types also. |
| 2642 | */ |
| 2643 | error = 0; |
Nathan Scott | c8ad20f | 2005-06-21 15:38:48 +1000 | [diff] [blame] | 2644 | if (buf_f->blf_flags & |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 2645 | (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) { |
Christoph Hellwig | 0c5e1ce | 2009-06-08 15:33:21 +0200 | [diff] [blame] | 2646 | if (item->ri_buf[i].i_addr == NULL) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2647 | xfs_alert(mp, |
Christoph Hellwig | 0c5e1ce | 2009-06-08 15:33:21 +0200 | [diff] [blame] | 2648 | "XFS: NULL dquot in %s.", __func__); |
| 2649 | goto next; |
| 2650 | } |
Jan Rekorajski | 8ec6dba | 2009-11-16 11:57:02 +0000 | [diff] [blame] | 2651 | if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2652 | xfs_alert(mp, |
Christoph Hellwig | 0c5e1ce | 2009-06-08 15:33:21 +0200 | [diff] [blame] | 2653 | "XFS: dquot too small (%d) in %s.", |
| 2654 | item->ri_buf[i].i_len, __func__); |
| 2655 | goto next; |
| 2656 | } |
Dave Chinner | 9aede1d | 2013-10-15 09:17:52 +1100 | [diff] [blame] | 2657 | error = xfs_dqcheck(mp, item->ri_buf[i].i_addr, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2658 | -1, 0, XFS_QMOPT_DOWARN, |
| 2659 | "dquot_buf_recover"); |
Christoph Hellwig | 0c5e1ce | 2009-06-08 15:33:21 +0200 | [diff] [blame] | 2660 | if (error) |
| 2661 | goto next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2662 | } |
Christoph Hellwig | 0c5e1ce | 2009-06-08 15:33:21 +0200 | [diff] [blame] | 2663 | |
| 2664 | memcpy(xfs_buf_offset(bp, |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 2665 | (uint)bit << XFS_BLF_SHIFT), /* dest */ |
Christoph Hellwig | 0c5e1ce | 2009-06-08 15:33:21 +0200 | [diff] [blame] | 2666 | item->ri_buf[i].i_addr, /* source */ |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 2667 | nbits<<XFS_BLF_SHIFT); /* length */ |
Christoph Hellwig | 0c5e1ce | 2009-06-08 15:33:21 +0200 | [diff] [blame] | 2668 | next: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2669 | i++; |
| 2670 | bit += nbits; |
| 2671 | } |
| 2672 | |
| 2673 | /* Shouldn't be any more regions */ |
| 2674 | ASSERT(i == item->ri_total); |
Christoph Hellwig | ee1a47a | 2013-04-21 14:53:46 -0500 | [diff] [blame] | 2675 | |
Brian Foster | 22db9af | 2016-09-26 08:32:07 +1000 | [diff] [blame] | 2676 | xlog_recover_validate_buf_type(mp, bp, buf_f, current_lsn); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2677 | } |
| 2678 | |
| 2679 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2680 | * Perform a dquot buffer recovery. |
Zhi Yong Wu | 8ba701e | 2013-08-12 03:15:01 +0000 | [diff] [blame] | 2681 | * 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] | 2682 | * (ie. USR or GRP), then just toss this buffer away; don't recover it. |
| 2683 | * Else, treat it as a regular buffer and do recovery. |
Dave Chinner | ad3714b | 2014-08-04 12:59:31 +1000 | [diff] [blame] | 2684 | * |
| 2685 | * Return false if the buffer was tossed and true if we recovered the buffer to |
| 2686 | * indicate to the caller if the buffer needs writing. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2687 | */ |
Dave Chinner | ad3714b | 2014-08-04 12:59:31 +1000 | [diff] [blame] | 2688 | STATIC bool |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2689 | xlog_recover_do_dquot_buffer( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 2690 | struct xfs_mount *mp, |
| 2691 | struct xlog *log, |
| 2692 | struct xlog_recover_item *item, |
| 2693 | struct xfs_buf *bp, |
| 2694 | struct xfs_buf_log_format *buf_f) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2695 | { |
| 2696 | uint type; |
| 2697 | |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 2698 | trace_xfs_log_recover_buf_dquot_buf(log, buf_f); |
| 2699 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2700 | /* |
| 2701 | * Filesystems are required to send in quota flags at mount time. |
| 2702 | */ |
Dave Chinner | ad3714b | 2014-08-04 12:59:31 +1000 | [diff] [blame] | 2703 | if (!mp->m_qflags) |
| 2704 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2705 | |
| 2706 | type = 0; |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 2707 | if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2708 | type |= XFS_DQ_USER; |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 2709 | if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF) |
Nathan Scott | c8ad20f | 2005-06-21 15:38:48 +1000 | [diff] [blame] | 2710 | type |= XFS_DQ_PROJ; |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 2711 | if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2712 | type |= XFS_DQ_GROUP; |
| 2713 | /* |
| 2714 | * This type of quotas was turned off, so ignore this buffer |
| 2715 | */ |
| 2716 | if (log->l_quotaoffs_flag & type) |
Dave Chinner | ad3714b | 2014-08-04 12:59:31 +1000 | [diff] [blame] | 2717 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2718 | |
Brian Foster | 22db9af | 2016-09-26 08:32:07 +1000 | [diff] [blame] | 2719 | xlog_recover_do_reg_buffer(mp, item, bp, buf_f, NULLCOMMITLSN); |
Dave Chinner | ad3714b | 2014-08-04 12:59:31 +1000 | [diff] [blame] | 2720 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2721 | } |
| 2722 | |
| 2723 | /* |
| 2724 | * This routine replays a modification made to a buffer at runtime. |
| 2725 | * There are actually two types of buffer, regular and inode, which |
| 2726 | * are handled differently. Inode buffers are handled differently |
| 2727 | * in that we only recover a specific set of data from them, namely |
| 2728 | * the inode di_next_unlinked fields. This is because all other inode |
| 2729 | * data is actually logged via inode records and any data we replay |
| 2730 | * here which overlaps that may be stale. |
| 2731 | * |
| 2732 | * 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] | 2733 | * with the XFS_BLF_CANCEL bit set to indicate that previous copies |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2734 | * of the buffer in the log should not be replayed at recovery time. |
| 2735 | * This is so that if the blocks covered by the buffer are reused for |
| 2736 | * file data before we crash we don't end up replaying old, freed |
| 2737 | * meta-data into a user's file. |
| 2738 | * |
| 2739 | * To handle the cancellation of buffer log items, we make two passes |
| 2740 | * over the log during recovery. During the first we build a table of |
| 2741 | * those buffers which have been cancelled, and during the second we |
| 2742 | * only replay those buffers which do not have corresponding cancel |
Zhi Yong Wu | 34be5ff | 2013-08-07 10:11:07 +0000 | [diff] [blame] | 2743 | * records in the table. See xlog_recover_buffer_pass[1,2] above |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2744 | * for more details on the implementation of the table of cancel records. |
| 2745 | */ |
| 2746 | STATIC int |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2747 | xlog_recover_buffer_pass2( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 2748 | struct xlog *log, |
| 2749 | struct list_head *buffer_list, |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2750 | struct xlog_recover_item *item, |
| 2751 | xfs_lsn_t current_lsn) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2752 | { |
Christoph Hellwig | 4e0d5f9 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 2753 | 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] | 2754 | xfs_mount_t *mp = log->l_mp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2755 | xfs_buf_t *bp; |
| 2756 | int error; |
Christoph Hellwig | 6ad112b | 2009-11-24 18:02:23 +0000 | [diff] [blame] | 2757 | uint buf_flags; |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2758 | xfs_lsn_t lsn; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2759 | |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2760 | /* |
| 2761 | * In this pass we only want to recover all the buffers which have |
| 2762 | * not been cancelled and are not cancellation buffers themselves. |
| 2763 | */ |
| 2764 | if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno, |
| 2765 | buf_f->blf_len, buf_f->blf_flags)) { |
| 2766 | trace_xfs_log_recover_buf_cancel(log, buf_f); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2767 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2768 | } |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2769 | |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 2770 | trace_xfs_log_recover_buf_recover(log, buf_f); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2771 | |
Dave Chinner | a8acad7 | 2012-04-23 15:58:54 +1000 | [diff] [blame] | 2772 | buf_flags = 0; |
Dave Chinner | 611c994 | 2012-04-23 15:59:07 +1000 | [diff] [blame] | 2773 | if (buf_f->blf_flags & XFS_BLF_INODE_BUF) |
| 2774 | buf_flags |= XBF_UNMAPPED; |
Christoph Hellwig | 6ad112b | 2009-11-24 18:02:23 +0000 | [diff] [blame] | 2775 | |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2776 | 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] | 2777 | buf_flags, NULL); |
Chandra Seetharaman | ac4d688 | 2011-08-03 02:18:29 +0000 | [diff] [blame] | 2778 | if (!bp) |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 2779 | return -ENOMEM; |
Chandra Seetharaman | e570280 | 2011-08-03 02:18:34 +0000 | [diff] [blame] | 2780 | error = bp->b_error; |
Chandra Seetharaman | 5a52c2a58 | 2011-07-22 23:39:51 +0000 | [diff] [blame] | 2781 | if (error) { |
Christoph Hellwig | 901796a | 2011-10-10 16:52:49 +0000 | [diff] [blame] | 2782 | xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)"); |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2783 | goto out_release; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2784 | } |
| 2785 | |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2786 | /* |
Dave Chinner | 67dc288 | 2014-08-04 12:43:06 +1000 | [diff] [blame] | 2787 | * 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] | 2788 | * the lsn of the transaction we are replaying. |
Dave Chinner | 67dc288 | 2014-08-04 12:43:06 +1000 | [diff] [blame] | 2789 | * |
| 2790 | * Note that we have to be extremely careful of readahead here. |
| 2791 | * Readahead does not attach verfiers to the buffers so if we don't |
| 2792 | * actually do any replay after readahead because of the LSN we found |
| 2793 | * in the buffer if more recent than that current transaction then we |
| 2794 | * need to attach the verifier directly. Failure to do so can lead to |
| 2795 | * future recovery actions (e.g. EFI and unlinked list recovery) can |
| 2796 | * operate on the buffers and they won't get the verifier attached. This |
| 2797 | * can lead to blocks on disk having the correct content but a stale |
| 2798 | * CRC. |
| 2799 | * |
| 2800 | * It is safe to assume these clean buffers are currently up to date. |
| 2801 | * If the buffer is dirtied by a later transaction being replayed, then |
| 2802 | * the verifier will be reset to match whatever recover turns that |
| 2803 | * buffer into. |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2804 | */ |
| 2805 | lsn = xlog_recover_get_buf_lsn(mp, bp); |
Dave Chinner | 67dc288 | 2014-08-04 12:43:06 +1000 | [diff] [blame] | 2806 | if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) { |
Brian Foster | 22db9af | 2016-09-26 08:32:07 +1000 | [diff] [blame] | 2807 | xlog_recover_validate_buf_type(mp, bp, buf_f, NULLCOMMITLSN); |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2808 | goto out_release; |
Dave Chinner | 67dc288 | 2014-08-04 12:43:06 +1000 | [diff] [blame] | 2809 | } |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2810 | |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2811 | if (buf_f->blf_flags & XFS_BLF_INODE_BUF) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2812 | error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f); |
Dave Chinner | ad3714b | 2014-08-04 12:59:31 +1000 | [diff] [blame] | 2813 | if (error) |
| 2814 | goto out_release; |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2815 | } else if (buf_f->blf_flags & |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 2816 | (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) { |
Dave Chinner | ad3714b | 2014-08-04 12:59:31 +1000 | [diff] [blame] | 2817 | bool dirty; |
| 2818 | |
| 2819 | dirty = xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f); |
| 2820 | if (!dirty) |
| 2821 | goto out_release; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2822 | } else { |
Brian Foster | 22db9af | 2016-09-26 08:32:07 +1000 | [diff] [blame] | 2823 | xlog_recover_do_reg_buffer(mp, item, bp, buf_f, current_lsn); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2824 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2825 | |
| 2826 | /* |
| 2827 | * Perform delayed write on the buffer. Asynchronous writes will be |
| 2828 | * slower when taking into account all the buffers to be flushed. |
| 2829 | * |
| 2830 | * Also make sure that only inode buffers with good sizes stay in |
| 2831 | * the buffer cache. The kernel moves inodes in buffers of 1 block |
Jie Liu | 0f49efd | 2013-12-13 15:51:48 +1100 | [diff] [blame] | 2832 | * or mp->m_inode_cluster_size bytes, whichever is bigger. The inode |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2833 | * buffers in the log can be a different size if the log was generated |
| 2834 | * by an older kernel using unclustered inode buffers or a newer kernel |
| 2835 | * running with a different inode cluster size. Regardless, if the |
Jie Liu | 0f49efd | 2013-12-13 15:51:48 +1100 | [diff] [blame] | 2836 | * the inode buffer size isn't MAX(blocksize, mp->m_inode_cluster_size) |
| 2837 | * 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] | 2838 | * the buffer out of the buffer cache so that the buffer won't |
| 2839 | * overlap with future reads of those inodes. |
| 2840 | */ |
| 2841 | if (XFS_DINODE_MAGIC == |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 2842 | be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) && |
Dave Chinner | aa0e883 | 2012-04-23 15:58:52 +1000 | [diff] [blame] | 2843 | (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] | 2844 | (__uint32_t)log->l_mp->m_inode_cluster_size))) { |
Christoph Hellwig | c867cb6 | 2011-10-10 16:52:46 +0000 | [diff] [blame] | 2845 | xfs_buf_stale(bp); |
Christoph Hellwig | c2b006c | 2011-08-23 08:28:07 +0000 | [diff] [blame] | 2846 | error = xfs_bwrite(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2847 | } else { |
Dave Chinner | ebad861 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 2848 | ASSERT(bp->b_target->bt_mount == mp); |
Christoph Hellwig | cb669ca | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 2849 | bp->b_iodone = xlog_recover_iodone; |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 2850 | xfs_buf_delwri_queue(bp, buffer_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2851 | } |
| 2852 | |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2853 | out_release: |
Christoph Hellwig | c2b006c | 2011-08-23 08:28:07 +0000 | [diff] [blame] | 2854 | xfs_buf_relse(bp); |
| 2855 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2856 | } |
| 2857 | |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 2858 | /* |
| 2859 | * Inode fork owner changes |
| 2860 | * |
| 2861 | * If we have been told that we have to reparent the inode fork, it's because an |
| 2862 | * extent swap operation on a CRC enabled filesystem has been done and we are |
| 2863 | * replaying it. We need to walk the BMBT of the appropriate fork and change the |
| 2864 | * owners of it. |
| 2865 | * |
| 2866 | * The complexity here is that we don't have an inode context to work with, so |
| 2867 | * after we've replayed the inode we need to instantiate one. This is where the |
| 2868 | * fun begins. |
| 2869 | * |
| 2870 | * We are in the middle of log recovery, so we can't run transactions. That |
| 2871 | * means we cannot use cache coherent inode instantiation via xfs_iget(), as |
| 2872 | * that will result in the corresponding iput() running the inode through |
| 2873 | * xfs_inactive(). If we've just replayed an inode core that changes the link |
| 2874 | * count to zero (i.e. it's been unlinked), then xfs_inactive() will run |
| 2875 | * transactions (bad!). |
| 2876 | * |
| 2877 | * So, to avoid this, we instantiate an inode directly from the inode core we've |
| 2878 | * just recovered. We have the buffer still locked, and all we really need to |
| 2879 | * instantiate is the inode core and the forks being modified. We can do this |
| 2880 | * manually, then run the inode btree owner change, and then tear down the |
| 2881 | * xfs_inode without having to run any transactions at all. |
| 2882 | * |
| 2883 | * Also, because we don't have a transaction context available here but need to |
| 2884 | * gather all the buffers we modify for writeback so we pass the buffer_list |
| 2885 | * instead for the operation to use. |
| 2886 | */ |
| 2887 | |
| 2888 | STATIC int |
| 2889 | xfs_recover_inode_owner_change( |
| 2890 | struct xfs_mount *mp, |
| 2891 | struct xfs_dinode *dip, |
| 2892 | struct xfs_inode_log_format *in_f, |
| 2893 | struct list_head *buffer_list) |
| 2894 | { |
| 2895 | struct xfs_inode *ip; |
| 2896 | int error; |
| 2897 | |
| 2898 | ASSERT(in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER)); |
| 2899 | |
| 2900 | ip = xfs_inode_alloc(mp, in_f->ilf_ino); |
| 2901 | if (!ip) |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 2902 | return -ENOMEM; |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 2903 | |
| 2904 | /* instantiate the inode */ |
Dave Chinner | 3987848 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 2905 | xfs_inode_from_disk(ip, dip); |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 2906 | ASSERT(ip->i_d.di_version >= 3); |
| 2907 | |
| 2908 | error = xfs_iformat_fork(ip, dip); |
| 2909 | if (error) |
| 2910 | goto out_free_ip; |
| 2911 | |
| 2912 | |
| 2913 | if (in_f->ilf_fields & XFS_ILOG_DOWNER) { |
| 2914 | ASSERT(in_f->ilf_fields & XFS_ILOG_DBROOT); |
| 2915 | error = xfs_bmbt_change_owner(NULL, ip, XFS_DATA_FORK, |
| 2916 | ip->i_ino, buffer_list); |
| 2917 | if (error) |
| 2918 | goto out_free_ip; |
| 2919 | } |
| 2920 | |
| 2921 | if (in_f->ilf_fields & XFS_ILOG_AOWNER) { |
| 2922 | ASSERT(in_f->ilf_fields & XFS_ILOG_ABROOT); |
| 2923 | error = xfs_bmbt_change_owner(NULL, ip, XFS_ATTR_FORK, |
| 2924 | ip->i_ino, buffer_list); |
| 2925 | if (error) |
| 2926 | goto out_free_ip; |
| 2927 | } |
| 2928 | |
| 2929 | out_free_ip: |
| 2930 | xfs_inode_free(ip); |
| 2931 | return error; |
| 2932 | } |
| 2933 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2934 | STATIC int |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2935 | xlog_recover_inode_pass2( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 2936 | struct xlog *log, |
| 2937 | struct list_head *buffer_list, |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 2938 | struct xlog_recover_item *item, |
| 2939 | xfs_lsn_t current_lsn) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2940 | { |
| 2941 | xfs_inode_log_format_t *in_f; |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2942 | xfs_mount_t *mp = log->l_mp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2943 | xfs_buf_t *bp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2944 | xfs_dinode_t *dip; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2945 | int len; |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 2946 | char *src; |
| 2947 | char *dest; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2948 | int error; |
| 2949 | int attr_index; |
| 2950 | uint fields; |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 2951 | struct xfs_log_dinode *ldip; |
Christoph Hellwig | 93848a9 | 2013-04-03 16:11:17 +1100 | [diff] [blame] | 2952 | uint isize; |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2953 | int need_free = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2954 | |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2955 | 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] | 2956 | in_f = item->ri_buf[0].i_addr; |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2957 | } else { |
Christoph Hellwig | 4e0d5f9 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 2958 | in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP); |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2959 | need_free = 1; |
| 2960 | error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f); |
| 2961 | if (error) |
| 2962 | goto error; |
| 2963 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2964 | |
| 2965 | /* |
| 2966 | * Inode buffers can be freed, look out for it, |
| 2967 | * and do not replay the inode. |
| 2968 | */ |
Christoph Hellwig | a194189 | 2008-11-28 14:23:40 +1100 | [diff] [blame] | 2969 | if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno, |
| 2970 | in_f->ilf_len, 0)) { |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2971 | error = 0; |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 2972 | trace_xfs_log_recover_inode_cancel(log, in_f); |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2973 | goto error; |
| 2974 | } |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 2975 | trace_xfs_log_recover_inode_recover(log, in_f); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2976 | |
Dave Chinner | c3f8fc7 | 2012-11-12 22:54:01 +1100 | [diff] [blame] | 2977 | 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] | 2978 | &xfs_inode_buf_ops); |
Chandra Seetharaman | ac4d688 | 2011-08-03 02:18:29 +0000 | [diff] [blame] | 2979 | if (!bp) { |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 2980 | error = -ENOMEM; |
Chandra Seetharaman | ac4d688 | 2011-08-03 02:18:29 +0000 | [diff] [blame] | 2981 | goto error; |
| 2982 | } |
Chandra Seetharaman | e570280 | 2011-08-03 02:18:34 +0000 | [diff] [blame] | 2983 | error = bp->b_error; |
Chandra Seetharaman | 5a52c2a58 | 2011-07-22 23:39:51 +0000 | [diff] [blame] | 2984 | if (error) { |
Christoph Hellwig | 901796a | 2011-10-10 16:52:49 +0000 | [diff] [blame] | 2985 | xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)"); |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 2986 | goto out_release; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2987 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2988 | ASSERT(in_f->ilf_fields & XFS_ILOG_CORE); |
Christoph Hellwig | 88ee2df | 2015-06-22 09:44:29 +1000 | [diff] [blame] | 2989 | dip = xfs_buf_offset(bp, in_f->ilf_boffset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2990 | |
| 2991 | /* |
| 2992 | * Make sure the place we're flushing out to really looks |
| 2993 | * like an inode! |
| 2994 | */ |
Christoph Hellwig | 69ef921 | 2011-07-08 14:36:05 +0200 | [diff] [blame] | 2995 | if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2996 | xfs_alert(mp, |
| 2997 | "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld", |
| 2998 | __func__, dip, bp, in_f->ilf_ino); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2999 | XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3000 | XFS_ERRLEVEL_LOW, mp); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3001 | error = -EFSCORRUPTED; |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 3002 | goto out_release; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3003 | } |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3004 | ldip = item->ri_buf[1].i_addr; |
| 3005 | if (unlikely(ldip->di_magic != XFS_DINODE_MAGIC)) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 3006 | xfs_alert(mp, |
| 3007 | "%s: Bad inode log record, rec ptr 0x%p, ino %Ld", |
| 3008 | __func__, item, in_f->ilf_ino); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3009 | XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3010 | XFS_ERRLEVEL_LOW, mp); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3011 | error = -EFSCORRUPTED; |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 3012 | goto out_release; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3013 | } |
| 3014 | |
Dave Chinner | e60896d | 2013-07-24 15:47:30 +1000 | [diff] [blame] | 3015 | /* |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 3016 | * 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] | 3017 | * than the lsn of the transaction we are replaying. Note: we still |
| 3018 | * need to replay an owner change even though the inode is more recent |
| 3019 | * than the transaction as there is no guarantee that all the btree |
| 3020 | * blocks are more recent than this transaction, too. |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 3021 | */ |
| 3022 | if (dip->di_version >= 3) { |
| 3023 | xfs_lsn_t lsn = be64_to_cpu(dip->di_lsn); |
| 3024 | |
| 3025 | if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) { |
| 3026 | trace_xfs_log_recover_inode_skip(log, in_f); |
| 3027 | error = 0; |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 3028 | goto out_owner_change; |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 3029 | } |
| 3030 | } |
| 3031 | |
| 3032 | /* |
Dave Chinner | e60896d | 2013-07-24 15:47:30 +1000 | [diff] [blame] | 3033 | * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes |
| 3034 | * are transactional and if ordering is necessary we can determine that |
| 3035 | * more accurately by the LSN field in the V3 inode core. Don't trust |
| 3036 | * the inode versions we might be changing them here - use the |
| 3037 | * superblock flag to determine whether we need to look at di_flushiter |
| 3038 | * to skip replay when the on disk inode is newer than the log one |
| 3039 | */ |
| 3040 | if (!xfs_sb_version_hascrc(&mp->m_sb) && |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3041 | ldip->di_flushiter < be16_to_cpu(dip->di_flushiter)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3042 | /* |
| 3043 | * Deal with the wrap case, DI_MAX_FLUSH is less |
| 3044 | * than smaller numbers |
| 3045 | */ |
Christoph Hellwig | 81591fe | 2008-11-28 14:23:39 +1100 | [diff] [blame] | 3046 | if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH && |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3047 | ldip->di_flushiter < (DI_MAX_FLUSH >> 1)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3048 | /* do nothing */ |
| 3049 | } else { |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 3050 | trace_xfs_log_recover_inode_skip(log, in_f); |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 3051 | error = 0; |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 3052 | goto out_release; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3053 | } |
| 3054 | } |
Dave Chinner | e60896d | 2013-07-24 15:47:30 +1000 | [diff] [blame] | 3055 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3056 | /* Take the opportunity to reset the flush iteration count */ |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3057 | ldip->di_flushiter = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3058 | |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3059 | if (unlikely(S_ISREG(ldip->di_mode))) { |
| 3060 | if ((ldip->di_format != XFS_DINODE_FMT_EXTENTS) && |
| 3061 | (ldip->di_format != XFS_DINODE_FMT_BTREE)) { |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3062 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)", |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3063 | XFS_ERRLEVEL_LOW, mp, ldip); |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 3064 | xfs_alert(mp, |
| 3065 | "%s: Bad regular inode log record, rec ptr 0x%p, " |
| 3066 | "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld", |
| 3067 | __func__, item, dip, bp, in_f->ilf_ino); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3068 | error = -EFSCORRUPTED; |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 3069 | goto out_release; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3070 | } |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3071 | } else if (unlikely(S_ISDIR(ldip->di_mode))) { |
| 3072 | if ((ldip->di_format != XFS_DINODE_FMT_EXTENTS) && |
| 3073 | (ldip->di_format != XFS_DINODE_FMT_BTREE) && |
| 3074 | (ldip->di_format != XFS_DINODE_FMT_LOCAL)) { |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3075 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)", |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3076 | XFS_ERRLEVEL_LOW, mp, ldip); |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 3077 | xfs_alert(mp, |
| 3078 | "%s: Bad dir inode log record, rec ptr 0x%p, " |
| 3079 | "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld", |
| 3080 | __func__, item, dip, bp, in_f->ilf_ino); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3081 | error = -EFSCORRUPTED; |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 3082 | goto out_release; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3083 | } |
| 3084 | } |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3085 | if (unlikely(ldip->di_nextents + ldip->di_anextents > ldip->di_nblocks)){ |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3086 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)", |
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 inode log record, rec ptr 0x%p, dino ptr 0x%p, " |
| 3090 | "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld", |
| 3091 | __func__, item, dip, bp, in_f->ilf_ino, |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3092 | ldip->di_nextents + ldip->di_anextents, |
| 3093 | ldip->di_nblocks); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3094 | error = -EFSCORRUPTED; |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 3095 | goto out_release; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3096 | } |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3097 | if (unlikely(ldip->di_forkoff > mp->m_sb.sb_inodesize)) { |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3098 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)", |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3099 | XFS_ERRLEVEL_LOW, mp, ldip); |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 3100 | xfs_alert(mp, |
| 3101 | "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, " |
| 3102 | "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__, |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3103 | item, dip, bp, in_f->ilf_ino, ldip->di_forkoff); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3104 | error = -EFSCORRUPTED; |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 3105 | goto out_release; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3106 | } |
Dave Chinner | f8d55aa052 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3107 | isize = xfs_log_dinode_size(ldip->di_version); |
Christoph Hellwig | 93848a9 | 2013-04-03 16:11:17 +1100 | [diff] [blame] | 3108 | if (unlikely(item->ri_buf[1].i_len > isize)) { |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3109 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)", |
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 length %d, rec ptr 0x%p", |
| 3113 | __func__, item->ri_buf[1].i_len, item); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3114 | error = -EFSCORRUPTED; |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 3115 | goto out_release; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3116 | } |
| 3117 | |
Dave Chinner | 3987848 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 3118 | /* recover the log dinode inode into the on disk inode */ |
| 3119 | xfs_log_dinode_to_disk(ldip, dip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3120 | |
| 3121 | /* the rest is in on-disk format */ |
Christoph Hellwig | 93848a9 | 2013-04-03 16:11:17 +1100 | [diff] [blame] | 3122 | if (item->ri_buf[1].i_len > isize) { |
| 3123 | memcpy((char *)dip + isize, |
| 3124 | item->ri_buf[1].i_addr + isize, |
| 3125 | item->ri_buf[1].i_len - isize); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3126 | } |
| 3127 | |
| 3128 | fields = in_f->ilf_fields; |
| 3129 | switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) { |
| 3130 | case XFS_ILOG_DEV: |
Christoph Hellwig | 81591fe | 2008-11-28 14:23:39 +1100 | [diff] [blame] | 3131 | xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3132 | break; |
| 3133 | case XFS_ILOG_UUID: |
Christoph Hellwig | 81591fe | 2008-11-28 14:23:39 +1100 | [diff] [blame] | 3134 | memcpy(XFS_DFORK_DPTR(dip), |
| 3135 | &in_f->ilf_u.ilfu_uuid, |
| 3136 | sizeof(uuid_t)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3137 | break; |
| 3138 | } |
| 3139 | |
| 3140 | if (in_f->ilf_size == 2) |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 3141 | goto out_owner_change; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3142 | len = item->ri_buf[2].i_len; |
| 3143 | src = item->ri_buf[2].i_addr; |
| 3144 | ASSERT(in_f->ilf_size <= 4); |
| 3145 | ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK)); |
| 3146 | ASSERT(!(fields & XFS_ILOG_DFORK) || |
| 3147 | (len == in_f->ilf_dsize)); |
| 3148 | |
| 3149 | switch (fields & XFS_ILOG_DFORK) { |
| 3150 | case XFS_ILOG_DDATA: |
| 3151 | case XFS_ILOG_DEXT: |
Christoph Hellwig | 81591fe | 2008-11-28 14:23:39 +1100 | [diff] [blame] | 3152 | memcpy(XFS_DFORK_DPTR(dip), src, len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3153 | break; |
| 3154 | |
| 3155 | case XFS_ILOG_DBROOT: |
Christoph Hellwig | 7cc95a8 | 2008-10-30 17:14:34 +1100 | [diff] [blame] | 3156 | xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len, |
Christoph Hellwig | 81591fe | 2008-11-28 14:23:39 +1100 | [diff] [blame] | 3157 | (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3158 | XFS_DFORK_DSIZE(dip, mp)); |
| 3159 | break; |
| 3160 | |
| 3161 | default: |
| 3162 | /* |
| 3163 | * There are no data fork flags set. |
| 3164 | */ |
| 3165 | ASSERT((fields & XFS_ILOG_DFORK) == 0); |
| 3166 | break; |
| 3167 | } |
| 3168 | |
| 3169 | /* |
| 3170 | * If we logged any attribute data, recover it. There may or |
| 3171 | * may not have been any other non-core data logged in this |
| 3172 | * transaction. |
| 3173 | */ |
| 3174 | if (in_f->ilf_fields & XFS_ILOG_AFORK) { |
| 3175 | if (in_f->ilf_fields & XFS_ILOG_DFORK) { |
| 3176 | attr_index = 3; |
| 3177 | } else { |
| 3178 | attr_index = 2; |
| 3179 | } |
| 3180 | len = item->ri_buf[attr_index].i_len; |
| 3181 | src = item->ri_buf[attr_index].i_addr; |
| 3182 | ASSERT(len == in_f->ilf_asize); |
| 3183 | |
| 3184 | switch (in_f->ilf_fields & XFS_ILOG_AFORK) { |
| 3185 | case XFS_ILOG_ADATA: |
| 3186 | case XFS_ILOG_AEXT: |
| 3187 | dest = XFS_DFORK_APTR(dip); |
| 3188 | ASSERT(len <= XFS_DFORK_ASIZE(dip, mp)); |
| 3189 | memcpy(dest, src, len); |
| 3190 | break; |
| 3191 | |
| 3192 | case XFS_ILOG_ABROOT: |
| 3193 | dest = XFS_DFORK_APTR(dip); |
Christoph Hellwig | 7cc95a8 | 2008-10-30 17:14:34 +1100 | [diff] [blame] | 3194 | xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, |
| 3195 | len, (xfs_bmdr_block_t*)dest, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3196 | XFS_DFORK_ASIZE(dip, mp)); |
| 3197 | break; |
| 3198 | |
| 3199 | default: |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 3200 | xfs_warn(log->l_mp, "%s: Invalid flag", __func__); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3201 | ASSERT(0); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3202 | error = -EIO; |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 3203 | goto out_release; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3204 | } |
| 3205 | } |
| 3206 | |
Dave Chinner | 638f4416 | 2013-08-30 10:23:45 +1000 | [diff] [blame] | 3207 | out_owner_change: |
| 3208 | if (in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER)) |
| 3209 | error = xfs_recover_inode_owner_change(mp, dip, in_f, |
| 3210 | buffer_list); |
Christoph Hellwig | 93848a9 | 2013-04-03 16:11:17 +1100 | [diff] [blame] | 3211 | /* re-generate the checksum. */ |
| 3212 | xfs_dinode_calc_crc(log->l_mp, dip); |
| 3213 | |
Dave Chinner | ebad861 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 3214 | ASSERT(bp->b_target->bt_mount == mp); |
Christoph Hellwig | cb669ca | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 3215 | bp->b_iodone = xlog_recover_iodone; |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 3216 | xfs_buf_delwri_queue(bp, buffer_list); |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 3217 | |
| 3218 | out_release: |
Christoph Hellwig | 61551f1 | 2011-08-23 08:28:06 +0000 | [diff] [blame] | 3219 | xfs_buf_relse(bp); |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 3220 | error: |
| 3221 | if (need_free) |
Denys Vlasenko | f0e2d93 | 2008-05-19 16:31:57 +1000 | [diff] [blame] | 3222 | kmem_free(in_f); |
Eric Sandeen | b474c7a | 2014-06-22 15:04:54 +1000 | [diff] [blame] | 3223 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3224 | } |
| 3225 | |
| 3226 | /* |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 3227 | * 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] | 3228 | * structure, so that we know not to do any dquot item or dquot buffer recovery, |
| 3229 | * of that type. |
| 3230 | */ |
| 3231 | STATIC int |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3232 | xlog_recover_quotaoff_pass1( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 3233 | struct xlog *log, |
| 3234 | struct xlog_recover_item *item) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3235 | { |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3236 | xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3237 | ASSERT(qoff_f); |
| 3238 | |
| 3239 | /* |
| 3240 | * The logitem format's flag tells us if this was user quotaoff, |
Nathan Scott | 77a7cce | 2006-01-11 15:35:57 +1100 | [diff] [blame] | 3241 | * group/project quotaoff or both. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3242 | */ |
| 3243 | if (qoff_f->qf_flags & XFS_UQUOTA_ACCT) |
| 3244 | log->l_quotaoffs_flag |= XFS_DQ_USER; |
Nathan Scott | 77a7cce | 2006-01-11 15:35:57 +1100 | [diff] [blame] | 3245 | if (qoff_f->qf_flags & XFS_PQUOTA_ACCT) |
| 3246 | log->l_quotaoffs_flag |= XFS_DQ_PROJ; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3247 | if (qoff_f->qf_flags & XFS_GQUOTA_ACCT) |
| 3248 | log->l_quotaoffs_flag |= XFS_DQ_GROUP; |
| 3249 | |
Eric Sandeen | d99831f | 2014-06-22 15:03:54 +1000 | [diff] [blame] | 3250 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3251 | } |
| 3252 | |
| 3253 | /* |
| 3254 | * Recover a dquot record |
| 3255 | */ |
| 3256 | STATIC int |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3257 | xlog_recover_dquot_pass2( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 3258 | struct xlog *log, |
| 3259 | struct list_head *buffer_list, |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 3260 | struct xlog_recover_item *item, |
| 3261 | xfs_lsn_t current_lsn) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3262 | { |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3263 | xfs_mount_t *mp = log->l_mp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3264 | xfs_buf_t *bp; |
| 3265 | struct xfs_disk_dquot *ddq, *recddq; |
| 3266 | int error; |
| 3267 | xfs_dq_logformat_t *dq_f; |
| 3268 | uint type; |
| 3269 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3270 | |
| 3271 | /* |
| 3272 | * Filesystems are required to send in quota flags at mount time. |
| 3273 | */ |
| 3274 | if (mp->m_qflags == 0) |
Eric Sandeen | d99831f | 2014-06-22 15:03:54 +1000 | [diff] [blame] | 3275 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3276 | |
Christoph Hellwig | 4e0d5f9 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 3277 | recddq = item->ri_buf[1].i_addr; |
| 3278 | if (recddq == NULL) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 3279 | xfs_alert(log->l_mp, "NULL dquot in %s.", __func__); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3280 | return -EIO; |
Christoph Hellwig | 0c5e1ce | 2009-06-08 15:33:21 +0200 | [diff] [blame] | 3281 | } |
Jan Rekorajski | 8ec6dba | 2009-11-16 11:57:02 +0000 | [diff] [blame] | 3282 | if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 3283 | xfs_alert(log->l_mp, "dquot too small (%d) in %s.", |
Christoph Hellwig | 0c5e1ce | 2009-06-08 15:33:21 +0200 | [diff] [blame] | 3284 | item->ri_buf[1].i_len, __func__); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3285 | return -EIO; |
Christoph Hellwig | 0c5e1ce | 2009-06-08 15:33:21 +0200 | [diff] [blame] | 3286 | } |
| 3287 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3288 | /* |
| 3289 | * This type of quotas was turned off, so ignore this record. |
| 3290 | */ |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 3291 | 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] | 3292 | ASSERT(type); |
| 3293 | if (log->l_quotaoffs_flag & type) |
Eric Sandeen | d99831f | 2014-06-22 15:03:54 +1000 | [diff] [blame] | 3294 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3295 | |
| 3296 | /* |
| 3297 | * At this point we know that quota was _not_ turned off. |
| 3298 | * Since the mount flags are not indicating to us otherwise, this |
| 3299 | * must mean that quota is on, and the dquot needs to be replayed. |
| 3300 | * Remember that we may not have fully recovered the superblock yet, |
| 3301 | * so we can't do the usual trick of looking at the SB quota bits. |
| 3302 | * |
| 3303 | * The other possibility, of course, is that the quota subsystem was |
| 3304 | * removed since the last mount - ENOSYS. |
| 3305 | */ |
Christoph Hellwig | 4e0d5f9 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 3306 | dq_f = item->ri_buf[0].i_addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3307 | ASSERT(dq_f); |
Dave Chinner | 9aede1d | 2013-10-15 09:17:52 +1100 | [diff] [blame] | 3308 | 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] | 3309 | "xlog_recover_dquot_pass2 (log copy)"); |
| 3310 | if (error) |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3311 | return -EIO; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3312 | ASSERT(dq_f->qlf_len == 1); |
| 3313 | |
Dave Chinner | ad3714b | 2014-08-04 12:59:31 +1000 | [diff] [blame] | 3314 | /* |
| 3315 | * At this point we are assuming that the dquots have been allocated |
| 3316 | * and hence the buffer has valid dquots stamped in it. It should, |
| 3317 | * therefore, pass verifier validation. If the dquot is bad, then the |
| 3318 | * we'll return an error here, so we don't need to specifically check |
| 3319 | * the dquot in the buffer after the verifier has run. |
| 3320 | */ |
Dave Chinner | 7ca790a | 2012-04-23 15:58:55 +1000 | [diff] [blame] | 3321 | 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] | 3322 | XFS_FSB_TO_BB(mp, dq_f->qlf_len), 0, &bp, |
Dave Chinner | ad3714b | 2014-08-04 12:59:31 +1000 | [diff] [blame] | 3323 | &xfs_dquot_buf_ops); |
Dave Chinner | 7ca790a | 2012-04-23 15:58:55 +1000 | [diff] [blame] | 3324 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3325 | return error; |
Dave Chinner | 7ca790a | 2012-04-23 15:58:55 +1000 | [diff] [blame] | 3326 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3327 | ASSERT(bp); |
Christoph Hellwig | 88ee2df | 2015-06-22 09:44:29 +1000 | [diff] [blame] | 3328 | ddq = xfs_buf_offset(bp, dq_f->qlf_boffset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3329 | |
| 3330 | /* |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 3331 | * If the dquot has an LSN in it, recover the dquot only if it's less |
| 3332 | * than the lsn of the transaction we are replaying. |
| 3333 | */ |
| 3334 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
| 3335 | struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddq; |
| 3336 | xfs_lsn_t lsn = be64_to_cpu(dqb->dd_lsn); |
| 3337 | |
| 3338 | if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) { |
| 3339 | goto out_release; |
| 3340 | } |
| 3341 | } |
| 3342 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3343 | memcpy(ddq, recddq, item->ri_buf[1].i_len); |
Dave Chinner | 6fcdc59 | 2013-06-03 15:28:46 +1000 | [diff] [blame] | 3344 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
| 3345 | xfs_update_cksum((char *)ddq, sizeof(struct xfs_dqblk), |
| 3346 | XFS_DQUOT_CRC_OFF); |
| 3347 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3348 | |
| 3349 | ASSERT(dq_f->qlf_size == 2); |
Dave Chinner | ebad861 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 3350 | ASSERT(bp->b_target->bt_mount == mp); |
Christoph Hellwig | cb669ca | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 3351 | bp->b_iodone = xlog_recover_iodone; |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 3352 | xfs_buf_delwri_queue(bp, buffer_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3353 | |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 3354 | out_release: |
| 3355 | xfs_buf_relse(bp); |
| 3356 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3357 | } |
| 3358 | |
| 3359 | /* |
| 3360 | * This routine is called to create an in-core extent free intent |
| 3361 | * item from the efi format structure which was logged on disk. |
| 3362 | * It allocates an in-core efi, copies the extents from the format |
| 3363 | * structure into it, and adds the efi to the AIL with the given |
| 3364 | * LSN. |
| 3365 | */ |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 3366 | STATIC int |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3367 | xlog_recover_efi_pass2( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 3368 | struct xlog *log, |
| 3369 | struct xlog_recover_item *item, |
| 3370 | xfs_lsn_t lsn) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3371 | { |
Brian Foster | e32a1d1 | 2015-08-19 09:52:21 +1000 | [diff] [blame] | 3372 | int error; |
| 3373 | struct xfs_mount *mp = log->l_mp; |
| 3374 | struct xfs_efi_log_item *efip; |
| 3375 | struct xfs_efi_log_format *efi_formatp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3376 | |
Christoph Hellwig | 4e0d5f9 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 3377 | efi_formatp = item->ri_buf[0].i_addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3378 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3379 | efip = xfs_efi_init(mp, efi_formatp->efi_nextents); |
Brian Foster | e32a1d1 | 2015-08-19 09:52:21 +1000 | [diff] [blame] | 3380 | error = xfs_efi_copy_format(&item->ri_buf[0], &efip->efi_format); |
| 3381 | if (error) { |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 3382 | xfs_efi_item_free(efip); |
| 3383 | return error; |
| 3384 | } |
Dave Chinner | b199c8a | 2010-12-20 11:59:49 +1100 | [diff] [blame] | 3385 | atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3386 | |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 3387 | spin_lock(&log->l_ailp->xa_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3388 | /* |
Brian Foster | e32a1d1 | 2015-08-19 09:52:21 +1000 | [diff] [blame] | 3389 | * The EFI has two references. One for the EFD and one for EFI to ensure |
| 3390 | * it makes it into the AIL. Insert the EFI into the AIL directly and |
| 3391 | * drop the EFI reference. Note that xfs_trans_ail_update() drops the |
| 3392 | * AIL lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3393 | */ |
Dave Chinner | e605994 | 2010-12-20 12:34:26 +1100 | [diff] [blame] | 3394 | xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn); |
Brian Foster | e32a1d1 | 2015-08-19 09:52:21 +1000 | [diff] [blame] | 3395 | xfs_efi_release(efip); |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 3396 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3397 | } |
| 3398 | |
| 3399 | |
| 3400 | /* |
Brian Foster | e32a1d1 | 2015-08-19 09:52:21 +1000 | [diff] [blame] | 3401 | * This routine is called when an EFD format structure is found in a committed |
| 3402 | * transaction in the log. Its purpose is to cancel the corresponding EFI if it |
| 3403 | * was still in the log. To do this it searches the AIL for the EFI with an id |
| 3404 | * equal to that in the EFD format structure. If we find it we drop the EFD |
| 3405 | * reference, which removes the EFI from the AIL and frees it. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3406 | */ |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3407 | STATIC int |
| 3408 | xlog_recover_efd_pass2( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 3409 | struct xlog *log, |
| 3410 | struct xlog_recover_item *item) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3411 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3412 | xfs_efd_log_format_t *efd_formatp; |
| 3413 | xfs_efi_log_item_t *efip = NULL; |
| 3414 | xfs_log_item_t *lip; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3415 | __uint64_t efi_id; |
David Chinner | 27d8d5f | 2008-10-30 17:38:39 +1100 | [diff] [blame] | 3416 | struct xfs_ail_cursor cur; |
David Chinner | 783a2f6 | 2008-10-30 17:39:58 +1100 | [diff] [blame] | 3417 | struct xfs_ail *ailp = log->l_ailp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3418 | |
Christoph Hellwig | 4e0d5f9 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 3419 | efd_formatp = item->ri_buf[0].i_addr; |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 3420 | ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) + |
| 3421 | ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) || |
| 3422 | (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) + |
| 3423 | ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t))))); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3424 | efi_id = efd_formatp->efd_efi_id; |
| 3425 | |
| 3426 | /* |
Brian Foster | e32a1d1 | 2015-08-19 09:52:21 +1000 | [diff] [blame] | 3427 | * Search for the EFI with the id in the EFD format structure in the |
| 3428 | * AIL. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3429 | */ |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 3430 | spin_lock(&ailp->xa_lock); |
| 3431 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3432 | while (lip != NULL) { |
| 3433 | if (lip->li_type == XFS_LI_EFI) { |
| 3434 | efip = (xfs_efi_log_item_t *)lip; |
| 3435 | if (efip->efi_format.efi_id == efi_id) { |
| 3436 | /* |
Brian Foster | e32a1d1 | 2015-08-19 09:52:21 +1000 | [diff] [blame] | 3437 | * Drop the EFD reference to the EFI. This |
| 3438 | * removes the EFI from the AIL and frees it. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3439 | */ |
Brian Foster | e32a1d1 | 2015-08-19 09:52:21 +1000 | [diff] [blame] | 3440 | spin_unlock(&ailp->xa_lock); |
| 3441 | xfs_efi_release(efip); |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 3442 | spin_lock(&ailp->xa_lock); |
David Chinner | 27d8d5f | 2008-10-30 17:38:39 +1100 | [diff] [blame] | 3443 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3444 | } |
| 3445 | } |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 3446 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3447 | } |
Brian Foster | e32a1d1 | 2015-08-19 09:52:21 +1000 | [diff] [blame] | 3448 | |
Eric Sandeen | e4a1e29 | 2014-04-14 19:06:05 +1000 | [diff] [blame] | 3449 | xfs_trans_ail_cursor_done(&cur); |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 3450 | spin_unlock(&ailp->xa_lock); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3451 | |
| 3452 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3453 | } |
| 3454 | |
| 3455 | /* |
Darrick J. Wong | 9e88b5d | 2016-08-03 12:09:48 +1000 | [diff] [blame] | 3456 | * This routine is called to create an in-core extent rmap update |
| 3457 | * item from the rui format structure which was logged on disk. |
| 3458 | * It allocates an in-core rui, copies the extents from the format |
| 3459 | * structure into it, and adds the rui to the AIL with the given |
| 3460 | * LSN. |
| 3461 | */ |
| 3462 | STATIC int |
| 3463 | xlog_recover_rui_pass2( |
| 3464 | struct xlog *log, |
| 3465 | struct xlog_recover_item *item, |
| 3466 | xfs_lsn_t lsn) |
| 3467 | { |
| 3468 | int error; |
| 3469 | struct xfs_mount *mp = log->l_mp; |
| 3470 | struct xfs_rui_log_item *ruip; |
| 3471 | struct xfs_rui_log_format *rui_formatp; |
| 3472 | |
| 3473 | rui_formatp = item->ri_buf[0].i_addr; |
| 3474 | |
| 3475 | ruip = xfs_rui_init(mp, rui_formatp->rui_nextents); |
| 3476 | error = xfs_rui_copy_format(&item->ri_buf[0], &ruip->rui_format); |
| 3477 | if (error) { |
| 3478 | xfs_rui_item_free(ruip); |
| 3479 | return error; |
| 3480 | } |
| 3481 | atomic_set(&ruip->rui_next_extent, rui_formatp->rui_nextents); |
| 3482 | |
| 3483 | spin_lock(&log->l_ailp->xa_lock); |
| 3484 | /* |
| 3485 | * The RUI has two references. One for the RUD and one for RUI to ensure |
| 3486 | * it makes it into the AIL. Insert the RUI into the AIL directly and |
| 3487 | * drop the RUI reference. Note that xfs_trans_ail_update() drops the |
| 3488 | * AIL lock. |
| 3489 | */ |
| 3490 | xfs_trans_ail_update(log->l_ailp, &ruip->rui_item, lsn); |
| 3491 | xfs_rui_release(ruip); |
| 3492 | return 0; |
| 3493 | } |
| 3494 | |
| 3495 | |
| 3496 | /* |
| 3497 | * This routine is called when an RUD format structure is found in a committed |
| 3498 | * transaction in the log. Its purpose is to cancel the corresponding RUI if it |
| 3499 | * was still in the log. To do this it searches the AIL for the RUI with an id |
| 3500 | * equal to that in the RUD format structure. If we find it we drop the RUD |
| 3501 | * reference, which removes the RUI from the AIL and frees it. |
| 3502 | */ |
| 3503 | STATIC int |
| 3504 | xlog_recover_rud_pass2( |
| 3505 | struct xlog *log, |
| 3506 | struct xlog_recover_item *item) |
| 3507 | { |
| 3508 | struct xfs_rud_log_format *rud_formatp; |
| 3509 | struct xfs_rui_log_item *ruip = NULL; |
| 3510 | struct xfs_log_item *lip; |
| 3511 | __uint64_t rui_id; |
| 3512 | struct xfs_ail_cursor cur; |
| 3513 | struct xfs_ail *ailp = log->l_ailp; |
| 3514 | |
| 3515 | rud_formatp = item->ri_buf[0].i_addr; |
Darrick J. Wong | 722e251 | 2016-08-03 12:28:43 +1000 | [diff] [blame] | 3516 | 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] | 3517 | rui_id = rud_formatp->rud_rui_id; |
| 3518 | |
| 3519 | /* |
| 3520 | * Search for the RUI with the id in the RUD format structure in the |
| 3521 | * AIL. |
| 3522 | */ |
| 3523 | spin_lock(&ailp->xa_lock); |
| 3524 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); |
| 3525 | while (lip != NULL) { |
| 3526 | if (lip->li_type == XFS_LI_RUI) { |
| 3527 | ruip = (struct xfs_rui_log_item *)lip; |
| 3528 | if (ruip->rui_format.rui_id == rui_id) { |
| 3529 | /* |
| 3530 | * Drop the RUD reference to the RUI. This |
| 3531 | * removes the RUI from the AIL and frees it. |
| 3532 | */ |
| 3533 | spin_unlock(&ailp->xa_lock); |
| 3534 | xfs_rui_release(ruip); |
| 3535 | spin_lock(&ailp->xa_lock); |
| 3536 | break; |
| 3537 | } |
| 3538 | } |
| 3539 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
| 3540 | } |
| 3541 | |
| 3542 | xfs_trans_ail_cursor_done(&cur); |
| 3543 | spin_unlock(&ailp->xa_lock); |
| 3544 | |
| 3545 | return 0; |
| 3546 | } |
| 3547 | |
| 3548 | /* |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3549 | * This routine is called when an inode create format structure is found in a |
| 3550 | * committed transaction in the log. It's purpose is to initialise the inodes |
| 3551 | * being allocated on disk. This requires us to get inode cluster buffers that |
| 3552 | * match the range to be intialised, stamped with inode templates and written |
| 3553 | * by delayed write so that subsequent modifications will hit the cached buffer |
| 3554 | * and only need writing out at the end of recovery. |
| 3555 | */ |
| 3556 | STATIC int |
| 3557 | xlog_recover_do_icreate_pass2( |
| 3558 | struct xlog *log, |
| 3559 | struct list_head *buffer_list, |
| 3560 | xlog_recover_item_t *item) |
| 3561 | { |
| 3562 | struct xfs_mount *mp = log->l_mp; |
| 3563 | struct xfs_icreate_log *icl; |
| 3564 | xfs_agnumber_t agno; |
| 3565 | xfs_agblock_t agbno; |
| 3566 | unsigned int count; |
| 3567 | unsigned int isize; |
| 3568 | xfs_agblock_t length; |
Brian Foster | fc0d165 | 2015-08-19 09:59:38 +1000 | [diff] [blame] | 3569 | int blks_per_cluster; |
| 3570 | int bb_per_cluster; |
| 3571 | int cancel_count; |
| 3572 | int nbufs; |
| 3573 | int i; |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3574 | |
| 3575 | icl = (struct xfs_icreate_log *)item->ri_buf[0].i_addr; |
| 3576 | if (icl->icl_type != XFS_LI_ICREATE) { |
| 3577 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad type"); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3578 | return -EINVAL; |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3579 | } |
| 3580 | |
| 3581 | if (icl->icl_size != 1) { |
| 3582 | 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] | 3583 | return -EINVAL; |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3584 | } |
| 3585 | |
| 3586 | agno = be32_to_cpu(icl->icl_ag); |
| 3587 | if (agno >= mp->m_sb.sb_agcount) { |
| 3588 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agno"); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3589 | return -EINVAL; |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3590 | } |
| 3591 | agbno = be32_to_cpu(icl->icl_agbno); |
| 3592 | if (!agbno || agbno == NULLAGBLOCK || agbno >= mp->m_sb.sb_agblocks) { |
| 3593 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agbno"); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3594 | return -EINVAL; |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3595 | } |
| 3596 | isize = be32_to_cpu(icl->icl_isize); |
| 3597 | if (isize != mp->m_sb.sb_inodesize) { |
| 3598 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad isize"); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3599 | return -EINVAL; |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3600 | } |
| 3601 | count = be32_to_cpu(icl->icl_count); |
| 3602 | if (!count) { |
| 3603 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count"); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3604 | return -EINVAL; |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3605 | } |
| 3606 | length = be32_to_cpu(icl->icl_length); |
| 3607 | if (!length || length >= mp->m_sb.sb_agblocks) { |
| 3608 | xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad length"); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3609 | return -EINVAL; |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3610 | } |
| 3611 | |
Brian Foster | 7f43c90 | 2015-05-29 09:06:30 +1000 | [diff] [blame] | 3612 | /* |
| 3613 | * The inode chunk is either full or sparse and we only support |
| 3614 | * m_ialloc_min_blks sized sparse allocations at this time. |
| 3615 | */ |
| 3616 | if (length != mp->m_ialloc_blks && |
| 3617 | length != mp->m_ialloc_min_blks) { |
| 3618 | xfs_warn(log->l_mp, |
| 3619 | "%s: unsupported chunk length", __FUNCTION__); |
| 3620 | return -EINVAL; |
| 3621 | } |
| 3622 | |
| 3623 | /* verify inode count is consistent with extent length */ |
| 3624 | if ((count >> mp->m_sb.sb_inopblog) != length) { |
| 3625 | xfs_warn(log->l_mp, |
| 3626 | "%s: inconsistent inode count and chunk length", |
| 3627 | __FUNCTION__); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3628 | return -EINVAL; |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3629 | } |
| 3630 | |
| 3631 | /* |
Brian Foster | fc0d165 | 2015-08-19 09:59:38 +1000 | [diff] [blame] | 3632 | * The icreate transaction can cover multiple cluster buffers and these |
| 3633 | * buffers could have been freed and reused. Check the individual |
| 3634 | * buffers for cancellation so we don't overwrite anything written after |
| 3635 | * a cancellation. |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3636 | */ |
Brian Foster | fc0d165 | 2015-08-19 09:59:38 +1000 | [diff] [blame] | 3637 | blks_per_cluster = xfs_icluster_size_fsb(mp); |
| 3638 | bb_per_cluster = XFS_FSB_TO_BB(mp, blks_per_cluster); |
| 3639 | nbufs = length / blks_per_cluster; |
| 3640 | for (i = 0, cancel_count = 0; i < nbufs; i++) { |
| 3641 | xfs_daddr_t daddr; |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3642 | |
Brian Foster | fc0d165 | 2015-08-19 09:59:38 +1000 | [diff] [blame] | 3643 | daddr = XFS_AGB_TO_DADDR(mp, agno, |
| 3644 | agbno + i * blks_per_cluster); |
| 3645 | if (xlog_check_buffer_cancelled(log, daddr, bb_per_cluster, 0)) |
| 3646 | cancel_count++; |
| 3647 | } |
| 3648 | |
| 3649 | /* |
| 3650 | * We currently only use icreate for a single allocation at a time. This |
| 3651 | * means we should expect either all or none of the buffers to be |
| 3652 | * cancelled. Be conservative and skip replay if at least one buffer is |
| 3653 | * cancelled, but warn the user that something is awry if the buffers |
| 3654 | * are not consistent. |
| 3655 | * |
| 3656 | * XXX: This must be refined to only skip cancelled clusters once we use |
| 3657 | * icreate for multiple chunk allocations. |
| 3658 | */ |
| 3659 | ASSERT(!cancel_count || cancel_count == nbufs); |
| 3660 | if (cancel_count) { |
| 3661 | if (cancel_count != nbufs) |
| 3662 | xfs_warn(mp, |
| 3663 | "WARNING: partial inode chunk cancellation, skipped icreate."); |
Brian Foster | 78d57e4 | 2015-08-19 09:58:48 +1000 | [diff] [blame] | 3664 | trace_xfs_log_recover_icreate_cancel(log, icl); |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3665 | return 0; |
Brian Foster | 78d57e4 | 2015-08-19 09:58:48 +1000 | [diff] [blame] | 3666 | } |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3667 | |
Brian Foster | 78d57e4 | 2015-08-19 09:58:48 +1000 | [diff] [blame] | 3668 | trace_xfs_log_recover_icreate_recover(log, icl); |
Brian Foster | fc0d165 | 2015-08-19 09:59:38 +1000 | [diff] [blame] | 3669 | return xfs_ialloc_inode_init(mp, NULL, buffer_list, count, agno, agbno, |
| 3670 | length, be32_to_cpu(icl->icl_gen)); |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3671 | } |
| 3672 | |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 3673 | STATIC void |
| 3674 | xlog_recover_buffer_ra_pass2( |
| 3675 | struct xlog *log, |
| 3676 | struct xlog_recover_item *item) |
| 3677 | { |
| 3678 | struct xfs_buf_log_format *buf_f = item->ri_buf[0].i_addr; |
| 3679 | struct xfs_mount *mp = log->l_mp; |
| 3680 | |
Dave Chinner | 84a5b73 | 2013-08-27 08:10:53 +1000 | [diff] [blame] | 3681 | if (xlog_peek_buffer_cancelled(log, buf_f->blf_blkno, |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 3682 | buf_f->blf_len, buf_f->blf_flags)) { |
| 3683 | return; |
| 3684 | } |
| 3685 | |
| 3686 | xfs_buf_readahead(mp->m_ddev_targp, buf_f->blf_blkno, |
| 3687 | buf_f->blf_len, NULL); |
| 3688 | } |
| 3689 | |
| 3690 | STATIC void |
| 3691 | xlog_recover_inode_ra_pass2( |
| 3692 | struct xlog *log, |
| 3693 | struct xlog_recover_item *item) |
| 3694 | { |
| 3695 | struct xfs_inode_log_format ilf_buf; |
| 3696 | struct xfs_inode_log_format *ilfp; |
| 3697 | struct xfs_mount *mp = log->l_mp; |
| 3698 | int error; |
| 3699 | |
| 3700 | if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) { |
| 3701 | ilfp = item->ri_buf[0].i_addr; |
| 3702 | } else { |
| 3703 | ilfp = &ilf_buf; |
| 3704 | memset(ilfp, 0, sizeof(*ilfp)); |
| 3705 | error = xfs_inode_item_format_convert(&item->ri_buf[0], ilfp); |
| 3706 | if (error) |
| 3707 | return; |
| 3708 | } |
| 3709 | |
Dave Chinner | 84a5b73 | 2013-08-27 08:10:53 +1000 | [diff] [blame] | 3710 | 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] | 3711 | return; |
| 3712 | |
| 3713 | xfs_buf_readahead(mp->m_ddev_targp, ilfp->ilf_blkno, |
Dave Chinner | d891400 | 2013-08-27 11:39:37 +1000 | [diff] [blame] | 3714 | ilfp->ilf_len, &xfs_inode_buf_ra_ops); |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 3715 | } |
| 3716 | |
| 3717 | STATIC void |
| 3718 | xlog_recover_dquot_ra_pass2( |
| 3719 | struct xlog *log, |
| 3720 | struct xlog_recover_item *item) |
| 3721 | { |
| 3722 | struct xfs_mount *mp = log->l_mp; |
| 3723 | struct xfs_disk_dquot *recddq; |
| 3724 | struct xfs_dq_logformat *dq_f; |
| 3725 | uint type; |
Dave Chinner | 7d6a13f | 2016-01-12 07:04:01 +1100 | [diff] [blame] | 3726 | int len; |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 3727 | |
| 3728 | |
| 3729 | if (mp->m_qflags == 0) |
| 3730 | return; |
| 3731 | |
| 3732 | recddq = item->ri_buf[1].i_addr; |
| 3733 | if (recddq == NULL) |
| 3734 | return; |
| 3735 | if (item->ri_buf[1].i_len < sizeof(struct xfs_disk_dquot)) |
| 3736 | return; |
| 3737 | |
| 3738 | type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP); |
| 3739 | ASSERT(type); |
| 3740 | if (log->l_quotaoffs_flag & type) |
| 3741 | return; |
| 3742 | |
| 3743 | dq_f = item->ri_buf[0].i_addr; |
| 3744 | ASSERT(dq_f); |
| 3745 | ASSERT(dq_f->qlf_len == 1); |
| 3746 | |
Dave Chinner | 7d6a13f | 2016-01-12 07:04:01 +1100 | [diff] [blame] | 3747 | len = XFS_FSB_TO_BB(mp, dq_f->qlf_len); |
| 3748 | if (xlog_peek_buffer_cancelled(log, dq_f->qlf_blkno, len, 0)) |
| 3749 | return; |
| 3750 | |
| 3751 | xfs_buf_readahead(mp->m_ddev_targp, dq_f->qlf_blkno, len, |
| 3752 | &xfs_dquot_buf_ra_ops); |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 3753 | } |
| 3754 | |
| 3755 | STATIC void |
| 3756 | xlog_recover_ra_pass2( |
| 3757 | struct xlog *log, |
| 3758 | struct xlog_recover_item *item) |
| 3759 | { |
| 3760 | switch (ITEM_TYPE(item)) { |
| 3761 | case XFS_LI_BUF: |
| 3762 | xlog_recover_buffer_ra_pass2(log, item); |
| 3763 | break; |
| 3764 | case XFS_LI_INODE: |
| 3765 | xlog_recover_inode_ra_pass2(log, item); |
| 3766 | break; |
| 3767 | case XFS_LI_DQUOT: |
| 3768 | xlog_recover_dquot_ra_pass2(log, item); |
| 3769 | break; |
| 3770 | case XFS_LI_EFI: |
| 3771 | case XFS_LI_EFD: |
| 3772 | case XFS_LI_QUOTAOFF: |
Darrick J. Wong | 9e88b5d | 2016-08-03 12:09:48 +1000 | [diff] [blame] | 3773 | case XFS_LI_RUI: |
| 3774 | case XFS_LI_RUD: |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 3775 | default: |
| 3776 | break; |
| 3777 | } |
| 3778 | } |
| 3779 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3780 | STATIC int |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3781 | xlog_recover_commit_pass1( |
Mark Tinguely | ad223e6 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 3782 | struct xlog *log, |
| 3783 | struct xlog_recover *trans, |
| 3784 | struct xlog_recover_item *item) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3785 | { |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3786 | trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1); |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 3787 | |
| 3788 | switch (ITEM_TYPE(item)) { |
| 3789 | case XFS_LI_BUF: |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3790 | return xlog_recover_buffer_pass1(log, item); |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 3791 | case XFS_LI_QUOTAOFF: |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3792 | return xlog_recover_quotaoff_pass1(log, item); |
| 3793 | case XFS_LI_INODE: |
| 3794 | case XFS_LI_EFI: |
| 3795 | case XFS_LI_EFD: |
| 3796 | case XFS_LI_DQUOT: |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3797 | case XFS_LI_ICREATE: |
Darrick J. Wong | 9e88b5d | 2016-08-03 12:09:48 +1000 | [diff] [blame] | 3798 | case XFS_LI_RUI: |
| 3799 | case XFS_LI_RUD: |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3800 | /* nothing to do in pass 1 */ |
| 3801 | return 0; |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 3802 | default: |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 3803 | xfs_warn(log->l_mp, "%s: invalid item type (%d)", |
| 3804 | __func__, ITEM_TYPE(item)); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3805 | ASSERT(0); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3806 | return -EIO; |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3807 | } |
| 3808 | } |
| 3809 | |
| 3810 | STATIC int |
| 3811 | xlog_recover_commit_pass2( |
Mark Tinguely | ad223e6 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 3812 | struct xlog *log, |
| 3813 | struct xlog_recover *trans, |
| 3814 | struct list_head *buffer_list, |
| 3815 | struct xlog_recover_item *item) |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3816 | { |
| 3817 | trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2); |
| 3818 | |
| 3819 | switch (ITEM_TYPE(item)) { |
| 3820 | case XFS_LI_BUF: |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 3821 | return xlog_recover_buffer_pass2(log, buffer_list, item, |
| 3822 | trans->r_lsn); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3823 | case XFS_LI_INODE: |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 3824 | return xlog_recover_inode_pass2(log, buffer_list, item, |
| 3825 | trans->r_lsn); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3826 | case XFS_LI_EFI: |
| 3827 | return xlog_recover_efi_pass2(log, item, trans->r_lsn); |
| 3828 | case XFS_LI_EFD: |
| 3829 | return xlog_recover_efd_pass2(log, item); |
Darrick J. Wong | 9e88b5d | 2016-08-03 12:09:48 +1000 | [diff] [blame] | 3830 | case XFS_LI_RUI: |
| 3831 | return xlog_recover_rui_pass2(log, item, trans->r_lsn); |
| 3832 | case XFS_LI_RUD: |
| 3833 | return xlog_recover_rud_pass2(log, item); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3834 | case XFS_LI_DQUOT: |
Dave Chinner | 50d5c8d | 2013-08-28 21:22:47 +1000 | [diff] [blame] | 3835 | return xlog_recover_dquot_pass2(log, buffer_list, item, |
| 3836 | trans->r_lsn); |
Dave Chinner | 28c8e41 | 2013-06-27 16:04:55 +1000 | [diff] [blame] | 3837 | case XFS_LI_ICREATE: |
| 3838 | return xlog_recover_do_icreate_pass2(log, buffer_list, item); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3839 | case XFS_LI_QUOTAOFF: |
| 3840 | /* nothing to do in pass2 */ |
| 3841 | return 0; |
| 3842 | default: |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 3843 | xfs_warn(log->l_mp, "%s: invalid item type (%d)", |
| 3844 | __func__, ITEM_TYPE(item)); |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 3845 | ASSERT(0); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 3846 | return -EIO; |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 3847 | } |
| 3848 | } |
| 3849 | |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 3850 | STATIC int |
| 3851 | xlog_recover_items_pass2( |
| 3852 | struct xlog *log, |
| 3853 | struct xlog_recover *trans, |
| 3854 | struct list_head *buffer_list, |
| 3855 | struct list_head *item_list) |
| 3856 | { |
| 3857 | struct xlog_recover_item *item; |
| 3858 | int error = 0; |
| 3859 | |
| 3860 | list_for_each_entry(item, item_list, ri_list) { |
| 3861 | error = xlog_recover_commit_pass2(log, trans, |
| 3862 | buffer_list, item); |
| 3863 | if (error) |
| 3864 | return error; |
| 3865 | } |
| 3866 | |
| 3867 | return error; |
| 3868 | } |
| 3869 | |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 3870 | /* |
| 3871 | * Perform the transaction. |
| 3872 | * |
| 3873 | * If the transaction modifies a buffer or inode, do it now. Otherwise, |
| 3874 | * EFIs and EFDs get queued up by adding entries into the AIL for them. |
| 3875 | */ |
| 3876 | STATIC int |
| 3877 | xlog_recover_commit_trans( |
Mark Tinguely | ad223e6 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 3878 | struct xlog *log, |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 3879 | struct xlog_recover *trans, |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 3880 | int pass, |
| 3881 | struct list_head *buffer_list) |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 3882 | { |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 3883 | int error = 0; |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 3884 | int items_queued = 0; |
| 3885 | struct xlog_recover_item *item; |
| 3886 | struct xlog_recover_item *next; |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 3887 | LIST_HEAD (ra_list); |
| 3888 | LIST_HEAD (done_list); |
| 3889 | |
| 3890 | #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3891 | |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 3892 | hlist_del(&trans->r_list); |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 3893 | |
| 3894 | error = xlog_recover_reorder_trans(log, trans, pass); |
| 3895 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3896 | return error; |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 3897 | |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 3898 | list_for_each_entry_safe(item, next, &trans->r_itemq, ri_list) { |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 3899 | switch (pass) { |
| 3900 | case XLOG_RECOVER_PASS1: |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 3901 | error = xlog_recover_commit_pass1(log, trans, item); |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 3902 | break; |
| 3903 | case XLOG_RECOVER_PASS2: |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 3904 | xlog_recover_ra_pass2(log, item); |
| 3905 | list_move_tail(&item->ri_list, &ra_list); |
| 3906 | items_queued++; |
| 3907 | if (items_queued >= XLOG_RECOVER_COMMIT_QUEUE_MAX) { |
| 3908 | error = xlog_recover_items_pass2(log, trans, |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 3909 | buffer_list, &ra_list); |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 3910 | list_splice_tail_init(&ra_list, &done_list); |
| 3911 | items_queued = 0; |
| 3912 | } |
| 3913 | |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 3914 | break; |
| 3915 | default: |
| 3916 | ASSERT(0); |
| 3917 | } |
| 3918 | |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 3919 | if (error) |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 3920 | goto out; |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 3921 | } |
| 3922 | |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 3923 | out: |
| 3924 | if (!list_empty(&ra_list)) { |
| 3925 | if (!error) |
| 3926 | error = xlog_recover_items_pass2(log, trans, |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 3927 | buffer_list, &ra_list); |
Zhi Yong Wu | 00574da | 2013-08-14 15:16:03 +0800 | [diff] [blame] | 3928 | list_splice_tail_init(&ra_list, &done_list); |
| 3929 | } |
| 3930 | |
| 3931 | if (!list_empty(&done_list)) |
| 3932 | list_splice_init(&done_list, &trans->r_itemq); |
| 3933 | |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 3934 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3935 | } |
| 3936 | |
Dave Chinner | 7656066 | 2014-09-29 09:45:42 +1000 | [diff] [blame] | 3937 | STATIC void |
| 3938 | xlog_recover_add_item( |
| 3939 | struct list_head *head) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3940 | { |
Dave Chinner | 7656066 | 2014-09-29 09:45:42 +1000 | [diff] [blame] | 3941 | xlog_recover_item_t *item; |
| 3942 | |
| 3943 | item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP); |
| 3944 | INIT_LIST_HEAD(&item->ri_list); |
| 3945 | list_add_tail(&item->ri_list, head); |
| 3946 | } |
| 3947 | |
| 3948 | STATIC int |
| 3949 | xlog_recover_add_to_cont_trans( |
| 3950 | struct xlog *log, |
| 3951 | struct xlog_recover *trans, |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 3952 | char *dp, |
Dave Chinner | 7656066 | 2014-09-29 09:45:42 +1000 | [diff] [blame] | 3953 | int len) |
| 3954 | { |
| 3955 | xlog_recover_item_t *item; |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 3956 | char *ptr, *old_ptr; |
Dave Chinner | 7656066 | 2014-09-29 09:45:42 +1000 | [diff] [blame] | 3957 | int old_len; |
| 3958 | |
Brian Foster | 89cebc84 | 2015-07-29 11:51:10 +1000 | [diff] [blame] | 3959 | /* |
| 3960 | * If the transaction is empty, the header was split across this and the |
| 3961 | * previous record. Copy the rest of the header. |
| 3962 | */ |
Dave Chinner | 7656066 | 2014-09-29 09:45:42 +1000 | [diff] [blame] | 3963 | if (list_empty(&trans->r_itemq)) { |
Brian Foster | 848ccfc | 2015-11-10 10:10:33 +1100 | [diff] [blame] | 3964 | ASSERT(len <= sizeof(struct xfs_trans_header)); |
Brian Foster | 89cebc84 | 2015-07-29 11:51:10 +1000 | [diff] [blame] | 3965 | if (len > sizeof(struct xfs_trans_header)) { |
| 3966 | xfs_warn(log->l_mp, "%s: bad header length", __func__); |
| 3967 | return -EIO; |
| 3968 | } |
| 3969 | |
Dave Chinner | 7656066 | 2014-09-29 09:45:42 +1000 | [diff] [blame] | 3970 | xlog_recover_add_item(&trans->r_itemq); |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 3971 | ptr = (char *)&trans->r_theader + |
Brian Foster | 89cebc84 | 2015-07-29 11:51:10 +1000 | [diff] [blame] | 3972 | sizeof(struct xfs_trans_header) - len; |
Dave Chinner | 7656066 | 2014-09-29 09:45:42 +1000 | [diff] [blame] | 3973 | memcpy(ptr, dp, len); |
| 3974 | return 0; |
| 3975 | } |
Brian Foster | 89cebc84 | 2015-07-29 11:51:10 +1000 | [diff] [blame] | 3976 | |
Dave Chinner | 7656066 | 2014-09-29 09:45:42 +1000 | [diff] [blame] | 3977 | /* take the tail entry */ |
| 3978 | item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list); |
| 3979 | |
| 3980 | old_ptr = item->ri_buf[item->ri_cnt-1].i_addr; |
| 3981 | old_len = item->ri_buf[item->ri_cnt-1].i_len; |
| 3982 | |
Christoph Hellwig | 664b60f | 2016-04-06 09:47:01 +1000 | [diff] [blame] | 3983 | ptr = kmem_realloc(old_ptr, len + old_len, KM_SLEEP); |
Dave Chinner | 7656066 | 2014-09-29 09:45:42 +1000 | [diff] [blame] | 3984 | memcpy(&ptr[old_len], dp, len); |
| 3985 | item->ri_buf[item->ri_cnt-1].i_len += len; |
| 3986 | item->ri_buf[item->ri_cnt-1].i_addr = ptr; |
| 3987 | trace_xfs_log_recover_item_add_cont(log, trans, item, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3988 | return 0; |
| 3989 | } |
| 3990 | |
| 3991 | /* |
Dave Chinner | 7656066 | 2014-09-29 09:45:42 +1000 | [diff] [blame] | 3992 | * The next region to add is the start of a new region. It could be |
| 3993 | * a whole region or it could be the first part of a new region. Because |
| 3994 | * of this, the assumption here is that the type and size fields of all |
| 3995 | * format structures fit into the first 32 bits of the structure. |
| 3996 | * |
| 3997 | * This works because all regions must be 32 bit aligned. Therefore, we |
| 3998 | * either have both fields or we have neither field. In the case we have |
| 3999 | * neither field, the data part of the region is zero length. We only have |
| 4000 | * a log_op_header and can throw away the header since a new one will appear |
| 4001 | * later. If we have at least 4 bytes, then we can determine how many regions |
| 4002 | * will appear in the current log item. |
| 4003 | */ |
| 4004 | STATIC int |
| 4005 | xlog_recover_add_to_trans( |
| 4006 | struct xlog *log, |
| 4007 | struct xlog_recover *trans, |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 4008 | char *dp, |
Dave Chinner | 7656066 | 2014-09-29 09:45:42 +1000 | [diff] [blame] | 4009 | int len) |
| 4010 | { |
| 4011 | xfs_inode_log_format_t *in_f; /* any will do */ |
| 4012 | xlog_recover_item_t *item; |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 4013 | char *ptr; |
Dave Chinner | 7656066 | 2014-09-29 09:45:42 +1000 | [diff] [blame] | 4014 | |
| 4015 | if (!len) |
| 4016 | return 0; |
| 4017 | if (list_empty(&trans->r_itemq)) { |
| 4018 | /* we need to catch log corruptions here */ |
| 4019 | if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) { |
| 4020 | xfs_warn(log->l_mp, "%s: bad header magic number", |
| 4021 | __func__); |
| 4022 | ASSERT(0); |
| 4023 | return -EIO; |
| 4024 | } |
Brian Foster | 89cebc84 | 2015-07-29 11:51:10 +1000 | [diff] [blame] | 4025 | |
| 4026 | if (len > sizeof(struct xfs_trans_header)) { |
| 4027 | xfs_warn(log->l_mp, "%s: bad header length", __func__); |
| 4028 | ASSERT(0); |
| 4029 | return -EIO; |
| 4030 | } |
| 4031 | |
| 4032 | /* |
| 4033 | * The transaction header can be arbitrarily split across op |
| 4034 | * records. If we don't have the whole thing here, copy what we |
| 4035 | * do have and handle the rest in the next record. |
| 4036 | */ |
| 4037 | if (len == sizeof(struct xfs_trans_header)) |
Dave Chinner | 7656066 | 2014-09-29 09:45:42 +1000 | [diff] [blame] | 4038 | xlog_recover_add_item(&trans->r_itemq); |
| 4039 | memcpy(&trans->r_theader, dp, len); |
| 4040 | return 0; |
| 4041 | } |
| 4042 | |
| 4043 | ptr = kmem_alloc(len, KM_SLEEP); |
| 4044 | memcpy(ptr, dp, len); |
| 4045 | in_f = (xfs_inode_log_format_t *)ptr; |
| 4046 | |
| 4047 | /* take the tail entry */ |
| 4048 | item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list); |
| 4049 | if (item->ri_total != 0 && |
| 4050 | item->ri_total == item->ri_cnt) { |
| 4051 | /* tail item is in use, get a new one */ |
| 4052 | xlog_recover_add_item(&trans->r_itemq); |
| 4053 | item = list_entry(trans->r_itemq.prev, |
| 4054 | xlog_recover_item_t, ri_list); |
| 4055 | } |
| 4056 | |
| 4057 | if (item->ri_total == 0) { /* first region to be added */ |
| 4058 | if (in_f->ilf_size == 0 || |
| 4059 | in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) { |
| 4060 | xfs_warn(log->l_mp, |
| 4061 | "bad number of regions (%d) in inode log format", |
| 4062 | in_f->ilf_size); |
| 4063 | ASSERT(0); |
| 4064 | kmem_free(ptr); |
| 4065 | return -EIO; |
| 4066 | } |
| 4067 | |
| 4068 | item->ri_total = in_f->ilf_size; |
| 4069 | item->ri_buf = |
| 4070 | kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t), |
| 4071 | KM_SLEEP); |
| 4072 | } |
| 4073 | ASSERT(item->ri_total > item->ri_cnt); |
| 4074 | /* Description region is ri_buf[0] */ |
| 4075 | item->ri_buf[item->ri_cnt].i_addr = ptr; |
| 4076 | item->ri_buf[item->ri_cnt].i_len = len; |
| 4077 | item->ri_cnt++; |
| 4078 | trace_xfs_log_recover_item_add(log, trans, item, 0); |
| 4079 | return 0; |
| 4080 | } |
Dave Chinner | b818cca | 2014-09-29 09:45:54 +1000 | [diff] [blame] | 4081 | |
Dave Chinner | 7656066 | 2014-09-29 09:45:42 +1000 | [diff] [blame] | 4082 | /* |
| 4083 | * Free up any resources allocated by the transaction |
| 4084 | * |
| 4085 | * Remember that EFIs, EFDs, and IUNLINKs are handled later. |
| 4086 | */ |
| 4087 | STATIC void |
| 4088 | xlog_recover_free_trans( |
| 4089 | struct xlog_recover *trans) |
| 4090 | { |
| 4091 | xlog_recover_item_t *item, *n; |
| 4092 | int i; |
| 4093 | |
| 4094 | list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) { |
| 4095 | /* Free the regions in the item. */ |
| 4096 | list_del(&item->ri_list); |
| 4097 | for (i = 0; i < item->ri_cnt; i++) |
| 4098 | kmem_free(item->ri_buf[i].i_addr); |
| 4099 | /* Free the item itself */ |
| 4100 | kmem_free(item->ri_buf); |
| 4101 | kmem_free(item); |
| 4102 | } |
| 4103 | /* Free the transaction recover structure */ |
| 4104 | kmem_free(trans); |
| 4105 | } |
| 4106 | |
Dave Chinner | e9131e5 | 2014-09-29 09:45:18 +1000 | [diff] [blame] | 4107 | /* |
| 4108 | * On error or completion, trans is freed. |
| 4109 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4110 | STATIC int |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4111 | xlog_recovery_process_trans( |
| 4112 | struct xlog *log, |
| 4113 | struct xlog_recover *trans, |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 4114 | char *dp, |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4115 | unsigned int len, |
| 4116 | unsigned int flags, |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 4117 | int pass, |
| 4118 | struct list_head *buffer_list) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4119 | { |
Dave Chinner | e9131e5 | 2014-09-29 09:45:18 +1000 | [diff] [blame] | 4120 | int error = 0; |
| 4121 | bool freeit = false; |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4122 | |
| 4123 | /* mask off ophdr transaction container flags */ |
| 4124 | flags &= ~XLOG_END_TRANS; |
| 4125 | if (flags & XLOG_WAS_CONT_TRANS) |
| 4126 | flags &= ~XLOG_CONTINUE_TRANS; |
| 4127 | |
Dave Chinner | 88b863d | 2014-09-29 09:45:32 +1000 | [diff] [blame] | 4128 | /* |
| 4129 | * Callees must not free the trans structure. We'll decide if we need to |
| 4130 | * free it or not based on the operation being done and it's result. |
| 4131 | */ |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4132 | switch (flags) { |
| 4133 | /* expected flag values */ |
| 4134 | case 0: |
| 4135 | case XLOG_CONTINUE_TRANS: |
| 4136 | error = xlog_recover_add_to_trans(log, trans, dp, len); |
| 4137 | break; |
| 4138 | case XLOG_WAS_CONT_TRANS: |
| 4139 | error = xlog_recover_add_to_cont_trans(log, trans, dp, len); |
| 4140 | break; |
| 4141 | case XLOG_COMMIT_TRANS: |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 4142 | error = xlog_recover_commit_trans(log, trans, pass, |
| 4143 | buffer_list); |
Dave Chinner | 88b863d | 2014-09-29 09:45:32 +1000 | [diff] [blame] | 4144 | /* success or fail, we are now done with this transaction. */ |
| 4145 | freeit = true; |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4146 | break; |
| 4147 | |
| 4148 | /* unexpected flag values */ |
| 4149 | case XLOG_UNMOUNT_TRANS: |
Dave Chinner | e9131e5 | 2014-09-29 09:45:18 +1000 | [diff] [blame] | 4150 | /* just skip trans */ |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4151 | xfs_warn(log->l_mp, "%s: Unmount LR", __func__); |
Dave Chinner | e9131e5 | 2014-09-29 09:45:18 +1000 | [diff] [blame] | 4152 | freeit = true; |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4153 | break; |
| 4154 | case XLOG_START_TRANS: |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4155 | default: |
| 4156 | xfs_warn(log->l_mp, "%s: bad flag 0x%x", __func__, flags); |
| 4157 | ASSERT(0); |
Dave Chinner | e9131e5 | 2014-09-29 09:45:18 +1000 | [diff] [blame] | 4158 | error = -EIO; |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4159 | break; |
| 4160 | } |
Dave Chinner | e9131e5 | 2014-09-29 09:45:18 +1000 | [diff] [blame] | 4161 | if (error || freeit) |
| 4162 | xlog_recover_free_trans(trans); |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4163 | return error; |
| 4164 | } |
| 4165 | |
Dave Chinner | b818cca | 2014-09-29 09:45:54 +1000 | [diff] [blame] | 4166 | /* |
| 4167 | * Lookup the transaction recovery structure associated with the ID in the |
| 4168 | * current ophdr. If the transaction doesn't exist and the start flag is set in |
| 4169 | * the ophdr, then allocate a new transaction for future ID matches to find. |
| 4170 | * Either way, return what we found during the lookup - an existing transaction |
| 4171 | * or nothing. |
| 4172 | */ |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4173 | STATIC struct xlog_recover * |
| 4174 | xlog_recover_ophdr_to_trans( |
| 4175 | struct hlist_head rhash[], |
| 4176 | struct xlog_rec_header *rhead, |
| 4177 | struct xlog_op_header *ohead) |
| 4178 | { |
| 4179 | struct xlog_recover *trans; |
| 4180 | xlog_tid_t tid; |
| 4181 | struct hlist_head *rhp; |
| 4182 | |
| 4183 | tid = be32_to_cpu(ohead->oh_tid); |
| 4184 | rhp = &rhash[XLOG_RHASH(tid)]; |
Dave Chinner | b818cca | 2014-09-29 09:45:54 +1000 | [diff] [blame] | 4185 | hlist_for_each_entry(trans, rhp, r_list) { |
| 4186 | if (trans->r_log_tid == tid) |
| 4187 | return trans; |
| 4188 | } |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4189 | |
| 4190 | /* |
Dave Chinner | b818cca | 2014-09-29 09:45:54 +1000 | [diff] [blame] | 4191 | * skip over non-start transaction headers - we could be |
| 4192 | * processing slack space before the next transaction starts |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4193 | */ |
Dave Chinner | b818cca | 2014-09-29 09:45:54 +1000 | [diff] [blame] | 4194 | if (!(ohead->oh_flags & XLOG_START_TRANS)) |
| 4195 | return NULL; |
| 4196 | |
| 4197 | ASSERT(be32_to_cpu(ohead->oh_len) == 0); |
| 4198 | |
| 4199 | /* |
| 4200 | * This is a new transaction so allocate a new recovery container to |
| 4201 | * hold the recovery ops that will follow. |
| 4202 | */ |
| 4203 | trans = kmem_zalloc(sizeof(struct xlog_recover), KM_SLEEP); |
| 4204 | trans->r_log_tid = tid; |
| 4205 | trans->r_lsn = be64_to_cpu(rhead->h_lsn); |
| 4206 | INIT_LIST_HEAD(&trans->r_itemq); |
| 4207 | INIT_HLIST_NODE(&trans->r_list); |
| 4208 | hlist_add_head(&trans->r_list, rhp); |
| 4209 | |
| 4210 | /* |
| 4211 | * Nothing more to do for this ophdr. Items to be added to this new |
| 4212 | * transaction will be in subsequent ophdr containers. |
| 4213 | */ |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4214 | return NULL; |
| 4215 | } |
| 4216 | |
| 4217 | STATIC int |
| 4218 | xlog_recover_process_ophdr( |
| 4219 | struct xlog *log, |
| 4220 | struct hlist_head rhash[], |
| 4221 | struct xlog_rec_header *rhead, |
| 4222 | struct xlog_op_header *ohead, |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 4223 | char *dp, |
| 4224 | char *end, |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 4225 | int pass, |
| 4226 | struct list_head *buffer_list) |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4227 | { |
| 4228 | struct xlog_recover *trans; |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4229 | unsigned int len; |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 4230 | int error; |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4231 | |
| 4232 | /* Do we understand who wrote this op? */ |
| 4233 | if (ohead->oh_clientid != XFS_TRANSACTION && |
| 4234 | ohead->oh_clientid != XFS_LOG) { |
| 4235 | xfs_warn(log->l_mp, "%s: bad clientid 0x%x", |
| 4236 | __func__, ohead->oh_clientid); |
| 4237 | ASSERT(0); |
| 4238 | return -EIO; |
| 4239 | } |
| 4240 | |
| 4241 | /* |
| 4242 | * Check the ophdr contains all the data it is supposed to contain. |
| 4243 | */ |
| 4244 | len = be32_to_cpu(ohead->oh_len); |
| 4245 | if (dp + len > end) { |
| 4246 | xfs_warn(log->l_mp, "%s: bad length 0x%x", __func__, len); |
| 4247 | WARN_ON(1); |
| 4248 | return -EIO; |
| 4249 | } |
| 4250 | |
| 4251 | trans = xlog_recover_ophdr_to_trans(rhash, rhead, ohead); |
| 4252 | if (!trans) { |
| 4253 | /* nothing to do, so skip over this ophdr */ |
| 4254 | return 0; |
| 4255 | } |
| 4256 | |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 4257 | /* |
| 4258 | * The recovered buffer queue is drained only once we know that all |
| 4259 | * recovery items for the current LSN have been processed. This is |
| 4260 | * required because: |
| 4261 | * |
| 4262 | * - Buffer write submission updates the metadata LSN of the buffer. |
| 4263 | * - Log recovery skips items with a metadata LSN >= the current LSN of |
| 4264 | * the recovery item. |
| 4265 | * - Separate recovery items against the same metadata buffer can share |
| 4266 | * a current LSN. I.e., consider that the LSN of a recovery item is |
| 4267 | * defined as the starting LSN of the first record in which its |
| 4268 | * transaction appears, that a record can hold multiple transactions, |
| 4269 | * and/or that a transaction can span multiple records. |
| 4270 | * |
| 4271 | * In other words, we are allowed to submit a buffer from log recovery |
| 4272 | * once per current LSN. Otherwise, we may incorrectly skip recovery |
| 4273 | * items and cause corruption. |
| 4274 | * |
| 4275 | * We don't know up front whether buffers are updated multiple times per |
| 4276 | * LSN. Therefore, track the current LSN of each commit log record as it |
| 4277 | * is processed and drain the queue when it changes. Use commit records |
| 4278 | * because they are ordered correctly by the logging code. |
| 4279 | */ |
| 4280 | if (log->l_recovery_lsn != trans->r_lsn && |
| 4281 | ohead->oh_flags & XLOG_COMMIT_TRANS) { |
| 4282 | error = xfs_buf_delwri_submit(buffer_list); |
| 4283 | if (error) |
| 4284 | return error; |
| 4285 | log->l_recovery_lsn = trans->r_lsn; |
| 4286 | } |
| 4287 | |
Dave Chinner | e9131e5 | 2014-09-29 09:45:18 +1000 | [diff] [blame] | 4288 | return xlog_recovery_process_trans(log, trans, dp, len, |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 4289 | ohead->oh_flags, pass, buffer_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4290 | } |
| 4291 | |
| 4292 | /* |
| 4293 | * There are two valid states of the r_state field. 0 indicates that the |
| 4294 | * transaction structure is in a normal state. We have either seen the |
| 4295 | * start of the transaction or the last operation we added was not a partial |
| 4296 | * operation. If the last operation we added to the transaction was a |
| 4297 | * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS. |
| 4298 | * |
| 4299 | * NOTE: skip LRs with 0 data length. |
| 4300 | */ |
| 4301 | STATIC int |
| 4302 | xlog_recover_process_data( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 4303 | struct xlog *log, |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 4304 | struct hlist_head rhash[], |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 4305 | struct xlog_rec_header *rhead, |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 4306 | char *dp, |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 4307 | int pass, |
| 4308 | struct list_head *buffer_list) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4309 | { |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4310 | struct xlog_op_header *ohead; |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 4311 | char *end; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4312 | int num_logops; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4313 | int error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4314 | |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4315 | end = dp + be32_to_cpu(rhead->h_len); |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 4316 | num_logops = be32_to_cpu(rhead->h_num_logops); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4317 | |
| 4318 | /* check the log format matches our own - else we can't recover */ |
| 4319 | if (xlog_header_check_recover(log->l_mp, rhead)) |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 4320 | return -EIO; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4321 | |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4322 | while ((dp < end) && num_logops) { |
| 4323 | |
| 4324 | ohead = (struct xlog_op_header *)dp; |
| 4325 | dp += sizeof(*ohead); |
| 4326 | ASSERT(dp <= end); |
| 4327 | |
| 4328 | /* errors will abort recovery */ |
| 4329 | error = xlog_recover_process_ophdr(log, rhash, rhead, ohead, |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 4330 | dp, end, pass, buffer_list); |
Dave Chinner | eeb1168 | 2014-09-29 09:45:03 +1000 | [diff] [blame] | 4331 | if (error) |
| 4332 | return error; |
| 4333 | |
Christoph Hellwig | 67fcb7b | 2007-10-12 10:58:59 +1000 | [diff] [blame] | 4334 | dp += be32_to_cpu(ohead->oh_len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4335 | num_logops--; |
| 4336 | } |
| 4337 | return 0; |
| 4338 | } |
| 4339 | |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4340 | /* Recover the EFI if necessary. */ |
David Chinner | 3c1e2bb | 2008-04-10 12:21:11 +1000 | [diff] [blame] | 4341 | STATIC int |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4342 | xlog_recover_process_efi( |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4343 | struct xfs_mount *mp, |
| 4344 | struct xfs_ail *ailp, |
| 4345 | struct xfs_log_item *lip) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4346 | { |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4347 | struct xfs_efi_log_item *efip; |
| 4348 | int error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4349 | |
| 4350 | /* |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4351 | * Skip EFIs that we've already processed. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4352 | */ |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4353 | efip = container_of(lip, struct xfs_efi_log_item, efi_item); |
| 4354 | if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) |
| 4355 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4356 | |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4357 | spin_unlock(&ailp->xa_lock); |
| 4358 | error = xfs_efi_recover(mp, efip); |
| 4359 | spin_lock(&ailp->xa_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4360 | |
David Chinner | fc6149d | 2008-04-10 12:21:53 +1000 | [diff] [blame] | 4361 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4362 | } |
| 4363 | |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4364 | /* Release the EFI since we're cancelling everything. */ |
| 4365 | STATIC void |
| 4366 | xlog_recover_cancel_efi( |
| 4367 | struct xfs_mount *mp, |
| 4368 | struct xfs_ail *ailp, |
| 4369 | struct xfs_log_item *lip) |
| 4370 | { |
| 4371 | struct xfs_efi_log_item *efip; |
| 4372 | |
| 4373 | efip = container_of(lip, struct xfs_efi_log_item, efi_item); |
| 4374 | |
| 4375 | spin_unlock(&ailp->xa_lock); |
| 4376 | xfs_efi_release(efip); |
| 4377 | spin_lock(&ailp->xa_lock); |
| 4378 | } |
| 4379 | |
Darrick J. Wong | 9e88b5d | 2016-08-03 12:09:48 +1000 | [diff] [blame] | 4380 | /* Recover the RUI if necessary. */ |
| 4381 | STATIC int |
| 4382 | xlog_recover_process_rui( |
| 4383 | struct xfs_mount *mp, |
| 4384 | struct xfs_ail *ailp, |
| 4385 | struct xfs_log_item *lip) |
| 4386 | { |
| 4387 | struct xfs_rui_log_item *ruip; |
| 4388 | int error; |
| 4389 | |
| 4390 | /* |
| 4391 | * Skip RUIs that we've already processed. |
| 4392 | */ |
| 4393 | ruip = container_of(lip, struct xfs_rui_log_item, rui_item); |
| 4394 | if (test_bit(XFS_RUI_RECOVERED, &ruip->rui_flags)) |
| 4395 | return 0; |
| 4396 | |
| 4397 | spin_unlock(&ailp->xa_lock); |
| 4398 | error = xfs_rui_recover(mp, ruip); |
| 4399 | spin_lock(&ailp->xa_lock); |
| 4400 | |
| 4401 | return error; |
| 4402 | } |
| 4403 | |
| 4404 | /* Release the RUI since we're cancelling everything. */ |
| 4405 | STATIC void |
| 4406 | xlog_recover_cancel_rui( |
| 4407 | struct xfs_mount *mp, |
| 4408 | struct xfs_ail *ailp, |
| 4409 | struct xfs_log_item *lip) |
| 4410 | { |
| 4411 | struct xfs_rui_log_item *ruip; |
| 4412 | |
| 4413 | ruip = container_of(lip, struct xfs_rui_log_item, rui_item); |
| 4414 | |
| 4415 | spin_unlock(&ailp->xa_lock); |
| 4416 | xfs_rui_release(ruip); |
| 4417 | spin_lock(&ailp->xa_lock); |
| 4418 | } |
| 4419 | |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4420 | /* Is this log item a deferred action intent? */ |
| 4421 | static inline bool xlog_item_is_intent(struct xfs_log_item *lip) |
| 4422 | { |
| 4423 | switch (lip->li_type) { |
| 4424 | case XFS_LI_EFI: |
Darrick J. Wong | 9e88b5d | 2016-08-03 12:09:48 +1000 | [diff] [blame] | 4425 | case XFS_LI_RUI: |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4426 | return true; |
| 4427 | default: |
| 4428 | return false; |
| 4429 | } |
| 4430 | } |
| 4431 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4432 | /* |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4433 | * When this is called, all of the log intent items which did not have |
| 4434 | * corresponding log done items should be in the AIL. What we do now |
| 4435 | * is update the data structures associated with each one. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4436 | * |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4437 | * Since we process the log intent items in normal transactions, they |
| 4438 | * will be removed at some point after the commit. This prevents us |
| 4439 | * from just walking down the list processing each one. We'll use a |
| 4440 | * flag in the intent item to skip those that we've already processed |
| 4441 | * and use the AIL iteration mechanism's generation count to try to |
| 4442 | * speed this up at least a bit. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4443 | * |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4444 | * When we start, we know that the intents are the only things in the |
| 4445 | * AIL. As we process them, however, other items are added to the |
| 4446 | * AIL. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4447 | */ |
David Chinner | 3c1e2bb | 2008-04-10 12:21:11 +1000 | [diff] [blame] | 4448 | STATIC int |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4449 | xlog_recover_process_intents( |
Brian Foster | f0b2efa | 2015-08-19 09:58:36 +1000 | [diff] [blame] | 4450 | struct xlog *log) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4451 | { |
Brian Foster | f0b2efa | 2015-08-19 09:58:36 +1000 | [diff] [blame] | 4452 | struct xfs_log_item *lip; |
David Chinner | 3c1e2bb | 2008-04-10 12:21:11 +1000 | [diff] [blame] | 4453 | int error = 0; |
David Chinner | 27d8d5f | 2008-10-30 17:38:39 +1100 | [diff] [blame] | 4454 | struct xfs_ail_cursor cur; |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 4455 | struct xfs_ail *ailp; |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4456 | xfs_lsn_t last_lsn; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4457 | |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 4458 | ailp = log->l_ailp; |
| 4459 | spin_lock(&ailp->xa_lock); |
| 4460 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4461 | 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] | 4462 | while (lip != NULL) { |
| 4463 | /* |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4464 | * We're done when we see something other than an intent. |
| 4465 | * There should be no intents left in the AIL now. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4466 | */ |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4467 | if (!xlog_item_is_intent(lip)) { |
David Chinner | 27d8d5f | 2008-10-30 17:38:39 +1100 | [diff] [blame] | 4468 | #ifdef DEBUG |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 4469 | for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur)) |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4470 | ASSERT(!xlog_item_is_intent(lip)); |
David Chinner | 27d8d5f | 2008-10-30 17:38:39 +1100 | [diff] [blame] | 4471 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4472 | break; |
| 4473 | } |
| 4474 | |
| 4475 | /* |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4476 | * We should never see a redo item with a LSN higher than |
| 4477 | * the last transaction we found in the log at the start |
| 4478 | * of recovery. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4479 | */ |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4480 | ASSERT(XFS_LSN_CMP(last_lsn, lip->li_lsn) >= 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4481 | |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4482 | switch (lip->li_type) { |
| 4483 | case XFS_LI_EFI: |
| 4484 | error = xlog_recover_process_efi(log->l_mp, ailp, lip); |
| 4485 | break; |
Darrick J. Wong | 9e88b5d | 2016-08-03 12:09:48 +1000 | [diff] [blame] | 4486 | case XFS_LI_RUI: |
| 4487 | error = xlog_recover_process_rui(log->l_mp, ailp, lip); |
| 4488 | break; |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4489 | } |
David Chinner | 27d8d5f | 2008-10-30 17:38:39 +1100 | [diff] [blame] | 4490 | if (error) |
| 4491 | goto out; |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 4492 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4493 | } |
David Chinner | 27d8d5f | 2008-10-30 17:38:39 +1100 | [diff] [blame] | 4494 | out: |
Eric Sandeen | e4a1e29 | 2014-04-14 19:06:05 +1000 | [diff] [blame] | 4495 | xfs_trans_ail_cursor_done(&cur); |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 4496 | spin_unlock(&ailp->xa_lock); |
David Chinner | 3c1e2bb | 2008-04-10 12:21:11 +1000 | [diff] [blame] | 4497 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4498 | } |
| 4499 | |
| 4500 | /* |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4501 | * A cancel occurs when the mount has failed and we're bailing out. |
| 4502 | * 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] | 4503 | */ |
| 4504 | STATIC int |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4505 | xlog_recover_cancel_intents( |
Brian Foster | f0b2efa | 2015-08-19 09:58:36 +1000 | [diff] [blame] | 4506 | struct xlog *log) |
| 4507 | { |
| 4508 | struct xfs_log_item *lip; |
Brian Foster | f0b2efa | 2015-08-19 09:58:36 +1000 | [diff] [blame] | 4509 | int error = 0; |
| 4510 | struct xfs_ail_cursor cur; |
| 4511 | struct xfs_ail *ailp; |
| 4512 | |
| 4513 | ailp = log->l_ailp; |
| 4514 | spin_lock(&ailp->xa_lock); |
| 4515 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); |
| 4516 | while (lip != NULL) { |
| 4517 | /* |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4518 | * We're done when we see something other than an intent. |
| 4519 | * There should be no intents left in the AIL now. |
Brian Foster | f0b2efa | 2015-08-19 09:58:36 +1000 | [diff] [blame] | 4520 | */ |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4521 | if (!xlog_item_is_intent(lip)) { |
Brian Foster | f0b2efa | 2015-08-19 09:58:36 +1000 | [diff] [blame] | 4522 | #ifdef DEBUG |
| 4523 | for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur)) |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4524 | ASSERT(!xlog_item_is_intent(lip)); |
Brian Foster | f0b2efa | 2015-08-19 09:58:36 +1000 | [diff] [blame] | 4525 | #endif |
| 4526 | break; |
| 4527 | } |
| 4528 | |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4529 | switch (lip->li_type) { |
| 4530 | case XFS_LI_EFI: |
| 4531 | xlog_recover_cancel_efi(log->l_mp, ailp, lip); |
| 4532 | break; |
Darrick J. Wong | 9e88b5d | 2016-08-03 12:09:48 +1000 | [diff] [blame] | 4533 | case XFS_LI_RUI: |
| 4534 | xlog_recover_cancel_rui(log->l_mp, ailp, lip); |
| 4535 | break; |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 4536 | } |
Brian Foster | f0b2efa | 2015-08-19 09:58:36 +1000 | [diff] [blame] | 4537 | |
| 4538 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
| 4539 | } |
| 4540 | |
| 4541 | xfs_trans_ail_cursor_done(&cur); |
| 4542 | spin_unlock(&ailp->xa_lock); |
| 4543 | return error; |
| 4544 | } |
| 4545 | |
| 4546 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4547 | * This routine performs a transaction to null out a bad inode pointer |
| 4548 | * in an agi unlinked inode hash bucket. |
| 4549 | */ |
| 4550 | STATIC void |
| 4551 | xlog_recover_clear_agi_bucket( |
| 4552 | xfs_mount_t *mp, |
| 4553 | xfs_agnumber_t agno, |
| 4554 | int bucket) |
| 4555 | { |
| 4556 | xfs_trans_t *tp; |
| 4557 | xfs_agi_t *agi; |
| 4558 | xfs_buf_t *agibp; |
| 4559 | int offset; |
| 4560 | int error; |
| 4561 | |
Christoph Hellwig | 253f491 | 2016-04-06 09:19:55 +1000 | [diff] [blame] | 4562 | 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] | 4563 | if (error) |
Christoph Hellwig | 253f491 | 2016-04-06 09:19:55 +1000 | [diff] [blame] | 4564 | goto out_error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4565 | |
Christoph Hellwig | 5e1be0f | 2008-11-28 14:23:37 +1100 | [diff] [blame] | 4566 | error = xfs_read_agi(mp, tp, agno, &agibp); |
| 4567 | if (error) |
David Chinner | e5720ee | 2008-04-10 12:21:18 +1000 | [diff] [blame] | 4568 | goto out_abort; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4569 | |
Christoph Hellwig | 5e1be0f | 2008-11-28 14:23:37 +1100 | [diff] [blame] | 4570 | agi = XFS_BUF_TO_AGI(agibp); |
Christoph Hellwig | 16259e7 | 2005-11-02 15:11:25 +1100 | [diff] [blame] | 4571 | agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4572 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
| 4573 | (sizeof(xfs_agino_t) * bucket); |
| 4574 | xfs_trans_log_buf(tp, agibp, offset, |
| 4575 | (offset + sizeof(xfs_agino_t) - 1)); |
| 4576 | |
Christoph Hellwig | 7039331 | 2015-06-04 13:48:08 +1000 | [diff] [blame] | 4577 | error = xfs_trans_commit(tp); |
David Chinner | e5720ee | 2008-04-10 12:21:18 +1000 | [diff] [blame] | 4578 | if (error) |
| 4579 | goto out_error; |
| 4580 | return; |
| 4581 | |
| 4582 | out_abort: |
Christoph Hellwig | 4906e21 | 2015-06-04 13:47:56 +1000 | [diff] [blame] | 4583 | xfs_trans_cancel(tp); |
David Chinner | e5720ee | 2008-04-10 12:21:18 +1000 | [diff] [blame] | 4584 | out_error: |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 4585 | xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno); |
David Chinner | e5720ee | 2008-04-10 12:21:18 +1000 | [diff] [blame] | 4586 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4587 | } |
| 4588 | |
Christoph Hellwig | 23fac50 | 2008-11-28 14:23:40 +1100 | [diff] [blame] | 4589 | STATIC xfs_agino_t |
| 4590 | xlog_recover_process_one_iunlink( |
| 4591 | struct xfs_mount *mp, |
| 4592 | xfs_agnumber_t agno, |
| 4593 | xfs_agino_t agino, |
| 4594 | int bucket) |
| 4595 | { |
| 4596 | struct xfs_buf *ibp; |
| 4597 | struct xfs_dinode *dip; |
| 4598 | struct xfs_inode *ip; |
| 4599 | xfs_ino_t ino; |
| 4600 | int error; |
| 4601 | |
| 4602 | ino = XFS_AGINO_TO_INO(mp, agno, agino); |
Dave Chinner | 7b6259e | 2010-06-24 11:35:17 +1000 | [diff] [blame] | 4603 | error = xfs_iget(mp, NULL, ino, 0, 0, &ip); |
Christoph Hellwig | 23fac50 | 2008-11-28 14:23:40 +1100 | [diff] [blame] | 4604 | if (error) |
| 4605 | goto fail; |
| 4606 | |
| 4607 | /* |
| 4608 | * Get the on disk inode to find the next inode in the bucket. |
| 4609 | */ |
Christoph Hellwig | 475ee41 | 2012-07-03 12:21:22 -0400 | [diff] [blame] | 4610 | 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] | 4611 | if (error) |
Christoph Hellwig | 0e44667 | 2008-11-28 14:23:42 +1100 | [diff] [blame] | 4612 | goto fail_iput; |
Christoph Hellwig | 23fac50 | 2008-11-28 14:23:40 +1100 | [diff] [blame] | 4613 | |
Dave Chinner | 54d7b5c | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 4614 | ASSERT(VFS_I(ip)->i_nlink == 0); |
Dave Chinner | c19b3b05 | 2016-02-09 16:54:58 +1100 | [diff] [blame] | 4615 | ASSERT(VFS_I(ip)->i_mode != 0); |
Christoph Hellwig | 23fac50 | 2008-11-28 14:23:40 +1100 | [diff] [blame] | 4616 | |
| 4617 | /* setup for the next pass */ |
| 4618 | agino = be32_to_cpu(dip->di_next_unlinked); |
| 4619 | xfs_buf_relse(ibp); |
| 4620 | |
| 4621 | /* |
| 4622 | * Prevent any DMAPI event from being sent when the reference on |
| 4623 | * the inode is dropped. |
| 4624 | */ |
| 4625 | ip->i_d.di_dmevmask = 0; |
| 4626 | |
Christoph Hellwig | 0e44667 | 2008-11-28 14:23:42 +1100 | [diff] [blame] | 4627 | IRELE(ip); |
Christoph Hellwig | 23fac50 | 2008-11-28 14:23:40 +1100 | [diff] [blame] | 4628 | return agino; |
| 4629 | |
Christoph Hellwig | 0e44667 | 2008-11-28 14:23:42 +1100 | [diff] [blame] | 4630 | fail_iput: |
| 4631 | IRELE(ip); |
Christoph Hellwig | 23fac50 | 2008-11-28 14:23:40 +1100 | [diff] [blame] | 4632 | fail: |
| 4633 | /* |
| 4634 | * We can't read in the inode this bucket points to, or this inode |
| 4635 | * is messed up. Just ditch this bucket of inodes. We will lose |
| 4636 | * some inodes and space, but at least we won't hang. |
| 4637 | * |
| 4638 | * Call xlog_recover_clear_agi_bucket() to perform a transaction to |
| 4639 | * clear the inode pointer in the bucket. |
| 4640 | */ |
| 4641 | xlog_recover_clear_agi_bucket(mp, agno, bucket); |
| 4642 | return NULLAGINO; |
| 4643 | } |
| 4644 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4645 | /* |
| 4646 | * xlog_iunlink_recover |
| 4647 | * |
| 4648 | * This is called during recovery to process any inodes which |
| 4649 | * we unlinked but not freed when the system crashed. These |
| 4650 | * inodes will be on the lists in the AGI blocks. What we do |
| 4651 | * here is scan all the AGIs and fully truncate and free any |
| 4652 | * inodes found on the lists. Each inode is removed from the |
| 4653 | * lists when it has been fully truncated and is freed. The |
| 4654 | * freeing of the inode and its removal from the list must be |
| 4655 | * atomic. |
| 4656 | */ |
Eric Sandeen | d96f8f8 | 2009-07-02 00:09:33 -0500 | [diff] [blame] | 4657 | STATIC void |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4658 | xlog_recover_process_iunlinks( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 4659 | struct xlog *log) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4660 | { |
| 4661 | xfs_mount_t *mp; |
| 4662 | xfs_agnumber_t agno; |
| 4663 | xfs_agi_t *agi; |
| 4664 | xfs_buf_t *agibp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4665 | xfs_agino_t agino; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4666 | int bucket; |
| 4667 | int error; |
| 4668 | uint mp_dmevmask; |
| 4669 | |
| 4670 | mp = log->l_mp; |
| 4671 | |
| 4672 | /* |
| 4673 | * Prevent any DMAPI event from being sent while in this function. |
| 4674 | */ |
| 4675 | mp_dmevmask = mp->m_dmevmask; |
| 4676 | mp->m_dmevmask = 0; |
| 4677 | |
| 4678 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { |
| 4679 | /* |
| 4680 | * Find the agi for this ag. |
| 4681 | */ |
Christoph Hellwig | 5e1be0f | 2008-11-28 14:23:37 +1100 | [diff] [blame] | 4682 | error = xfs_read_agi(mp, NULL, agno, &agibp); |
| 4683 | if (error) { |
| 4684 | /* |
| 4685 | * AGI is b0rked. Don't process it. |
| 4686 | * |
| 4687 | * We should probably mark the filesystem as corrupt |
| 4688 | * after we've recovered all the ag's we can.... |
| 4689 | */ |
| 4690 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4691 | } |
Jan Kara | d97d32e | 2012-03-15 09:34:02 +0000 | [diff] [blame] | 4692 | /* |
| 4693 | * Unlock the buffer so that it can be acquired in the normal |
| 4694 | * course of the transaction to truncate and free each inode. |
| 4695 | * Because we are not racing with anyone else here for the AGI |
| 4696 | * buffer, we don't even need to hold it locked to read the |
| 4697 | * initial unlinked bucket entries out of the buffer. We keep |
| 4698 | * buffer reference though, so that it stays pinned in memory |
| 4699 | * while we need the buffer. |
| 4700 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4701 | agi = XFS_BUF_TO_AGI(agibp); |
Jan Kara | d97d32e | 2012-03-15 09:34:02 +0000 | [diff] [blame] | 4702 | xfs_buf_unlock(agibp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4703 | |
| 4704 | for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) { |
Christoph Hellwig | 16259e7 | 2005-11-02 15:11:25 +1100 | [diff] [blame] | 4705 | agino = be32_to_cpu(agi->agi_unlinked[bucket]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4706 | while (agino != NULLAGINO) { |
Christoph Hellwig | 23fac50 | 2008-11-28 14:23:40 +1100 | [diff] [blame] | 4707 | agino = xlog_recover_process_one_iunlink(mp, |
| 4708 | agno, agino, bucket); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4709 | } |
| 4710 | } |
Jan Kara | d97d32e | 2012-03-15 09:34:02 +0000 | [diff] [blame] | 4711 | xfs_buf_rele(agibp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4712 | } |
| 4713 | |
| 4714 | mp->m_dmevmask = mp_dmevmask; |
| 4715 | } |
| 4716 | |
Christoph Hellwig | 0e446be | 2012-11-12 22:54:24 +1100 | [diff] [blame] | 4717 | STATIC int |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4718 | xlog_unpack_data( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 4719 | struct xlog_rec_header *rhead, |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 4720 | char *dp, |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 4721 | struct xlog *log) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4722 | { |
| 4723 | int i, j, k; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4724 | |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 4725 | for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) && |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4726 | i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) { |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 4727 | *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4728 | dp += BBSIZE; |
| 4729 | } |
| 4730 | |
Eric Sandeen | 6211870 | 2008-03-06 13:44:28 +1100 | [diff] [blame] | 4731 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
Christoph Hellwig | b28708d | 2008-11-28 14:23:38 +1100 | [diff] [blame] | 4732 | xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead; |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 4733 | for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4734 | j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); |
| 4735 | k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 4736 | *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4737 | dp += BBSIZE; |
| 4738 | } |
| 4739 | } |
Christoph Hellwig | 0e446be | 2012-11-12 22:54:24 +1100 | [diff] [blame] | 4740 | |
| 4741 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4742 | } |
| 4743 | |
Brian Foster | 9d94901 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 4744 | /* |
Brian Foster | b94fb2d | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 4745 | * CRC check, unpack and process a log record. |
Brian Foster | 9d94901 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 4746 | */ |
| 4747 | STATIC int |
| 4748 | xlog_recover_process( |
| 4749 | struct xlog *log, |
| 4750 | struct hlist_head rhash[], |
| 4751 | struct xlog_rec_header *rhead, |
| 4752 | char *dp, |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 4753 | int pass, |
| 4754 | struct list_head *buffer_list) |
Brian Foster | 9d94901 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 4755 | { |
| 4756 | int error; |
Brian Foster | b94fb2d | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 4757 | __le32 crc; |
| 4758 | |
Brian Foster | b94fb2d | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 4759 | crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len)); |
Brian Foster | 6528250 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 4760 | |
| 4761 | /* |
| 4762 | * Nothing else to do if this is a CRC verification pass. Just return |
| 4763 | * if this a record with a non-zero crc. Unfortunately, mkfs always |
| 4764 | * sets h_crc to 0 so we must consider this valid even on v5 supers. |
| 4765 | * Otherwise, return EFSBADCRC on failure so the callers up the stack |
| 4766 | * know precisely what failed. |
| 4767 | */ |
| 4768 | if (pass == XLOG_RECOVER_CRCPASS) { |
Darrick J. Wong | 8e0bd49 | 2016-02-08 11:03:58 +1100 | [diff] [blame] | 4769 | if (rhead->h_crc && crc != rhead->h_crc) |
Brian Foster | 6528250 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 4770 | return -EFSBADCRC; |
| 4771 | return 0; |
| 4772 | } |
| 4773 | |
| 4774 | /* |
| 4775 | * We're in the normal recovery path. Issue a warning if and only if the |
| 4776 | * CRC in the header is non-zero. This is an advisory warning and the |
| 4777 | * zero CRC check prevents warnings from being emitted when upgrading |
| 4778 | * the kernel from one that does not add CRCs by default. |
| 4779 | */ |
Darrick J. Wong | 8e0bd49 | 2016-02-08 11:03:58 +1100 | [diff] [blame] | 4780 | if (crc != rhead->h_crc) { |
Brian Foster | b94fb2d | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 4781 | if (rhead->h_crc || xfs_sb_version_hascrc(&log->l_mp->m_sb)) { |
| 4782 | xfs_alert(log->l_mp, |
| 4783 | "log record CRC mismatch: found 0x%x, expected 0x%x.", |
| 4784 | le32_to_cpu(rhead->h_crc), |
| 4785 | le32_to_cpu(crc)); |
| 4786 | xfs_hex_dump(dp, 32); |
| 4787 | } |
| 4788 | |
| 4789 | /* |
| 4790 | * If the filesystem is CRC enabled, this mismatch becomes a |
| 4791 | * fatal log corruption failure. |
| 4792 | */ |
| 4793 | if (xfs_sb_version_hascrc(&log->l_mp->m_sb)) |
| 4794 | return -EFSCORRUPTED; |
| 4795 | } |
Brian Foster | 9d94901 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 4796 | |
| 4797 | error = xlog_unpack_data(rhead, dp, log); |
| 4798 | if (error) |
| 4799 | return error; |
| 4800 | |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 4801 | return xlog_recover_process_data(log, rhash, rhead, dp, pass, |
| 4802 | buffer_list); |
Brian Foster | 9d94901 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 4803 | } |
| 4804 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4805 | STATIC int |
| 4806 | xlog_valid_rec_header( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 4807 | struct xlog *log, |
| 4808 | struct xlog_rec_header *rhead, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4809 | xfs_daddr_t blkno) |
| 4810 | { |
| 4811 | int hlen; |
| 4812 | |
Christoph Hellwig | 69ef921 | 2011-07-08 14:36:05 +0200 | [diff] [blame] | 4813 | if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4814 | XFS_ERROR_REPORT("xlog_valid_rec_header(1)", |
| 4815 | XFS_ERRLEVEL_LOW, log->l_mp); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 4816 | return -EFSCORRUPTED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4817 | } |
| 4818 | if (unlikely( |
| 4819 | (!rhead->h_version || |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 4820 | (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 4821 | xfs_warn(log->l_mp, "%s: unrecognised log version (%d).", |
Harvey Harrison | 34a622b | 2008-04-10 12:19:21 +1000 | [diff] [blame] | 4822 | __func__, be32_to_cpu(rhead->h_version)); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 4823 | return -EIO; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4824 | } |
| 4825 | |
| 4826 | /* LR body must have data or it wouldn't have been written */ |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 4827 | hlen = be32_to_cpu(rhead->h_len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4828 | if (unlikely( hlen <= 0 || hlen > INT_MAX )) { |
| 4829 | XFS_ERROR_REPORT("xlog_valid_rec_header(2)", |
| 4830 | XFS_ERRLEVEL_LOW, log->l_mp); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 4831 | return -EFSCORRUPTED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4832 | } |
| 4833 | if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) { |
| 4834 | XFS_ERROR_REPORT("xlog_valid_rec_header(3)", |
| 4835 | XFS_ERRLEVEL_LOW, log->l_mp); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 4836 | return -EFSCORRUPTED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4837 | } |
| 4838 | return 0; |
| 4839 | } |
| 4840 | |
| 4841 | /* |
| 4842 | * Read the log from tail to head and process the log records found. |
| 4843 | * Handle the two cases where the tail and head are in the same cycle |
| 4844 | * and where the active portion of the log wraps around the end of |
| 4845 | * the physical log separately. The pass parameter is passed through |
| 4846 | * to the routines called to process the data and is not looked at |
| 4847 | * here. |
| 4848 | */ |
| 4849 | STATIC int |
| 4850 | xlog_do_recovery_pass( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 4851 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4852 | xfs_daddr_t head_blk, |
| 4853 | xfs_daddr_t tail_blk, |
Brian Foster | d7f3769 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 4854 | int pass, |
| 4855 | xfs_daddr_t *first_bad) /* out: first bad log rec */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4856 | { |
| 4857 | xlog_rec_header_t *rhead; |
| 4858 | xfs_daddr_t blk_no; |
Brian Foster | d7f3769 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 4859 | xfs_daddr_t rhead_blk; |
Christoph Hellwig | b2a922c | 2015-06-22 09:45:10 +1000 | [diff] [blame] | 4860 | char *offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4861 | xfs_buf_t *hbp, *dbp; |
Brian Foster | a70f9fe | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 4862 | int error = 0, h_size, h_len; |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 4863 | int error2 = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4864 | int bblks, split_bblks; |
| 4865 | int hblks, split_hblks, wrapped_hblks; |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 4866 | struct hlist_head rhash[XLOG_RHASH_SIZE]; |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 4867 | LIST_HEAD (buffer_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4868 | |
| 4869 | ASSERT(head_blk != tail_blk); |
Brian Foster | d7f3769 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 4870 | rhead_blk = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4871 | |
| 4872 | /* |
| 4873 | * Read the header of the tail block and get the iclog buffer size from |
| 4874 | * h_size. Use this to tell how many sectors make up the log header. |
| 4875 | */ |
Eric Sandeen | 6211870 | 2008-03-06 13:44:28 +1100 | [diff] [blame] | 4876 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4877 | /* |
| 4878 | * When using variable length iclogs, read first sector of |
| 4879 | * iclog header and extract the header size from it. Get a |
| 4880 | * new hbp that is the correct size. |
| 4881 | */ |
| 4882 | hbp = xlog_get_bp(log, 1); |
| 4883 | if (!hbp) |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 4884 | return -ENOMEM; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 4885 | |
| 4886 | error = xlog_bread(log, tail_blk, 1, hbp, &offset); |
| 4887 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4888 | goto bread_err1; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 4889 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4890 | rhead = (xlog_rec_header_t *)offset; |
| 4891 | error = xlog_valid_rec_header(log, rhead, tail_blk); |
| 4892 | if (error) |
| 4893 | goto bread_err1; |
Brian Foster | a70f9fe | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 4894 | |
| 4895 | /* |
| 4896 | * xfsprogs has a bug where record length is based on lsunit but |
| 4897 | * h_size (iclog size) is hardcoded to 32k. Now that we |
| 4898 | * unconditionally CRC verify the unmount record, this means the |
| 4899 | * log buffer can be too small for the record and cause an |
| 4900 | * overrun. |
| 4901 | * |
| 4902 | * Detect this condition here. Use lsunit for the buffer size as |
| 4903 | * long as this looks like the mkfs case. Otherwise, return an |
| 4904 | * error to avoid a buffer overrun. |
| 4905 | */ |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 4906 | h_size = be32_to_cpu(rhead->h_size); |
Brian Foster | a70f9fe | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 4907 | h_len = be32_to_cpu(rhead->h_len); |
| 4908 | if (h_len > h_size) { |
| 4909 | if (h_len <= log->l_mp->m_logbsize && |
| 4910 | be32_to_cpu(rhead->h_num_logops) == 1) { |
| 4911 | xfs_warn(log->l_mp, |
| 4912 | "invalid iclog size (%d bytes), using lsunit (%d bytes)", |
| 4913 | h_size, log->l_mp->m_logbsize); |
| 4914 | h_size = log->l_mp->m_logbsize; |
| 4915 | } else |
| 4916 | return -EFSCORRUPTED; |
| 4917 | } |
| 4918 | |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 4919 | if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) && |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4920 | (h_size > XLOG_HEADER_CYCLE_SIZE)) { |
| 4921 | hblks = h_size / XLOG_HEADER_CYCLE_SIZE; |
| 4922 | if (h_size % XLOG_HEADER_CYCLE_SIZE) |
| 4923 | hblks++; |
| 4924 | xlog_put_bp(hbp); |
| 4925 | hbp = xlog_get_bp(log, hblks); |
| 4926 | } else { |
| 4927 | hblks = 1; |
| 4928 | } |
| 4929 | } else { |
Alex Elder | 69ce58f | 2010-04-20 17:09:59 +1000 | [diff] [blame] | 4930 | ASSERT(log->l_sectBBsize == 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4931 | hblks = 1; |
| 4932 | hbp = xlog_get_bp(log, 1); |
| 4933 | h_size = XLOG_BIG_RECORD_BSIZE; |
| 4934 | } |
| 4935 | |
| 4936 | if (!hbp) |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 4937 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4938 | dbp = xlog_get_bp(log, BTOBB(h_size)); |
| 4939 | if (!dbp) { |
| 4940 | xlog_put_bp(hbp); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 4941 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4942 | } |
| 4943 | |
| 4944 | memset(rhash, 0, sizeof(rhash)); |
Brian Foster | d7f3769 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 4945 | blk_no = rhead_blk = tail_blk; |
Eric Sandeen | 970fd3f | 2014-09-09 11:57:29 +1000 | [diff] [blame] | 4946 | if (tail_blk > head_blk) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4947 | /* |
| 4948 | * Perform recovery around the end of the physical log. |
| 4949 | * 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] | 4950 | * we can't do a sequential recovery. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4951 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4952 | while (blk_no < log->l_logBBsize) { |
| 4953 | /* |
| 4954 | * Check for header wrapping around physical end-of-log |
| 4955 | */ |
Chandra Seetharaman | 6292604 | 2011-07-22 23:40:15 +0000 | [diff] [blame] | 4956 | offset = hbp->b_addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4957 | split_hblks = 0; |
| 4958 | wrapped_hblks = 0; |
| 4959 | if (blk_no + hblks <= log->l_logBBsize) { |
| 4960 | /* Read header in one read */ |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 4961 | error = xlog_bread(log, blk_no, hblks, hbp, |
| 4962 | &offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4963 | if (error) |
| 4964 | goto bread_err2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4965 | } else { |
| 4966 | /* This LR is split across physical log end */ |
| 4967 | if (blk_no != log->l_logBBsize) { |
| 4968 | /* some data before physical log end */ |
| 4969 | ASSERT(blk_no <= INT_MAX); |
| 4970 | split_hblks = log->l_logBBsize - (int)blk_no; |
| 4971 | ASSERT(split_hblks > 0); |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 4972 | error = xlog_bread(log, blk_no, |
| 4973 | split_hblks, hbp, |
| 4974 | &offset); |
| 4975 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4976 | goto bread_err2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4977 | } |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 4978 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4979 | /* |
| 4980 | * Note: this black magic still works with |
| 4981 | * large sector sizes (non-512) only because: |
| 4982 | * - we increased the buffer size originally |
| 4983 | * by 1 sector giving us enough extra space |
| 4984 | * for the second read; |
| 4985 | * - the log start is guaranteed to be sector |
| 4986 | * aligned; |
| 4987 | * - we read the log end (LR header start) |
| 4988 | * _first_, then the log start (LR header end) |
| 4989 | * - order is important. |
| 4990 | */ |
David Chinner | 234f56a | 2008-04-10 12:24:24 +1000 | [diff] [blame] | 4991 | wrapped_hblks = hblks - split_hblks; |
Dave Chinner | 4439647 | 2011-04-21 09:34:27 +0000 | [diff] [blame] | 4992 | error = xlog_bread_offset(log, 0, |
| 4993 | wrapped_hblks, hbp, |
| 4994 | offset + BBTOB(split_hblks)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4995 | if (error) |
| 4996 | goto bread_err2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4997 | } |
| 4998 | rhead = (xlog_rec_header_t *)offset; |
| 4999 | error = xlog_valid_rec_header(log, rhead, |
| 5000 | split_hblks ? blk_no : 0); |
| 5001 | if (error) |
| 5002 | goto bread_err2; |
| 5003 | |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 5004 | bblks = (int)BTOBB(be32_to_cpu(rhead->h_len)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5005 | blk_no += hblks; |
| 5006 | |
| 5007 | /* Read in data for log record */ |
| 5008 | if (blk_no + bblks <= log->l_logBBsize) { |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 5009 | error = xlog_bread(log, blk_no, bblks, dbp, |
| 5010 | &offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5011 | if (error) |
| 5012 | goto bread_err2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5013 | } else { |
| 5014 | /* This log record is split across the |
| 5015 | * physical end of log */ |
Chandra Seetharaman | 6292604 | 2011-07-22 23:40:15 +0000 | [diff] [blame] | 5016 | offset = dbp->b_addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5017 | split_bblks = 0; |
| 5018 | if (blk_no != log->l_logBBsize) { |
| 5019 | /* some data is before the physical |
| 5020 | * end of log */ |
| 5021 | ASSERT(!wrapped_hblks); |
| 5022 | ASSERT(blk_no <= INT_MAX); |
| 5023 | split_bblks = |
| 5024 | log->l_logBBsize - (int)blk_no; |
| 5025 | ASSERT(split_bblks > 0); |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 5026 | error = xlog_bread(log, blk_no, |
| 5027 | split_bblks, dbp, |
| 5028 | &offset); |
| 5029 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5030 | goto bread_err2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5031 | } |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 5032 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5033 | /* |
| 5034 | * Note: this black magic still works with |
| 5035 | * large sector sizes (non-512) only because: |
| 5036 | * - we increased the buffer size originally |
| 5037 | * by 1 sector giving us enough extra space |
| 5038 | * for the second read; |
| 5039 | * - the log start is guaranteed to be sector |
| 5040 | * aligned; |
| 5041 | * - we read the log end (LR header start) |
| 5042 | * _first_, then the log start (LR header end) |
| 5043 | * - order is important. |
| 5044 | */ |
Dave Chinner | 4439647 | 2011-04-21 09:34:27 +0000 | [diff] [blame] | 5045 | error = xlog_bread_offset(log, 0, |
Dave Chinner | 009507b | 2012-11-02 11:38:44 +1100 | [diff] [blame] | 5046 | bblks - split_bblks, dbp, |
Dave Chinner | 4439647 | 2011-04-21 09:34:27 +0000 | [diff] [blame] | 5047 | offset + BBTOB(split_bblks)); |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 5048 | if (error) |
| 5049 | goto bread_err2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5050 | } |
Christoph Hellwig | 0e446be | 2012-11-12 22:54:24 +1100 | [diff] [blame] | 5051 | |
Brian Foster | 9d94901 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5052 | error = xlog_recover_process(log, rhash, rhead, offset, |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 5053 | pass, &buffer_list); |
Christoph Hellwig | 0e446be | 2012-11-12 22:54:24 +1100 | [diff] [blame] | 5054 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5055 | goto bread_err2; |
Brian Foster | d7f3769 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5056 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5057 | blk_no += bblks; |
Brian Foster | d7f3769 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5058 | rhead_blk = blk_no; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5059 | } |
| 5060 | |
| 5061 | ASSERT(blk_no >= log->l_logBBsize); |
| 5062 | blk_no -= log->l_logBBsize; |
Brian Foster | d7f3769 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5063 | rhead_blk = blk_no; |
Eric Sandeen | 970fd3f | 2014-09-09 11:57:29 +1000 | [diff] [blame] | 5064 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5065 | |
Eric Sandeen | 970fd3f | 2014-09-09 11:57:29 +1000 | [diff] [blame] | 5066 | /* read first part of physical log */ |
| 5067 | while (blk_no < head_blk) { |
| 5068 | error = xlog_bread(log, blk_no, hblks, hbp, &offset); |
| 5069 | if (error) |
| 5070 | goto bread_err2; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 5071 | |
Eric Sandeen | 970fd3f | 2014-09-09 11:57:29 +1000 | [diff] [blame] | 5072 | rhead = (xlog_rec_header_t *)offset; |
| 5073 | error = xlog_valid_rec_header(log, rhead, blk_no); |
| 5074 | if (error) |
| 5075 | goto bread_err2; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 5076 | |
Eric Sandeen | 970fd3f | 2014-09-09 11:57:29 +1000 | [diff] [blame] | 5077 | /* blocks in data section */ |
| 5078 | bblks = (int)BTOBB(be32_to_cpu(rhead->h_len)); |
| 5079 | error = xlog_bread(log, blk_no+hblks, bblks, dbp, |
| 5080 | &offset); |
| 5081 | if (error) |
| 5082 | goto bread_err2; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 5083 | |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 5084 | error = xlog_recover_process(log, rhash, rhead, offset, pass, |
| 5085 | &buffer_list); |
Eric Sandeen | 970fd3f | 2014-09-09 11:57:29 +1000 | [diff] [blame] | 5086 | if (error) |
| 5087 | goto bread_err2; |
Brian Foster | d7f3769 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5088 | |
Eric Sandeen | 970fd3f | 2014-09-09 11:57:29 +1000 | [diff] [blame] | 5089 | blk_no += bblks + hblks; |
Brian Foster | d7f3769 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5090 | rhead_blk = blk_no; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5091 | } |
| 5092 | |
| 5093 | bread_err2: |
| 5094 | xlog_put_bp(dbp); |
| 5095 | bread_err1: |
| 5096 | xlog_put_bp(hbp); |
Brian Foster | d7f3769 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5097 | |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 5098 | /* |
| 5099 | * Submit buffers that have been added from the last record processed, |
| 5100 | * regardless of error status. |
| 5101 | */ |
| 5102 | if (!list_empty(&buffer_list)) |
| 5103 | error2 = xfs_buf_delwri_submit(&buffer_list); |
| 5104 | |
Brian Foster | d7f3769 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5105 | if (error && first_bad) |
| 5106 | *first_bad = rhead_blk; |
| 5107 | |
Brian Foster | 12818d2 | 2016-09-26 08:22:16 +1000 | [diff] [blame] | 5108 | return error ? error : error2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5109 | } |
| 5110 | |
| 5111 | /* |
| 5112 | * Do the recovery of the log. We actually do this in two phases. |
| 5113 | * The two passes are necessary in order to implement the function |
| 5114 | * of cancelling a record written into the log. The first pass |
| 5115 | * determines those things which have been cancelled, and the |
| 5116 | * second pass replays log items normally except for those which |
| 5117 | * have been cancelled. The handling of the replay and cancellations |
| 5118 | * takes place in the log item type specific routines. |
| 5119 | * |
| 5120 | * The table of items which have cancel records in the log is allocated |
| 5121 | * and freed at this level, since only here do we know when all of |
| 5122 | * the log recovery has been completed. |
| 5123 | */ |
| 5124 | STATIC int |
| 5125 | xlog_do_log_recovery( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 5126 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5127 | xfs_daddr_t head_blk, |
| 5128 | xfs_daddr_t tail_blk) |
| 5129 | { |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 5130 | int error, i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5131 | |
| 5132 | ASSERT(head_blk != tail_blk); |
| 5133 | |
| 5134 | /* |
| 5135 | * First do a pass to find all of the cancelled buf log items. |
| 5136 | * Store them in the buf_cancel_table for use in the second pass. |
| 5137 | */ |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 5138 | log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE * |
| 5139 | sizeof(struct list_head), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5140 | KM_SLEEP); |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 5141 | for (i = 0; i < XLOG_BC_TABLE_SIZE; i++) |
| 5142 | INIT_LIST_HEAD(&log->l_buf_cancel_table[i]); |
| 5143 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5144 | error = xlog_do_recovery_pass(log, head_blk, tail_blk, |
Brian Foster | d7f3769 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5145 | XLOG_RECOVER_PASS1, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5146 | if (error != 0) { |
Denys Vlasenko | f0e2d93 | 2008-05-19 16:31:57 +1000 | [diff] [blame] | 5147 | kmem_free(log->l_buf_cancel_table); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5148 | log->l_buf_cancel_table = NULL; |
| 5149 | return error; |
| 5150 | } |
| 5151 | /* |
| 5152 | * Then do a second pass to actually recover the items in the log. |
| 5153 | * When it is complete free the table of buf cancel items. |
| 5154 | */ |
| 5155 | error = xlog_do_recovery_pass(log, head_blk, tail_blk, |
Brian Foster | d7f3769 | 2016-01-04 15:55:10 +1100 | [diff] [blame] | 5156 | XLOG_RECOVER_PASS2, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5157 | #ifdef DEBUG |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 5158 | if (!error) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5159 | int i; |
| 5160 | |
| 5161 | for (i = 0; i < XLOG_BC_TABLE_SIZE; i++) |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 5162 | ASSERT(list_empty(&log->l_buf_cancel_table[i])); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5163 | } |
| 5164 | #endif /* DEBUG */ |
| 5165 | |
Denys Vlasenko | f0e2d93 | 2008-05-19 16:31:57 +1000 | [diff] [blame] | 5166 | kmem_free(log->l_buf_cancel_table); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5167 | log->l_buf_cancel_table = NULL; |
| 5168 | |
| 5169 | return error; |
| 5170 | } |
| 5171 | |
| 5172 | /* |
| 5173 | * Do the actual recovery |
| 5174 | */ |
| 5175 | STATIC int |
| 5176 | xlog_do_recover( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 5177 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5178 | xfs_daddr_t head_blk, |
| 5179 | xfs_daddr_t tail_blk) |
| 5180 | { |
Dave Chinner | a798011 | 2016-03-07 08:39:36 +1100 | [diff] [blame] | 5181 | struct xfs_mount *mp = log->l_mp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5182 | int error; |
| 5183 | xfs_buf_t *bp; |
| 5184 | xfs_sb_t *sbp; |
| 5185 | |
| 5186 | /* |
| 5187 | * First replay the images in the log. |
| 5188 | */ |
| 5189 | error = xlog_do_log_recovery(log, head_blk, tail_blk); |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 5190 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5191 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5192 | |
| 5193 | /* |
| 5194 | * If IO errors happened during recovery, bail out. |
| 5195 | */ |
Dave Chinner | a798011 | 2016-03-07 08:39:36 +1100 | [diff] [blame] | 5196 | if (XFS_FORCED_SHUTDOWN(mp)) { |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 5197 | return -EIO; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5198 | } |
| 5199 | |
| 5200 | /* |
| 5201 | * We now update the tail_lsn since much of the recovery has completed |
| 5202 | * and there may be space available to use. If there were no extent |
| 5203 | * or iunlinks, we can free up the entire log and set the tail_lsn to |
| 5204 | * be the last_sync_lsn. This was set in xlog_find_tail to be the |
| 5205 | * lsn of the last known good LR on disk. If there are extent frees |
| 5206 | * or iunlinks they will have some entries in the AIL; so we look at |
| 5207 | * the AIL to determine how to set the tail_lsn. |
| 5208 | */ |
Dave Chinner | a798011 | 2016-03-07 08:39:36 +1100 | [diff] [blame] | 5209 | xlog_assign_tail_lsn(mp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5210 | |
| 5211 | /* |
| 5212 | * Now that we've finished replaying all buffer and inode |
Dave Chinner | 9802182 | 2012-11-12 22:54:03 +1100 | [diff] [blame] | 5213 | * updates, re-read in the superblock and reverify it. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5214 | */ |
Dave Chinner | a798011 | 2016-03-07 08:39:36 +1100 | [diff] [blame] | 5215 | bp = xfs_getsb(mp, 0); |
Dave Chinner | 1157b32c | 2016-02-10 15:01:11 +1100 | [diff] [blame] | 5216 | bp->b_flags &= ~(XBF_DONE | XBF_ASYNC); |
Dave Chinner | b68c082 | 2016-02-10 15:01:11 +1100 | [diff] [blame] | 5217 | ASSERT(!(bp->b_flags & XBF_WRITE)); |
Dave Chinner | 0cac682 | 2016-02-10 15:01:11 +1100 | [diff] [blame] | 5218 | bp->b_flags |= XBF_READ; |
Dave Chinner | 1813dd6 | 2012-11-14 17:54:40 +1100 | [diff] [blame] | 5219 | bp->b_ops = &xfs_sb_buf_ops; |
Christoph Hellwig | 83a0adc | 2013-12-17 00:03:52 -0800 | [diff] [blame] | 5220 | |
Dave Chinner | 595bff7 | 2014-10-02 09:05:14 +1000 | [diff] [blame] | 5221 | error = xfs_buf_submit_wait(bp); |
David Chinner | d64e31a | 2008-04-10 12:22:17 +1000 | [diff] [blame] | 5222 | if (error) { |
Dave Chinner | a798011 | 2016-03-07 08:39:36 +1100 | [diff] [blame] | 5223 | if (!XFS_FORCED_SHUTDOWN(mp)) { |
Dave Chinner | 595bff7 | 2014-10-02 09:05:14 +1000 | [diff] [blame] | 5224 | xfs_buf_ioerror_alert(bp, __func__); |
| 5225 | ASSERT(0); |
| 5226 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5227 | xfs_buf_relse(bp); |
| 5228 | return error; |
| 5229 | } |
| 5230 | |
| 5231 | /* Convert superblock from on-disk format */ |
Dave Chinner | a798011 | 2016-03-07 08:39:36 +1100 | [diff] [blame] | 5232 | sbp = &mp->m_sb; |
Dave Chinner | 9802182 | 2012-11-12 22:54:03 +1100 | [diff] [blame] | 5233 | xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5234 | xfs_buf_relse(bp); |
| 5235 | |
Dave Chinner | a798011 | 2016-03-07 08:39:36 +1100 | [diff] [blame] | 5236 | /* re-initialise in-core superblock and geometry structures */ |
| 5237 | xfs_reinit_percpu_counters(mp); |
| 5238 | error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi); |
| 5239 | if (error) { |
| 5240 | xfs_warn(mp, "Failed post-recovery per-ag init: %d", error); |
| 5241 | return error; |
| 5242 | } |
Darrick J. Wong | 5254885 | 2016-08-03 11:38:24 +1000 | [diff] [blame] | 5243 | mp->m_alloc_set_aside = xfs_alloc_set_aside(mp); |
Lachlan McIlroy | 5478eea | 2007-02-10 18:36:29 +1100 | [diff] [blame] | 5244 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5245 | xlog_recover_check_summary(log); |
| 5246 | |
| 5247 | /* Normal transactions can now occur */ |
| 5248 | log->l_flags &= ~XLOG_ACTIVE_RECOVERY; |
| 5249 | return 0; |
| 5250 | } |
| 5251 | |
| 5252 | /* |
| 5253 | * Perform recovery and re-initialize some log variables in xlog_find_tail. |
| 5254 | * |
| 5255 | * Return error or zero. |
| 5256 | */ |
| 5257 | int |
| 5258 | xlog_recover( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 5259 | struct xlog *log) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5260 | { |
| 5261 | xfs_daddr_t head_blk, tail_blk; |
| 5262 | int error; |
| 5263 | |
| 5264 | /* find the tail of the log */ |
Brian Foster | a45086e | 2015-10-12 15:59:25 +1100 | [diff] [blame] | 5265 | error = xlog_find_tail(log, &head_blk, &tail_blk); |
| 5266 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5267 | return error; |
| 5268 | |
Brian Foster | a45086e | 2015-10-12 15:59:25 +1100 | [diff] [blame] | 5269 | /* |
| 5270 | * The superblock was read before the log was available and thus the LSN |
| 5271 | * could not be verified. Check the superblock LSN against the current |
| 5272 | * LSN now that it's known. |
| 5273 | */ |
| 5274 | if (xfs_sb_version_hascrc(&log->l_mp->m_sb) && |
| 5275 | !xfs_log_check_lsn(log->l_mp, log->l_mp->m_sb.sb_lsn)) |
| 5276 | return -EINVAL; |
| 5277 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5278 | if (tail_blk != head_blk) { |
| 5279 | /* There used to be a comment here: |
| 5280 | * |
| 5281 | * disallow recovery on read-only mounts. note -- mount |
| 5282 | * checks for ENOSPC and turns it into an intelligent |
| 5283 | * error message. |
| 5284 | * ...but this is no longer true. Now, unless you specify |
| 5285 | * NORECOVERY (in which case this function would never be |
| 5286 | * called), we just go ahead and recover. We do this all |
| 5287 | * under the vfs layer, so we can get away with it unless |
| 5288 | * the device itself is read-only, in which case we fail. |
| 5289 | */ |
Utako Kusaka | 3a02ee1 | 2007-05-08 13:50:06 +1000 | [diff] [blame] | 5290 | if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5291 | return error; |
| 5292 | } |
| 5293 | |
Dave Chinner | e721f50 | 2013-04-03 16:11:32 +1100 | [diff] [blame] | 5294 | /* |
| 5295 | * Version 5 superblock log feature mask validation. We know the |
| 5296 | * log is dirty so check if there are any unknown log features |
| 5297 | * in what we need to recover. If there are unknown features |
| 5298 | * (e.g. unsupported transactions, then simply reject the |
| 5299 | * attempt at recovery before touching anything. |
| 5300 | */ |
| 5301 | if (XFS_SB_VERSION_NUM(&log->l_mp->m_sb) == XFS_SB_VERSION_5 && |
| 5302 | xfs_sb_has_incompat_log_feature(&log->l_mp->m_sb, |
| 5303 | XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) { |
| 5304 | xfs_warn(log->l_mp, |
Joe Perches | f41febd | 2015-07-29 11:52:04 +1000 | [diff] [blame] | 5305 | "Superblock has unknown incompatible log features (0x%x) enabled.", |
Dave Chinner | e721f50 | 2013-04-03 16:11:32 +1100 | [diff] [blame] | 5306 | (log->l_mp->m_sb.sb_features_log_incompat & |
| 5307 | XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)); |
Joe Perches | f41febd | 2015-07-29 11:52:04 +1000 | [diff] [blame] | 5308 | xfs_warn(log->l_mp, |
| 5309 | "The log can not be fully and/or safely recovered by this kernel."); |
| 5310 | xfs_warn(log->l_mp, |
| 5311 | "Please recover the log on a kernel that supports the unknown features."); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 5312 | return -EINVAL; |
Dave Chinner | e721f50 | 2013-04-03 16:11:32 +1100 | [diff] [blame] | 5313 | } |
| 5314 | |
Brian Foster | 2e22717 | 2014-09-09 11:56:13 +1000 | [diff] [blame] | 5315 | /* |
| 5316 | * Delay log recovery if the debug hook is set. This is debug |
| 5317 | * instrumention to coordinate simulation of I/O failures with |
| 5318 | * log recovery. |
| 5319 | */ |
| 5320 | if (xfs_globals.log_recovery_delay) { |
| 5321 | xfs_notice(log->l_mp, |
| 5322 | "Delaying log recovery for %d seconds.", |
| 5323 | xfs_globals.log_recovery_delay); |
| 5324 | msleep(xfs_globals.log_recovery_delay * 1000); |
| 5325 | } |
| 5326 | |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 5327 | xfs_notice(log->l_mp, "Starting recovery (logdev: %s)", |
| 5328 | log->l_mp->m_logname ? log->l_mp->m_logname |
| 5329 | : "internal"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5330 | |
| 5331 | error = xlog_do_recover(log, head_blk, tail_blk); |
| 5332 | log->l_flags |= XLOG_RECOVERY_NEEDED; |
| 5333 | } |
| 5334 | return error; |
| 5335 | } |
| 5336 | |
| 5337 | /* |
| 5338 | * In the first part of recovery we replay inodes and buffers and build |
| 5339 | * up the list of extent free items which need to be processed. Here |
| 5340 | * we process the extent free items and clean up the on disk unlinked |
| 5341 | * inode lists. This is separated from the first part of recovery so |
| 5342 | * that the root and real-time bitmap inodes can be read in from disk in |
| 5343 | * between the two stages. This is necessary so that we can free space |
| 5344 | * in the real-time portion of the file system. |
| 5345 | */ |
| 5346 | int |
| 5347 | xlog_recover_finish( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 5348 | struct xlog *log) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5349 | { |
| 5350 | /* |
| 5351 | * Now we're ready to do the transactions needed for the |
| 5352 | * rest of recovery. Start with completing all the extent |
| 5353 | * free intent records and then process the unlinked inode |
| 5354 | * lists. At this point, we essentially run in normal mode |
| 5355 | * except that we're still performing recovery actions |
| 5356 | * rather than accepting new requests. |
| 5357 | */ |
| 5358 | if (log->l_flags & XLOG_RECOVERY_NEEDED) { |
David Chinner | 3c1e2bb | 2008-04-10 12:21:11 +1000 | [diff] [blame] | 5359 | int error; |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 5360 | error = xlog_recover_process_intents(log); |
David Chinner | 3c1e2bb | 2008-04-10 12:21:11 +1000 | [diff] [blame] | 5361 | if (error) { |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 5362 | xfs_alert(log->l_mp, "Failed to recover intents"); |
David Chinner | 3c1e2bb | 2008-04-10 12:21:11 +1000 | [diff] [blame] | 5363 | return error; |
| 5364 | } |
Darrick J. Wong | 9e88b5d | 2016-08-03 12:09:48 +1000 | [diff] [blame] | 5365 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5366 | /* |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 5367 | * Sync the log to get all the intents out of the AIL. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5368 | * This isn't absolutely necessary, but it helps in |
| 5369 | * case the unlink transactions would have problems |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 5370 | * pushing the intents out of the way. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5371 | */ |
Christoph Hellwig | a14a348 | 2010-01-19 09:56:46 +0000 | [diff] [blame] | 5372 | xfs_log_force(log->l_mp, XFS_LOG_SYNC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5373 | |
Christoph Hellwig | 4249023 | 2008-08-13 16:49:32 +1000 | [diff] [blame] | 5374 | xlog_recover_process_iunlinks(log); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5375 | |
| 5376 | xlog_recover_check_summary(log); |
| 5377 | |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 5378 | xfs_notice(log->l_mp, "Ending recovery (logdev: %s)", |
| 5379 | log->l_mp->m_logname ? log->l_mp->m_logname |
| 5380 | : "internal"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5381 | log->l_flags &= ~XLOG_RECOVERY_NEEDED; |
| 5382 | } else { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 5383 | xfs_info(log->l_mp, "Ending clean mount"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5384 | } |
| 5385 | return 0; |
| 5386 | } |
| 5387 | |
Brian Foster | f0b2efa | 2015-08-19 09:58:36 +1000 | [diff] [blame] | 5388 | int |
| 5389 | xlog_recover_cancel( |
| 5390 | struct xlog *log) |
| 5391 | { |
| 5392 | int error = 0; |
| 5393 | |
| 5394 | if (log->l_flags & XLOG_RECOVERY_NEEDED) |
Darrick J. Wong | dc42375 | 2016-08-03 11:23:49 +1000 | [diff] [blame] | 5395 | error = xlog_recover_cancel_intents(log); |
Brian Foster | f0b2efa | 2015-08-19 09:58:36 +1000 | [diff] [blame] | 5396 | |
| 5397 | return error; |
| 5398 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5399 | |
| 5400 | #if defined(DEBUG) |
| 5401 | /* |
| 5402 | * Read all of the agf and agi counters and check that they |
| 5403 | * are consistent with the superblock counters. |
| 5404 | */ |
| 5405 | void |
| 5406 | xlog_recover_check_summary( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 5407 | struct xlog *log) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5408 | { |
| 5409 | xfs_mount_t *mp; |
| 5410 | xfs_agf_t *agfp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5411 | xfs_buf_t *agfbp; |
| 5412 | xfs_buf_t *agibp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5413 | xfs_agnumber_t agno; |
| 5414 | __uint64_t freeblks; |
| 5415 | __uint64_t itotal; |
| 5416 | __uint64_t ifree; |
Christoph Hellwig | 5e1be0f | 2008-11-28 14:23:37 +1100 | [diff] [blame] | 5417 | int error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5418 | |
| 5419 | mp = log->l_mp; |
| 5420 | |
| 5421 | freeblks = 0LL; |
| 5422 | itotal = 0LL; |
| 5423 | ifree = 0LL; |
| 5424 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { |
From: Christoph Hellwig | 4805621 | 2008-11-28 14:23:38 +1100 | [diff] [blame] | 5425 | error = xfs_read_agf(mp, NULL, agno, 0, &agfbp); |
| 5426 | if (error) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 5427 | xfs_alert(mp, "%s agf read failed agno %d error %d", |
| 5428 | __func__, agno, error); |
From: Christoph Hellwig | 4805621 | 2008-11-28 14:23:38 +1100 | [diff] [blame] | 5429 | } else { |
| 5430 | agfp = XFS_BUF_TO_AGF(agfbp); |
| 5431 | freeblks += be32_to_cpu(agfp->agf_freeblks) + |
| 5432 | be32_to_cpu(agfp->agf_flcount); |
| 5433 | xfs_buf_relse(agfbp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5434 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5435 | |
Christoph Hellwig | 5e1be0f | 2008-11-28 14:23:37 +1100 | [diff] [blame] | 5436 | error = xfs_read_agi(mp, NULL, agno, &agibp); |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 5437 | if (error) { |
| 5438 | xfs_alert(mp, "%s agi read failed agno %d error %d", |
| 5439 | __func__, agno, error); |
| 5440 | } else { |
Christoph Hellwig | 5e1be0f | 2008-11-28 14:23:37 +1100 | [diff] [blame] | 5441 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5442 | |
Christoph Hellwig | 5e1be0f | 2008-11-28 14:23:37 +1100 | [diff] [blame] | 5443 | itotal += be32_to_cpu(agi->agi_count); |
| 5444 | ifree += be32_to_cpu(agi->agi_freecount); |
| 5445 | xfs_buf_relse(agibp); |
| 5446 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5447 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5448 | } |
| 5449 | #endif /* DEBUG */ |