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" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 20 | #include "xfs_types.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 21 | #include "xfs_bit.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 22 | #include "xfs_log.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 23 | #include "xfs_inum.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 24 | #include "xfs_trans.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 25 | #include "xfs_sb.h" |
| 26 | #include "xfs_ag.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 27 | #include "xfs_mount.h" |
| 28 | #include "xfs_error.h" |
| 29 | #include "xfs_bmap_btree.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 30 | #include "xfs_alloc_btree.h" |
| 31 | #include "xfs_ialloc_btree.h" |
Christoph Hellwig | ee1a47a | 2013-04-21 14:53:46 -0500 | [diff] [blame] | 32 | #include "xfs_btree.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 33 | #include "xfs_dinode.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 34 | #include "xfs_inode.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 35 | #include "xfs_inode_item.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 36 | #include "xfs_alloc.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 37 | #include "xfs_ialloc.h" |
| 38 | #include "xfs_log_priv.h" |
| 39 | #include "xfs_buf_item.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 40 | #include "xfs_log_recover.h" |
| 41 | #include "xfs_extfree_item.h" |
| 42 | #include "xfs_trans_priv.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 43 | #include "xfs_quota.h" |
Christoph Hellwig | 4335509 | 2008-03-27 18:01:08 +1100 | [diff] [blame] | 44 | #include "xfs_utils.h" |
Christoph Hellwig | 0e446be | 2012-11-12 22:54:24 +1100 | [diff] [blame] | 45 | #include "xfs_cksum.h" |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 46 | #include "xfs_trace.h" |
Dave Chinner | 33479e0 | 2012-10-08 21:56:11 +1100 | [diff] [blame] | 47 | #include "xfs_icache.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 48 | |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 49 | STATIC int |
| 50 | xlog_find_zeroed( |
| 51 | struct xlog *, |
| 52 | xfs_daddr_t *); |
| 53 | STATIC int |
| 54 | xlog_clear_stale_blocks( |
| 55 | struct xlog *, |
| 56 | xfs_lsn_t); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 57 | #if defined(DEBUG) |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 58 | STATIC void |
| 59 | xlog_recover_check_summary( |
| 60 | struct xlog *); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 61 | #else |
| 62 | #define xlog_recover_check_summary(log) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 63 | #endif |
| 64 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 65 | /* |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 66 | * This structure is used during recovery to record the buf log items which |
| 67 | * have been canceled and should not be replayed. |
| 68 | */ |
| 69 | struct xfs_buf_cancel { |
| 70 | xfs_daddr_t bc_blkno; |
| 71 | uint bc_len; |
| 72 | int bc_refcount; |
| 73 | struct list_head bc_list; |
| 74 | }; |
| 75 | |
| 76 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 77 | * Sector aligned buffer routines for buffer create/read/write/access |
| 78 | */ |
| 79 | |
Alex Elder | ff30a62 | 2010-04-13 15:22:58 +1000 | [diff] [blame] | 80 | /* |
| 81 | * Verify the given count of basic blocks is valid number of blocks |
| 82 | * to specify for an operation involving the given XFS log buffer. |
| 83 | * Returns nonzero if the count is valid, 0 otherwise. |
| 84 | */ |
| 85 | |
| 86 | static inline int |
| 87 | xlog_buf_bbcount_valid( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 88 | struct xlog *log, |
Alex Elder | ff30a62 | 2010-04-13 15:22:58 +1000 | [diff] [blame] | 89 | int bbcount) |
| 90 | { |
| 91 | return bbcount > 0 && bbcount <= log->l_logBBsize; |
| 92 | } |
| 93 | |
Alex Elder | 36adecf | 2010-04-13 15:21:13 +1000 | [diff] [blame] | 94 | /* |
| 95 | * Allocate a buffer to hold log data. The buffer needs to be able |
| 96 | * to map to a range of nbblks basic blocks at any valid (basic |
| 97 | * block) offset within the log. |
| 98 | */ |
Eric Sandeen | 5d77c0d | 2009-11-19 15:52:00 +0000 | [diff] [blame] | 99 | STATIC xfs_buf_t * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 100 | xlog_get_bp( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 101 | struct xlog *log, |
Dave Chinner | 3228149 | 2009-01-22 15:37:47 +1100 | [diff] [blame] | 102 | int nbblks) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 103 | { |
Christoph Hellwig | c8da0fa | 2011-07-08 14:36:25 +0200 | [diff] [blame] | 104 | struct xfs_buf *bp; |
| 105 | |
Alex Elder | ff30a62 | 2010-04-13 15:22:58 +1000 | [diff] [blame] | 106 | if (!xlog_buf_bbcount_valid(log, nbblks)) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 107 | xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer", |
Alex Elder | ff30a62 | 2010-04-13 15:22:58 +1000 | [diff] [blame] | 108 | nbblks); |
| 109 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp); |
Dave Chinner | 3228149 | 2009-01-22 15:37:47 +1100 | [diff] [blame] | 110 | return NULL; |
| 111 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 112 | |
Alex Elder | 36adecf | 2010-04-13 15:21:13 +1000 | [diff] [blame] | 113 | /* |
| 114 | * We do log I/O in units of log sectors (a power-of-2 |
| 115 | * multiple of the basic block size), so we round up the |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 116 | * requested size to accommodate the basic blocks required |
Alex Elder | 36adecf | 2010-04-13 15:21:13 +1000 | [diff] [blame] | 117 | * for complete log sectors. |
| 118 | * |
| 119 | * In addition, the buffer may be used for a non-sector- |
| 120 | * aligned block offset, in which case an I/O of the |
| 121 | * requested size could extend beyond the end of the |
| 122 | * buffer. If the requested size is only 1 basic block it |
| 123 | * will never straddle a sector boundary, so this won't be |
| 124 | * an issue. Nor will this be a problem if the log I/O is |
| 125 | * done in basic blocks (sector size 1). But otherwise we |
| 126 | * extend the buffer by one extra log sector to ensure |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 127 | * there's space to accommodate this possibility. |
Alex Elder | 36adecf | 2010-04-13 15:21:13 +1000 | [diff] [blame] | 128 | */ |
Alex Elder | 69ce58f | 2010-04-20 17:09:59 +1000 | [diff] [blame] | 129 | if (nbblks > 1 && log->l_sectBBsize > 1) |
| 130 | nbblks += log->l_sectBBsize; |
| 131 | nbblks = round_up(nbblks, log->l_sectBBsize); |
Alex Elder | 36adecf | 2010-04-13 15:21:13 +1000 | [diff] [blame] | 132 | |
Dave Chinner | e70b73f | 2012-04-23 15:58:49 +1000 | [diff] [blame] | 133 | 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] | 134 | if (bp) |
| 135 | xfs_buf_unlock(bp); |
| 136 | return bp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 137 | } |
| 138 | |
Eric Sandeen | 5d77c0d | 2009-11-19 15:52:00 +0000 | [diff] [blame] | 139 | STATIC void |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 140 | xlog_put_bp( |
| 141 | xfs_buf_t *bp) |
| 142 | { |
| 143 | xfs_buf_free(bp); |
| 144 | } |
| 145 | |
Alex Elder | 48389ef | 2010-04-20 17:10:21 +1000 | [diff] [blame] | 146 | /* |
| 147 | * Return the address of the start of the given block number's data |
| 148 | * in a log buffer. The buffer covers a log sector-aligned region. |
| 149 | */ |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 150 | STATIC xfs_caddr_t |
| 151 | xlog_align( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 152 | struct xlog *log, |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 153 | xfs_daddr_t blk_no, |
| 154 | int nbblks, |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 155 | struct xfs_buf *bp) |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 156 | { |
Christoph Hellwig | fdc07f4 | 2010-05-10 17:28:14 +0000 | [diff] [blame] | 157 | 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] | 158 | |
Dave Chinner | 4e94b71 | 2012-04-23 15:58:51 +1000 | [diff] [blame] | 159 | ASSERT(offset + nbblks <= bp->b_length); |
Chandra Seetharaman | 6292604 | 2011-07-22 23:40:15 +0000 | [diff] [blame] | 160 | return bp->b_addr + BBTOB(offset); |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 161 | } |
| 162 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 163 | |
| 164 | /* |
| 165 | * nbblks should be uint, but oh well. Just want to catch that 32-bit length. |
| 166 | */ |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 167 | STATIC int |
| 168 | xlog_bread_noalign( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 169 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 170 | xfs_daddr_t blk_no, |
| 171 | int nbblks, |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 172 | struct xfs_buf *bp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 173 | { |
| 174 | int error; |
| 175 | |
Alex Elder | ff30a62 | 2010-04-13 15:22:58 +1000 | [diff] [blame] | 176 | if (!xlog_buf_bbcount_valid(log, nbblks)) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 177 | xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer", |
Alex Elder | ff30a62 | 2010-04-13 15:22:58 +1000 | [diff] [blame] | 178 | nbblks); |
| 179 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp); |
Dave Chinner | 3228149 | 2009-01-22 15:37:47 +1100 | [diff] [blame] | 180 | return EFSCORRUPTED; |
| 181 | } |
| 182 | |
Alex Elder | 69ce58f | 2010-04-20 17:09:59 +1000 | [diff] [blame] | 183 | blk_no = round_down(blk_no, log->l_sectBBsize); |
| 184 | nbblks = round_up(nbblks, log->l_sectBBsize); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 185 | |
| 186 | ASSERT(nbblks > 0); |
Dave Chinner | 4e94b71 | 2012-04-23 15:58:51 +1000 | [diff] [blame] | 187 | ASSERT(nbblks <= bp->b_length); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 188 | |
| 189 | XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no); |
| 190 | XFS_BUF_READ(bp); |
Dave Chinner | aa0e883 | 2012-04-23 15:58:52 +1000 | [diff] [blame] | 191 | bp->b_io_length = nbblks; |
Dave Chinner | 0e95f19 | 2012-04-23 15:58:46 +1000 | [diff] [blame] | 192 | bp->b_error = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 193 | |
| 194 | xfsbdstrat(log->l_mp, bp); |
Christoph Hellwig | 1a1a3e9 | 2010-10-06 18:41:18 +0000 | [diff] [blame] | 195 | error = xfs_buf_iowait(bp); |
David Chinner | d64e31a | 2008-04-10 12:22:17 +1000 | [diff] [blame] | 196 | if (error) |
Christoph Hellwig | 901796a | 2011-10-10 16:52:49 +0000 | [diff] [blame] | 197 | xfs_buf_ioerror_alert(bp, __func__); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 198 | return error; |
| 199 | } |
| 200 | |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 201 | STATIC int |
| 202 | xlog_bread( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 203 | struct xlog *log, |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 204 | xfs_daddr_t blk_no, |
| 205 | int nbblks, |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 206 | struct xfs_buf *bp, |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 207 | xfs_caddr_t *offset) |
| 208 | { |
| 209 | int error; |
| 210 | |
| 211 | error = xlog_bread_noalign(log, blk_no, nbblks, bp); |
| 212 | if (error) |
| 213 | return error; |
| 214 | |
| 215 | *offset = xlog_align(log, blk_no, nbblks, bp); |
| 216 | return 0; |
| 217 | } |
| 218 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 219 | /* |
Dave Chinner | 4439647 | 2011-04-21 09:34:27 +0000 | [diff] [blame] | 220 | * Read at an offset into the buffer. Returns with the buffer in it's original |
| 221 | * state regardless of the result of the read. |
| 222 | */ |
| 223 | STATIC int |
| 224 | xlog_bread_offset( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 225 | struct xlog *log, |
Dave Chinner | 4439647 | 2011-04-21 09:34:27 +0000 | [diff] [blame] | 226 | xfs_daddr_t blk_no, /* block to read from */ |
| 227 | int nbblks, /* blocks to read */ |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 228 | struct xfs_buf *bp, |
Dave Chinner | 4439647 | 2011-04-21 09:34:27 +0000 | [diff] [blame] | 229 | xfs_caddr_t offset) |
| 230 | { |
Chandra Seetharaman | 6292604 | 2011-07-22 23:40:15 +0000 | [diff] [blame] | 231 | xfs_caddr_t orig_offset = bp->b_addr; |
Dave Chinner | 4e94b71 | 2012-04-23 15:58:51 +1000 | [diff] [blame] | 232 | int orig_len = BBTOB(bp->b_length); |
Dave Chinner | 4439647 | 2011-04-21 09:34:27 +0000 | [diff] [blame] | 233 | int error, error2; |
| 234 | |
Chandra Seetharaman | 02fe03d | 2011-07-22 23:40:22 +0000 | [diff] [blame] | 235 | error = xfs_buf_associate_memory(bp, offset, BBTOB(nbblks)); |
Dave Chinner | 4439647 | 2011-04-21 09:34:27 +0000 | [diff] [blame] | 236 | if (error) |
| 237 | return error; |
| 238 | |
| 239 | error = xlog_bread_noalign(log, blk_no, nbblks, bp); |
| 240 | |
| 241 | /* must reset buffer pointer even on error */ |
Chandra Seetharaman | 02fe03d | 2011-07-22 23:40:22 +0000 | [diff] [blame] | 242 | error2 = xfs_buf_associate_memory(bp, orig_offset, orig_len); |
Dave Chinner | 4439647 | 2011-04-21 09:34:27 +0000 | [diff] [blame] | 243 | if (error) |
| 244 | return error; |
| 245 | return error2; |
| 246 | } |
| 247 | |
| 248 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 249 | * Write out the buffer at the given block for the given number of blocks. |
| 250 | * The buffer is kept locked across the write and is returned locked. |
| 251 | * This can only be used for synchronous log writes. |
| 252 | */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 253 | STATIC int |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 254 | xlog_bwrite( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 255 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 256 | xfs_daddr_t blk_no, |
| 257 | int nbblks, |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 258 | struct xfs_buf *bp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 259 | { |
| 260 | int error; |
| 261 | |
Alex Elder | ff30a62 | 2010-04-13 15:22:58 +1000 | [diff] [blame] | 262 | if (!xlog_buf_bbcount_valid(log, nbblks)) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 263 | xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer", |
Alex Elder | ff30a62 | 2010-04-13 15:22:58 +1000 | [diff] [blame] | 264 | nbblks); |
| 265 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp); |
Dave Chinner | 3228149 | 2009-01-22 15:37:47 +1100 | [diff] [blame] | 266 | return EFSCORRUPTED; |
| 267 | } |
| 268 | |
Alex Elder | 69ce58f | 2010-04-20 17:09:59 +1000 | [diff] [blame] | 269 | blk_no = round_down(blk_no, log->l_sectBBsize); |
| 270 | nbblks = round_up(nbblks, log->l_sectBBsize); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 271 | |
| 272 | ASSERT(nbblks > 0); |
Dave Chinner | 4e94b71 | 2012-04-23 15:58:51 +1000 | [diff] [blame] | 273 | ASSERT(nbblks <= bp->b_length); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 274 | |
| 275 | XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no); |
| 276 | XFS_BUF_ZEROFLAGS(bp); |
Chandra Seetharaman | 72790aa | 2011-07-22 23:40:04 +0000 | [diff] [blame] | 277 | xfs_buf_hold(bp); |
Christoph Hellwig | 0c842ad | 2011-07-08 14:36:19 +0200 | [diff] [blame] | 278 | xfs_buf_lock(bp); |
Dave Chinner | aa0e883 | 2012-04-23 15:58:52 +1000 | [diff] [blame] | 279 | bp->b_io_length = nbblks; |
Dave Chinner | 0e95f19 | 2012-04-23 15:58:46 +1000 | [diff] [blame] | 280 | bp->b_error = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 281 | |
Christoph Hellwig | c2b006c | 2011-08-23 08:28:07 +0000 | [diff] [blame] | 282 | error = xfs_bwrite(bp); |
Christoph Hellwig | 901796a | 2011-10-10 16:52:49 +0000 | [diff] [blame] | 283 | if (error) |
| 284 | xfs_buf_ioerror_alert(bp, __func__); |
Christoph Hellwig | c2b006c | 2011-08-23 08:28:07 +0000 | [diff] [blame] | 285 | xfs_buf_relse(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 286 | return error; |
| 287 | } |
| 288 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 289 | #ifdef DEBUG |
| 290 | /* |
| 291 | * dump debug superblock and log record information |
| 292 | */ |
| 293 | STATIC void |
| 294 | xlog_header_check_dump( |
| 295 | xfs_mount_t *mp, |
| 296 | xlog_rec_header_t *head) |
| 297 | { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 298 | xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d\n", |
Joe Perches | 03daa57 | 2009-12-14 18:01:10 -0800 | [diff] [blame] | 299 | __func__, &mp->m_sb.sb_uuid, XLOG_FMT); |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 300 | xfs_debug(mp, " log : uuid = %pU, fmt = %d\n", |
Joe Perches | 03daa57 | 2009-12-14 18:01:10 -0800 | [diff] [blame] | 301 | &head->h_fs_uuid, be32_to_cpu(head->h_fmt)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 302 | } |
| 303 | #else |
| 304 | #define xlog_header_check_dump(mp, head) |
| 305 | #endif |
| 306 | |
| 307 | /* |
| 308 | * check log record header for recovery |
| 309 | */ |
| 310 | STATIC int |
| 311 | xlog_header_check_recover( |
| 312 | xfs_mount_t *mp, |
| 313 | xlog_rec_header_t *head) |
| 314 | { |
Christoph Hellwig | 69ef921 | 2011-07-08 14:36:05 +0200 | [diff] [blame] | 315 | ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 316 | |
| 317 | /* |
| 318 | * IRIX doesn't write the h_fmt field and leaves it zeroed |
| 319 | * (XLOG_FMT_UNKNOWN). This stops us from trying to recover |
| 320 | * a dirty log created in IRIX. |
| 321 | */ |
Christoph Hellwig | 69ef921 | 2011-07-08 14:36:05 +0200 | [diff] [blame] | 322 | if (unlikely(head->h_fmt != cpu_to_be32(XLOG_FMT))) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 323 | xfs_warn(mp, |
| 324 | "dirty log written in incompatible format - can't recover"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 325 | xlog_header_check_dump(mp, head); |
| 326 | XFS_ERROR_REPORT("xlog_header_check_recover(1)", |
| 327 | XFS_ERRLEVEL_HIGH, mp); |
| 328 | return XFS_ERROR(EFSCORRUPTED); |
| 329 | } 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] | 330 | xfs_warn(mp, |
| 331 | "dirty log entry has mismatched uuid - can't recover"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 332 | xlog_header_check_dump(mp, head); |
| 333 | XFS_ERROR_REPORT("xlog_header_check_recover(2)", |
| 334 | XFS_ERRLEVEL_HIGH, mp); |
| 335 | return XFS_ERROR(EFSCORRUPTED); |
| 336 | } |
| 337 | return 0; |
| 338 | } |
| 339 | |
| 340 | /* |
| 341 | * read the head block of the log and check the header |
| 342 | */ |
| 343 | STATIC int |
| 344 | xlog_header_check_mount( |
| 345 | xfs_mount_t *mp, |
| 346 | xlog_rec_header_t *head) |
| 347 | { |
Christoph Hellwig | 69ef921 | 2011-07-08 14:36:05 +0200 | [diff] [blame] | 348 | ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 349 | |
| 350 | if (uuid_is_nil(&head->h_fs_uuid)) { |
| 351 | /* |
| 352 | * IRIX doesn't write the h_fs_uuid or h_fmt fields. If |
| 353 | * h_fs_uuid is nil, we assume this log was last mounted |
| 354 | * by IRIX and continue. |
| 355 | */ |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 356 | xfs_warn(mp, "nil uuid in log - IRIX style log"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 357 | } 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] | 358 | xfs_warn(mp, "log has mismatched uuid - can't recover"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 359 | xlog_header_check_dump(mp, head); |
| 360 | XFS_ERROR_REPORT("xlog_header_check_mount", |
| 361 | XFS_ERRLEVEL_HIGH, mp); |
| 362 | return XFS_ERROR(EFSCORRUPTED); |
| 363 | } |
| 364 | return 0; |
| 365 | } |
| 366 | |
| 367 | STATIC void |
| 368 | xlog_recover_iodone( |
| 369 | struct xfs_buf *bp) |
| 370 | { |
Chandra Seetharaman | 5a52c2a58 | 2011-07-22 23:39:51 +0000 | [diff] [blame] | 371 | if (bp->b_error) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 372 | /* |
| 373 | * We're not going to bother about retrying |
| 374 | * this during recovery. One strike! |
| 375 | */ |
Christoph Hellwig | 901796a | 2011-10-10 16:52:49 +0000 | [diff] [blame] | 376 | xfs_buf_ioerror_alert(bp, __func__); |
Dave Chinner | ebad861 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 377 | xfs_force_shutdown(bp->b_target->bt_mount, |
| 378 | SHUTDOWN_META_IO_ERROR); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 379 | } |
Christoph Hellwig | cb669ca | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 380 | bp->b_iodone = NULL; |
Christoph Hellwig | 1a1a3e9 | 2010-10-06 18:41:18 +0000 | [diff] [blame] | 381 | xfs_buf_ioend(bp, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 382 | } |
| 383 | |
| 384 | /* |
| 385 | * This routine finds (to an approximation) the first block in the physical |
| 386 | * log which contains the given cycle. It uses a binary search algorithm. |
| 387 | * Note that the algorithm can not be perfect because the disk will not |
| 388 | * necessarily be perfect. |
| 389 | */ |
David Chinner | a8272ce | 2007-11-23 16:28:09 +1100 | [diff] [blame] | 390 | STATIC int |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 391 | xlog_find_cycle_start( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 392 | struct xlog *log, |
| 393 | struct xfs_buf *bp, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 394 | xfs_daddr_t first_blk, |
| 395 | xfs_daddr_t *last_blk, |
| 396 | uint cycle) |
| 397 | { |
| 398 | xfs_caddr_t offset; |
| 399 | xfs_daddr_t mid_blk; |
Alex Elder | e3bb2e3 | 2010-04-15 18:17:30 +0000 | [diff] [blame] | 400 | xfs_daddr_t end_blk; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 401 | uint mid_cycle; |
| 402 | int error; |
| 403 | |
Alex Elder | e3bb2e3 | 2010-04-15 18:17:30 +0000 | [diff] [blame] | 404 | end_blk = *last_blk; |
| 405 | mid_blk = BLK_AVG(first_blk, end_blk); |
| 406 | while (mid_blk != first_blk && mid_blk != end_blk) { |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 407 | error = xlog_bread(log, mid_blk, 1, bp, &offset); |
| 408 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 409 | return error; |
Christoph Hellwig | 03bea6f | 2007-10-12 10:58:05 +1000 | [diff] [blame] | 410 | mid_cycle = xlog_get_cycle(offset); |
Alex Elder | e3bb2e3 | 2010-04-15 18:17:30 +0000 | [diff] [blame] | 411 | if (mid_cycle == cycle) |
| 412 | end_blk = mid_blk; /* last_half_cycle == mid_cycle */ |
| 413 | else |
| 414 | first_blk = mid_blk; /* first_half_cycle == mid_cycle */ |
| 415 | mid_blk = BLK_AVG(first_blk, end_blk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 416 | } |
Alex Elder | e3bb2e3 | 2010-04-15 18:17:30 +0000 | [diff] [blame] | 417 | ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) || |
| 418 | (mid_blk == end_blk && mid_blk-1 == first_blk)); |
| 419 | |
| 420 | *last_blk = end_blk; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 421 | |
| 422 | return 0; |
| 423 | } |
| 424 | |
| 425 | /* |
Alex Elder | 3f943d8 | 2010-04-15 18:17:34 +0000 | [diff] [blame] | 426 | * Check that a range of blocks does not contain stop_on_cycle_no. |
| 427 | * Fill in *new_blk with the block offset where such a block is |
| 428 | * found, or with -1 (an invalid block number) if there is no such |
| 429 | * block in the range. The scan needs to occur from front to back |
| 430 | * and the pointer into the region must be updated since a later |
| 431 | * routine will need to perform another test. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 432 | */ |
| 433 | STATIC int |
| 434 | xlog_find_verify_cycle( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 435 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 436 | xfs_daddr_t start_blk, |
| 437 | int nbblks, |
| 438 | uint stop_on_cycle_no, |
| 439 | xfs_daddr_t *new_blk) |
| 440 | { |
| 441 | xfs_daddr_t i, j; |
| 442 | uint cycle; |
| 443 | xfs_buf_t *bp; |
| 444 | xfs_daddr_t bufblks; |
| 445 | xfs_caddr_t buf = NULL; |
| 446 | int error = 0; |
| 447 | |
Alex Elder | 6881a22 | 2010-04-13 15:22:29 +1000 | [diff] [blame] | 448 | /* |
| 449 | * Greedily allocate a buffer big enough to handle the full |
| 450 | * range of basic blocks we'll be examining. If that fails, |
| 451 | * try a smaller size. We need to be able to read at least |
| 452 | * a log sector, or we're out of luck. |
| 453 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 454 | bufblks = 1 << ffs(nbblks); |
Dave Chinner | 81158e0 | 2012-04-27 19:45:22 +1000 | [diff] [blame] | 455 | while (bufblks > log->l_logBBsize) |
| 456 | bufblks >>= 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 457 | while (!(bp = xlog_get_bp(log, bufblks))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 458 | bufblks >>= 1; |
Alex Elder | 69ce58f | 2010-04-20 17:09:59 +1000 | [diff] [blame] | 459 | if (bufblks < log->l_sectBBsize) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 460 | return ENOMEM; |
| 461 | } |
| 462 | |
| 463 | for (i = start_blk; i < start_blk + nbblks; i += bufblks) { |
| 464 | int bcount; |
| 465 | |
| 466 | bcount = min(bufblks, (start_blk + nbblks - i)); |
| 467 | |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 468 | error = xlog_bread(log, i, bcount, bp, &buf); |
| 469 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 470 | goto out; |
| 471 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 472 | for (j = 0; j < bcount; j++) { |
Christoph Hellwig | 03bea6f | 2007-10-12 10:58:05 +1000 | [diff] [blame] | 473 | cycle = xlog_get_cycle(buf); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 474 | if (cycle == stop_on_cycle_no) { |
| 475 | *new_blk = i+j; |
| 476 | goto out; |
| 477 | } |
| 478 | |
| 479 | buf += BBSIZE; |
| 480 | } |
| 481 | } |
| 482 | |
| 483 | *new_blk = -1; |
| 484 | |
| 485 | out: |
| 486 | xlog_put_bp(bp); |
| 487 | return error; |
| 488 | } |
| 489 | |
| 490 | /* |
| 491 | * Potentially backup over partial log record write. |
| 492 | * |
| 493 | * In the typical case, last_blk is the number of the block directly after |
| 494 | * a good log record. Therefore, we subtract one to get the block number |
| 495 | * of the last block in the given buffer. extra_bblks contains the number |
| 496 | * of blocks we would have read on a previous read. This happens when the |
| 497 | * last log record is split over the end of the physical log. |
| 498 | * |
| 499 | * extra_bblks is the number of blocks potentially verified on a previous |
| 500 | * call to this routine. |
| 501 | */ |
| 502 | STATIC int |
| 503 | xlog_find_verify_log_record( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 504 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 505 | xfs_daddr_t start_blk, |
| 506 | xfs_daddr_t *last_blk, |
| 507 | int extra_bblks) |
| 508 | { |
| 509 | xfs_daddr_t i; |
| 510 | xfs_buf_t *bp; |
| 511 | xfs_caddr_t offset = NULL; |
| 512 | xlog_rec_header_t *head = NULL; |
| 513 | int error = 0; |
| 514 | int smallmem = 0; |
| 515 | int num_blks = *last_blk - start_blk; |
| 516 | int xhdrs; |
| 517 | |
| 518 | ASSERT(start_blk != 0 || *last_blk != start_blk); |
| 519 | |
| 520 | if (!(bp = xlog_get_bp(log, num_blks))) { |
| 521 | if (!(bp = xlog_get_bp(log, 1))) |
| 522 | return ENOMEM; |
| 523 | smallmem = 1; |
| 524 | } else { |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 525 | error = xlog_bread(log, start_blk, num_blks, bp, &offset); |
| 526 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 527 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 528 | offset += ((num_blks - 1) << BBSHIFT); |
| 529 | } |
| 530 | |
| 531 | for (i = (*last_blk) - 1; i >= 0; i--) { |
| 532 | if (i < start_blk) { |
| 533 | /* valid log record not found */ |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 534 | xfs_warn(log->l_mp, |
| 535 | "Log inconsistent (didn't find previous header)"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 536 | ASSERT(0); |
| 537 | error = XFS_ERROR(EIO); |
| 538 | goto out; |
| 539 | } |
| 540 | |
| 541 | if (smallmem) { |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 542 | error = xlog_bread(log, i, 1, bp, &offset); |
| 543 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 544 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 545 | } |
| 546 | |
| 547 | head = (xlog_rec_header_t *)offset; |
| 548 | |
Christoph Hellwig | 69ef921 | 2011-07-08 14:36:05 +0200 | [diff] [blame] | 549 | if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 550 | break; |
| 551 | |
| 552 | if (!smallmem) |
| 553 | offset -= BBSIZE; |
| 554 | } |
| 555 | |
| 556 | /* |
| 557 | * We hit the beginning of the physical log & still no header. Return |
| 558 | * to caller. If caller can handle a return of -1, then this routine |
| 559 | * will be called again for the end of the physical log. |
| 560 | */ |
| 561 | if (i == -1) { |
| 562 | error = -1; |
| 563 | goto out; |
| 564 | } |
| 565 | |
| 566 | /* |
| 567 | * We have the final block of the good log (the first block |
| 568 | * of the log record _before_ the head. So we check the uuid. |
| 569 | */ |
| 570 | if ((error = xlog_header_check_mount(log->l_mp, head))) |
| 571 | goto out; |
| 572 | |
| 573 | /* |
| 574 | * We may have found a log record header before we expected one. |
| 575 | * last_blk will be the 1st block # with a given cycle #. We may end |
| 576 | * up reading an entire log record. In this case, we don't want to |
| 577 | * reset last_blk. Only when last_blk points in the middle of a log |
| 578 | * record do we update last_blk. |
| 579 | */ |
Eric Sandeen | 6211870 | 2008-03-06 13:44:28 +1100 | [diff] [blame] | 580 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 581 | uint h_size = be32_to_cpu(head->h_size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 582 | |
| 583 | xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE; |
| 584 | if (h_size % XLOG_HEADER_CYCLE_SIZE) |
| 585 | xhdrs++; |
| 586 | } else { |
| 587 | xhdrs = 1; |
| 588 | } |
| 589 | |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 590 | if (*last_blk - i + extra_bblks != |
| 591 | BTOBB(be32_to_cpu(head->h_len)) + xhdrs) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 592 | *last_blk = i; |
| 593 | |
| 594 | out: |
| 595 | xlog_put_bp(bp); |
| 596 | return error; |
| 597 | } |
| 598 | |
| 599 | /* |
| 600 | * Head is defined to be the point of the log where the next log write |
| 601 | * write could go. This means that incomplete LR writes at the end are |
| 602 | * eliminated when calculating the head. We aren't guaranteed that previous |
| 603 | * LR have complete transactions. We only know that a cycle number of |
| 604 | * current cycle number -1 won't be present in the log if we start writing |
| 605 | * from our current block number. |
| 606 | * |
| 607 | * last_blk contains the block number of the first block with a given |
| 608 | * cycle number. |
| 609 | * |
| 610 | * Return: zero if normal, non-zero if error. |
| 611 | */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 612 | STATIC int |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 613 | xlog_find_head( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 614 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 615 | xfs_daddr_t *return_head_blk) |
| 616 | { |
| 617 | xfs_buf_t *bp; |
| 618 | xfs_caddr_t offset; |
| 619 | xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk; |
| 620 | int num_scan_bblks; |
| 621 | uint first_half_cycle, last_half_cycle; |
| 622 | uint stop_on_cycle; |
| 623 | int error, log_bbnum = log->l_logBBsize; |
| 624 | |
| 625 | /* Is the end of the log device zeroed? */ |
| 626 | if ((error = xlog_find_zeroed(log, &first_blk)) == -1) { |
| 627 | *return_head_blk = first_blk; |
| 628 | |
| 629 | /* Is the whole lot zeroed? */ |
| 630 | if (!first_blk) { |
| 631 | /* Linux XFS shouldn't generate totally zeroed logs - |
| 632 | * mkfs etc write a dummy unmount record to a fresh |
| 633 | * log so we can store the uuid in there |
| 634 | */ |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 635 | xfs_warn(log->l_mp, "totally zeroed log"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 636 | } |
| 637 | |
| 638 | return 0; |
| 639 | } else if (error) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 640 | xfs_warn(log->l_mp, "empty log check failed"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 641 | return error; |
| 642 | } |
| 643 | |
| 644 | first_blk = 0; /* get cycle # of 1st block */ |
| 645 | bp = xlog_get_bp(log, 1); |
| 646 | if (!bp) |
| 647 | return ENOMEM; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 648 | |
| 649 | error = xlog_bread(log, 0, 1, bp, &offset); |
| 650 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 651 | goto bp_err; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 652 | |
Christoph Hellwig | 03bea6f | 2007-10-12 10:58:05 +1000 | [diff] [blame] | 653 | first_half_cycle = xlog_get_cycle(offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 654 | |
| 655 | last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */ |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 656 | error = xlog_bread(log, last_blk, 1, bp, &offset); |
| 657 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 658 | goto bp_err; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 659 | |
Christoph Hellwig | 03bea6f | 2007-10-12 10:58:05 +1000 | [diff] [blame] | 660 | last_half_cycle = xlog_get_cycle(offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 661 | ASSERT(last_half_cycle != 0); |
| 662 | |
| 663 | /* |
| 664 | * If the 1st half cycle number is equal to the last half cycle number, |
| 665 | * then the entire log is stamped with the same cycle number. In this |
| 666 | * case, head_blk can't be set to zero (which makes sense). The below |
| 667 | * math doesn't work out properly with head_blk equal to zero. Instead, |
| 668 | * we set it to log_bbnum which is an invalid block number, but this |
| 669 | * value makes the math correct. If head_blk doesn't changed through |
| 670 | * all the tests below, *head_blk is set to zero at the very end rather |
| 671 | * than log_bbnum. In a sense, log_bbnum and zero are the same block |
| 672 | * in a circular file. |
| 673 | */ |
| 674 | if (first_half_cycle == last_half_cycle) { |
| 675 | /* |
| 676 | * In this case we believe that the entire log should have |
| 677 | * cycle number last_half_cycle. We need to scan backwards |
| 678 | * from the end verifying that there are no holes still |
| 679 | * containing last_half_cycle - 1. If we find such a hole, |
| 680 | * then the start of that hole will be the new head. The |
| 681 | * simple case looks like |
| 682 | * x | x ... | x - 1 | x |
| 683 | * Another case that fits this picture would be |
| 684 | * x | x + 1 | x ... | x |
Nathan Scott | c41564b | 2006-03-29 08:55:14 +1000 | [diff] [blame] | 685 | * 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] | 686 | * log, as one of the latest writes at the beginning was |
| 687 | * incomplete. |
| 688 | * One more case is |
| 689 | * x | x + 1 | x ... | x - 1 | x |
| 690 | * This is really the combination of the above two cases, and |
| 691 | * the head has to end up at the start of the x-1 hole at the |
| 692 | * end of the log. |
| 693 | * |
| 694 | * In the 256k log case, we will read from the beginning to the |
| 695 | * end of the log and search for cycle numbers equal to x-1. |
| 696 | * We don't worry about the x+1 blocks that we encounter, |
| 697 | * because we know that they cannot be the head since the log |
| 698 | * started with x. |
| 699 | */ |
| 700 | head_blk = log_bbnum; |
| 701 | stop_on_cycle = last_half_cycle - 1; |
| 702 | } else { |
| 703 | /* |
| 704 | * In this case we want to find the first block with cycle |
| 705 | * number matching last_half_cycle. We expect the log to be |
| 706 | * some variation on |
Alex Elder | 3f943d8 | 2010-04-15 18:17:34 +0000 | [diff] [blame] | 707 | * x + 1 ... | x ... | x |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 708 | * The first block with cycle number x (last_half_cycle) will |
| 709 | * be where the new head belongs. First we do a binary search |
| 710 | * for the first occurrence of last_half_cycle. The binary |
| 711 | * search may not be totally accurate, so then we scan back |
| 712 | * from there looking for occurrences of last_half_cycle before |
| 713 | * us. If that backwards scan wraps around the beginning of |
| 714 | * the log, then we look for occurrences of last_half_cycle - 1 |
| 715 | * at the end of the log. The cases we're looking for look |
| 716 | * like |
Alex Elder | 3f943d8 | 2010-04-15 18:17:34 +0000 | [diff] [blame] | 717 | * v binary search stopped here |
| 718 | * x + 1 ... | x | x + 1 | x ... | x |
| 719 | * ^ but we want to locate this spot |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 720 | * or |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 721 | * <---------> less than scan distance |
Alex Elder | 3f943d8 | 2010-04-15 18:17:34 +0000 | [diff] [blame] | 722 | * x + 1 ... | x ... | x - 1 | x |
| 723 | * ^ we want to locate this spot |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 724 | */ |
| 725 | stop_on_cycle = last_half_cycle; |
| 726 | if ((error = xlog_find_cycle_start(log, bp, first_blk, |
| 727 | &head_blk, last_half_cycle))) |
| 728 | goto bp_err; |
| 729 | } |
| 730 | |
| 731 | /* |
| 732 | * Now validate the answer. Scan back some number of maximum possible |
| 733 | * blocks and make sure each one has the expected cycle number. The |
| 734 | * maximum is determined by the total possible amount of buffering |
| 735 | * in the in-core log. The following number can be made tighter if |
| 736 | * we actually look at the block size of the filesystem. |
| 737 | */ |
| 738 | num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log); |
| 739 | if (head_blk >= num_scan_bblks) { |
| 740 | /* |
| 741 | * We are guaranteed that the entire check can be performed |
| 742 | * in one buffer. |
| 743 | */ |
| 744 | start_blk = head_blk - num_scan_bblks; |
| 745 | if ((error = xlog_find_verify_cycle(log, |
| 746 | start_blk, num_scan_bblks, |
| 747 | stop_on_cycle, &new_blk))) |
| 748 | goto bp_err; |
| 749 | if (new_blk != -1) |
| 750 | head_blk = new_blk; |
| 751 | } else { /* need to read 2 parts of log */ |
| 752 | /* |
| 753 | * We are going to scan backwards in the log in two parts. |
| 754 | * First we scan the physical end of the log. In this part |
| 755 | * of the log, we are looking for blocks with cycle number |
| 756 | * last_half_cycle - 1. |
| 757 | * If we find one, then we know that the log starts there, as |
| 758 | * we've found a hole that didn't get written in going around |
| 759 | * the end of the physical log. The simple case for this is |
| 760 | * x + 1 ... | x ... | x - 1 | x |
| 761 | * <---------> less than scan distance |
| 762 | * If all of the blocks at the end of the log have cycle number |
| 763 | * last_half_cycle, then we check the blocks at the start of |
| 764 | * the log looking for occurrences of last_half_cycle. If we |
| 765 | * find one, then our current estimate for the location of the |
| 766 | * first occurrence of last_half_cycle is wrong and we move |
| 767 | * back to the hole we've found. This case looks like |
| 768 | * x + 1 ... | x | x + 1 | x ... |
| 769 | * ^ binary search stopped here |
| 770 | * Another case we need to handle that only occurs in 256k |
| 771 | * logs is |
| 772 | * x + 1 ... | x ... | x+1 | x ... |
| 773 | * ^ binary search stops here |
| 774 | * In a 256k log, the scan at the end of the log will see the |
| 775 | * x + 1 blocks. We need to skip past those since that is |
| 776 | * certainly not the head of the log. By searching for |
| 777 | * last_half_cycle-1 we accomplish that. |
| 778 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 779 | ASSERT(head_blk <= INT_MAX && |
Alex Elder | 3f943d8 | 2010-04-15 18:17:34 +0000 | [diff] [blame] | 780 | (xfs_daddr_t) num_scan_bblks >= head_blk); |
| 781 | start_blk = log_bbnum - (num_scan_bblks - head_blk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 782 | if ((error = xlog_find_verify_cycle(log, start_blk, |
| 783 | num_scan_bblks - (int)head_blk, |
| 784 | (stop_on_cycle - 1), &new_blk))) |
| 785 | goto bp_err; |
| 786 | if (new_blk != -1) { |
| 787 | head_blk = new_blk; |
Alex Elder | 9db127e | 2010-04-15 18:17:26 +0000 | [diff] [blame] | 788 | goto validate_head; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 789 | } |
| 790 | |
| 791 | /* |
| 792 | * Scan beginning of log now. The last part of the physical |
| 793 | * log is good. This scan needs to verify that it doesn't find |
| 794 | * the last_half_cycle. |
| 795 | */ |
| 796 | start_blk = 0; |
| 797 | ASSERT(head_blk <= INT_MAX); |
| 798 | if ((error = xlog_find_verify_cycle(log, |
| 799 | start_blk, (int)head_blk, |
| 800 | stop_on_cycle, &new_blk))) |
| 801 | goto bp_err; |
| 802 | if (new_blk != -1) |
| 803 | head_blk = new_blk; |
| 804 | } |
| 805 | |
Alex Elder | 9db127e | 2010-04-15 18:17:26 +0000 | [diff] [blame] | 806 | validate_head: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 807 | /* |
| 808 | * Now we need to make sure head_blk is not pointing to a block in |
| 809 | * the middle of a log record. |
| 810 | */ |
| 811 | num_scan_bblks = XLOG_REC_SHIFT(log); |
| 812 | if (head_blk >= num_scan_bblks) { |
| 813 | start_blk = head_blk - num_scan_bblks; /* don't read head_blk */ |
| 814 | |
| 815 | /* start ptr at last block ptr before head_blk */ |
| 816 | if ((error = xlog_find_verify_log_record(log, start_blk, |
| 817 | &head_blk, 0)) == -1) { |
| 818 | error = XFS_ERROR(EIO); |
| 819 | goto bp_err; |
| 820 | } else if (error) |
| 821 | goto bp_err; |
| 822 | } else { |
| 823 | start_blk = 0; |
| 824 | ASSERT(head_blk <= INT_MAX); |
| 825 | if ((error = xlog_find_verify_log_record(log, start_blk, |
| 826 | &head_blk, 0)) == -1) { |
| 827 | /* We hit the beginning of the log during our search */ |
Alex Elder | 3f943d8 | 2010-04-15 18:17:34 +0000 | [diff] [blame] | 828 | start_blk = log_bbnum - (num_scan_bblks - head_blk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 829 | new_blk = log_bbnum; |
| 830 | ASSERT(start_blk <= INT_MAX && |
| 831 | (xfs_daddr_t) log_bbnum-start_blk >= 0); |
| 832 | ASSERT(head_blk <= INT_MAX); |
| 833 | if ((error = xlog_find_verify_log_record(log, |
| 834 | start_blk, &new_blk, |
| 835 | (int)head_blk)) == -1) { |
| 836 | error = XFS_ERROR(EIO); |
| 837 | goto bp_err; |
| 838 | } else if (error) |
| 839 | goto bp_err; |
| 840 | if (new_blk != log_bbnum) |
| 841 | head_blk = new_blk; |
| 842 | } else if (error) |
| 843 | goto bp_err; |
| 844 | } |
| 845 | |
| 846 | xlog_put_bp(bp); |
| 847 | if (head_blk == log_bbnum) |
| 848 | *return_head_blk = 0; |
| 849 | else |
| 850 | *return_head_blk = head_blk; |
| 851 | /* |
| 852 | * When returning here, we have a good block number. Bad block |
| 853 | * means that during a previous crash, we didn't have a clean break |
| 854 | * from cycle number N to cycle number N-1. In this case, we need |
| 855 | * to find the first block with cycle number N-1. |
| 856 | */ |
| 857 | return 0; |
| 858 | |
| 859 | bp_err: |
| 860 | xlog_put_bp(bp); |
| 861 | |
| 862 | if (error) |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 863 | xfs_warn(log->l_mp, "failed to find log head"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 864 | return error; |
| 865 | } |
| 866 | |
| 867 | /* |
| 868 | * Find the sync block number or the tail of the log. |
| 869 | * |
| 870 | * This will be the block number of the last record to have its |
| 871 | * associated buffers synced to disk. Every log record header has |
| 872 | * a sync lsn embedded in it. LSNs hold block numbers, so it is easy |
| 873 | * to get a sync block number. The only concern is to figure out which |
| 874 | * log record header to believe. |
| 875 | * |
| 876 | * The following algorithm uses the log record header with the largest |
| 877 | * lsn. The entire log record does not need to be valid. We only care |
| 878 | * that the header is valid. |
| 879 | * |
| 880 | * We could speed up search by using current head_blk buffer, but it is not |
| 881 | * available. |
| 882 | */ |
Eric Sandeen | 5d77c0d | 2009-11-19 15:52:00 +0000 | [diff] [blame] | 883 | STATIC int |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 884 | xlog_find_tail( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 885 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 886 | xfs_daddr_t *head_blk, |
Eric Sandeen | 65be605 | 2006-01-11 15:34:19 +1100 | [diff] [blame] | 887 | xfs_daddr_t *tail_blk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 888 | { |
| 889 | xlog_rec_header_t *rhead; |
| 890 | xlog_op_header_t *op_head; |
| 891 | xfs_caddr_t offset = NULL; |
| 892 | xfs_buf_t *bp; |
| 893 | int error, i, found; |
| 894 | xfs_daddr_t umount_data_blk; |
| 895 | xfs_daddr_t after_umount_blk; |
| 896 | xfs_lsn_t tail_lsn; |
| 897 | int hblks; |
| 898 | |
| 899 | found = 0; |
| 900 | |
| 901 | /* |
| 902 | * Find previous log record |
| 903 | */ |
| 904 | if ((error = xlog_find_head(log, head_blk))) |
| 905 | return error; |
| 906 | |
| 907 | bp = xlog_get_bp(log, 1); |
| 908 | if (!bp) |
| 909 | return ENOMEM; |
| 910 | if (*head_blk == 0) { /* special case */ |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 911 | error = xlog_bread(log, 0, 1, bp, &offset); |
| 912 | if (error) |
Alex Elder | 9db127e | 2010-04-15 18:17:26 +0000 | [diff] [blame] | 913 | goto done; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 914 | |
Christoph Hellwig | 03bea6f | 2007-10-12 10:58:05 +1000 | [diff] [blame] | 915 | if (xlog_get_cycle(offset) == 0) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 916 | *tail_blk = 0; |
| 917 | /* leave all other log inited values alone */ |
Alex Elder | 9db127e | 2010-04-15 18:17:26 +0000 | [diff] [blame] | 918 | goto done; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 919 | } |
| 920 | } |
| 921 | |
| 922 | /* |
| 923 | * Search backwards looking for log record header block |
| 924 | */ |
| 925 | ASSERT(*head_blk < INT_MAX); |
| 926 | for (i = (int)(*head_blk) - 1; i >= 0; i--) { |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 927 | error = xlog_bread(log, i, 1, bp, &offset); |
| 928 | if (error) |
Alex Elder | 9db127e | 2010-04-15 18:17:26 +0000 | [diff] [blame] | 929 | goto done; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 930 | |
Christoph Hellwig | 69ef921 | 2011-07-08 14:36:05 +0200 | [diff] [blame] | 931 | if (*(__be32 *)offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 932 | found = 1; |
| 933 | break; |
| 934 | } |
| 935 | } |
| 936 | /* |
| 937 | * If we haven't found the log record header block, start looking |
| 938 | * again from the end of the physical log. XXXmiken: There should be |
| 939 | * a check here to make sure we didn't search more than N blocks in |
| 940 | * the previous code. |
| 941 | */ |
| 942 | if (!found) { |
| 943 | for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) { |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 944 | error = xlog_bread(log, i, 1, bp, &offset); |
| 945 | if (error) |
Alex Elder | 9db127e | 2010-04-15 18:17:26 +0000 | [diff] [blame] | 946 | goto done; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 947 | |
Christoph Hellwig | 69ef921 | 2011-07-08 14:36:05 +0200 | [diff] [blame] | 948 | if (*(__be32 *)offset == |
| 949 | cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 950 | found = 2; |
| 951 | break; |
| 952 | } |
| 953 | } |
| 954 | } |
| 955 | if (!found) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 956 | xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 957 | ASSERT(0); |
| 958 | return XFS_ERROR(EIO); |
| 959 | } |
| 960 | |
| 961 | /* find blk_no of tail of log */ |
| 962 | rhead = (xlog_rec_header_t *)offset; |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 963 | *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 964 | |
| 965 | /* |
| 966 | * Reset log values according to the state of the log when we |
| 967 | * crashed. In the case where head_blk == 0, we bump curr_cycle |
| 968 | * one because the next write starts a new cycle rather than |
| 969 | * continuing the cycle of the last good log record. At this |
| 970 | * point we have guaranteed that all partial log records have been |
| 971 | * accounted for. Therefore, we know that the last good log record |
| 972 | * written was complete and ended exactly on the end boundary |
| 973 | * of the physical log. |
| 974 | */ |
| 975 | log->l_prev_block = i; |
| 976 | log->l_curr_block = (int)*head_blk; |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 977 | log->l_curr_cycle = be32_to_cpu(rhead->h_cycle); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 978 | if (found == 2) |
| 979 | log->l_curr_cycle++; |
Dave Chinner | 1c3cb9e | 2010-12-21 12:28:39 +1100 | [diff] [blame] | 980 | atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn)); |
Dave Chinner | 84f3c68 | 2010-12-03 22:11:29 +1100 | [diff] [blame] | 981 | atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn)); |
Christoph Hellwig | 2849696 | 2012-02-20 02:31:25 +0000 | [diff] [blame] | 982 | xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle, |
Dave Chinner | a69ed03 | 2010-12-21 12:08:20 +1100 | [diff] [blame] | 983 | BBTOB(log->l_curr_block)); |
Christoph Hellwig | 2849696 | 2012-02-20 02:31:25 +0000 | [diff] [blame] | 984 | xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle, |
Dave Chinner | a69ed03 | 2010-12-21 12:08:20 +1100 | [diff] [blame] | 985 | BBTOB(log->l_curr_block)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 986 | |
| 987 | /* |
| 988 | * Look for unmount record. If we find it, then we know there |
| 989 | * was a clean unmount. Since 'i' could be the last block in |
| 990 | * the physical log, we convert to a log block before comparing |
| 991 | * to the head_blk. |
| 992 | * |
| 993 | * Save the current tail lsn to use to pass to |
| 994 | * xlog_clear_stale_blocks() below. We won't want to clear the |
| 995 | * unmount record if there is one, so we pass the lsn of the |
| 996 | * unmount record rather than the block after it. |
| 997 | */ |
Eric Sandeen | 6211870 | 2008-03-06 13:44:28 +1100 | [diff] [blame] | 998 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 999 | int h_size = be32_to_cpu(rhead->h_size); |
| 1000 | int h_version = be32_to_cpu(rhead->h_version); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1001 | |
| 1002 | if ((h_version & XLOG_VERSION_2) && |
| 1003 | (h_size > XLOG_HEADER_CYCLE_SIZE)) { |
| 1004 | hblks = h_size / XLOG_HEADER_CYCLE_SIZE; |
| 1005 | if (h_size % XLOG_HEADER_CYCLE_SIZE) |
| 1006 | hblks++; |
| 1007 | } else { |
| 1008 | hblks = 1; |
| 1009 | } |
| 1010 | } else { |
| 1011 | hblks = 1; |
| 1012 | } |
| 1013 | after_umount_blk = (i + hblks + (int) |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 1014 | BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize; |
Dave Chinner | 1c3cb9e | 2010-12-21 12:28:39 +1100 | [diff] [blame] | 1015 | tail_lsn = atomic64_read(&log->l_tail_lsn); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1016 | if (*head_blk == after_umount_blk && |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 1017 | be32_to_cpu(rhead->h_num_logops) == 1) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1018 | umount_data_blk = (i + hblks) % log->l_logBBsize; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1019 | error = xlog_bread(log, umount_data_blk, 1, bp, &offset); |
| 1020 | if (error) |
Alex Elder | 9db127e | 2010-04-15 18:17:26 +0000 | [diff] [blame] | 1021 | goto done; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1022 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1023 | op_head = (xlog_op_header_t *)offset; |
| 1024 | if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) { |
| 1025 | /* |
| 1026 | * Set tail and last sync so that newly written |
| 1027 | * log records will point recovery to after the |
| 1028 | * current unmount record. |
| 1029 | */ |
Dave Chinner | 1c3cb9e | 2010-12-21 12:28:39 +1100 | [diff] [blame] | 1030 | xlog_assign_atomic_lsn(&log->l_tail_lsn, |
| 1031 | log->l_curr_cycle, after_umount_blk); |
| 1032 | xlog_assign_atomic_lsn(&log->l_last_sync_lsn, |
| 1033 | log->l_curr_cycle, after_umount_blk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1034 | *tail_blk = after_umount_blk; |
David Chinner | 92821e2 | 2007-05-24 15:26:31 +1000 | [diff] [blame] | 1035 | |
| 1036 | /* |
| 1037 | * Note that the unmount was clean. If the unmount |
| 1038 | * was not clean, we need to know this to rebuild the |
| 1039 | * superblock counters from the perag headers if we |
| 1040 | * have a filesystem using non-persistent counters. |
| 1041 | */ |
| 1042 | log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1043 | } |
| 1044 | } |
| 1045 | |
| 1046 | /* |
| 1047 | * Make sure that there are no blocks in front of the head |
| 1048 | * with the same cycle number as the head. This can happen |
| 1049 | * because we allow multiple outstanding log writes concurrently, |
| 1050 | * and the later writes might make it out before earlier ones. |
| 1051 | * |
| 1052 | * We use the lsn from before modifying it so that we'll never |
| 1053 | * overwrite the unmount record after a clean unmount. |
| 1054 | * |
| 1055 | * Do this only if we are going to recover the filesystem |
| 1056 | * |
| 1057 | * NOTE: This used to say "if (!readonly)" |
| 1058 | * However on Linux, we can & do recover a read-only filesystem. |
| 1059 | * We only skip recovery if NORECOVERY is specified on mount, |
| 1060 | * in which case we would not be here. |
| 1061 | * |
| 1062 | * But... if the -device- itself is readonly, just skip this. |
| 1063 | * We can't recover this device anyway, so it won't matter. |
| 1064 | */ |
Alex Elder | 9db127e | 2010-04-15 18:17:26 +0000 | [diff] [blame] | 1065 | if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1066 | error = xlog_clear_stale_blocks(log, tail_lsn); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1067 | |
Alex Elder | 9db127e | 2010-04-15 18:17:26 +0000 | [diff] [blame] | 1068 | done: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1069 | xlog_put_bp(bp); |
| 1070 | |
| 1071 | if (error) |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 1072 | xfs_warn(log->l_mp, "failed to locate log tail"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1073 | return error; |
| 1074 | } |
| 1075 | |
| 1076 | /* |
| 1077 | * Is the log zeroed at all? |
| 1078 | * |
| 1079 | * The last binary search should be changed to perform an X block read |
| 1080 | * once X becomes small enough. You can then search linearly through |
| 1081 | * the X blocks. This will cut down on the number of reads we need to do. |
| 1082 | * |
| 1083 | * If the log is partially zeroed, this routine will pass back the blkno |
| 1084 | * of the first block with cycle number 0. It won't have a complete LR |
| 1085 | * preceding it. |
| 1086 | * |
| 1087 | * Return: |
| 1088 | * 0 => the log is completely written to |
| 1089 | * -1 => use *blk_no as the first block of the log |
| 1090 | * >0 => error has occurred |
| 1091 | */ |
David Chinner | a8272ce | 2007-11-23 16:28:09 +1100 | [diff] [blame] | 1092 | STATIC int |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1093 | xlog_find_zeroed( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 1094 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1095 | xfs_daddr_t *blk_no) |
| 1096 | { |
| 1097 | xfs_buf_t *bp; |
| 1098 | xfs_caddr_t offset; |
| 1099 | uint first_cycle, last_cycle; |
| 1100 | xfs_daddr_t new_blk, last_blk, start_blk; |
| 1101 | xfs_daddr_t num_scan_bblks; |
| 1102 | int error, log_bbnum = log->l_logBBsize; |
| 1103 | |
Nathan Scott | 6fdf8cc | 2006-06-28 10:13:52 +1000 | [diff] [blame] | 1104 | *blk_no = 0; |
| 1105 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1106 | /* check totally zeroed log */ |
| 1107 | bp = xlog_get_bp(log, 1); |
| 1108 | if (!bp) |
| 1109 | return ENOMEM; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1110 | error = xlog_bread(log, 0, 1, bp, &offset); |
| 1111 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1112 | goto bp_err; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1113 | |
Christoph Hellwig | 03bea6f | 2007-10-12 10:58:05 +1000 | [diff] [blame] | 1114 | first_cycle = xlog_get_cycle(offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1115 | if (first_cycle == 0) { /* completely zeroed log */ |
| 1116 | *blk_no = 0; |
| 1117 | xlog_put_bp(bp); |
| 1118 | return -1; |
| 1119 | } |
| 1120 | |
| 1121 | /* check partially zeroed log */ |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1122 | error = xlog_bread(log, log_bbnum-1, 1, bp, &offset); |
| 1123 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1124 | goto bp_err; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1125 | |
Christoph Hellwig | 03bea6f | 2007-10-12 10:58:05 +1000 | [diff] [blame] | 1126 | last_cycle = xlog_get_cycle(offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1127 | if (last_cycle != 0) { /* log completely written to */ |
| 1128 | xlog_put_bp(bp); |
| 1129 | return 0; |
| 1130 | } else if (first_cycle != 1) { |
| 1131 | /* |
| 1132 | * If the cycle of the last block is zero, the cycle of |
| 1133 | * the first block must be 1. If it's not, maybe we're |
| 1134 | * not looking at a log... Bail out. |
| 1135 | */ |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 1136 | xfs_warn(log->l_mp, |
| 1137 | "Log inconsistent or not a log (last==0, first!=1)"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1138 | return XFS_ERROR(EINVAL); |
| 1139 | } |
| 1140 | |
| 1141 | /* we have a partially zeroed log */ |
| 1142 | last_blk = log_bbnum-1; |
| 1143 | if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0))) |
| 1144 | goto bp_err; |
| 1145 | |
| 1146 | /* |
| 1147 | * Validate the answer. Because there is no way to guarantee that |
| 1148 | * the entire log is made up of log records which are the same size, |
| 1149 | * we scan over the defined maximum blocks. At this point, the maximum |
| 1150 | * is not chosen to mean anything special. XXXmiken |
| 1151 | */ |
| 1152 | num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log); |
| 1153 | ASSERT(num_scan_bblks <= INT_MAX); |
| 1154 | |
| 1155 | if (last_blk < num_scan_bblks) |
| 1156 | num_scan_bblks = last_blk; |
| 1157 | start_blk = last_blk - num_scan_bblks; |
| 1158 | |
| 1159 | /* |
| 1160 | * We search for any instances of cycle number 0 that occur before |
| 1161 | * our current estimate of the head. What we're trying to detect is |
| 1162 | * 1 ... | 0 | 1 | 0... |
| 1163 | * ^ binary search ends here |
| 1164 | */ |
| 1165 | if ((error = xlog_find_verify_cycle(log, start_blk, |
| 1166 | (int)num_scan_bblks, 0, &new_blk))) |
| 1167 | goto bp_err; |
| 1168 | if (new_blk != -1) |
| 1169 | last_blk = new_blk; |
| 1170 | |
| 1171 | /* |
| 1172 | * Potentially backup over partial log record write. We don't need |
| 1173 | * to search the end of the log because we know it is zero. |
| 1174 | */ |
| 1175 | if ((error = xlog_find_verify_log_record(log, start_blk, |
| 1176 | &last_blk, 0)) == -1) { |
| 1177 | error = XFS_ERROR(EIO); |
| 1178 | goto bp_err; |
| 1179 | } else if (error) |
| 1180 | goto bp_err; |
| 1181 | |
| 1182 | *blk_no = last_blk; |
| 1183 | bp_err: |
| 1184 | xlog_put_bp(bp); |
| 1185 | if (error) |
| 1186 | return error; |
| 1187 | return -1; |
| 1188 | } |
| 1189 | |
| 1190 | /* |
| 1191 | * These are simple subroutines used by xlog_clear_stale_blocks() below |
| 1192 | * to initialize a buffer full of empty log record headers and write |
| 1193 | * them into the log. |
| 1194 | */ |
| 1195 | STATIC void |
| 1196 | xlog_add_record( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 1197 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1198 | xfs_caddr_t buf, |
| 1199 | int cycle, |
| 1200 | int block, |
| 1201 | int tail_cycle, |
| 1202 | int tail_block) |
| 1203 | { |
| 1204 | xlog_rec_header_t *recp = (xlog_rec_header_t *)buf; |
| 1205 | |
| 1206 | memset(buf, 0, BBSIZE); |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 1207 | recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM); |
| 1208 | recp->h_cycle = cpu_to_be32(cycle); |
| 1209 | recp->h_version = cpu_to_be32( |
Eric Sandeen | 6211870 | 2008-03-06 13:44:28 +1100 | [diff] [blame] | 1210 | xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1); |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 1211 | recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block)); |
| 1212 | recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block)); |
| 1213 | recp->h_fmt = cpu_to_be32(XLOG_FMT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1214 | memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t)); |
| 1215 | } |
| 1216 | |
| 1217 | STATIC int |
| 1218 | xlog_write_log_records( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 1219 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1220 | int cycle, |
| 1221 | int start_block, |
| 1222 | int blocks, |
| 1223 | int tail_cycle, |
| 1224 | int tail_block) |
| 1225 | { |
| 1226 | xfs_caddr_t offset; |
| 1227 | xfs_buf_t *bp; |
| 1228 | int balign, ealign; |
Alex Elder | 69ce58f | 2010-04-20 17:09:59 +1000 | [diff] [blame] | 1229 | int sectbb = log->l_sectBBsize; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1230 | int end_block = start_block + blocks; |
| 1231 | int bufblks; |
| 1232 | int error = 0; |
| 1233 | int i, j = 0; |
| 1234 | |
Alex Elder | 6881a22 | 2010-04-13 15:22:29 +1000 | [diff] [blame] | 1235 | /* |
| 1236 | * Greedily allocate a buffer big enough to handle the full |
| 1237 | * range of basic blocks to be written. If that fails, try |
| 1238 | * a smaller size. We need to be able to write at least a |
| 1239 | * log sector, or we're out of luck. |
| 1240 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1241 | bufblks = 1 << ffs(blocks); |
Dave Chinner | 81158e0 | 2012-04-27 19:45:22 +1000 | [diff] [blame] | 1242 | while (bufblks > log->l_logBBsize) |
| 1243 | bufblks >>= 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1244 | while (!(bp = xlog_get_bp(log, bufblks))) { |
| 1245 | bufblks >>= 1; |
Alex Elder | 69ce58f | 2010-04-20 17:09:59 +1000 | [diff] [blame] | 1246 | if (bufblks < sectbb) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1247 | return ENOMEM; |
| 1248 | } |
| 1249 | |
| 1250 | /* We may need to do a read at the start to fill in part of |
| 1251 | * the buffer in the starting sector not covered by the first |
| 1252 | * write below. |
| 1253 | */ |
Alex Elder | 5c17f53 | 2010-04-13 15:22:48 +1000 | [diff] [blame] | 1254 | balign = round_down(start_block, sectbb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1255 | if (balign != start_block) { |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1256 | error = xlog_bread_noalign(log, start_block, 1, bp); |
| 1257 | if (error) |
| 1258 | goto out_put_bp; |
| 1259 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1260 | j = start_block - balign; |
| 1261 | } |
| 1262 | |
| 1263 | for (i = start_block; i < end_block; i += bufblks) { |
| 1264 | int bcount, endcount; |
| 1265 | |
| 1266 | bcount = min(bufblks, end_block - start_block); |
| 1267 | endcount = bcount - j; |
| 1268 | |
| 1269 | /* We may need to do a read at the end to fill in part of |
| 1270 | * the buffer in the final sector not covered by the write. |
| 1271 | * If this is the same sector as the above read, skip it. |
| 1272 | */ |
Alex Elder | 5c17f53 | 2010-04-13 15:22:48 +1000 | [diff] [blame] | 1273 | ealign = round_down(end_block, sectbb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1274 | if (j == 0 && (start_block + endcount > ealign)) { |
Chandra Seetharaman | 6292604 | 2011-07-22 23:40:15 +0000 | [diff] [blame] | 1275 | offset = bp->b_addr + BBTOB(ealign - start_block); |
Dave Chinner | 4439647 | 2011-04-21 09:34:27 +0000 | [diff] [blame] | 1276 | error = xlog_bread_offset(log, ealign, sectbb, |
| 1277 | bp, offset); |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1278 | if (error) |
| 1279 | break; |
| 1280 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1281 | } |
| 1282 | |
| 1283 | offset = xlog_align(log, start_block, endcount, bp); |
| 1284 | for (; j < endcount; j++) { |
| 1285 | xlog_add_record(log, offset, cycle, i+j, |
| 1286 | tail_cycle, tail_block); |
| 1287 | offset += BBSIZE; |
| 1288 | } |
| 1289 | error = xlog_bwrite(log, start_block, endcount, bp); |
| 1290 | if (error) |
| 1291 | break; |
| 1292 | start_block += endcount; |
| 1293 | j = 0; |
| 1294 | } |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 1295 | |
| 1296 | out_put_bp: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1297 | xlog_put_bp(bp); |
| 1298 | return error; |
| 1299 | } |
| 1300 | |
| 1301 | /* |
| 1302 | * This routine is called to blow away any incomplete log writes out |
| 1303 | * in front of the log head. We do this so that we won't become confused |
| 1304 | * if we come up, write only a little bit more, and then crash again. |
| 1305 | * If we leave the partial log records out there, this situation could |
| 1306 | * cause us to think those partial writes are valid blocks since they |
| 1307 | * have the current cycle number. We get rid of them by overwriting them |
| 1308 | * with empty log records with the old cycle number rather than the |
| 1309 | * current one. |
| 1310 | * |
| 1311 | * The tail lsn is passed in rather than taken from |
| 1312 | * the log so that we will not write over the unmount record after a |
| 1313 | * clean unmount in a 512 block log. Doing so would leave the log without |
| 1314 | * any valid log records in it until a new one was written. If we crashed |
| 1315 | * during that time we would not be able to recover. |
| 1316 | */ |
| 1317 | STATIC int |
| 1318 | xlog_clear_stale_blocks( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 1319 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1320 | xfs_lsn_t tail_lsn) |
| 1321 | { |
| 1322 | int tail_cycle, head_cycle; |
| 1323 | int tail_block, head_block; |
| 1324 | int tail_distance, max_distance; |
| 1325 | int distance; |
| 1326 | int error; |
| 1327 | |
| 1328 | tail_cycle = CYCLE_LSN(tail_lsn); |
| 1329 | tail_block = BLOCK_LSN(tail_lsn); |
| 1330 | head_cycle = log->l_curr_cycle; |
| 1331 | head_block = log->l_curr_block; |
| 1332 | |
| 1333 | /* |
| 1334 | * Figure out the distance between the new head of the log |
| 1335 | * and the tail. We want to write over any blocks beyond the |
| 1336 | * head that we may have written just before the crash, but |
| 1337 | * we don't want to overwrite the tail of the log. |
| 1338 | */ |
| 1339 | if (head_cycle == tail_cycle) { |
| 1340 | /* |
| 1341 | * The tail is behind the head in the physical log, |
| 1342 | * so the distance from the head to the tail is the |
| 1343 | * distance from the head to the end of the log plus |
| 1344 | * the distance from the beginning of the log to the |
| 1345 | * tail. |
| 1346 | */ |
| 1347 | if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) { |
| 1348 | XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)", |
| 1349 | XFS_ERRLEVEL_LOW, log->l_mp); |
| 1350 | return XFS_ERROR(EFSCORRUPTED); |
| 1351 | } |
| 1352 | tail_distance = tail_block + (log->l_logBBsize - head_block); |
| 1353 | } else { |
| 1354 | /* |
| 1355 | * The head is behind the tail in the physical log, |
| 1356 | * so the distance from the head to the tail is just |
| 1357 | * the tail block minus the head block. |
| 1358 | */ |
| 1359 | if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){ |
| 1360 | XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)", |
| 1361 | XFS_ERRLEVEL_LOW, log->l_mp); |
| 1362 | return XFS_ERROR(EFSCORRUPTED); |
| 1363 | } |
| 1364 | tail_distance = tail_block - head_block; |
| 1365 | } |
| 1366 | |
| 1367 | /* |
| 1368 | * If the head is right up against the tail, we can't clear |
| 1369 | * anything. |
| 1370 | */ |
| 1371 | if (tail_distance <= 0) { |
| 1372 | ASSERT(tail_distance == 0); |
| 1373 | return 0; |
| 1374 | } |
| 1375 | |
| 1376 | max_distance = XLOG_TOTAL_REC_SHIFT(log); |
| 1377 | /* |
| 1378 | * Take the smaller of the maximum amount of outstanding I/O |
| 1379 | * we could have and the distance to the tail to clear out. |
| 1380 | * We take the smaller so that we don't overwrite the tail and |
| 1381 | * we don't waste all day writing from the head to the tail |
| 1382 | * for no reason. |
| 1383 | */ |
| 1384 | max_distance = MIN(max_distance, tail_distance); |
| 1385 | |
| 1386 | if ((head_block + max_distance) <= log->l_logBBsize) { |
| 1387 | /* |
| 1388 | * We can stomp all the blocks we need to without |
| 1389 | * wrapping around the end of the log. Just do it |
| 1390 | * in a single write. Use the cycle number of the |
| 1391 | * current cycle minus one so that the log will look like: |
| 1392 | * n ... | n - 1 ... |
| 1393 | */ |
| 1394 | error = xlog_write_log_records(log, (head_cycle - 1), |
| 1395 | head_block, max_distance, tail_cycle, |
| 1396 | tail_block); |
| 1397 | if (error) |
| 1398 | return error; |
| 1399 | } else { |
| 1400 | /* |
| 1401 | * We need to wrap around the end of the physical log in |
| 1402 | * order to clear all the blocks. Do it in two separate |
| 1403 | * I/Os. The first write should be from the head to the |
| 1404 | * end of the physical log, and it should use the current |
| 1405 | * cycle number minus one just like above. |
| 1406 | */ |
| 1407 | distance = log->l_logBBsize - head_block; |
| 1408 | error = xlog_write_log_records(log, (head_cycle - 1), |
| 1409 | head_block, distance, tail_cycle, |
| 1410 | tail_block); |
| 1411 | |
| 1412 | if (error) |
| 1413 | return error; |
| 1414 | |
| 1415 | /* |
| 1416 | * Now write the blocks at the start of the physical log. |
| 1417 | * This writes the remainder of the blocks we want to clear. |
| 1418 | * It uses the current cycle number since we're now on the |
| 1419 | * same cycle as the head so that we get: |
| 1420 | * n ... n ... | n - 1 ... |
| 1421 | * ^^^^^ blocks we're writing |
| 1422 | */ |
| 1423 | distance = max_distance - (log->l_logBBsize - head_block); |
| 1424 | error = xlog_write_log_records(log, head_cycle, 0, distance, |
| 1425 | tail_cycle, tail_block); |
| 1426 | if (error) |
| 1427 | return error; |
| 1428 | } |
| 1429 | |
| 1430 | return 0; |
| 1431 | } |
| 1432 | |
| 1433 | /****************************************************************************** |
| 1434 | * |
| 1435 | * Log recover routines |
| 1436 | * |
| 1437 | ****************************************************************************** |
| 1438 | */ |
| 1439 | |
| 1440 | STATIC xlog_recover_t * |
| 1441 | xlog_recover_find_tid( |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1442 | struct hlist_head *head, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1443 | xlog_tid_t tid) |
| 1444 | { |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1445 | xlog_recover_t *trans; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1446 | |
Sasha Levin | b67bfe0 | 2013-02-27 17:06:00 -0800 | [diff] [blame] | 1447 | hlist_for_each_entry(trans, head, r_list) { |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1448 | if (trans->r_log_tid == tid) |
| 1449 | return trans; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1450 | } |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1451 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1452 | } |
| 1453 | |
| 1454 | STATIC void |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1455 | xlog_recover_new_tid( |
| 1456 | struct hlist_head *head, |
| 1457 | xlog_tid_t tid, |
| 1458 | xfs_lsn_t lsn) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1459 | { |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1460 | xlog_recover_t *trans; |
| 1461 | |
| 1462 | trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP); |
| 1463 | trans->r_log_tid = tid; |
| 1464 | trans->r_lsn = lsn; |
| 1465 | INIT_LIST_HEAD(&trans->r_itemq); |
| 1466 | |
| 1467 | INIT_HLIST_NODE(&trans->r_list); |
| 1468 | hlist_add_head(&trans->r_list, head); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1469 | } |
| 1470 | |
| 1471 | STATIC void |
| 1472 | xlog_recover_add_item( |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1473 | struct list_head *head) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1474 | { |
| 1475 | xlog_recover_item_t *item; |
| 1476 | |
| 1477 | item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP); |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1478 | INIT_LIST_HEAD(&item->ri_list); |
| 1479 | list_add_tail(&item->ri_list, head); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1480 | } |
| 1481 | |
| 1482 | STATIC int |
| 1483 | xlog_recover_add_to_cont_trans( |
Mark Tinguely | ad223e6 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 1484 | struct xlog *log, |
| 1485 | struct xlog_recover *trans, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1486 | xfs_caddr_t dp, |
| 1487 | int len) |
| 1488 | { |
| 1489 | xlog_recover_item_t *item; |
| 1490 | xfs_caddr_t ptr, old_ptr; |
| 1491 | int old_len; |
| 1492 | |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1493 | if (list_empty(&trans->r_itemq)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1494 | /* finish copying rest of trans header */ |
| 1495 | xlog_recover_add_item(&trans->r_itemq); |
| 1496 | ptr = (xfs_caddr_t) &trans->r_theader + |
| 1497 | sizeof(xfs_trans_header_t) - len; |
| 1498 | memcpy(ptr, dp, len); /* d, s, l */ |
| 1499 | return 0; |
| 1500 | } |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1501 | /* take the tail entry */ |
| 1502 | item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1503 | |
| 1504 | old_ptr = item->ri_buf[item->ri_cnt-1].i_addr; |
| 1505 | old_len = item->ri_buf[item->ri_cnt-1].i_len; |
| 1506 | |
Mitsuo Hayasaka | 4505360 | 2012-01-27 06:37:26 +0000 | [diff] [blame] | 1507 | ptr = kmem_realloc(old_ptr, len+old_len, old_len, KM_SLEEP); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1508 | memcpy(&ptr[old_len], dp, len); /* d, s, l */ |
| 1509 | item->ri_buf[item->ri_cnt-1].i_len += len; |
| 1510 | item->ri_buf[item->ri_cnt-1].i_addr = ptr; |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 1511 | trace_xfs_log_recover_item_add_cont(log, trans, item, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1512 | return 0; |
| 1513 | } |
| 1514 | |
| 1515 | /* |
| 1516 | * The next region to add is the start of a new region. It could be |
| 1517 | * a whole region or it could be the first part of a new region. Because |
| 1518 | * of this, the assumption here is that the type and size fields of all |
| 1519 | * format structures fit into the first 32 bits of the structure. |
| 1520 | * |
| 1521 | * This works because all regions must be 32 bit aligned. Therefore, we |
| 1522 | * either have both fields or we have neither field. In the case we have |
| 1523 | * neither field, the data part of the region is zero length. We only have |
| 1524 | * a log_op_header and can throw away the header since a new one will appear |
| 1525 | * later. If we have at least 4 bytes, then we can determine how many regions |
| 1526 | * will appear in the current log item. |
| 1527 | */ |
| 1528 | STATIC int |
| 1529 | xlog_recover_add_to_trans( |
Mark Tinguely | ad223e6 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 1530 | struct xlog *log, |
| 1531 | struct xlog_recover *trans, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1532 | xfs_caddr_t dp, |
| 1533 | int len) |
| 1534 | { |
| 1535 | xfs_inode_log_format_t *in_f; /* any will do */ |
| 1536 | xlog_recover_item_t *item; |
| 1537 | xfs_caddr_t ptr; |
| 1538 | |
| 1539 | if (!len) |
| 1540 | return 0; |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1541 | if (list_empty(&trans->r_itemq)) { |
David Chinner | 5a792c4 | 2008-10-30 17:40:09 +1100 | [diff] [blame] | 1542 | /* we need to catch log corruptions here */ |
| 1543 | if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 1544 | xfs_warn(log->l_mp, "%s: bad header magic number", |
| 1545 | __func__); |
David Chinner | 5a792c4 | 2008-10-30 17:40:09 +1100 | [diff] [blame] | 1546 | ASSERT(0); |
| 1547 | return XFS_ERROR(EIO); |
| 1548 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1549 | if (len == sizeof(xfs_trans_header_t)) |
| 1550 | xlog_recover_add_item(&trans->r_itemq); |
| 1551 | memcpy(&trans->r_theader, dp, len); /* d, s, l */ |
| 1552 | return 0; |
| 1553 | } |
| 1554 | |
| 1555 | ptr = kmem_alloc(len, KM_SLEEP); |
| 1556 | memcpy(ptr, dp, len); |
| 1557 | in_f = (xfs_inode_log_format_t *)ptr; |
| 1558 | |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1559 | /* take the tail entry */ |
| 1560 | item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list); |
| 1561 | if (item->ri_total != 0 && |
| 1562 | item->ri_total == item->ri_cnt) { |
| 1563 | /* tail item is in use, get a new one */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1564 | xlog_recover_add_item(&trans->r_itemq); |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1565 | item = list_entry(trans->r_itemq.prev, |
| 1566 | xlog_recover_item_t, ri_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1567 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1568 | |
| 1569 | if (item->ri_total == 0) { /* first region to be added */ |
Christoph Hellwig | e8fa6b4 | 2009-03-03 14:48:36 -0500 | [diff] [blame] | 1570 | if (in_f->ilf_size == 0 || |
| 1571 | in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 1572 | xfs_warn(log->l_mp, |
| 1573 | "bad number of regions (%d) in inode log format", |
Christoph Hellwig | e8fa6b4 | 2009-03-03 14:48:36 -0500 | [diff] [blame] | 1574 | in_f->ilf_size); |
| 1575 | ASSERT(0); |
| 1576 | return XFS_ERROR(EIO); |
| 1577 | } |
| 1578 | |
| 1579 | item->ri_total = in_f->ilf_size; |
| 1580 | item->ri_buf = |
| 1581 | kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t), |
| 1582 | KM_SLEEP); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1583 | } |
| 1584 | ASSERT(item->ri_total > item->ri_cnt); |
| 1585 | /* Description region is ri_buf[0] */ |
| 1586 | item->ri_buf[item->ri_cnt].i_addr = ptr; |
| 1587 | item->ri_buf[item->ri_cnt].i_len = len; |
| 1588 | item->ri_cnt++; |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 1589 | trace_xfs_log_recover_item_add(log, trans, item, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1590 | return 0; |
| 1591 | } |
| 1592 | |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1593 | /* |
| 1594 | * Sort the log items in the transaction. Cancelled buffers need |
| 1595 | * to be put first so they are processed before any items that might |
| 1596 | * modify the buffers. If they are cancelled, then the modifications |
| 1597 | * don't need to be replayed. |
| 1598 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1599 | STATIC int |
| 1600 | xlog_recover_reorder_trans( |
Mark Tinguely | ad223e6 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 1601 | struct xlog *log, |
| 1602 | struct xlog_recover *trans, |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 1603 | int pass) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1604 | { |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1605 | xlog_recover_item_t *item, *n; |
| 1606 | LIST_HEAD(sort_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1607 | |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1608 | list_splice_init(&trans->r_itemq, &sort_list); |
| 1609 | list_for_each_entry_safe(item, n, &sort_list, ri_list) { |
Christoph Hellwig | 4e0d5f9 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 1610 | 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] | 1611 | |
| 1612 | switch (ITEM_TYPE(item)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1613 | case XFS_LI_BUF: |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 1614 | if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) { |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 1615 | trace_xfs_log_recover_item_reorder_head(log, |
| 1616 | trans, item, pass); |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1617 | list_move(&item->ri_list, &trans->r_itemq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1618 | break; |
| 1619 | } |
| 1620 | case XFS_LI_INODE: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1621 | case XFS_LI_DQUOT: |
| 1622 | case XFS_LI_QUOTAOFF: |
| 1623 | case XFS_LI_EFD: |
| 1624 | case XFS_LI_EFI: |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 1625 | trace_xfs_log_recover_item_reorder_tail(log, |
| 1626 | trans, item, pass); |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1627 | list_move_tail(&item->ri_list, &trans->r_itemq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1628 | break; |
| 1629 | default: |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 1630 | xfs_warn(log->l_mp, |
| 1631 | "%s: unrecognized type of log operation", |
| 1632 | __func__); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1633 | ASSERT(0); |
| 1634 | return XFS_ERROR(EIO); |
| 1635 | } |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 1636 | } |
| 1637 | ASSERT(list_empty(&sort_list)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1638 | return 0; |
| 1639 | } |
| 1640 | |
| 1641 | /* |
| 1642 | * Build up the table of buf cancel records so that we don't replay |
| 1643 | * cancelled data in the second pass. For buffer records that are |
| 1644 | * not cancel records, there is nothing to do here so we just return. |
| 1645 | * |
| 1646 | * If we get a cancel record which is already in the table, this indicates |
| 1647 | * that the buffer was cancelled multiple times. In order to ensure |
| 1648 | * that during pass 2 we keep the record in the table until we reach its |
| 1649 | * last occurrence in the log, we keep a reference count in the cancel |
| 1650 | * record in the table to tell us how many times we expect to see this |
| 1651 | * record during the second pass. |
| 1652 | */ |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 1653 | STATIC int |
| 1654 | xlog_recover_buffer_pass1( |
Mark Tinguely | ad223e6 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 1655 | struct xlog *log, |
| 1656 | struct xlog_recover_item *item) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1657 | { |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 1658 | 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] | 1659 | struct list_head *bucket; |
| 1660 | struct xfs_buf_cancel *bcp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1661 | |
| 1662 | /* |
| 1663 | * If this isn't a cancel buffer item, then just return. |
| 1664 | */ |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 1665 | if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) { |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 1666 | trace_xfs_log_recover_buf_not_cancel(log, buf_f); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 1667 | return 0; |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 1668 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1669 | |
| 1670 | /* |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 1671 | * Insert an xfs_buf_cancel record into the hash table of them. |
| 1672 | * If there is already an identical record, bump its reference count. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1673 | */ |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 1674 | bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno); |
| 1675 | list_for_each_entry(bcp, bucket, bc_list) { |
| 1676 | if (bcp->bc_blkno == buf_f->blf_blkno && |
| 1677 | bcp->bc_len == buf_f->blf_len) { |
| 1678 | bcp->bc_refcount++; |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 1679 | trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 1680 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1681 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1682 | } |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 1683 | |
| 1684 | bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP); |
| 1685 | bcp->bc_blkno = buf_f->blf_blkno; |
| 1686 | bcp->bc_len = buf_f->blf_len; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1687 | bcp->bc_refcount = 1; |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 1688 | list_add_tail(&bcp->bc_list, bucket); |
| 1689 | |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 1690 | trace_xfs_log_recover_buf_cancel_add(log, buf_f); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 1691 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1692 | } |
| 1693 | |
| 1694 | /* |
| 1695 | * Check to see whether the buffer being recovered has a corresponding |
| 1696 | * entry in the buffer cancel record table. If it does then return 1 |
| 1697 | * so that it will be cancelled, otherwise return 0. If the buffer is |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 1698 | * actually a buffer cancel item (XFS_BLF_CANCEL is set), then decrement |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1699 | * the refcount on the entry in the table and remove it from the table |
| 1700 | * if this is the last reference. |
| 1701 | * |
| 1702 | * We remove the cancel record from the table when we encounter its |
| 1703 | * last occurrence in the log so that if the same buffer is re-used |
| 1704 | * again after its last cancellation we actually replay the changes |
| 1705 | * made at that point. |
| 1706 | */ |
| 1707 | STATIC int |
| 1708 | xlog_check_buffer_cancelled( |
Mark Tinguely | ad223e6 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 1709 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1710 | xfs_daddr_t blkno, |
| 1711 | uint len, |
| 1712 | ushort flags) |
| 1713 | { |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 1714 | struct list_head *bucket; |
| 1715 | struct xfs_buf_cancel *bcp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1716 | |
| 1717 | if (log->l_buf_cancel_table == NULL) { |
| 1718 | /* |
| 1719 | * There is nothing in the table built in pass one, |
| 1720 | * so this buffer must not be cancelled. |
| 1721 | */ |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 1722 | ASSERT(!(flags & XFS_BLF_CANCEL)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1723 | return 0; |
| 1724 | } |
| 1725 | |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 1726 | /* |
| 1727 | * Search for an entry in the cancel table that matches our buffer. |
| 1728 | */ |
| 1729 | bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno); |
| 1730 | list_for_each_entry(bcp, bucket, bc_list) { |
| 1731 | if (bcp->bc_blkno == blkno && bcp->bc_len == len) |
| 1732 | goto found; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1733 | } |
| 1734 | |
| 1735 | /* |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 1736 | * We didn't find a corresponding entry in the table, so return 0 so |
| 1737 | * that the buffer is NOT cancelled. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1738 | */ |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 1739 | ASSERT(!(flags & XFS_BLF_CANCEL)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1740 | return 0; |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 1741 | |
| 1742 | found: |
| 1743 | /* |
| 1744 | * We've go a match, so return 1 so that the recovery of this buffer |
| 1745 | * is cancelled. If this buffer is actually a buffer cancel log |
| 1746 | * item, then decrement the refcount on the one in the table and |
| 1747 | * remove it if this is the last reference. |
| 1748 | */ |
| 1749 | if (flags & XFS_BLF_CANCEL) { |
| 1750 | if (--bcp->bc_refcount == 0) { |
| 1751 | list_del(&bcp->bc_list); |
| 1752 | kmem_free(bcp); |
| 1753 | } |
| 1754 | } |
| 1755 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1756 | } |
| 1757 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1758 | /* |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 1759 | * Perform recovery for a buffer full of inodes. In these buffers, the only |
| 1760 | * data which should be recovered is that which corresponds to the |
| 1761 | * di_next_unlinked pointers in the on disk inode structures. The rest of the |
| 1762 | * data for the inodes is always logged through the inodes themselves rather |
| 1763 | * than the inode buffer and is recovered in xlog_recover_inode_pass2(). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1764 | * |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 1765 | * The only time when buffers full of inodes are fully recovered is when the |
| 1766 | * buffer is full of newly allocated inodes. In this case the buffer will |
| 1767 | * not be marked as an inode buffer and so will be sent to |
| 1768 | * xlog_recover_do_reg_buffer() below during recovery. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1769 | */ |
| 1770 | STATIC int |
| 1771 | xlog_recover_do_inode_buffer( |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 1772 | struct xfs_mount *mp, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1773 | xlog_recover_item_t *item, |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 1774 | struct xfs_buf *bp, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1775 | xfs_buf_log_format_t *buf_f) |
| 1776 | { |
| 1777 | int i; |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 1778 | int item_index = 0; |
| 1779 | int bit = 0; |
| 1780 | int nbits = 0; |
| 1781 | int reg_buf_offset = 0; |
| 1782 | int reg_buf_bytes = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1783 | int next_unlinked_offset; |
| 1784 | int inodes_per_buf; |
| 1785 | xfs_agino_t *logged_nextp; |
| 1786 | xfs_agino_t *buffer_nextp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1787 | |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 1788 | trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f); |
Christoph Hellwig | 93848a9 | 2013-04-03 16:11:17 +1100 | [diff] [blame] | 1789 | bp->b_ops = &xfs_inode_buf_ops; |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 1790 | |
Dave Chinner | aa0e883 | 2012-04-23 15:58:52 +1000 | [diff] [blame] | 1791 | 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] | 1792 | for (i = 0; i < inodes_per_buf; i++) { |
| 1793 | next_unlinked_offset = (i * mp->m_sb.sb_inodesize) + |
| 1794 | offsetof(xfs_dinode_t, di_next_unlinked); |
| 1795 | |
| 1796 | while (next_unlinked_offset >= |
| 1797 | (reg_buf_offset + reg_buf_bytes)) { |
| 1798 | /* |
| 1799 | * The next di_next_unlinked field is beyond |
| 1800 | * the current logged region. Find the next |
| 1801 | * logged region that contains or is beyond |
| 1802 | * the current di_next_unlinked field. |
| 1803 | */ |
| 1804 | bit += nbits; |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 1805 | bit = xfs_next_bit(buf_f->blf_data_map, |
| 1806 | buf_f->blf_map_size, bit); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1807 | |
| 1808 | /* |
| 1809 | * If there are no more logged regions in the |
| 1810 | * buffer, then we're done. |
| 1811 | */ |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 1812 | if (bit == -1) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1813 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1814 | |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 1815 | nbits = xfs_contig_bits(buf_f->blf_data_map, |
| 1816 | buf_f->blf_map_size, bit); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1817 | ASSERT(nbits > 0); |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 1818 | reg_buf_offset = bit << XFS_BLF_SHIFT; |
| 1819 | reg_buf_bytes = nbits << XFS_BLF_SHIFT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1820 | item_index++; |
| 1821 | } |
| 1822 | |
| 1823 | /* |
| 1824 | * If the current logged region starts after the current |
| 1825 | * di_next_unlinked field, then move on to the next |
| 1826 | * di_next_unlinked field. |
| 1827 | */ |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 1828 | if (next_unlinked_offset < reg_buf_offset) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1829 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1830 | |
| 1831 | ASSERT(item->ri_buf[item_index].i_addr != NULL); |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 1832 | ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0); |
Dave Chinner | aa0e883 | 2012-04-23 15:58:52 +1000 | [diff] [blame] | 1833 | ASSERT((reg_buf_offset + reg_buf_bytes) <= |
| 1834 | BBTOB(bp->b_io_length)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1835 | |
| 1836 | /* |
| 1837 | * The current logged region contains a copy of the |
| 1838 | * current di_next_unlinked field. Extract its value |
| 1839 | * and copy it to the buffer copy. |
| 1840 | */ |
Christoph Hellwig | 4e0d5f9 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 1841 | logged_nextp = item->ri_buf[item_index].i_addr + |
| 1842 | next_unlinked_offset - reg_buf_offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1843 | if (unlikely(*logged_nextp == 0)) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 1844 | xfs_alert(mp, |
| 1845 | "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). " |
| 1846 | "Trying to replay bad (0) inode di_next_unlinked field.", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1847 | item, bp); |
| 1848 | XFS_ERROR_REPORT("xlog_recover_do_inode_buf", |
| 1849 | XFS_ERRLEVEL_LOW, mp); |
| 1850 | return XFS_ERROR(EFSCORRUPTED); |
| 1851 | } |
| 1852 | |
| 1853 | buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp, |
| 1854 | next_unlinked_offset); |
Tim Shimmin | 87c199c | 2006-06-09 14:56:16 +1000 | [diff] [blame] | 1855 | *buffer_nextp = *logged_nextp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1856 | } |
| 1857 | |
| 1858 | return 0; |
| 1859 | } |
| 1860 | |
| 1861 | /* |
| 1862 | * Perform a 'normal' buffer recovery. Each logged region of the |
| 1863 | * buffer should be copied over the corresponding region in the |
| 1864 | * given buffer. The bitmap in the buf log format structure indicates |
| 1865 | * where to place the logged data. |
| 1866 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1867 | STATIC void |
| 1868 | xlog_recover_do_reg_buffer( |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 1869 | struct xfs_mount *mp, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1870 | xlog_recover_item_t *item, |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 1871 | struct xfs_buf *bp, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1872 | xfs_buf_log_format_t *buf_f) |
| 1873 | { |
| 1874 | int i; |
| 1875 | int bit; |
| 1876 | int nbits; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1877 | int error; |
| 1878 | |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 1879 | trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f); |
| 1880 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1881 | bit = 0; |
| 1882 | i = 1; /* 0 is the buf format structure */ |
| 1883 | while (1) { |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 1884 | bit = xfs_next_bit(buf_f->blf_data_map, |
| 1885 | buf_f->blf_map_size, bit); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1886 | if (bit == -1) |
| 1887 | break; |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 1888 | nbits = xfs_contig_bits(buf_f->blf_data_map, |
| 1889 | buf_f->blf_map_size, bit); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1890 | ASSERT(nbits > 0); |
Christoph Hellwig | 4b80916 | 2007-08-16 15:37:36 +1000 | [diff] [blame] | 1891 | ASSERT(item->ri_buf[i].i_addr != NULL); |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 1892 | ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0); |
Dave Chinner | aa0e883 | 2012-04-23 15:58:52 +1000 | [diff] [blame] | 1893 | ASSERT(BBTOB(bp->b_io_length) >= |
| 1894 | ((uint)bit << XFS_BLF_SHIFT) + (nbits << XFS_BLF_SHIFT)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1895 | |
| 1896 | /* |
| 1897 | * Do a sanity check if this is a dquot buffer. Just checking |
| 1898 | * the first dquot in the buffer should do. XXXThis is |
| 1899 | * probably a good thing to do for other buf types also. |
| 1900 | */ |
| 1901 | error = 0; |
Nathan Scott | c8ad20f | 2005-06-21 15:38:48 +1000 | [diff] [blame] | 1902 | if (buf_f->blf_flags & |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 1903 | (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) { |
Christoph Hellwig | 0c5e1ce | 2009-06-08 15:33:21 +0200 | [diff] [blame] | 1904 | if (item->ri_buf[i].i_addr == NULL) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 1905 | xfs_alert(mp, |
Christoph Hellwig | 0c5e1ce | 2009-06-08 15:33:21 +0200 | [diff] [blame] | 1906 | "XFS: NULL dquot in %s.", __func__); |
| 1907 | goto next; |
| 1908 | } |
Jan Rekorajski | 8ec6dba | 2009-11-16 11:57:02 +0000 | [diff] [blame] | 1909 | if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 1910 | xfs_alert(mp, |
Christoph Hellwig | 0c5e1ce | 2009-06-08 15:33:21 +0200 | [diff] [blame] | 1911 | "XFS: dquot too small (%d) in %s.", |
| 1912 | item->ri_buf[i].i_len, __func__); |
| 1913 | goto next; |
| 1914 | } |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 1915 | error = xfs_qm_dqcheck(mp, item->ri_buf[i].i_addr, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1916 | -1, 0, XFS_QMOPT_DOWARN, |
| 1917 | "dquot_buf_recover"); |
Christoph Hellwig | 0c5e1ce | 2009-06-08 15:33:21 +0200 | [diff] [blame] | 1918 | if (error) |
| 1919 | goto next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1920 | } |
Christoph Hellwig | 0c5e1ce | 2009-06-08 15:33:21 +0200 | [diff] [blame] | 1921 | |
| 1922 | memcpy(xfs_buf_offset(bp, |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 1923 | (uint)bit << XFS_BLF_SHIFT), /* dest */ |
Christoph Hellwig | 0c5e1ce | 2009-06-08 15:33:21 +0200 | [diff] [blame] | 1924 | item->ri_buf[i].i_addr, /* source */ |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 1925 | nbits<<XFS_BLF_SHIFT); /* length */ |
Christoph Hellwig | 0c5e1ce | 2009-06-08 15:33:21 +0200 | [diff] [blame] | 1926 | next: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1927 | i++; |
| 1928 | bit += nbits; |
| 1929 | } |
| 1930 | |
| 1931 | /* Shouldn't be any more regions */ |
| 1932 | ASSERT(i == item->ri_total); |
Christoph Hellwig | ee1a47a | 2013-04-21 14:53:46 -0500 | [diff] [blame] | 1933 | |
| 1934 | switch (buf_f->blf_flags & XFS_BLF_TYPE_MASK) { |
| 1935 | case XFS_BLF_BTREE_BUF: |
| 1936 | switch (be32_to_cpu(*(__be32 *)bp->b_addr)) { |
| 1937 | case XFS_ABTB_CRC_MAGIC: |
| 1938 | case XFS_ABTC_CRC_MAGIC: |
| 1939 | case XFS_ABTB_MAGIC: |
| 1940 | case XFS_ABTC_MAGIC: |
| 1941 | bp->b_ops = &xfs_allocbt_buf_ops; |
| 1942 | break; |
| 1943 | case XFS_IBT_CRC_MAGIC: |
| 1944 | case XFS_IBT_MAGIC: |
| 1945 | bp->b_ops = &xfs_inobt_buf_ops; |
| 1946 | break; |
| 1947 | case XFS_BMAP_CRC_MAGIC: |
| 1948 | case XFS_BMAP_MAGIC: |
| 1949 | bp->b_ops = &xfs_bmbt_buf_ops; |
| 1950 | break; |
| 1951 | default: |
| 1952 | xfs_warn(mp, "Bad btree block magic!"); |
| 1953 | ASSERT(0); |
| 1954 | break; |
| 1955 | } |
| 1956 | break; |
Dave Chinner | 4e0e604 | 2013-04-03 16:11:13 +1100 | [diff] [blame] | 1957 | case XFS_BLF_AGF_BUF: |
| 1958 | if (*(__be32 *)bp->b_addr != cpu_to_be32(XFS_AGF_MAGIC)) { |
| 1959 | xfs_warn(mp, "Bad AGF block magic!"); |
| 1960 | ASSERT(0); |
| 1961 | break; |
| 1962 | } |
| 1963 | bp->b_ops = &xfs_agf_buf_ops; |
| 1964 | break; |
Christoph Hellwig | 77c95bb | 2013-04-03 16:11:14 +1100 | [diff] [blame] | 1965 | case XFS_BLF_AGFL_BUF: |
| 1966 | if (!xfs_sb_version_hascrc(&mp->m_sb)) |
| 1967 | break; |
| 1968 | if (*(__be32 *)bp->b_addr != cpu_to_be32(XFS_AGFL_MAGIC)) { |
| 1969 | xfs_warn(mp, "Bad AGFL block magic!"); |
| 1970 | ASSERT(0); |
| 1971 | break; |
| 1972 | } |
| 1973 | bp->b_ops = &xfs_agfl_buf_ops; |
| 1974 | break; |
Dave Chinner | 983d09f | 2013-04-03 16:11:15 +1100 | [diff] [blame] | 1975 | case XFS_BLF_AGI_BUF: |
| 1976 | if (*(__be32 *)bp->b_addr != cpu_to_be32(XFS_AGI_MAGIC)) { |
| 1977 | xfs_warn(mp, "Bad AGI block magic!"); |
| 1978 | ASSERT(0); |
| 1979 | break; |
| 1980 | } |
| 1981 | bp->b_ops = &xfs_agi_buf_ops; |
| 1982 | break; |
Christoph Hellwig | 3fe58f3 | 2013-04-03 16:11:16 +1100 | [diff] [blame] | 1983 | case XFS_BLF_UDQUOT_BUF: |
| 1984 | case XFS_BLF_PDQUOT_BUF: |
| 1985 | case XFS_BLF_GDQUOT_BUF: |
| 1986 | if (*(__be16 *)bp->b_addr != cpu_to_be16(XFS_DQUOT_MAGIC)) { |
| 1987 | xfs_warn(mp, "Bad DQUOT block magic!"); |
| 1988 | ASSERT(0); |
| 1989 | break; |
| 1990 | } |
| 1991 | bp->b_ops = &xfs_dquot_buf_ops; |
| 1992 | break; |
Christoph Hellwig | 93848a9 | 2013-04-03 16:11:17 +1100 | [diff] [blame] | 1993 | case XFS_BLF_DINO_BUF: |
| 1994 | /* |
| 1995 | * we get here with inode allocation buffers, not buffers that |
| 1996 | * track unlinked list changes. |
| 1997 | */ |
| 1998 | if (*(__be16 *)bp->b_addr != cpu_to_be16(XFS_DINODE_MAGIC)) { |
| 1999 | xfs_warn(mp, "Bad INODE block magic!"); |
| 2000 | ASSERT(0); |
| 2001 | break; |
| 2002 | } |
| 2003 | bp->b_ops = &xfs_inode_buf_ops; |
| 2004 | break; |
Christoph Hellwig | ee1a47a | 2013-04-21 14:53:46 -0500 | [diff] [blame] | 2005 | default: |
| 2006 | break; |
| 2007 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2008 | } |
| 2009 | |
| 2010 | /* |
| 2011 | * Do some primitive error checking on ondisk dquot data structures. |
| 2012 | */ |
| 2013 | int |
| 2014 | xfs_qm_dqcheck( |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2015 | struct xfs_mount *mp, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2016 | xfs_disk_dquot_t *ddq, |
| 2017 | xfs_dqid_t id, |
| 2018 | uint type, /* used only when IO_dorepair is true */ |
| 2019 | uint flags, |
| 2020 | char *str) |
| 2021 | { |
| 2022 | xfs_dqblk_t *d = (xfs_dqblk_t *)ddq; |
| 2023 | int errs = 0; |
| 2024 | |
| 2025 | /* |
| 2026 | * We can encounter an uninitialized dquot buffer for 2 reasons: |
| 2027 | * 1. If we crash while deleting the quotainode(s), and those blks got |
| 2028 | * used for user data. This is because we take the path of regular |
| 2029 | * file deletion; however, the size field of quotainodes is never |
| 2030 | * updated, so all the tricks that we play in itruncate_finish |
| 2031 | * don't quite matter. |
| 2032 | * |
| 2033 | * 2. We don't play the quota buffers when there's a quotaoff logitem. |
| 2034 | * But the allocation will be replayed so we'll end up with an |
| 2035 | * uninitialized quota block. |
| 2036 | * |
| 2037 | * This is all fine; things are still consistent, and we haven't lost |
| 2038 | * any quota information. Just don't complain about bad dquot blks. |
| 2039 | */ |
Christoph Hellwig | 69ef921 | 2011-07-08 14:36:05 +0200 | [diff] [blame] | 2040 | if (ddq->d_magic != cpu_to_be16(XFS_DQUOT_MAGIC)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2041 | if (flags & XFS_QMOPT_DOWARN) |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2042 | xfs_alert(mp, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2043 | "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x", |
Christoph Hellwig | 1149d96 | 2005-11-02 15:01:12 +1100 | [diff] [blame] | 2044 | str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2045 | errs++; |
| 2046 | } |
Christoph Hellwig | 1149d96 | 2005-11-02 15:01:12 +1100 | [diff] [blame] | 2047 | if (ddq->d_version != XFS_DQUOT_VERSION) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2048 | if (flags & XFS_QMOPT_DOWARN) |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2049 | xfs_alert(mp, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2050 | "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x", |
Christoph Hellwig | 1149d96 | 2005-11-02 15:01:12 +1100 | [diff] [blame] | 2051 | str, id, ddq->d_version, XFS_DQUOT_VERSION); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2052 | errs++; |
| 2053 | } |
| 2054 | |
Christoph Hellwig | 1149d96 | 2005-11-02 15:01:12 +1100 | [diff] [blame] | 2055 | if (ddq->d_flags != XFS_DQ_USER && |
| 2056 | ddq->d_flags != XFS_DQ_PROJ && |
| 2057 | ddq->d_flags != XFS_DQ_GROUP) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2058 | if (flags & XFS_QMOPT_DOWARN) |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2059 | xfs_alert(mp, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2060 | "%s : XFS dquot ID 0x%x, unknown flags 0x%x", |
Christoph Hellwig | 1149d96 | 2005-11-02 15:01:12 +1100 | [diff] [blame] | 2061 | str, id, ddq->d_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2062 | errs++; |
| 2063 | } |
| 2064 | |
Christoph Hellwig | 1149d96 | 2005-11-02 15:01:12 +1100 | [diff] [blame] | 2065 | if (id != -1 && id != be32_to_cpu(ddq->d_id)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2066 | if (flags & XFS_QMOPT_DOWARN) |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2067 | xfs_alert(mp, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2068 | "%s : ondisk-dquot 0x%p, ID mismatch: " |
| 2069 | "0x%x expected, found id 0x%x", |
Christoph Hellwig | 1149d96 | 2005-11-02 15:01:12 +1100 | [diff] [blame] | 2070 | str, ddq, id, be32_to_cpu(ddq->d_id)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2071 | errs++; |
| 2072 | } |
| 2073 | |
| 2074 | if (!errs && ddq->d_id) { |
Christoph Hellwig | 1149d96 | 2005-11-02 15:01:12 +1100 | [diff] [blame] | 2075 | if (ddq->d_blk_softlimit && |
Mitsuo Hayasaka | d0a3fe6 | 2012-02-06 12:50:07 +0000 | [diff] [blame] | 2076 | be64_to_cpu(ddq->d_bcount) > |
Christoph Hellwig | 1149d96 | 2005-11-02 15:01:12 +1100 | [diff] [blame] | 2077 | be64_to_cpu(ddq->d_blk_softlimit)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2078 | if (!ddq->d_btimer) { |
| 2079 | if (flags & XFS_QMOPT_DOWARN) |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2080 | xfs_alert(mp, |
| 2081 | "%s : Dquot ID 0x%x (0x%p) BLK TIMER NOT STARTED", |
Christoph Hellwig | 1149d96 | 2005-11-02 15:01:12 +1100 | [diff] [blame] | 2082 | str, (int)be32_to_cpu(ddq->d_id), ddq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2083 | errs++; |
| 2084 | } |
| 2085 | } |
Christoph Hellwig | 1149d96 | 2005-11-02 15:01:12 +1100 | [diff] [blame] | 2086 | if (ddq->d_ino_softlimit && |
Mitsuo Hayasaka | d0a3fe6 | 2012-02-06 12:50:07 +0000 | [diff] [blame] | 2087 | be64_to_cpu(ddq->d_icount) > |
Christoph Hellwig | 1149d96 | 2005-11-02 15:01:12 +1100 | [diff] [blame] | 2088 | be64_to_cpu(ddq->d_ino_softlimit)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2089 | if (!ddq->d_itimer) { |
| 2090 | if (flags & XFS_QMOPT_DOWARN) |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2091 | xfs_alert(mp, |
| 2092 | "%s : Dquot ID 0x%x (0x%p) INODE TIMER NOT STARTED", |
Christoph Hellwig | 1149d96 | 2005-11-02 15:01:12 +1100 | [diff] [blame] | 2093 | str, (int)be32_to_cpu(ddq->d_id), ddq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2094 | errs++; |
| 2095 | } |
| 2096 | } |
Christoph Hellwig | 1149d96 | 2005-11-02 15:01:12 +1100 | [diff] [blame] | 2097 | if (ddq->d_rtb_softlimit && |
Mitsuo Hayasaka | d0a3fe6 | 2012-02-06 12:50:07 +0000 | [diff] [blame] | 2098 | be64_to_cpu(ddq->d_rtbcount) > |
Christoph Hellwig | 1149d96 | 2005-11-02 15:01:12 +1100 | [diff] [blame] | 2099 | be64_to_cpu(ddq->d_rtb_softlimit)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2100 | if (!ddq->d_rtbtimer) { |
| 2101 | if (flags & XFS_QMOPT_DOWARN) |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2102 | xfs_alert(mp, |
| 2103 | "%s : Dquot ID 0x%x (0x%p) RTBLK TIMER NOT STARTED", |
Christoph Hellwig | 1149d96 | 2005-11-02 15:01:12 +1100 | [diff] [blame] | 2104 | str, (int)be32_to_cpu(ddq->d_id), ddq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2105 | errs++; |
| 2106 | } |
| 2107 | } |
| 2108 | } |
| 2109 | |
| 2110 | if (!errs || !(flags & XFS_QMOPT_DQREPAIR)) |
| 2111 | return errs; |
| 2112 | |
| 2113 | if (flags & XFS_QMOPT_DOWARN) |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2114 | xfs_notice(mp, "Re-initializing dquot ID 0x%x", id); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2115 | |
| 2116 | /* |
| 2117 | * Typically, a repair is only requested by quotacheck. |
| 2118 | */ |
| 2119 | ASSERT(id != -1); |
| 2120 | ASSERT(flags & XFS_QMOPT_DQREPAIR); |
| 2121 | memset(d, 0, sizeof(xfs_dqblk_t)); |
Christoph Hellwig | 1149d96 | 2005-11-02 15:01:12 +1100 | [diff] [blame] | 2122 | |
| 2123 | d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC); |
| 2124 | d->dd_diskdq.d_version = XFS_DQUOT_VERSION; |
| 2125 | d->dd_diskdq.d_flags = type; |
| 2126 | d->dd_diskdq.d_id = cpu_to_be32(id); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2127 | |
| 2128 | return errs; |
| 2129 | } |
| 2130 | |
| 2131 | /* |
| 2132 | * Perform a dquot buffer recovery. |
| 2133 | * Simple algorithm: if we have found a QUOTAOFF logitem of the same type |
| 2134 | * (ie. USR or GRP), then just toss this buffer away; don't recover it. |
| 2135 | * Else, treat it as a regular buffer and do recovery. |
| 2136 | */ |
| 2137 | STATIC void |
| 2138 | xlog_recover_do_dquot_buffer( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 2139 | struct xfs_mount *mp, |
| 2140 | struct xlog *log, |
| 2141 | struct xlog_recover_item *item, |
| 2142 | struct xfs_buf *bp, |
| 2143 | struct xfs_buf_log_format *buf_f) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2144 | { |
| 2145 | uint type; |
| 2146 | |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 2147 | trace_xfs_log_recover_buf_dquot_buf(log, buf_f); |
| 2148 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2149 | /* |
| 2150 | * Filesystems are required to send in quota flags at mount time. |
| 2151 | */ |
| 2152 | if (mp->m_qflags == 0) { |
| 2153 | return; |
| 2154 | } |
| 2155 | |
| 2156 | type = 0; |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 2157 | if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2158 | type |= XFS_DQ_USER; |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 2159 | if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF) |
Nathan Scott | c8ad20f | 2005-06-21 15:38:48 +1000 | [diff] [blame] | 2160 | type |= XFS_DQ_PROJ; |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 2161 | if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2162 | type |= XFS_DQ_GROUP; |
| 2163 | /* |
| 2164 | * This type of quotas was turned off, so ignore this buffer |
| 2165 | */ |
| 2166 | if (log->l_quotaoffs_flag & type) |
| 2167 | return; |
| 2168 | |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 2169 | xlog_recover_do_reg_buffer(mp, item, bp, buf_f); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2170 | } |
| 2171 | |
| 2172 | /* |
| 2173 | * This routine replays a modification made to a buffer at runtime. |
| 2174 | * There are actually two types of buffer, regular and inode, which |
| 2175 | * are handled differently. Inode buffers are handled differently |
| 2176 | * in that we only recover a specific set of data from them, namely |
| 2177 | * the inode di_next_unlinked fields. This is because all other inode |
| 2178 | * data is actually logged via inode records and any data we replay |
| 2179 | * here which overlaps that may be stale. |
| 2180 | * |
| 2181 | * 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] | 2182 | * with the XFS_BLF_CANCEL bit set to indicate that previous copies |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2183 | * of the buffer in the log should not be replayed at recovery time. |
| 2184 | * This is so that if the blocks covered by the buffer are reused for |
| 2185 | * file data before we crash we don't end up replaying old, freed |
| 2186 | * meta-data into a user's file. |
| 2187 | * |
| 2188 | * To handle the cancellation of buffer log items, we make two passes |
| 2189 | * over the log during recovery. During the first we build a table of |
| 2190 | * those buffers which have been cancelled, and during the second we |
| 2191 | * only replay those buffers which do not have corresponding cancel |
| 2192 | * records in the table. See xlog_recover_do_buffer_pass[1,2] above |
| 2193 | * for more details on the implementation of the table of cancel records. |
| 2194 | */ |
| 2195 | STATIC int |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2196 | xlog_recover_buffer_pass2( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 2197 | struct xlog *log, |
| 2198 | struct list_head *buffer_list, |
| 2199 | struct xlog_recover_item *item) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2200 | { |
Christoph Hellwig | 4e0d5f9 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 2201 | 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] | 2202 | xfs_mount_t *mp = log->l_mp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2203 | xfs_buf_t *bp; |
| 2204 | int error; |
Christoph Hellwig | 6ad112b | 2009-11-24 18:02:23 +0000 | [diff] [blame] | 2205 | uint buf_flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2206 | |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2207 | /* |
| 2208 | * In this pass we only want to recover all the buffers which have |
| 2209 | * not been cancelled and are not cancellation buffers themselves. |
| 2210 | */ |
| 2211 | if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno, |
| 2212 | buf_f->blf_len, buf_f->blf_flags)) { |
| 2213 | trace_xfs_log_recover_buf_cancel(log, buf_f); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2214 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2215 | } |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2216 | |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 2217 | trace_xfs_log_recover_buf_recover(log, buf_f); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2218 | |
Dave Chinner | a8acad7 | 2012-04-23 15:58:54 +1000 | [diff] [blame] | 2219 | buf_flags = 0; |
Dave Chinner | 611c994 | 2012-04-23 15:59:07 +1000 | [diff] [blame] | 2220 | if (buf_f->blf_flags & XFS_BLF_INODE_BUF) |
| 2221 | buf_flags |= XBF_UNMAPPED; |
Christoph Hellwig | 6ad112b | 2009-11-24 18:02:23 +0000 | [diff] [blame] | 2222 | |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2223 | 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] | 2224 | buf_flags, NULL); |
Chandra Seetharaman | ac4d688 | 2011-08-03 02:18:29 +0000 | [diff] [blame] | 2225 | if (!bp) |
| 2226 | return XFS_ERROR(ENOMEM); |
Chandra Seetharaman | e570280 | 2011-08-03 02:18:34 +0000 | [diff] [blame] | 2227 | error = bp->b_error; |
Chandra Seetharaman | 5a52c2a58 | 2011-07-22 23:39:51 +0000 | [diff] [blame] | 2228 | if (error) { |
Christoph Hellwig | 901796a | 2011-10-10 16:52:49 +0000 | [diff] [blame] | 2229 | xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2230 | xfs_buf_relse(bp); |
| 2231 | return error; |
| 2232 | } |
| 2233 | |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2234 | if (buf_f->blf_flags & XFS_BLF_INODE_BUF) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2235 | error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f); |
Christoph Hellwig | e2714bf | 2010-12-01 22:06:21 +0000 | [diff] [blame] | 2236 | } else if (buf_f->blf_flags & |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 2237 | (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2238 | xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f); |
| 2239 | } else { |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 2240 | xlog_recover_do_reg_buffer(mp, item, bp, buf_f); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2241 | } |
| 2242 | if (error) |
| 2243 | return XFS_ERROR(error); |
| 2244 | |
| 2245 | /* |
| 2246 | * Perform delayed write on the buffer. Asynchronous writes will be |
| 2247 | * slower when taking into account all the buffers to be flushed. |
| 2248 | * |
| 2249 | * Also make sure that only inode buffers with good sizes stay in |
| 2250 | * the buffer cache. The kernel moves inodes in buffers of 1 block |
| 2251 | * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode |
| 2252 | * buffers in the log can be a different size if the log was generated |
| 2253 | * by an older kernel using unclustered inode buffers or a newer kernel |
| 2254 | * running with a different inode cluster size. Regardless, if the |
| 2255 | * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE) |
| 2256 | * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep |
| 2257 | * the buffer out of the buffer cache so that the buffer won't |
| 2258 | * overlap with future reads of those inodes. |
| 2259 | */ |
| 2260 | if (XFS_DINODE_MAGIC == |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 2261 | be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) && |
Dave Chinner | aa0e883 | 2012-04-23 15:58:52 +1000 | [diff] [blame] | 2262 | (BBTOB(bp->b_io_length) != MAX(log->l_mp->m_sb.sb_blocksize, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2263 | (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) { |
Christoph Hellwig | c867cb6 | 2011-10-10 16:52:46 +0000 | [diff] [blame] | 2264 | xfs_buf_stale(bp); |
Christoph Hellwig | c2b006c | 2011-08-23 08:28:07 +0000 | [diff] [blame] | 2265 | error = xfs_bwrite(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2266 | } else { |
Dave Chinner | ebad861 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 2267 | ASSERT(bp->b_target->bt_mount == mp); |
Christoph Hellwig | cb669ca | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 2268 | bp->b_iodone = xlog_recover_iodone; |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 2269 | xfs_buf_delwri_queue(bp, buffer_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2270 | } |
| 2271 | |
Christoph Hellwig | c2b006c | 2011-08-23 08:28:07 +0000 | [diff] [blame] | 2272 | xfs_buf_relse(bp); |
| 2273 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2274 | } |
| 2275 | |
| 2276 | STATIC int |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2277 | xlog_recover_inode_pass2( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 2278 | struct xlog *log, |
| 2279 | struct list_head *buffer_list, |
| 2280 | struct xlog_recover_item *item) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2281 | { |
| 2282 | xfs_inode_log_format_t *in_f; |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2283 | xfs_mount_t *mp = log->l_mp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2284 | xfs_buf_t *bp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2285 | xfs_dinode_t *dip; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2286 | int len; |
| 2287 | xfs_caddr_t src; |
| 2288 | xfs_caddr_t dest; |
| 2289 | int error; |
| 2290 | int attr_index; |
| 2291 | uint fields; |
Christoph Hellwig | 347d1c0 | 2007-08-28 13:57:51 +1000 | [diff] [blame] | 2292 | xfs_icdinode_t *dicp; |
Christoph Hellwig | 93848a9 | 2013-04-03 16:11:17 +1100 | [diff] [blame] | 2293 | uint isize; |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2294 | int need_free = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2295 | |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2296 | 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] | 2297 | in_f = item->ri_buf[0].i_addr; |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2298 | } else { |
Christoph Hellwig | 4e0d5f9 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 2299 | in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP); |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2300 | need_free = 1; |
| 2301 | error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f); |
| 2302 | if (error) |
| 2303 | goto error; |
| 2304 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2305 | |
| 2306 | /* |
| 2307 | * Inode buffers can be freed, look out for it, |
| 2308 | * and do not replay the inode. |
| 2309 | */ |
Christoph Hellwig | a194189 | 2008-11-28 14:23:40 +1100 | [diff] [blame] | 2310 | if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno, |
| 2311 | in_f->ilf_len, 0)) { |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2312 | error = 0; |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 2313 | trace_xfs_log_recover_inode_cancel(log, in_f); |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2314 | goto error; |
| 2315 | } |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 2316 | trace_xfs_log_recover_inode_recover(log, in_f); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2317 | |
Dave Chinner | c3f8fc7 | 2012-11-12 22:54:01 +1100 | [diff] [blame] | 2318 | 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] | 2319 | &xfs_inode_buf_ops); |
Chandra Seetharaman | ac4d688 | 2011-08-03 02:18:29 +0000 | [diff] [blame] | 2320 | if (!bp) { |
| 2321 | error = ENOMEM; |
| 2322 | goto error; |
| 2323 | } |
Chandra Seetharaman | e570280 | 2011-08-03 02:18:34 +0000 | [diff] [blame] | 2324 | error = bp->b_error; |
Chandra Seetharaman | 5a52c2a58 | 2011-07-22 23:39:51 +0000 | [diff] [blame] | 2325 | if (error) { |
Christoph Hellwig | 901796a | 2011-10-10 16:52:49 +0000 | [diff] [blame] | 2326 | xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2327 | xfs_buf_relse(bp); |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2328 | goto error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2329 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2330 | ASSERT(in_f->ilf_fields & XFS_ILOG_CORE); |
Christoph Hellwig | a194189 | 2008-11-28 14:23:40 +1100 | [diff] [blame] | 2331 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, in_f->ilf_boffset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2332 | |
| 2333 | /* |
| 2334 | * Make sure the place we're flushing out to really looks |
| 2335 | * like an inode! |
| 2336 | */ |
Christoph Hellwig | 69ef921 | 2011-07-08 14:36:05 +0200 | [diff] [blame] | 2337 | if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2338 | xfs_buf_relse(bp); |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2339 | xfs_alert(mp, |
| 2340 | "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld", |
| 2341 | __func__, dip, bp, in_f->ilf_ino); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2342 | XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2343 | XFS_ERRLEVEL_LOW, mp); |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2344 | error = EFSCORRUPTED; |
| 2345 | goto error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2346 | } |
Christoph Hellwig | 4e0d5f9 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 2347 | dicp = item->ri_buf[1].i_addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2348 | if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) { |
| 2349 | xfs_buf_relse(bp); |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2350 | xfs_alert(mp, |
| 2351 | "%s: Bad inode log record, rec ptr 0x%p, ino %Ld", |
| 2352 | __func__, item, in_f->ilf_ino); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2353 | XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2354 | XFS_ERRLEVEL_LOW, mp); |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2355 | error = EFSCORRUPTED; |
| 2356 | goto error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2357 | } |
| 2358 | |
| 2359 | /* Skip replay when the on disk inode is newer than the log one */ |
Christoph Hellwig | 81591fe | 2008-11-28 14:23:39 +1100 | [diff] [blame] | 2360 | if (dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2361 | /* |
| 2362 | * Deal with the wrap case, DI_MAX_FLUSH is less |
| 2363 | * than smaller numbers |
| 2364 | */ |
Christoph Hellwig | 81591fe | 2008-11-28 14:23:39 +1100 | [diff] [blame] | 2365 | if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH && |
Christoph Hellwig | 347d1c0 | 2007-08-28 13:57:51 +1000 | [diff] [blame] | 2366 | dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2367 | /* do nothing */ |
| 2368 | } else { |
| 2369 | xfs_buf_relse(bp); |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 2370 | trace_xfs_log_recover_inode_skip(log, in_f); |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2371 | error = 0; |
| 2372 | goto error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2373 | } |
| 2374 | } |
| 2375 | /* Take the opportunity to reset the flush iteration count */ |
| 2376 | dicp->di_flushiter = 0; |
| 2377 | |
Al Viro | abbede1 | 2011-07-26 02:31:30 -0400 | [diff] [blame] | 2378 | if (unlikely(S_ISREG(dicp->di_mode))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2379 | if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) && |
| 2380 | (dicp->di_format != XFS_DINODE_FMT_BTREE)) { |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2381 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2382 | XFS_ERRLEVEL_LOW, mp, dicp); |
| 2383 | xfs_buf_relse(bp); |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2384 | xfs_alert(mp, |
| 2385 | "%s: Bad regular inode log record, rec ptr 0x%p, " |
| 2386 | "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld", |
| 2387 | __func__, item, dip, bp, in_f->ilf_ino); |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2388 | error = EFSCORRUPTED; |
| 2389 | goto error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2390 | } |
Al Viro | abbede1 | 2011-07-26 02:31:30 -0400 | [diff] [blame] | 2391 | } else if (unlikely(S_ISDIR(dicp->di_mode))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2392 | if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) && |
| 2393 | (dicp->di_format != XFS_DINODE_FMT_BTREE) && |
| 2394 | (dicp->di_format != XFS_DINODE_FMT_LOCAL)) { |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2395 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2396 | XFS_ERRLEVEL_LOW, mp, dicp); |
| 2397 | xfs_buf_relse(bp); |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2398 | xfs_alert(mp, |
| 2399 | "%s: Bad dir inode log record, rec ptr 0x%p, " |
| 2400 | "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld", |
| 2401 | __func__, item, dip, bp, in_f->ilf_ino); |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2402 | error = EFSCORRUPTED; |
| 2403 | goto error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2404 | } |
| 2405 | } |
| 2406 | if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){ |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2407 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2408 | XFS_ERRLEVEL_LOW, mp, dicp); |
| 2409 | xfs_buf_relse(bp); |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2410 | xfs_alert(mp, |
| 2411 | "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, " |
| 2412 | "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld", |
| 2413 | __func__, item, dip, bp, in_f->ilf_ino, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2414 | dicp->di_nextents + dicp->di_anextents, |
| 2415 | dicp->di_nblocks); |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2416 | error = EFSCORRUPTED; |
| 2417 | goto error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2418 | } |
| 2419 | if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) { |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2420 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2421 | XFS_ERRLEVEL_LOW, mp, dicp); |
| 2422 | xfs_buf_relse(bp); |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2423 | xfs_alert(mp, |
| 2424 | "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, " |
| 2425 | "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__, |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2426 | item, dip, bp, in_f->ilf_ino, dicp->di_forkoff); |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2427 | error = EFSCORRUPTED; |
| 2428 | goto error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2429 | } |
Christoph Hellwig | 93848a9 | 2013-04-03 16:11:17 +1100 | [diff] [blame] | 2430 | isize = xfs_icdinode_size(dicp->di_version); |
| 2431 | if (unlikely(item->ri_buf[1].i_len > isize)) { |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2432 | XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2433 | XFS_ERRLEVEL_LOW, mp, dicp); |
| 2434 | xfs_buf_relse(bp); |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2435 | xfs_alert(mp, |
| 2436 | "%s: Bad inode log record length %d, rec ptr 0x%p", |
| 2437 | __func__, item->ri_buf[1].i_len, item); |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2438 | error = EFSCORRUPTED; |
| 2439 | goto error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2440 | } |
| 2441 | |
| 2442 | /* The core is in in-core format */ |
Christoph Hellwig | 93848a9 | 2013-04-03 16:11:17 +1100 | [diff] [blame] | 2443 | xfs_dinode_to_disk(dip, dicp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2444 | |
| 2445 | /* the rest is in on-disk format */ |
Christoph Hellwig | 93848a9 | 2013-04-03 16:11:17 +1100 | [diff] [blame] | 2446 | if (item->ri_buf[1].i_len > isize) { |
| 2447 | memcpy((char *)dip + isize, |
| 2448 | item->ri_buf[1].i_addr + isize, |
| 2449 | item->ri_buf[1].i_len - isize); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2450 | } |
| 2451 | |
| 2452 | fields = in_f->ilf_fields; |
| 2453 | switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) { |
| 2454 | case XFS_ILOG_DEV: |
Christoph Hellwig | 81591fe | 2008-11-28 14:23:39 +1100 | [diff] [blame] | 2455 | xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2456 | break; |
| 2457 | case XFS_ILOG_UUID: |
Christoph Hellwig | 81591fe | 2008-11-28 14:23:39 +1100 | [diff] [blame] | 2458 | memcpy(XFS_DFORK_DPTR(dip), |
| 2459 | &in_f->ilf_u.ilfu_uuid, |
| 2460 | sizeof(uuid_t)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2461 | break; |
| 2462 | } |
| 2463 | |
| 2464 | if (in_f->ilf_size == 2) |
| 2465 | goto write_inode_buffer; |
| 2466 | len = item->ri_buf[2].i_len; |
| 2467 | src = item->ri_buf[2].i_addr; |
| 2468 | ASSERT(in_f->ilf_size <= 4); |
| 2469 | ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK)); |
| 2470 | ASSERT(!(fields & XFS_ILOG_DFORK) || |
| 2471 | (len == in_f->ilf_dsize)); |
| 2472 | |
| 2473 | switch (fields & XFS_ILOG_DFORK) { |
| 2474 | case XFS_ILOG_DDATA: |
| 2475 | case XFS_ILOG_DEXT: |
Christoph Hellwig | 81591fe | 2008-11-28 14:23:39 +1100 | [diff] [blame] | 2476 | memcpy(XFS_DFORK_DPTR(dip), src, len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2477 | break; |
| 2478 | |
| 2479 | case XFS_ILOG_DBROOT: |
Christoph Hellwig | 7cc95a8 | 2008-10-30 17:14:34 +1100 | [diff] [blame] | 2480 | xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len, |
Christoph Hellwig | 81591fe | 2008-11-28 14:23:39 +1100 | [diff] [blame] | 2481 | (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2482 | XFS_DFORK_DSIZE(dip, mp)); |
| 2483 | break; |
| 2484 | |
| 2485 | default: |
| 2486 | /* |
| 2487 | * There are no data fork flags set. |
| 2488 | */ |
| 2489 | ASSERT((fields & XFS_ILOG_DFORK) == 0); |
| 2490 | break; |
| 2491 | } |
| 2492 | |
| 2493 | /* |
| 2494 | * If we logged any attribute data, recover it. There may or |
| 2495 | * may not have been any other non-core data logged in this |
| 2496 | * transaction. |
| 2497 | */ |
| 2498 | if (in_f->ilf_fields & XFS_ILOG_AFORK) { |
| 2499 | if (in_f->ilf_fields & XFS_ILOG_DFORK) { |
| 2500 | attr_index = 3; |
| 2501 | } else { |
| 2502 | attr_index = 2; |
| 2503 | } |
| 2504 | len = item->ri_buf[attr_index].i_len; |
| 2505 | src = item->ri_buf[attr_index].i_addr; |
| 2506 | ASSERT(len == in_f->ilf_asize); |
| 2507 | |
| 2508 | switch (in_f->ilf_fields & XFS_ILOG_AFORK) { |
| 2509 | case XFS_ILOG_ADATA: |
| 2510 | case XFS_ILOG_AEXT: |
| 2511 | dest = XFS_DFORK_APTR(dip); |
| 2512 | ASSERT(len <= XFS_DFORK_ASIZE(dip, mp)); |
| 2513 | memcpy(dest, src, len); |
| 2514 | break; |
| 2515 | |
| 2516 | case XFS_ILOG_ABROOT: |
| 2517 | dest = XFS_DFORK_APTR(dip); |
Christoph Hellwig | 7cc95a8 | 2008-10-30 17:14:34 +1100 | [diff] [blame] | 2518 | xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, |
| 2519 | len, (xfs_bmdr_block_t*)dest, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2520 | XFS_DFORK_ASIZE(dip, mp)); |
| 2521 | break; |
| 2522 | |
| 2523 | default: |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2524 | xfs_warn(log->l_mp, "%s: Invalid flag", __func__); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2525 | ASSERT(0); |
| 2526 | xfs_buf_relse(bp); |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2527 | error = EIO; |
| 2528 | goto error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2529 | } |
| 2530 | } |
| 2531 | |
| 2532 | write_inode_buffer: |
Christoph Hellwig | 93848a9 | 2013-04-03 16:11:17 +1100 | [diff] [blame] | 2533 | /* re-generate the checksum. */ |
| 2534 | xfs_dinode_calc_crc(log->l_mp, dip); |
| 2535 | |
Dave Chinner | ebad861 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 2536 | ASSERT(bp->b_target->bt_mount == mp); |
Christoph Hellwig | cb669ca | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 2537 | bp->b_iodone = xlog_recover_iodone; |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 2538 | xfs_buf_delwri_queue(bp, buffer_list); |
Christoph Hellwig | 61551f1 | 2011-08-23 08:28:06 +0000 | [diff] [blame] | 2539 | xfs_buf_relse(bp); |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2540 | error: |
| 2541 | if (need_free) |
Denys Vlasenko | f0e2d93 | 2008-05-19 16:31:57 +1000 | [diff] [blame] | 2542 | kmem_free(in_f); |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2543 | return XFS_ERROR(error); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2544 | } |
| 2545 | |
| 2546 | /* |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 2547 | * 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] | 2548 | * structure, so that we know not to do any dquot item or dquot buffer recovery, |
| 2549 | * of that type. |
| 2550 | */ |
| 2551 | STATIC int |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2552 | xlog_recover_quotaoff_pass1( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 2553 | struct xlog *log, |
| 2554 | struct xlog_recover_item *item) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2555 | { |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2556 | xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2557 | ASSERT(qoff_f); |
| 2558 | |
| 2559 | /* |
| 2560 | * The logitem format's flag tells us if this was user quotaoff, |
Nathan Scott | 77a7cce | 2006-01-11 15:35:57 +1100 | [diff] [blame] | 2561 | * group/project quotaoff or both. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2562 | */ |
| 2563 | if (qoff_f->qf_flags & XFS_UQUOTA_ACCT) |
| 2564 | log->l_quotaoffs_flag |= XFS_DQ_USER; |
Nathan Scott | 77a7cce | 2006-01-11 15:35:57 +1100 | [diff] [blame] | 2565 | if (qoff_f->qf_flags & XFS_PQUOTA_ACCT) |
| 2566 | log->l_quotaoffs_flag |= XFS_DQ_PROJ; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2567 | if (qoff_f->qf_flags & XFS_GQUOTA_ACCT) |
| 2568 | log->l_quotaoffs_flag |= XFS_DQ_GROUP; |
| 2569 | |
| 2570 | return (0); |
| 2571 | } |
| 2572 | |
| 2573 | /* |
| 2574 | * Recover a dquot record |
| 2575 | */ |
| 2576 | STATIC int |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2577 | xlog_recover_dquot_pass2( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 2578 | struct xlog *log, |
| 2579 | struct list_head *buffer_list, |
| 2580 | struct xlog_recover_item *item) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2581 | { |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2582 | xfs_mount_t *mp = log->l_mp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2583 | xfs_buf_t *bp; |
| 2584 | struct xfs_disk_dquot *ddq, *recddq; |
| 2585 | int error; |
| 2586 | xfs_dq_logformat_t *dq_f; |
| 2587 | uint type; |
| 2588 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2589 | |
| 2590 | /* |
| 2591 | * Filesystems are required to send in quota flags at mount time. |
| 2592 | */ |
| 2593 | if (mp->m_qflags == 0) |
| 2594 | return (0); |
| 2595 | |
Christoph Hellwig | 4e0d5f9 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 2596 | recddq = item->ri_buf[1].i_addr; |
| 2597 | if (recddq == NULL) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2598 | xfs_alert(log->l_mp, "NULL dquot in %s.", __func__); |
Christoph Hellwig | 0c5e1ce | 2009-06-08 15:33:21 +0200 | [diff] [blame] | 2599 | return XFS_ERROR(EIO); |
| 2600 | } |
Jan Rekorajski | 8ec6dba | 2009-11-16 11:57:02 +0000 | [diff] [blame] | 2601 | if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2602 | xfs_alert(log->l_mp, "dquot too small (%d) in %s.", |
Christoph Hellwig | 0c5e1ce | 2009-06-08 15:33:21 +0200 | [diff] [blame] | 2603 | item->ri_buf[1].i_len, __func__); |
| 2604 | return XFS_ERROR(EIO); |
| 2605 | } |
| 2606 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2607 | /* |
| 2608 | * This type of quotas was turned off, so ignore this record. |
| 2609 | */ |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 2610 | 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] | 2611 | ASSERT(type); |
| 2612 | if (log->l_quotaoffs_flag & type) |
| 2613 | return (0); |
| 2614 | |
| 2615 | /* |
| 2616 | * At this point we know that quota was _not_ turned off. |
| 2617 | * Since the mount flags are not indicating to us otherwise, this |
| 2618 | * must mean that quota is on, and the dquot needs to be replayed. |
| 2619 | * Remember that we may not have fully recovered the superblock yet, |
| 2620 | * so we can't do the usual trick of looking at the SB quota bits. |
| 2621 | * |
| 2622 | * The other possibility, of course, is that the quota subsystem was |
| 2623 | * removed since the last mount - ENOSYS. |
| 2624 | */ |
Christoph Hellwig | 4e0d5f9 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 2625 | dq_f = item->ri_buf[0].i_addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2626 | ASSERT(dq_f); |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2627 | error = xfs_qm_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN, |
| 2628 | "xlog_recover_dquot_pass2 (log copy)"); |
| 2629 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2630 | return XFS_ERROR(EIO); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2631 | ASSERT(dq_f->qlf_len == 1); |
| 2632 | |
Dave Chinner | 7ca790a | 2012-04-23 15:58:55 +1000 | [diff] [blame] | 2633 | 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] | 2634 | XFS_FSB_TO_BB(mp, dq_f->qlf_len), 0, &bp, |
| 2635 | NULL); |
Dave Chinner | 7ca790a | 2012-04-23 15:58:55 +1000 | [diff] [blame] | 2636 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2637 | return error; |
Dave Chinner | 7ca790a | 2012-04-23 15:58:55 +1000 | [diff] [blame] | 2638 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2639 | ASSERT(bp); |
| 2640 | ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset); |
| 2641 | |
| 2642 | /* |
| 2643 | * At least the magic num portion should be on disk because this |
| 2644 | * was among a chunk of dquots created earlier, and we did some |
| 2645 | * minimal initialization then. |
| 2646 | */ |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2647 | error = xfs_qm_dqcheck(mp, ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN, |
| 2648 | "xlog_recover_dquot_pass2"); |
| 2649 | if (error) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2650 | xfs_buf_relse(bp); |
| 2651 | return XFS_ERROR(EIO); |
| 2652 | } |
| 2653 | |
| 2654 | memcpy(ddq, recddq, item->ri_buf[1].i_len); |
| 2655 | |
| 2656 | ASSERT(dq_f->qlf_size == 2); |
Dave Chinner | ebad861 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 2657 | ASSERT(bp->b_target->bt_mount == mp); |
Christoph Hellwig | cb669ca | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 2658 | bp->b_iodone = xlog_recover_iodone; |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 2659 | xfs_buf_delwri_queue(bp, buffer_list); |
Christoph Hellwig | 61551f1 | 2011-08-23 08:28:06 +0000 | [diff] [blame] | 2660 | xfs_buf_relse(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2661 | |
| 2662 | return (0); |
| 2663 | } |
| 2664 | |
| 2665 | /* |
| 2666 | * This routine is called to create an in-core extent free intent |
| 2667 | * item from the efi format structure which was logged on disk. |
| 2668 | * It allocates an in-core efi, copies the extents from the format |
| 2669 | * structure into it, and adds the efi to the AIL with the given |
| 2670 | * LSN. |
| 2671 | */ |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2672 | STATIC int |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2673 | xlog_recover_efi_pass2( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 2674 | struct xlog *log, |
| 2675 | struct xlog_recover_item *item, |
| 2676 | xfs_lsn_t lsn) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2677 | { |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2678 | int error; |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2679 | xfs_mount_t *mp = log->l_mp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2680 | xfs_efi_log_item_t *efip; |
| 2681 | xfs_efi_log_format_t *efi_formatp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2682 | |
Christoph Hellwig | 4e0d5f9 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 2683 | efi_formatp = item->ri_buf[0].i_addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2684 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2685 | efip = xfs_efi_init(mp, efi_formatp->efi_nextents); |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2686 | if ((error = xfs_efi_copy_format(&(item->ri_buf[0]), |
| 2687 | &(efip->efi_format)))) { |
| 2688 | xfs_efi_item_free(efip); |
| 2689 | return error; |
| 2690 | } |
Dave Chinner | b199c8a | 2010-12-20 11:59:49 +1100 | [diff] [blame] | 2691 | atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2692 | |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 2693 | spin_lock(&log->l_ailp->xa_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2694 | /* |
David Chinner | 783a2f6 | 2008-10-30 17:39:58 +1100 | [diff] [blame] | 2695 | * xfs_trans_ail_update() drops the AIL lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2696 | */ |
Dave Chinner | e605994 | 2010-12-20 12:34:26 +1100 | [diff] [blame] | 2697 | xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn); |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2698 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2699 | } |
| 2700 | |
| 2701 | |
| 2702 | /* |
| 2703 | * This routine is called when an efd format structure is found in |
| 2704 | * a committed transaction in the log. It's purpose is to cancel |
| 2705 | * the corresponding efi if it was still in the log. To do this |
| 2706 | * it searches the AIL for the efi with an id equal to that in the |
| 2707 | * efd format structure. If we find it, we remove the efi from the |
| 2708 | * AIL and free it. |
| 2709 | */ |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2710 | STATIC int |
| 2711 | xlog_recover_efd_pass2( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 2712 | struct xlog *log, |
| 2713 | struct xlog_recover_item *item) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2714 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2715 | xfs_efd_log_format_t *efd_formatp; |
| 2716 | xfs_efi_log_item_t *efip = NULL; |
| 2717 | xfs_log_item_t *lip; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2718 | __uint64_t efi_id; |
David Chinner | 27d8d5f | 2008-10-30 17:38:39 +1100 | [diff] [blame] | 2719 | struct xfs_ail_cursor cur; |
David Chinner | 783a2f6 | 2008-10-30 17:39:58 +1100 | [diff] [blame] | 2720 | struct xfs_ail *ailp = log->l_ailp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2721 | |
Christoph Hellwig | 4e0d5f9 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 2722 | efd_formatp = item->ri_buf[0].i_addr; |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 2723 | ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) + |
| 2724 | ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) || |
| 2725 | (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) + |
| 2726 | ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t))))); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2727 | efi_id = efd_formatp->efd_efi_id; |
| 2728 | |
| 2729 | /* |
| 2730 | * Search for the efi with the id in the efd format structure |
| 2731 | * in the AIL. |
| 2732 | */ |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 2733 | spin_lock(&ailp->xa_lock); |
| 2734 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2735 | while (lip != NULL) { |
| 2736 | if (lip->li_type == XFS_LI_EFI) { |
| 2737 | efip = (xfs_efi_log_item_t *)lip; |
| 2738 | if (efip->efi_format.efi_id == efi_id) { |
| 2739 | /* |
David Chinner | 783a2f6 | 2008-10-30 17:39:58 +1100 | [diff] [blame] | 2740 | * xfs_trans_ail_delete() drops the |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2741 | * AIL lock. |
| 2742 | */ |
Dave Chinner | 04913fd | 2012-04-23 15:58:41 +1000 | [diff] [blame] | 2743 | xfs_trans_ail_delete(ailp, lip, |
| 2744 | SHUTDOWN_CORRUPT_INCORE); |
David Chinner | 8ae2c0f | 2007-11-23 16:28:17 +1100 | [diff] [blame] | 2745 | xfs_efi_item_free(efip); |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 2746 | spin_lock(&ailp->xa_lock); |
David Chinner | 27d8d5f | 2008-10-30 17:38:39 +1100 | [diff] [blame] | 2747 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2748 | } |
| 2749 | } |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 2750 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2751 | } |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 2752 | xfs_trans_ail_cursor_done(ailp, &cur); |
| 2753 | spin_unlock(&ailp->xa_lock); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2754 | |
| 2755 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2756 | } |
| 2757 | |
| 2758 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2759 | * Free up any resources allocated by the transaction |
| 2760 | * |
| 2761 | * Remember that EFIs, EFDs, and IUNLINKs are handled later. |
| 2762 | */ |
| 2763 | STATIC void |
| 2764 | xlog_recover_free_trans( |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 2765 | struct xlog_recover *trans) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2766 | { |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 2767 | xlog_recover_item_t *item, *n; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2768 | int i; |
| 2769 | |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 2770 | list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) { |
| 2771 | /* Free the regions in the item. */ |
| 2772 | list_del(&item->ri_list); |
| 2773 | for (i = 0; i < item->ri_cnt; i++) |
| 2774 | kmem_free(item->ri_buf[i].i_addr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2775 | /* Free the item itself */ |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 2776 | kmem_free(item->ri_buf); |
| 2777 | kmem_free(item); |
| 2778 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2779 | /* Free the transaction recover structure */ |
Denys Vlasenko | f0e2d93 | 2008-05-19 16:31:57 +1000 | [diff] [blame] | 2780 | kmem_free(trans); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2781 | } |
| 2782 | |
| 2783 | STATIC int |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2784 | xlog_recover_commit_pass1( |
Mark Tinguely | ad223e6 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 2785 | struct xlog *log, |
| 2786 | struct xlog_recover *trans, |
| 2787 | struct xlog_recover_item *item) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2788 | { |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2789 | trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1); |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 2790 | |
| 2791 | switch (ITEM_TYPE(item)) { |
| 2792 | case XFS_LI_BUF: |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2793 | return xlog_recover_buffer_pass1(log, item); |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 2794 | case XFS_LI_QUOTAOFF: |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2795 | return xlog_recover_quotaoff_pass1(log, item); |
| 2796 | case XFS_LI_INODE: |
| 2797 | case XFS_LI_EFI: |
| 2798 | case XFS_LI_EFD: |
| 2799 | case XFS_LI_DQUOT: |
| 2800 | /* nothing to do in pass 1 */ |
| 2801 | return 0; |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 2802 | default: |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2803 | xfs_warn(log->l_mp, "%s: invalid item type (%d)", |
| 2804 | __func__, ITEM_TYPE(item)); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2805 | ASSERT(0); |
| 2806 | return XFS_ERROR(EIO); |
| 2807 | } |
| 2808 | } |
| 2809 | |
| 2810 | STATIC int |
| 2811 | xlog_recover_commit_pass2( |
Mark Tinguely | ad223e6 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 2812 | struct xlog *log, |
| 2813 | struct xlog_recover *trans, |
| 2814 | struct list_head *buffer_list, |
| 2815 | struct xlog_recover_item *item) |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2816 | { |
| 2817 | trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2); |
| 2818 | |
| 2819 | switch (ITEM_TYPE(item)) { |
| 2820 | case XFS_LI_BUF: |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 2821 | return xlog_recover_buffer_pass2(log, buffer_list, item); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2822 | case XFS_LI_INODE: |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 2823 | return xlog_recover_inode_pass2(log, buffer_list, item); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2824 | case XFS_LI_EFI: |
| 2825 | return xlog_recover_efi_pass2(log, item, trans->r_lsn); |
| 2826 | case XFS_LI_EFD: |
| 2827 | return xlog_recover_efd_pass2(log, item); |
| 2828 | case XFS_LI_DQUOT: |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 2829 | return xlog_recover_dquot_pass2(log, buffer_list, item); |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2830 | case XFS_LI_QUOTAOFF: |
| 2831 | /* nothing to do in pass2 */ |
| 2832 | return 0; |
| 2833 | default: |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2834 | xfs_warn(log->l_mp, "%s: invalid item type (%d)", |
| 2835 | __func__, ITEM_TYPE(item)); |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 2836 | ASSERT(0); |
| 2837 | return XFS_ERROR(EIO); |
| 2838 | } |
| 2839 | } |
| 2840 | |
| 2841 | /* |
| 2842 | * Perform the transaction. |
| 2843 | * |
| 2844 | * If the transaction modifies a buffer or inode, do it now. Otherwise, |
| 2845 | * EFIs and EFDs get queued up by adding entries into the AIL for them. |
| 2846 | */ |
| 2847 | STATIC int |
| 2848 | xlog_recover_commit_trans( |
Mark Tinguely | ad223e6 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 2849 | struct xlog *log, |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 2850 | struct xlog_recover *trans, |
| 2851 | int pass) |
| 2852 | { |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 2853 | int error = 0, error2; |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 2854 | xlog_recover_item_t *item; |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 2855 | LIST_HEAD (buffer_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2856 | |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 2857 | hlist_del(&trans->r_list); |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 2858 | |
| 2859 | error = xlog_recover_reorder_trans(log, trans, pass); |
| 2860 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2861 | return error; |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 2862 | |
| 2863 | list_for_each_entry(item, &trans->r_itemq, ri_list) { |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 2864 | switch (pass) { |
| 2865 | case XLOG_RECOVER_PASS1: |
Christoph Hellwig | c9f71f5 | 2010-12-01 22:06:24 +0000 | [diff] [blame] | 2866 | error = xlog_recover_commit_pass1(log, trans, item); |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 2867 | break; |
| 2868 | case XLOG_RECOVER_PASS2: |
| 2869 | error = xlog_recover_commit_pass2(log, trans, |
| 2870 | &buffer_list, item); |
| 2871 | break; |
| 2872 | default: |
| 2873 | ASSERT(0); |
| 2874 | } |
| 2875 | |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 2876 | if (error) |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 2877 | goto out; |
Christoph Hellwig | d045094 | 2010-12-01 22:06:23 +0000 | [diff] [blame] | 2878 | } |
| 2879 | |
| 2880 | xlog_recover_free_trans(trans); |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 2881 | |
| 2882 | out: |
| 2883 | error2 = xfs_buf_delwri_submit(&buffer_list); |
| 2884 | return error ? error : error2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2885 | } |
| 2886 | |
| 2887 | STATIC int |
| 2888 | xlog_recover_unmount_trans( |
Mark Tinguely | ad223e6 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 2889 | struct xlog *log, |
| 2890 | struct xlog_recover *trans) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2891 | { |
| 2892 | /* Do nothing now */ |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2893 | xfs_warn(log->l_mp, "%s: Unmount LR", __func__); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2894 | return 0; |
| 2895 | } |
| 2896 | |
| 2897 | /* |
| 2898 | * There are two valid states of the r_state field. 0 indicates that the |
| 2899 | * transaction structure is in a normal state. We have either seen the |
| 2900 | * start of the transaction or the last operation we added was not a partial |
| 2901 | * operation. If the last operation we added to the transaction was a |
| 2902 | * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS. |
| 2903 | * |
| 2904 | * NOTE: skip LRs with 0 data length. |
| 2905 | */ |
| 2906 | STATIC int |
| 2907 | xlog_recover_process_data( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 2908 | struct xlog *log, |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 2909 | struct hlist_head rhash[], |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 2910 | struct xlog_rec_header *rhead, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2911 | xfs_caddr_t dp, |
| 2912 | int pass) |
| 2913 | { |
| 2914 | xfs_caddr_t lp; |
| 2915 | int num_logops; |
| 2916 | xlog_op_header_t *ohead; |
| 2917 | xlog_recover_t *trans; |
| 2918 | xlog_tid_t tid; |
| 2919 | int error; |
| 2920 | unsigned long hash; |
| 2921 | uint flags; |
| 2922 | |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 2923 | lp = dp + be32_to_cpu(rhead->h_len); |
| 2924 | num_logops = be32_to_cpu(rhead->h_num_logops); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2925 | |
| 2926 | /* check the log format matches our own - else we can't recover */ |
| 2927 | if (xlog_header_check_recover(log->l_mp, rhead)) |
| 2928 | return (XFS_ERROR(EIO)); |
| 2929 | |
| 2930 | while ((dp < lp) && num_logops) { |
| 2931 | ASSERT(dp + sizeof(xlog_op_header_t) <= lp); |
| 2932 | ohead = (xlog_op_header_t *)dp; |
| 2933 | dp += sizeof(xlog_op_header_t); |
| 2934 | if (ohead->oh_clientid != XFS_TRANSACTION && |
| 2935 | ohead->oh_clientid != XFS_LOG) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2936 | xfs_warn(log->l_mp, "%s: bad clientid 0x%x", |
| 2937 | __func__, ohead->oh_clientid); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2938 | ASSERT(0); |
| 2939 | return (XFS_ERROR(EIO)); |
| 2940 | } |
Christoph Hellwig | 67fcb7b | 2007-10-12 10:58:59 +1000 | [diff] [blame] | 2941 | tid = be32_to_cpu(ohead->oh_tid); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2942 | hash = XLOG_RHASH(tid); |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 2943 | trans = xlog_recover_find_tid(&rhash[hash], tid); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2944 | if (trans == NULL) { /* not found; add new tid */ |
| 2945 | if (ohead->oh_flags & XLOG_START_TRANS) |
| 2946 | xlog_recover_new_tid(&rhash[hash], tid, |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 2947 | be64_to_cpu(rhead->h_lsn)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2948 | } else { |
Lachlan McIlroy | 9742bb9 | 2008-01-10 16:43:36 +1100 | [diff] [blame] | 2949 | if (dp + be32_to_cpu(ohead->oh_len) > lp) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2950 | xfs_warn(log->l_mp, "%s: bad length 0x%x", |
| 2951 | __func__, be32_to_cpu(ohead->oh_len)); |
Lachlan McIlroy | 9742bb9 | 2008-01-10 16:43:36 +1100 | [diff] [blame] | 2952 | WARN_ON(1); |
| 2953 | return (XFS_ERROR(EIO)); |
| 2954 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2955 | flags = ohead->oh_flags & ~XLOG_END_TRANS; |
| 2956 | if (flags & XLOG_WAS_CONT_TRANS) |
| 2957 | flags &= ~XLOG_CONTINUE_TRANS; |
| 2958 | switch (flags) { |
| 2959 | case XLOG_COMMIT_TRANS: |
| 2960 | error = xlog_recover_commit_trans(log, |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 2961 | trans, pass); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2962 | break; |
| 2963 | case XLOG_UNMOUNT_TRANS: |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2964 | error = xlog_recover_unmount_trans(log, trans); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2965 | break; |
| 2966 | case XLOG_WAS_CONT_TRANS: |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 2967 | error = xlog_recover_add_to_cont_trans(log, |
| 2968 | trans, dp, |
| 2969 | be32_to_cpu(ohead->oh_len)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2970 | break; |
| 2971 | case XLOG_START_TRANS: |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2972 | xfs_warn(log->l_mp, "%s: bad transaction", |
| 2973 | __func__); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2974 | ASSERT(0); |
| 2975 | error = XFS_ERROR(EIO); |
| 2976 | break; |
| 2977 | case 0: |
| 2978 | case XLOG_CONTINUE_TRANS: |
Dave Chinner | 9abbc53 | 2010-04-13 15:06:46 +1000 | [diff] [blame] | 2979 | error = xlog_recover_add_to_trans(log, trans, |
Christoph Hellwig | 67fcb7b | 2007-10-12 10:58:59 +1000 | [diff] [blame] | 2980 | dp, be32_to_cpu(ohead->oh_len)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2981 | break; |
| 2982 | default: |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 2983 | xfs_warn(log->l_mp, "%s: bad flag 0x%x", |
| 2984 | __func__, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2985 | ASSERT(0); |
| 2986 | error = XFS_ERROR(EIO); |
| 2987 | break; |
| 2988 | } |
| 2989 | if (error) |
| 2990 | return error; |
| 2991 | } |
Christoph Hellwig | 67fcb7b | 2007-10-12 10:58:59 +1000 | [diff] [blame] | 2992 | dp += be32_to_cpu(ohead->oh_len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2993 | num_logops--; |
| 2994 | } |
| 2995 | return 0; |
| 2996 | } |
| 2997 | |
| 2998 | /* |
| 2999 | * Process an extent free intent item that was recovered from |
| 3000 | * the log. We need to free the extents that it describes. |
| 3001 | */ |
David Chinner | 3c1e2bb | 2008-04-10 12:21:11 +1000 | [diff] [blame] | 3002 | STATIC int |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3003 | xlog_recover_process_efi( |
| 3004 | xfs_mount_t *mp, |
| 3005 | xfs_efi_log_item_t *efip) |
| 3006 | { |
| 3007 | xfs_efd_log_item_t *efdp; |
| 3008 | xfs_trans_t *tp; |
| 3009 | int i; |
David Chinner | 3c1e2bb | 2008-04-10 12:21:11 +1000 | [diff] [blame] | 3010 | int error = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3011 | xfs_extent_t *extp; |
| 3012 | xfs_fsblock_t startblock_fsb; |
| 3013 | |
Dave Chinner | b199c8a | 2010-12-20 11:59:49 +1100 | [diff] [blame] | 3014 | ASSERT(!test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3015 | |
| 3016 | /* |
| 3017 | * First check the validity of the extents described by the |
| 3018 | * EFI. If any are bad, then assume that all are bad and |
| 3019 | * just toss the EFI. |
| 3020 | */ |
| 3021 | for (i = 0; i < efip->efi_format.efi_nextents; i++) { |
| 3022 | extp = &(efip->efi_format.efi_extents[i]); |
| 3023 | startblock_fsb = XFS_BB_TO_FSB(mp, |
| 3024 | XFS_FSB_TO_DADDR(mp, extp->ext_start)); |
| 3025 | if ((startblock_fsb == 0) || |
| 3026 | (extp->ext_len == 0) || |
| 3027 | (startblock_fsb >= mp->m_sb.sb_dblocks) || |
| 3028 | (extp->ext_len >= mp->m_sb.sb_agblocks)) { |
| 3029 | /* |
| 3030 | * This will pull the EFI from the AIL and |
| 3031 | * free the memory associated with it. |
| 3032 | */ |
Dave Chinner | 666d644 | 2013-04-03 14:09:21 +1100 | [diff] [blame] | 3033 | set_bit(XFS_EFI_RECOVERED, &efip->efi_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3034 | xfs_efi_release(efip, efip->efi_format.efi_nextents); |
David Chinner | 3c1e2bb | 2008-04-10 12:21:11 +1000 | [diff] [blame] | 3035 | return XFS_ERROR(EIO); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3036 | } |
| 3037 | } |
| 3038 | |
| 3039 | tp = xfs_trans_alloc(mp, 0); |
David Chinner | 3c1e2bb | 2008-04-10 12:21:11 +1000 | [diff] [blame] | 3040 | error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 0, 0); |
David Chinner | fc6149d | 2008-04-10 12:21:53 +1000 | [diff] [blame] | 3041 | if (error) |
| 3042 | goto abort_error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3043 | efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents); |
| 3044 | |
| 3045 | for (i = 0; i < efip->efi_format.efi_nextents; i++) { |
| 3046 | extp = &(efip->efi_format.efi_extents[i]); |
David Chinner | fc6149d | 2008-04-10 12:21:53 +1000 | [diff] [blame] | 3047 | error = xfs_free_extent(tp, extp->ext_start, extp->ext_len); |
| 3048 | if (error) |
| 3049 | goto abort_error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3050 | xfs_trans_log_efd_extent(tp, efdp, extp->ext_start, |
| 3051 | extp->ext_len); |
| 3052 | } |
| 3053 | |
Dave Chinner | b199c8a | 2010-12-20 11:59:49 +1100 | [diff] [blame] | 3054 | set_bit(XFS_EFI_RECOVERED, &efip->efi_flags); |
David Chinner | e5720ee | 2008-04-10 12:21:18 +1000 | [diff] [blame] | 3055 | error = xfs_trans_commit(tp, 0); |
David Chinner | 3c1e2bb | 2008-04-10 12:21:11 +1000 | [diff] [blame] | 3056 | return error; |
David Chinner | fc6149d | 2008-04-10 12:21:53 +1000 | [diff] [blame] | 3057 | |
| 3058 | abort_error: |
| 3059 | xfs_trans_cancel(tp, XFS_TRANS_ABORT); |
| 3060 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3061 | } |
| 3062 | |
| 3063 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3064 | * When this is called, all of the EFIs which did not have |
| 3065 | * corresponding EFDs should be in the AIL. What we do now |
| 3066 | * is free the extents associated with each one. |
| 3067 | * |
| 3068 | * Since we process the EFIs in normal transactions, they |
| 3069 | * will be removed at some point after the commit. This prevents |
| 3070 | * us from just walking down the list processing each one. |
| 3071 | * We'll use a flag in the EFI to skip those that we've already |
| 3072 | * processed and use the AIL iteration mechanism's generation |
| 3073 | * count to try to speed this up at least a bit. |
| 3074 | * |
| 3075 | * When we start, we know that the EFIs are the only things in |
| 3076 | * the AIL. As we process them, however, other items are added |
| 3077 | * to the AIL. Since everything added to the AIL must come after |
| 3078 | * everything already in the AIL, we stop processing as soon as |
| 3079 | * we see something other than an EFI in the AIL. |
| 3080 | */ |
David Chinner | 3c1e2bb | 2008-04-10 12:21:11 +1000 | [diff] [blame] | 3081 | STATIC int |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3082 | xlog_recover_process_efis( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 3083 | struct xlog *log) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3084 | { |
| 3085 | xfs_log_item_t *lip; |
| 3086 | xfs_efi_log_item_t *efip; |
David Chinner | 3c1e2bb | 2008-04-10 12:21:11 +1000 | [diff] [blame] | 3087 | int error = 0; |
David Chinner | 27d8d5f | 2008-10-30 17:38:39 +1100 | [diff] [blame] | 3088 | struct xfs_ail_cursor cur; |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 3089 | struct xfs_ail *ailp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3090 | |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 3091 | ailp = log->l_ailp; |
| 3092 | spin_lock(&ailp->xa_lock); |
| 3093 | lip = xfs_trans_ail_cursor_first(ailp, &cur, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3094 | while (lip != NULL) { |
| 3095 | /* |
| 3096 | * We're done when we see something other than an EFI. |
David Chinner | 27d8d5f | 2008-10-30 17:38:39 +1100 | [diff] [blame] | 3097 | * There should be no EFIs left in the AIL now. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3098 | */ |
| 3099 | if (lip->li_type != XFS_LI_EFI) { |
David Chinner | 27d8d5f | 2008-10-30 17:38:39 +1100 | [diff] [blame] | 3100 | #ifdef DEBUG |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 3101 | for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur)) |
David Chinner | 27d8d5f | 2008-10-30 17:38:39 +1100 | [diff] [blame] | 3102 | ASSERT(lip->li_type != XFS_LI_EFI); |
| 3103 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3104 | break; |
| 3105 | } |
| 3106 | |
| 3107 | /* |
| 3108 | * Skip EFIs that we've already processed. |
| 3109 | */ |
| 3110 | efip = (xfs_efi_log_item_t *)lip; |
Dave Chinner | b199c8a | 2010-12-20 11:59:49 +1100 | [diff] [blame] | 3111 | if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) { |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 3112 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3113 | continue; |
| 3114 | } |
| 3115 | |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 3116 | spin_unlock(&ailp->xa_lock); |
| 3117 | error = xlog_recover_process_efi(log->l_mp, efip); |
| 3118 | spin_lock(&ailp->xa_lock); |
David Chinner | 27d8d5f | 2008-10-30 17:38:39 +1100 | [diff] [blame] | 3119 | if (error) |
| 3120 | goto out; |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 3121 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3122 | } |
David Chinner | 27d8d5f | 2008-10-30 17:38:39 +1100 | [diff] [blame] | 3123 | out: |
David Chinner | a9c21c1 | 2008-10-30 17:39:35 +1100 | [diff] [blame] | 3124 | xfs_trans_ail_cursor_done(ailp, &cur); |
| 3125 | spin_unlock(&ailp->xa_lock); |
David Chinner | 3c1e2bb | 2008-04-10 12:21:11 +1000 | [diff] [blame] | 3126 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3127 | } |
| 3128 | |
| 3129 | /* |
| 3130 | * This routine performs a transaction to null out a bad inode pointer |
| 3131 | * in an agi unlinked inode hash bucket. |
| 3132 | */ |
| 3133 | STATIC void |
| 3134 | xlog_recover_clear_agi_bucket( |
| 3135 | xfs_mount_t *mp, |
| 3136 | xfs_agnumber_t agno, |
| 3137 | int bucket) |
| 3138 | { |
| 3139 | xfs_trans_t *tp; |
| 3140 | xfs_agi_t *agi; |
| 3141 | xfs_buf_t *agibp; |
| 3142 | int offset; |
| 3143 | int error; |
| 3144 | |
| 3145 | tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET); |
Christoph Hellwig | 5e1be0f | 2008-11-28 14:23:37 +1100 | [diff] [blame] | 3146 | error = xfs_trans_reserve(tp, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp), |
| 3147 | 0, 0, 0); |
David Chinner | e5720ee | 2008-04-10 12:21:18 +1000 | [diff] [blame] | 3148 | if (error) |
| 3149 | goto out_abort; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3150 | |
Christoph Hellwig | 5e1be0f | 2008-11-28 14:23:37 +1100 | [diff] [blame] | 3151 | error = xfs_read_agi(mp, tp, agno, &agibp); |
| 3152 | if (error) |
David Chinner | e5720ee | 2008-04-10 12:21:18 +1000 | [diff] [blame] | 3153 | goto out_abort; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3154 | |
Christoph Hellwig | 5e1be0f | 2008-11-28 14:23:37 +1100 | [diff] [blame] | 3155 | agi = XFS_BUF_TO_AGI(agibp); |
Christoph Hellwig | 16259e7 | 2005-11-02 15:11:25 +1100 | [diff] [blame] | 3156 | agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3157 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
| 3158 | (sizeof(xfs_agino_t) * bucket); |
| 3159 | xfs_trans_log_buf(tp, agibp, offset, |
| 3160 | (offset + sizeof(xfs_agino_t) - 1)); |
| 3161 | |
David Chinner | e5720ee | 2008-04-10 12:21:18 +1000 | [diff] [blame] | 3162 | error = xfs_trans_commit(tp, 0); |
| 3163 | if (error) |
| 3164 | goto out_error; |
| 3165 | return; |
| 3166 | |
| 3167 | out_abort: |
| 3168 | xfs_trans_cancel(tp, XFS_TRANS_ABORT); |
| 3169 | out_error: |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 3170 | xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno); |
David Chinner | e5720ee | 2008-04-10 12:21:18 +1000 | [diff] [blame] | 3171 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3172 | } |
| 3173 | |
Christoph Hellwig | 23fac50 | 2008-11-28 14:23:40 +1100 | [diff] [blame] | 3174 | STATIC xfs_agino_t |
| 3175 | xlog_recover_process_one_iunlink( |
| 3176 | struct xfs_mount *mp, |
| 3177 | xfs_agnumber_t agno, |
| 3178 | xfs_agino_t agino, |
| 3179 | int bucket) |
| 3180 | { |
| 3181 | struct xfs_buf *ibp; |
| 3182 | struct xfs_dinode *dip; |
| 3183 | struct xfs_inode *ip; |
| 3184 | xfs_ino_t ino; |
| 3185 | int error; |
| 3186 | |
| 3187 | ino = XFS_AGINO_TO_INO(mp, agno, agino); |
Dave Chinner | 7b6259e | 2010-06-24 11:35:17 +1000 | [diff] [blame] | 3188 | error = xfs_iget(mp, NULL, ino, 0, 0, &ip); |
Christoph Hellwig | 23fac50 | 2008-11-28 14:23:40 +1100 | [diff] [blame] | 3189 | if (error) |
| 3190 | goto fail; |
| 3191 | |
| 3192 | /* |
| 3193 | * Get the on disk inode to find the next inode in the bucket. |
| 3194 | */ |
Christoph Hellwig | 475ee41 | 2012-07-03 12:21:22 -0400 | [diff] [blame] | 3195 | 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] | 3196 | if (error) |
Christoph Hellwig | 0e44667 | 2008-11-28 14:23:42 +1100 | [diff] [blame] | 3197 | goto fail_iput; |
Christoph Hellwig | 23fac50 | 2008-11-28 14:23:40 +1100 | [diff] [blame] | 3198 | |
Christoph Hellwig | 23fac50 | 2008-11-28 14:23:40 +1100 | [diff] [blame] | 3199 | ASSERT(ip->i_d.di_nlink == 0); |
Christoph Hellwig | 0e44667 | 2008-11-28 14:23:42 +1100 | [diff] [blame] | 3200 | ASSERT(ip->i_d.di_mode != 0); |
Christoph Hellwig | 23fac50 | 2008-11-28 14:23:40 +1100 | [diff] [blame] | 3201 | |
| 3202 | /* setup for the next pass */ |
| 3203 | agino = be32_to_cpu(dip->di_next_unlinked); |
| 3204 | xfs_buf_relse(ibp); |
| 3205 | |
| 3206 | /* |
| 3207 | * Prevent any DMAPI event from being sent when the reference on |
| 3208 | * the inode is dropped. |
| 3209 | */ |
| 3210 | ip->i_d.di_dmevmask = 0; |
| 3211 | |
Christoph Hellwig | 0e44667 | 2008-11-28 14:23:42 +1100 | [diff] [blame] | 3212 | IRELE(ip); |
Christoph Hellwig | 23fac50 | 2008-11-28 14:23:40 +1100 | [diff] [blame] | 3213 | return agino; |
| 3214 | |
Christoph Hellwig | 0e44667 | 2008-11-28 14:23:42 +1100 | [diff] [blame] | 3215 | fail_iput: |
| 3216 | IRELE(ip); |
Christoph Hellwig | 23fac50 | 2008-11-28 14:23:40 +1100 | [diff] [blame] | 3217 | fail: |
| 3218 | /* |
| 3219 | * We can't read in the inode this bucket points to, or this inode |
| 3220 | * is messed up. Just ditch this bucket of inodes. We will lose |
| 3221 | * some inodes and space, but at least we won't hang. |
| 3222 | * |
| 3223 | * Call xlog_recover_clear_agi_bucket() to perform a transaction to |
| 3224 | * clear the inode pointer in the bucket. |
| 3225 | */ |
| 3226 | xlog_recover_clear_agi_bucket(mp, agno, bucket); |
| 3227 | return NULLAGINO; |
| 3228 | } |
| 3229 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3230 | /* |
| 3231 | * xlog_iunlink_recover |
| 3232 | * |
| 3233 | * This is called during recovery to process any inodes which |
| 3234 | * we unlinked but not freed when the system crashed. These |
| 3235 | * inodes will be on the lists in the AGI blocks. What we do |
| 3236 | * here is scan all the AGIs and fully truncate and free any |
| 3237 | * inodes found on the lists. Each inode is removed from the |
| 3238 | * lists when it has been fully truncated and is freed. The |
| 3239 | * freeing of the inode and its removal from the list must be |
| 3240 | * atomic. |
| 3241 | */ |
Eric Sandeen | d96f8f8 | 2009-07-02 00:09:33 -0500 | [diff] [blame] | 3242 | STATIC void |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3243 | xlog_recover_process_iunlinks( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 3244 | struct xlog *log) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3245 | { |
| 3246 | xfs_mount_t *mp; |
| 3247 | xfs_agnumber_t agno; |
| 3248 | xfs_agi_t *agi; |
| 3249 | xfs_buf_t *agibp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3250 | xfs_agino_t agino; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3251 | int bucket; |
| 3252 | int error; |
| 3253 | uint mp_dmevmask; |
| 3254 | |
| 3255 | mp = log->l_mp; |
| 3256 | |
| 3257 | /* |
| 3258 | * Prevent any DMAPI event from being sent while in this function. |
| 3259 | */ |
| 3260 | mp_dmevmask = mp->m_dmevmask; |
| 3261 | mp->m_dmevmask = 0; |
| 3262 | |
| 3263 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { |
| 3264 | /* |
| 3265 | * Find the agi for this ag. |
| 3266 | */ |
Christoph Hellwig | 5e1be0f | 2008-11-28 14:23:37 +1100 | [diff] [blame] | 3267 | error = xfs_read_agi(mp, NULL, agno, &agibp); |
| 3268 | if (error) { |
| 3269 | /* |
| 3270 | * AGI is b0rked. Don't process it. |
| 3271 | * |
| 3272 | * We should probably mark the filesystem as corrupt |
| 3273 | * after we've recovered all the ag's we can.... |
| 3274 | */ |
| 3275 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3276 | } |
Jan Kara | d97d32e | 2012-03-15 09:34:02 +0000 | [diff] [blame] | 3277 | /* |
| 3278 | * Unlock the buffer so that it can be acquired in the normal |
| 3279 | * course of the transaction to truncate and free each inode. |
| 3280 | * Because we are not racing with anyone else here for the AGI |
| 3281 | * buffer, we don't even need to hold it locked to read the |
| 3282 | * initial unlinked bucket entries out of the buffer. We keep |
| 3283 | * buffer reference though, so that it stays pinned in memory |
| 3284 | * while we need the buffer. |
| 3285 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3286 | agi = XFS_BUF_TO_AGI(agibp); |
Jan Kara | d97d32e | 2012-03-15 09:34:02 +0000 | [diff] [blame] | 3287 | xfs_buf_unlock(agibp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3288 | |
| 3289 | for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) { |
Christoph Hellwig | 16259e7 | 2005-11-02 15:11:25 +1100 | [diff] [blame] | 3290 | agino = be32_to_cpu(agi->agi_unlinked[bucket]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3291 | while (agino != NULLAGINO) { |
Christoph Hellwig | 23fac50 | 2008-11-28 14:23:40 +1100 | [diff] [blame] | 3292 | agino = xlog_recover_process_one_iunlink(mp, |
| 3293 | agno, agino, bucket); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3294 | } |
| 3295 | } |
Jan Kara | d97d32e | 2012-03-15 09:34:02 +0000 | [diff] [blame] | 3296 | xfs_buf_rele(agibp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3297 | } |
| 3298 | |
| 3299 | mp->m_dmevmask = mp_dmevmask; |
| 3300 | } |
| 3301 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3302 | /* |
Christoph Hellwig | 0e446be | 2012-11-12 22:54:24 +1100 | [diff] [blame] | 3303 | * Upack the log buffer data and crc check it. If the check fails, issue a |
| 3304 | * warning if and only if the CRC in the header is non-zero. This makes the |
| 3305 | * check an advisory warning, and the zero CRC check will prevent failure |
| 3306 | * warnings from being emitted when upgrading the kernel from one that does not |
| 3307 | * add CRCs by default. |
| 3308 | * |
| 3309 | * When filesystems are CRC enabled, this CRC mismatch becomes a fatal log |
| 3310 | * corruption failure |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3311 | */ |
Christoph Hellwig | 0e446be | 2012-11-12 22:54:24 +1100 | [diff] [blame] | 3312 | STATIC int |
| 3313 | xlog_unpack_data_crc( |
| 3314 | struct xlog_rec_header *rhead, |
| 3315 | xfs_caddr_t dp, |
| 3316 | struct xlog *log) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3317 | { |
Dave Chinner | f9668a0 | 2012-11-28 13:01:03 +1100 | [diff] [blame] | 3318 | __le32 crc; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3319 | |
Christoph Hellwig | 0e446be | 2012-11-12 22:54:24 +1100 | [diff] [blame] | 3320 | crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len)); |
| 3321 | if (crc != rhead->h_crc) { |
| 3322 | if (rhead->h_crc || xfs_sb_version_hascrc(&log->l_mp->m_sb)) { |
| 3323 | xfs_alert(log->l_mp, |
| 3324 | "log record CRC mismatch: found 0x%x, expected 0x%x.\n", |
Dave Chinner | f9668a0 | 2012-11-28 13:01:03 +1100 | [diff] [blame] | 3325 | le32_to_cpu(rhead->h_crc), |
| 3326 | le32_to_cpu(crc)); |
Christoph Hellwig | 0e446be | 2012-11-12 22:54:24 +1100 | [diff] [blame] | 3327 | xfs_hex_dump(dp, 32); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3328 | } |
| 3329 | |
Christoph Hellwig | 0e446be | 2012-11-12 22:54:24 +1100 | [diff] [blame] | 3330 | /* |
| 3331 | * If we've detected a log record corruption, then we can't |
| 3332 | * recover past this point. Abort recovery if we are enforcing |
| 3333 | * CRC protection by punting an error back up the stack. |
| 3334 | */ |
| 3335 | if (xfs_sb_version_hascrc(&log->l_mp->m_sb)) |
| 3336 | return EFSCORRUPTED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3337 | } |
Christoph Hellwig | 0e446be | 2012-11-12 22:54:24 +1100 | [diff] [blame] | 3338 | |
| 3339 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3340 | } |
| 3341 | |
Christoph Hellwig | 0e446be | 2012-11-12 22:54:24 +1100 | [diff] [blame] | 3342 | STATIC int |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3343 | xlog_unpack_data( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 3344 | struct xlog_rec_header *rhead, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3345 | xfs_caddr_t dp, |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 3346 | struct xlog *log) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3347 | { |
| 3348 | int i, j, k; |
Christoph Hellwig | 0e446be | 2012-11-12 22:54:24 +1100 | [diff] [blame] | 3349 | int error; |
| 3350 | |
| 3351 | error = xlog_unpack_data_crc(rhead, dp, log); |
| 3352 | if (error) |
| 3353 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3354 | |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 3355 | for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) && |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3356 | i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) { |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 3357 | *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3358 | dp += BBSIZE; |
| 3359 | } |
| 3360 | |
Eric Sandeen | 6211870 | 2008-03-06 13:44:28 +1100 | [diff] [blame] | 3361 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
Christoph Hellwig | b28708d | 2008-11-28 14:23:38 +1100 | [diff] [blame] | 3362 | xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead; |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 3363 | for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3364 | j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); |
| 3365 | k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 3366 | *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3367 | dp += BBSIZE; |
| 3368 | } |
| 3369 | } |
Christoph Hellwig | 0e446be | 2012-11-12 22:54:24 +1100 | [diff] [blame] | 3370 | |
| 3371 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3372 | } |
| 3373 | |
| 3374 | STATIC int |
| 3375 | xlog_valid_rec_header( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 3376 | struct xlog *log, |
| 3377 | struct xlog_rec_header *rhead, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3378 | xfs_daddr_t blkno) |
| 3379 | { |
| 3380 | int hlen; |
| 3381 | |
Christoph Hellwig | 69ef921 | 2011-07-08 14:36:05 +0200 | [diff] [blame] | 3382 | if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3383 | XFS_ERROR_REPORT("xlog_valid_rec_header(1)", |
| 3384 | XFS_ERRLEVEL_LOW, log->l_mp); |
| 3385 | return XFS_ERROR(EFSCORRUPTED); |
| 3386 | } |
| 3387 | if (unlikely( |
| 3388 | (!rhead->h_version || |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 3389 | (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 3390 | xfs_warn(log->l_mp, "%s: unrecognised log version (%d).", |
Harvey Harrison | 34a622b | 2008-04-10 12:19:21 +1000 | [diff] [blame] | 3391 | __func__, be32_to_cpu(rhead->h_version)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3392 | return XFS_ERROR(EIO); |
| 3393 | } |
| 3394 | |
| 3395 | /* LR body must have data or it wouldn't have been written */ |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 3396 | hlen = be32_to_cpu(rhead->h_len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3397 | if (unlikely( hlen <= 0 || hlen > INT_MAX )) { |
| 3398 | XFS_ERROR_REPORT("xlog_valid_rec_header(2)", |
| 3399 | XFS_ERRLEVEL_LOW, log->l_mp); |
| 3400 | return XFS_ERROR(EFSCORRUPTED); |
| 3401 | } |
| 3402 | if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) { |
| 3403 | XFS_ERROR_REPORT("xlog_valid_rec_header(3)", |
| 3404 | XFS_ERRLEVEL_LOW, log->l_mp); |
| 3405 | return XFS_ERROR(EFSCORRUPTED); |
| 3406 | } |
| 3407 | return 0; |
| 3408 | } |
| 3409 | |
| 3410 | /* |
| 3411 | * Read the log from tail to head and process the log records found. |
| 3412 | * Handle the two cases where the tail and head are in the same cycle |
| 3413 | * and where the active portion of the log wraps around the end of |
| 3414 | * the physical log separately. The pass parameter is passed through |
| 3415 | * to the routines called to process the data and is not looked at |
| 3416 | * here. |
| 3417 | */ |
| 3418 | STATIC int |
| 3419 | xlog_do_recovery_pass( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 3420 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3421 | xfs_daddr_t head_blk, |
| 3422 | xfs_daddr_t tail_blk, |
| 3423 | int pass) |
| 3424 | { |
| 3425 | xlog_rec_header_t *rhead; |
| 3426 | xfs_daddr_t blk_no; |
Andy Poling | fc5bc4c | 2009-11-03 17:26:47 +0000 | [diff] [blame] | 3427 | xfs_caddr_t offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3428 | xfs_buf_t *hbp, *dbp; |
| 3429 | int error = 0, h_size; |
| 3430 | int bblks, split_bblks; |
| 3431 | int hblks, split_hblks, wrapped_hblks; |
Dave Chinner | f0a7695 | 2010-01-11 11:49:57 +0000 | [diff] [blame] | 3432 | struct hlist_head rhash[XLOG_RHASH_SIZE]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3433 | |
| 3434 | ASSERT(head_blk != tail_blk); |
| 3435 | |
| 3436 | /* |
| 3437 | * Read the header of the tail block and get the iclog buffer size from |
| 3438 | * h_size. Use this to tell how many sectors make up the log header. |
| 3439 | */ |
Eric Sandeen | 6211870 | 2008-03-06 13:44:28 +1100 | [diff] [blame] | 3440 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3441 | /* |
| 3442 | * When using variable length iclogs, read first sector of |
| 3443 | * iclog header and extract the header size from it. Get a |
| 3444 | * new hbp that is the correct size. |
| 3445 | */ |
| 3446 | hbp = xlog_get_bp(log, 1); |
| 3447 | if (!hbp) |
| 3448 | return ENOMEM; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 3449 | |
| 3450 | error = xlog_bread(log, tail_blk, 1, hbp, &offset); |
| 3451 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3452 | goto bread_err1; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 3453 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3454 | rhead = (xlog_rec_header_t *)offset; |
| 3455 | error = xlog_valid_rec_header(log, rhead, tail_blk); |
| 3456 | if (error) |
| 3457 | goto bread_err1; |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 3458 | h_size = be32_to_cpu(rhead->h_size); |
| 3459 | if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) && |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3460 | (h_size > XLOG_HEADER_CYCLE_SIZE)) { |
| 3461 | hblks = h_size / XLOG_HEADER_CYCLE_SIZE; |
| 3462 | if (h_size % XLOG_HEADER_CYCLE_SIZE) |
| 3463 | hblks++; |
| 3464 | xlog_put_bp(hbp); |
| 3465 | hbp = xlog_get_bp(log, hblks); |
| 3466 | } else { |
| 3467 | hblks = 1; |
| 3468 | } |
| 3469 | } else { |
Alex Elder | 69ce58f | 2010-04-20 17:09:59 +1000 | [diff] [blame] | 3470 | ASSERT(log->l_sectBBsize == 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3471 | hblks = 1; |
| 3472 | hbp = xlog_get_bp(log, 1); |
| 3473 | h_size = XLOG_BIG_RECORD_BSIZE; |
| 3474 | } |
| 3475 | |
| 3476 | if (!hbp) |
| 3477 | return ENOMEM; |
| 3478 | dbp = xlog_get_bp(log, BTOBB(h_size)); |
| 3479 | if (!dbp) { |
| 3480 | xlog_put_bp(hbp); |
| 3481 | return ENOMEM; |
| 3482 | } |
| 3483 | |
| 3484 | memset(rhash, 0, sizeof(rhash)); |
| 3485 | if (tail_blk <= head_blk) { |
| 3486 | for (blk_no = tail_blk; blk_no < head_blk; ) { |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 3487 | error = xlog_bread(log, blk_no, hblks, hbp, &offset); |
| 3488 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3489 | goto bread_err2; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 3490 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3491 | rhead = (xlog_rec_header_t *)offset; |
| 3492 | error = xlog_valid_rec_header(log, rhead, blk_no); |
| 3493 | if (error) |
| 3494 | goto bread_err2; |
| 3495 | |
| 3496 | /* blocks in data section */ |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 3497 | bblks = (int)BTOBB(be32_to_cpu(rhead->h_len)); |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 3498 | error = xlog_bread(log, blk_no + hblks, bblks, dbp, |
| 3499 | &offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3500 | if (error) |
| 3501 | goto bread_err2; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 3502 | |
Christoph Hellwig | 0e446be | 2012-11-12 22:54:24 +1100 | [diff] [blame] | 3503 | error = xlog_unpack_data(rhead, offset, log); |
| 3504 | if (error) |
| 3505 | goto bread_err2; |
| 3506 | |
| 3507 | error = xlog_recover_process_data(log, |
| 3508 | rhash, rhead, offset, pass); |
| 3509 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3510 | goto bread_err2; |
| 3511 | blk_no += bblks + hblks; |
| 3512 | } |
| 3513 | } else { |
| 3514 | /* |
| 3515 | * Perform recovery around the end of the physical log. |
| 3516 | * When the head is not on the same cycle number as the tail, |
| 3517 | * we can't do a sequential recovery as above. |
| 3518 | */ |
| 3519 | blk_no = tail_blk; |
| 3520 | while (blk_no < log->l_logBBsize) { |
| 3521 | /* |
| 3522 | * Check for header wrapping around physical end-of-log |
| 3523 | */ |
Chandra Seetharaman | 6292604 | 2011-07-22 23:40:15 +0000 | [diff] [blame] | 3524 | offset = hbp->b_addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3525 | split_hblks = 0; |
| 3526 | wrapped_hblks = 0; |
| 3527 | if (blk_no + hblks <= log->l_logBBsize) { |
| 3528 | /* Read header in one read */ |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 3529 | error = xlog_bread(log, blk_no, hblks, hbp, |
| 3530 | &offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3531 | if (error) |
| 3532 | goto bread_err2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3533 | } else { |
| 3534 | /* This LR is split across physical log end */ |
| 3535 | if (blk_no != log->l_logBBsize) { |
| 3536 | /* some data before physical log end */ |
| 3537 | ASSERT(blk_no <= INT_MAX); |
| 3538 | split_hblks = log->l_logBBsize - (int)blk_no; |
| 3539 | ASSERT(split_hblks > 0); |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 3540 | error = xlog_bread(log, blk_no, |
| 3541 | split_hblks, hbp, |
| 3542 | &offset); |
| 3543 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3544 | goto bread_err2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3545 | } |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 3546 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3547 | /* |
| 3548 | * Note: this black magic still works with |
| 3549 | * large sector sizes (non-512) only because: |
| 3550 | * - we increased the buffer size originally |
| 3551 | * by 1 sector giving us enough extra space |
| 3552 | * for the second read; |
| 3553 | * - the log start is guaranteed to be sector |
| 3554 | * aligned; |
| 3555 | * - we read the log end (LR header start) |
| 3556 | * _first_, then the log start (LR header end) |
| 3557 | * - order is important. |
| 3558 | */ |
David Chinner | 234f56a | 2008-04-10 12:24:24 +1000 | [diff] [blame] | 3559 | wrapped_hblks = hblks - split_hblks; |
Dave Chinner | 4439647 | 2011-04-21 09:34:27 +0000 | [diff] [blame] | 3560 | error = xlog_bread_offset(log, 0, |
| 3561 | wrapped_hblks, hbp, |
| 3562 | offset + BBTOB(split_hblks)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3563 | if (error) |
| 3564 | goto bread_err2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3565 | } |
| 3566 | rhead = (xlog_rec_header_t *)offset; |
| 3567 | error = xlog_valid_rec_header(log, rhead, |
| 3568 | split_hblks ? blk_no : 0); |
| 3569 | if (error) |
| 3570 | goto bread_err2; |
| 3571 | |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 3572 | bblks = (int)BTOBB(be32_to_cpu(rhead->h_len)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3573 | blk_no += hblks; |
| 3574 | |
| 3575 | /* Read in data for log record */ |
| 3576 | if (blk_no + bblks <= log->l_logBBsize) { |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 3577 | error = xlog_bread(log, blk_no, bblks, dbp, |
| 3578 | &offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3579 | if (error) |
| 3580 | goto bread_err2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3581 | } else { |
| 3582 | /* This log record is split across the |
| 3583 | * physical end of log */ |
Chandra Seetharaman | 6292604 | 2011-07-22 23:40:15 +0000 | [diff] [blame] | 3584 | offset = dbp->b_addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3585 | split_bblks = 0; |
| 3586 | if (blk_no != log->l_logBBsize) { |
| 3587 | /* some data is before the physical |
| 3588 | * end of log */ |
| 3589 | ASSERT(!wrapped_hblks); |
| 3590 | ASSERT(blk_no <= INT_MAX); |
| 3591 | split_bblks = |
| 3592 | log->l_logBBsize - (int)blk_no; |
| 3593 | ASSERT(split_bblks > 0); |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 3594 | error = xlog_bread(log, blk_no, |
| 3595 | split_bblks, dbp, |
| 3596 | &offset); |
| 3597 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3598 | goto bread_err2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3599 | } |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 3600 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3601 | /* |
| 3602 | * Note: this black magic still works with |
| 3603 | * large sector sizes (non-512) only because: |
| 3604 | * - we increased the buffer size originally |
| 3605 | * by 1 sector giving us enough extra space |
| 3606 | * for the second read; |
| 3607 | * - the log start is guaranteed to be sector |
| 3608 | * aligned; |
| 3609 | * - we read the log end (LR header start) |
| 3610 | * _first_, then the log start (LR header end) |
| 3611 | * - order is important. |
| 3612 | */ |
Dave Chinner | 4439647 | 2011-04-21 09:34:27 +0000 | [diff] [blame] | 3613 | error = xlog_bread_offset(log, 0, |
Dave Chinner | 009507b | 2012-11-02 11:38:44 +1100 | [diff] [blame] | 3614 | bblks - split_bblks, dbp, |
Dave Chinner | 4439647 | 2011-04-21 09:34:27 +0000 | [diff] [blame] | 3615 | offset + BBTOB(split_bblks)); |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 3616 | if (error) |
| 3617 | goto bread_err2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3618 | } |
Christoph Hellwig | 0e446be | 2012-11-12 22:54:24 +1100 | [diff] [blame] | 3619 | |
| 3620 | error = xlog_unpack_data(rhead, offset, log); |
| 3621 | if (error) |
| 3622 | goto bread_err2; |
| 3623 | |
| 3624 | error = xlog_recover_process_data(log, rhash, |
| 3625 | rhead, offset, pass); |
| 3626 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3627 | goto bread_err2; |
| 3628 | blk_no += bblks; |
| 3629 | } |
| 3630 | |
| 3631 | ASSERT(blk_no >= log->l_logBBsize); |
| 3632 | blk_no -= log->l_logBBsize; |
| 3633 | |
| 3634 | /* read first part of physical log */ |
| 3635 | while (blk_no < head_blk) { |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 3636 | error = xlog_bread(log, blk_no, hblks, hbp, &offset); |
| 3637 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3638 | goto bread_err2; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 3639 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3640 | rhead = (xlog_rec_header_t *)offset; |
| 3641 | error = xlog_valid_rec_header(log, rhead, blk_no); |
| 3642 | if (error) |
| 3643 | goto bread_err2; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 3644 | |
Christoph Hellwig | b53e675 | 2007-10-12 10:59:34 +1000 | [diff] [blame] | 3645 | bblks = (int)BTOBB(be32_to_cpu(rhead->h_len)); |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 3646 | error = xlog_bread(log, blk_no+hblks, bblks, dbp, |
| 3647 | &offset); |
| 3648 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3649 | goto bread_err2; |
Christoph Hellwig | 076e6ac | 2009-03-16 08:24:13 +0100 | [diff] [blame] | 3650 | |
Christoph Hellwig | 0e446be | 2012-11-12 22:54:24 +1100 | [diff] [blame] | 3651 | error = xlog_unpack_data(rhead, offset, log); |
| 3652 | if (error) |
| 3653 | goto bread_err2; |
| 3654 | |
| 3655 | error = xlog_recover_process_data(log, rhash, |
| 3656 | rhead, offset, pass); |
| 3657 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3658 | goto bread_err2; |
| 3659 | blk_no += bblks + hblks; |
| 3660 | } |
| 3661 | } |
| 3662 | |
| 3663 | bread_err2: |
| 3664 | xlog_put_bp(dbp); |
| 3665 | bread_err1: |
| 3666 | xlog_put_bp(hbp); |
| 3667 | return error; |
| 3668 | } |
| 3669 | |
| 3670 | /* |
| 3671 | * Do the recovery of the log. We actually do this in two phases. |
| 3672 | * The two passes are necessary in order to implement the function |
| 3673 | * of cancelling a record written into the log. The first pass |
| 3674 | * determines those things which have been cancelled, and the |
| 3675 | * second pass replays log items normally except for those which |
| 3676 | * have been cancelled. The handling of the replay and cancellations |
| 3677 | * takes place in the log item type specific routines. |
| 3678 | * |
| 3679 | * The table of items which have cancel records in the log is allocated |
| 3680 | * and freed at this level, since only here do we know when all of |
| 3681 | * the log recovery has been completed. |
| 3682 | */ |
| 3683 | STATIC int |
| 3684 | xlog_do_log_recovery( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 3685 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3686 | xfs_daddr_t head_blk, |
| 3687 | xfs_daddr_t tail_blk) |
| 3688 | { |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 3689 | int error, i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3690 | |
| 3691 | ASSERT(head_blk != tail_blk); |
| 3692 | |
| 3693 | /* |
| 3694 | * First do a pass to find all of the cancelled buf log items. |
| 3695 | * Store them in the buf_cancel_table for use in the second pass. |
| 3696 | */ |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 3697 | log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE * |
| 3698 | sizeof(struct list_head), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3699 | KM_SLEEP); |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 3700 | for (i = 0; i < XLOG_BC_TABLE_SIZE; i++) |
| 3701 | INIT_LIST_HEAD(&log->l_buf_cancel_table[i]); |
| 3702 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3703 | error = xlog_do_recovery_pass(log, head_blk, tail_blk, |
| 3704 | XLOG_RECOVER_PASS1); |
| 3705 | if (error != 0) { |
Denys Vlasenko | f0e2d93 | 2008-05-19 16:31:57 +1000 | [diff] [blame] | 3706 | kmem_free(log->l_buf_cancel_table); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3707 | log->l_buf_cancel_table = NULL; |
| 3708 | return error; |
| 3709 | } |
| 3710 | /* |
| 3711 | * Then do a second pass to actually recover the items in the log. |
| 3712 | * When it is complete free the table of buf cancel items. |
| 3713 | */ |
| 3714 | error = xlog_do_recovery_pass(log, head_blk, tail_blk, |
| 3715 | XLOG_RECOVER_PASS2); |
| 3716 | #ifdef DEBUG |
Tim Shimmin | 6d192a9 | 2006-06-09 14:55:38 +1000 | [diff] [blame] | 3717 | if (!error) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3718 | int i; |
| 3719 | |
| 3720 | for (i = 0; i < XLOG_BC_TABLE_SIZE; i++) |
Christoph Hellwig | d5689ea | 2010-12-01 22:06:22 +0000 | [diff] [blame] | 3721 | ASSERT(list_empty(&log->l_buf_cancel_table[i])); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3722 | } |
| 3723 | #endif /* DEBUG */ |
| 3724 | |
Denys Vlasenko | f0e2d93 | 2008-05-19 16:31:57 +1000 | [diff] [blame] | 3725 | kmem_free(log->l_buf_cancel_table); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3726 | log->l_buf_cancel_table = NULL; |
| 3727 | |
| 3728 | return error; |
| 3729 | } |
| 3730 | |
| 3731 | /* |
| 3732 | * Do the actual recovery |
| 3733 | */ |
| 3734 | STATIC int |
| 3735 | xlog_do_recover( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 3736 | struct xlog *log, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3737 | xfs_daddr_t head_blk, |
| 3738 | xfs_daddr_t tail_blk) |
| 3739 | { |
| 3740 | int error; |
| 3741 | xfs_buf_t *bp; |
| 3742 | xfs_sb_t *sbp; |
| 3743 | |
| 3744 | /* |
| 3745 | * First replay the images in the log. |
| 3746 | */ |
| 3747 | error = xlog_do_log_recovery(log, head_blk, tail_blk); |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 3748 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3749 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3750 | |
| 3751 | /* |
| 3752 | * If IO errors happened during recovery, bail out. |
| 3753 | */ |
| 3754 | if (XFS_FORCED_SHUTDOWN(log->l_mp)) { |
| 3755 | return (EIO); |
| 3756 | } |
| 3757 | |
| 3758 | /* |
| 3759 | * We now update the tail_lsn since much of the recovery has completed |
| 3760 | * and there may be space available to use. If there were no extent |
| 3761 | * or iunlinks, we can free up the entire log and set the tail_lsn to |
| 3762 | * be the last_sync_lsn. This was set in xlog_find_tail to be the |
| 3763 | * lsn of the last known good LR on disk. If there are extent frees |
| 3764 | * or iunlinks they will have some entries in the AIL; so we look at |
| 3765 | * the AIL to determine how to set the tail_lsn. |
| 3766 | */ |
| 3767 | xlog_assign_tail_lsn(log->l_mp); |
| 3768 | |
| 3769 | /* |
| 3770 | * Now that we've finished replaying all buffer and inode |
Dave Chinner | 9802182 | 2012-11-12 22:54:03 +1100 | [diff] [blame] | 3771 | * updates, re-read in the superblock and reverify it. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3772 | */ |
| 3773 | bp = xfs_getsb(log->l_mp, 0); |
| 3774 | XFS_BUF_UNDONE(bp); |
Lachlan McIlroy | bebf963 | 2007-10-15 13:18:02 +1000 | [diff] [blame] | 3775 | ASSERT(!(XFS_BUF_ISWRITE(bp))); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3776 | XFS_BUF_READ(bp); |
Lachlan McIlroy | bebf963 | 2007-10-15 13:18:02 +1000 | [diff] [blame] | 3777 | XFS_BUF_UNASYNC(bp); |
Dave Chinner | 1813dd6 | 2012-11-14 17:54:40 +1100 | [diff] [blame] | 3778 | bp->b_ops = &xfs_sb_buf_ops; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3779 | xfsbdstrat(log->l_mp, bp); |
Christoph Hellwig | 1a1a3e9 | 2010-10-06 18:41:18 +0000 | [diff] [blame] | 3780 | error = xfs_buf_iowait(bp); |
David Chinner | d64e31a | 2008-04-10 12:22:17 +1000 | [diff] [blame] | 3781 | if (error) { |
Christoph Hellwig | 901796a | 2011-10-10 16:52:49 +0000 | [diff] [blame] | 3782 | xfs_buf_ioerror_alert(bp, __func__); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3783 | ASSERT(0); |
| 3784 | xfs_buf_relse(bp); |
| 3785 | return error; |
| 3786 | } |
| 3787 | |
| 3788 | /* Convert superblock from on-disk format */ |
| 3789 | sbp = &log->l_mp->m_sb; |
Dave Chinner | 9802182 | 2012-11-12 22:54:03 +1100 | [diff] [blame] | 3790 | xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3791 | ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC); |
Eric Sandeen | 6211870 | 2008-03-06 13:44:28 +1100 | [diff] [blame] | 3792 | ASSERT(xfs_sb_good_version(sbp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3793 | xfs_buf_relse(bp); |
| 3794 | |
Lachlan McIlroy | 5478eea | 2007-02-10 18:36:29 +1100 | [diff] [blame] | 3795 | /* We've re-read the superblock so re-initialize per-cpu counters */ |
| 3796 | xfs_icsb_reinit_counters(log->l_mp); |
| 3797 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3798 | xlog_recover_check_summary(log); |
| 3799 | |
| 3800 | /* Normal transactions can now occur */ |
| 3801 | log->l_flags &= ~XLOG_ACTIVE_RECOVERY; |
| 3802 | return 0; |
| 3803 | } |
| 3804 | |
| 3805 | /* |
| 3806 | * Perform recovery and re-initialize some log variables in xlog_find_tail. |
| 3807 | * |
| 3808 | * Return error or zero. |
| 3809 | */ |
| 3810 | int |
| 3811 | xlog_recover( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 3812 | struct xlog *log) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3813 | { |
| 3814 | xfs_daddr_t head_blk, tail_blk; |
| 3815 | int error; |
| 3816 | |
| 3817 | /* find the tail of the log */ |
Eric Sandeen | 65be605 | 2006-01-11 15:34:19 +1100 | [diff] [blame] | 3818 | if ((error = xlog_find_tail(log, &head_blk, &tail_blk))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3819 | return error; |
| 3820 | |
| 3821 | if (tail_blk != head_blk) { |
| 3822 | /* There used to be a comment here: |
| 3823 | * |
| 3824 | * disallow recovery on read-only mounts. note -- mount |
| 3825 | * checks for ENOSPC and turns it into an intelligent |
| 3826 | * error message. |
| 3827 | * ...but this is no longer true. Now, unless you specify |
| 3828 | * NORECOVERY (in which case this function would never be |
| 3829 | * called), we just go ahead and recover. We do this all |
| 3830 | * under the vfs layer, so we can get away with it unless |
| 3831 | * the device itself is read-only, in which case we fail. |
| 3832 | */ |
Utako Kusaka | 3a02ee1 | 2007-05-08 13:50:06 +1000 | [diff] [blame] | 3833 | if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3834 | return error; |
| 3835 | } |
| 3836 | |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 3837 | xfs_notice(log->l_mp, "Starting recovery (logdev: %s)", |
| 3838 | log->l_mp->m_logname ? log->l_mp->m_logname |
| 3839 | : "internal"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3840 | |
| 3841 | error = xlog_do_recover(log, head_blk, tail_blk); |
| 3842 | log->l_flags |= XLOG_RECOVERY_NEEDED; |
| 3843 | } |
| 3844 | return error; |
| 3845 | } |
| 3846 | |
| 3847 | /* |
| 3848 | * In the first part of recovery we replay inodes and buffers and build |
| 3849 | * up the list of extent free items which need to be processed. Here |
| 3850 | * we process the extent free items and clean up the on disk unlinked |
| 3851 | * inode lists. This is separated from the first part of recovery so |
| 3852 | * that the root and real-time bitmap inodes can be read in from disk in |
| 3853 | * between the two stages. This is necessary so that we can free space |
| 3854 | * in the real-time portion of the file system. |
| 3855 | */ |
| 3856 | int |
| 3857 | xlog_recover_finish( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 3858 | struct xlog *log) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3859 | { |
| 3860 | /* |
| 3861 | * Now we're ready to do the transactions needed for the |
| 3862 | * rest of recovery. Start with completing all the extent |
| 3863 | * free intent records and then process the unlinked inode |
| 3864 | * lists. At this point, we essentially run in normal mode |
| 3865 | * except that we're still performing recovery actions |
| 3866 | * rather than accepting new requests. |
| 3867 | */ |
| 3868 | if (log->l_flags & XLOG_RECOVERY_NEEDED) { |
David Chinner | 3c1e2bb | 2008-04-10 12:21:11 +1000 | [diff] [blame] | 3869 | int error; |
| 3870 | error = xlog_recover_process_efis(log); |
| 3871 | if (error) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 3872 | xfs_alert(log->l_mp, "Failed to recover EFIs"); |
David Chinner | 3c1e2bb | 2008-04-10 12:21:11 +1000 | [diff] [blame] | 3873 | return error; |
| 3874 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3875 | /* |
| 3876 | * Sync the log to get all the EFIs out of the AIL. |
| 3877 | * This isn't absolutely necessary, but it helps in |
| 3878 | * case the unlink transactions would have problems |
| 3879 | * pushing the EFIs out of the way. |
| 3880 | */ |
Christoph Hellwig | a14a348 | 2010-01-19 09:56:46 +0000 | [diff] [blame] | 3881 | xfs_log_force(log->l_mp, XFS_LOG_SYNC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3882 | |
Christoph Hellwig | 4249023 | 2008-08-13 16:49:32 +1000 | [diff] [blame] | 3883 | xlog_recover_process_iunlinks(log); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3884 | |
| 3885 | xlog_recover_check_summary(log); |
| 3886 | |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 3887 | xfs_notice(log->l_mp, "Ending recovery (logdev: %s)", |
| 3888 | log->l_mp->m_logname ? log->l_mp->m_logname |
| 3889 | : "internal"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3890 | log->l_flags &= ~XLOG_RECOVERY_NEEDED; |
| 3891 | } else { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 3892 | xfs_info(log->l_mp, "Ending clean mount"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3893 | } |
| 3894 | return 0; |
| 3895 | } |
| 3896 | |
| 3897 | |
| 3898 | #if defined(DEBUG) |
| 3899 | /* |
| 3900 | * Read all of the agf and agi counters and check that they |
| 3901 | * are consistent with the superblock counters. |
| 3902 | */ |
| 3903 | void |
| 3904 | xlog_recover_check_summary( |
Mark Tinguely | 9a8d2fd | 2012-06-14 09:22:16 -0500 | [diff] [blame] | 3905 | struct xlog *log) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3906 | { |
| 3907 | xfs_mount_t *mp; |
| 3908 | xfs_agf_t *agfp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3909 | xfs_buf_t *agfbp; |
| 3910 | xfs_buf_t *agibp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3911 | xfs_agnumber_t agno; |
| 3912 | __uint64_t freeblks; |
| 3913 | __uint64_t itotal; |
| 3914 | __uint64_t ifree; |
Christoph Hellwig | 5e1be0f | 2008-11-28 14:23:37 +1100 | [diff] [blame] | 3915 | int error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3916 | |
| 3917 | mp = log->l_mp; |
| 3918 | |
| 3919 | freeblks = 0LL; |
| 3920 | itotal = 0LL; |
| 3921 | ifree = 0LL; |
| 3922 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { |
From: Christoph Hellwig | 4805621 | 2008-11-28 14:23:38 +1100 | [diff] [blame] | 3923 | error = xfs_read_agf(mp, NULL, agno, 0, &agfbp); |
| 3924 | if (error) { |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 3925 | xfs_alert(mp, "%s agf read failed agno %d error %d", |
| 3926 | __func__, agno, error); |
From: Christoph Hellwig | 4805621 | 2008-11-28 14:23:38 +1100 | [diff] [blame] | 3927 | } else { |
| 3928 | agfp = XFS_BUF_TO_AGF(agfbp); |
| 3929 | freeblks += be32_to_cpu(agfp->agf_freeblks) + |
| 3930 | be32_to_cpu(agfp->agf_flcount); |
| 3931 | xfs_buf_relse(agfbp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3932 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3933 | |
Christoph Hellwig | 5e1be0f | 2008-11-28 14:23:37 +1100 | [diff] [blame] | 3934 | error = xfs_read_agi(mp, NULL, agno, &agibp); |
Dave Chinner | a0fa2b6 | 2011-03-07 10:01:35 +1100 | [diff] [blame] | 3935 | if (error) { |
| 3936 | xfs_alert(mp, "%s agi read failed agno %d error %d", |
| 3937 | __func__, agno, error); |
| 3938 | } else { |
Christoph Hellwig | 5e1be0f | 2008-11-28 14:23:37 +1100 | [diff] [blame] | 3939 | struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3940 | |
Christoph Hellwig | 5e1be0f | 2008-11-28 14:23:37 +1100 | [diff] [blame] | 3941 | itotal += be32_to_cpu(agi->agi_count); |
| 3942 | ifree += be32_to_cpu(agi->agi_freecount); |
| 3943 | xfs_buf_relse(agibp); |
| 3944 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3945 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3946 | } |
| 3947 | #endif /* DEBUG */ |