blob: da37beb76f6e67faf90a658d55f49a06c6fc4152 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Tim Shimmin87c199c2006-06-09 14:56:16 +10002 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
Nathan Scott7b718762005-11-02 14:58:39 +11003 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * 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 Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * 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 Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * 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 Torvalds1da177e2005-04-16 15:20:36 -070017 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110019#include "xfs_fs.h"
Dave Chinner70a9883c2013-10-23 10:36:05 +110020#include "xfs_shared.h"
Dave Chinner239880e2013-10-23 10:50:10 +110021#include "xfs_format.h"
22#include "xfs_log_format.h"
23#include "xfs_trans_resv.h"
Nathan Scotta844f452005-11-02 14:38:42 +110024#include "xfs_bit.h"
Nathan Scotta844f452005-11-02 14:38:42 +110025#include "xfs_sb.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include "xfs_mount.h"
Dave Chinner57062782013-10-15 09:17:51 +110027#include "xfs_da_format.h"
Dave Chinner9a2cc412014-12-04 09:43:17 +110028#include "xfs_da_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include "xfs_inode.h"
Dave Chinner239880e2013-10-23 10:50:10 +110030#include "xfs_trans.h"
Dave Chinner239880e2013-10-23 10:50:10 +110031#include "xfs_log.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include "xfs_log_priv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_log_recover.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110034#include "xfs_inode_item.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include "xfs_extfree_item.h"
36#include "xfs_trans_priv.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110037#include "xfs_alloc.h"
38#include "xfs_ialloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include "xfs_quota.h"
Christoph Hellwig0e446be2012-11-12 22:54:24 +110040#include "xfs_cksum.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000041#include "xfs_trace.h"
Dave Chinner33479e02012-10-08 21:56:11 +110042#include "xfs_icache.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110043#include "xfs_bmap_btree.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110044#include "xfs_error.h"
Dave Chinner2b9ab5a2013-08-12 20:49:37 +100045#include "xfs_dir2.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
Dave Chinnerfc06c6d2013-08-12 20:49:22 +100047#define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
48
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -050049STATIC int
50xlog_find_zeroed(
51 struct xlog *,
52 xfs_daddr_t *);
53STATIC int
54xlog_clear_stale_blocks(
55 struct xlog *,
56 xfs_lsn_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -070057#if defined(DEBUG)
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -050058STATIC void
59xlog_recover_check_summary(
60 struct xlog *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070061#else
62#define xlog_recover_check_summary(log)
Linus Torvalds1da177e2005-04-16 15:20:36 -070063#endif
Brian Foster7088c412016-01-05 07:40:16 +110064STATIC int
65xlog_do_recovery_pass(
66 struct xlog *, xfs_daddr_t, xfs_daddr_t, int, xfs_daddr_t *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070067
Linus Torvalds1da177e2005-04-16 15:20:36 -070068/*
Christoph Hellwigd5689ea2010-12-01 22:06:22 +000069 * This structure is used during recovery to record the buf log items which
70 * have been canceled and should not be replayed.
71 */
72struct xfs_buf_cancel {
73 xfs_daddr_t bc_blkno;
74 uint bc_len;
75 int bc_refcount;
76 struct list_head bc_list;
77};
78
79/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 * Sector aligned buffer routines for buffer create/read/write/access
81 */
82
Alex Elderff30a622010-04-13 15:22:58 +100083/*
84 * Verify the given count of basic blocks is valid number of blocks
85 * to specify for an operation involving the given XFS log buffer.
86 * Returns nonzero if the count is valid, 0 otherwise.
87 */
88
89static inline int
90xlog_buf_bbcount_valid(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -050091 struct xlog *log,
Alex Elderff30a622010-04-13 15:22:58 +100092 int bbcount)
93{
94 return bbcount > 0 && bbcount <= log->l_logBBsize;
95}
96
Alex Elder36adecf2010-04-13 15:21:13 +100097/*
98 * Allocate a buffer to hold log data. The buffer needs to be able
99 * to map to a range of nbblks basic blocks at any valid (basic
100 * block) offset within the log.
101 */
Eric Sandeen5d77c0d2009-11-19 15:52:00 +0000102STATIC xfs_buf_t *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103xlog_get_bp(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500104 struct xlog *log,
Dave Chinner32281492009-01-22 15:37:47 +1100105 int nbblks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106{
Christoph Hellwigc8da0fa2011-07-08 14:36:25 +0200107 struct xfs_buf *bp;
108
Alex Elderff30a622010-04-13 15:22:58 +1000109 if (!xlog_buf_bbcount_valid(log, nbblks)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100110 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
Alex Elderff30a622010-04-13 15:22:58 +1000111 nbblks);
112 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
Dave Chinner32281492009-01-22 15:37:47 +1100113 return NULL;
114 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115
Alex Elder36adecf2010-04-13 15:21:13 +1000116 /*
117 * We do log I/O in units of log sectors (a power-of-2
118 * multiple of the basic block size), so we round up the
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300119 * requested size to accommodate the basic blocks required
Alex Elder36adecf2010-04-13 15:21:13 +1000120 * for complete log sectors.
121 *
122 * In addition, the buffer may be used for a non-sector-
123 * aligned block offset, in which case an I/O of the
124 * requested size could extend beyond the end of the
125 * buffer. If the requested size is only 1 basic block it
126 * will never straddle a sector boundary, so this won't be
127 * an issue. Nor will this be a problem if the log I/O is
128 * done in basic blocks (sector size 1). But otherwise we
129 * extend the buffer by one extra log sector to ensure
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300130 * there's space to accommodate this possibility.
Alex Elder36adecf2010-04-13 15:21:13 +1000131 */
Alex Elder69ce58f2010-04-20 17:09:59 +1000132 if (nbblks > 1 && log->l_sectBBsize > 1)
133 nbblks += log->l_sectBBsize;
134 nbblks = round_up(nbblks, log->l_sectBBsize);
Alex Elder36adecf2010-04-13 15:21:13 +1000135
Dave Chinnere70b73f2012-04-23 15:58:49 +1000136 bp = xfs_buf_get_uncached(log->l_mp->m_logdev_targp, nbblks, 0);
Christoph Hellwigc8da0fa2011-07-08 14:36:25 +0200137 if (bp)
138 xfs_buf_unlock(bp);
139 return bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140}
141
Eric Sandeen5d77c0d2009-11-19 15:52:00 +0000142STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143xlog_put_bp(
144 xfs_buf_t *bp)
145{
146 xfs_buf_free(bp);
147}
148
Alex Elder48389ef2010-04-20 17:10:21 +1000149/*
150 * Return the address of the start of the given block number's data
151 * in a log buffer. The buffer covers a log sector-aligned region.
152 */
Christoph Hellwigb2a922c2015-06-22 09:45:10 +1000153STATIC char *
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100154xlog_align(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500155 struct xlog *log,
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100156 xfs_daddr_t blk_no,
157 int nbblks,
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500158 struct xfs_buf *bp)
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100159{
Christoph Hellwigfdc07f42010-05-10 17:28:14 +0000160 xfs_daddr_t offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100161
Dave Chinner4e94b712012-04-23 15:58:51 +1000162 ASSERT(offset + nbblks <= bp->b_length);
Chandra Seetharaman62926042011-07-22 23:40:15 +0000163 return bp->b_addr + BBTOB(offset);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100164}
165
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166
167/*
168 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
169 */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100170STATIC int
171xlog_bread_noalign(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500172 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 xfs_daddr_t blk_no,
174 int nbblks,
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500175 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176{
177 int error;
178
Alex Elderff30a622010-04-13 15:22:58 +1000179 if (!xlog_buf_bbcount_valid(log, nbblks)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100180 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
Alex Elderff30a622010-04-13 15:22:58 +1000181 nbblks);
182 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
Dave Chinner24513372014-06-25 14:58:08 +1000183 return -EFSCORRUPTED;
Dave Chinner32281492009-01-22 15:37:47 +1100184 }
185
Alex Elder69ce58f2010-04-20 17:09:59 +1000186 blk_no = round_down(blk_no, log->l_sectBBsize);
187 nbblks = round_up(nbblks, log->l_sectBBsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
189 ASSERT(nbblks > 0);
Dave Chinner4e94b712012-04-23 15:58:51 +1000190 ASSERT(nbblks <= bp->b_length);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191
192 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
193 XFS_BUF_READ(bp);
Dave Chinneraa0e8832012-04-23 15:58:52 +1000194 bp->b_io_length = nbblks;
Dave Chinner0e95f192012-04-23 15:58:46 +1000195 bp->b_error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196
Dave Chinner595bff72014-10-02 09:05:14 +1000197 error = xfs_buf_submit_wait(bp);
198 if (error && !XFS_FORCED_SHUTDOWN(log->l_mp))
Christoph Hellwig901796a2011-10-10 16:52:49 +0000199 xfs_buf_ioerror_alert(bp, __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 return error;
201}
202
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100203STATIC int
204xlog_bread(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500205 struct xlog *log,
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100206 xfs_daddr_t blk_no,
207 int nbblks,
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500208 struct xfs_buf *bp,
Christoph Hellwigb2a922c2015-06-22 09:45:10 +1000209 char **offset)
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100210{
211 int error;
212
213 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
214 if (error)
215 return error;
216
217 *offset = xlog_align(log, blk_no, nbblks, bp);
218 return 0;
219}
220
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221/*
Dave Chinner44396472011-04-21 09:34:27 +0000222 * Read at an offset into the buffer. Returns with the buffer in it's original
223 * state regardless of the result of the read.
224 */
225STATIC int
226xlog_bread_offset(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500227 struct xlog *log,
Dave Chinner44396472011-04-21 09:34:27 +0000228 xfs_daddr_t blk_no, /* block to read from */
229 int nbblks, /* blocks to read */
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500230 struct xfs_buf *bp,
Christoph Hellwigb2a922c2015-06-22 09:45:10 +1000231 char *offset)
Dave Chinner44396472011-04-21 09:34:27 +0000232{
Christoph Hellwigb2a922c2015-06-22 09:45:10 +1000233 char *orig_offset = bp->b_addr;
Dave Chinner4e94b712012-04-23 15:58:51 +1000234 int orig_len = BBTOB(bp->b_length);
Dave Chinner44396472011-04-21 09:34:27 +0000235 int error, error2;
236
Chandra Seetharaman02fe03d2011-07-22 23:40:22 +0000237 error = xfs_buf_associate_memory(bp, offset, BBTOB(nbblks));
Dave Chinner44396472011-04-21 09:34:27 +0000238 if (error)
239 return error;
240
241 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
242
243 /* must reset buffer pointer even on error */
Chandra Seetharaman02fe03d2011-07-22 23:40:22 +0000244 error2 = xfs_buf_associate_memory(bp, orig_offset, orig_len);
Dave Chinner44396472011-04-21 09:34:27 +0000245 if (error)
246 return error;
247 return error2;
248}
249
250/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 * Write out the buffer at the given block for the given number of blocks.
252 * The buffer is kept locked across the write and is returned locked.
253 * This can only be used for synchronous log writes.
254 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000255STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256xlog_bwrite(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500257 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 xfs_daddr_t blk_no,
259 int nbblks,
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500260 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261{
262 int error;
263
Alex Elderff30a622010-04-13 15:22:58 +1000264 if (!xlog_buf_bbcount_valid(log, nbblks)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100265 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
Alex Elderff30a622010-04-13 15:22:58 +1000266 nbblks);
267 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
Dave Chinner24513372014-06-25 14:58:08 +1000268 return -EFSCORRUPTED;
Dave Chinner32281492009-01-22 15:37:47 +1100269 }
270
Alex Elder69ce58f2010-04-20 17:09:59 +1000271 blk_no = round_down(blk_no, log->l_sectBBsize);
272 nbblks = round_up(nbblks, log->l_sectBBsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273
274 ASSERT(nbblks > 0);
Dave Chinner4e94b712012-04-23 15:58:51 +1000275 ASSERT(nbblks <= bp->b_length);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276
277 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
278 XFS_BUF_ZEROFLAGS(bp);
Chandra Seetharaman72790aa2011-07-22 23:40:04 +0000279 xfs_buf_hold(bp);
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200280 xfs_buf_lock(bp);
Dave Chinneraa0e8832012-04-23 15:58:52 +1000281 bp->b_io_length = nbblks;
Dave Chinner0e95f192012-04-23 15:58:46 +1000282 bp->b_error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283
Christoph Hellwigc2b006c2011-08-23 08:28:07 +0000284 error = xfs_bwrite(bp);
Christoph Hellwig901796a2011-10-10 16:52:49 +0000285 if (error)
286 xfs_buf_ioerror_alert(bp, __func__);
Christoph Hellwigc2b006c2011-08-23 08:28:07 +0000287 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 return error;
289}
290
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291#ifdef DEBUG
292/*
293 * dump debug superblock and log record information
294 */
295STATIC void
296xlog_header_check_dump(
297 xfs_mount_t *mp,
298 xlog_rec_header_t *head)
299{
Eric Sandeen08e96e12013-10-11 20:59:05 -0500300 xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d",
Joe Perches03daa572009-12-14 18:01:10 -0800301 __func__, &mp->m_sb.sb_uuid, XLOG_FMT);
Eric Sandeen08e96e12013-10-11 20:59:05 -0500302 xfs_debug(mp, " log : uuid = %pU, fmt = %d",
Joe Perches03daa572009-12-14 18:01:10 -0800303 &head->h_fs_uuid, be32_to_cpu(head->h_fmt));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304}
305#else
306#define xlog_header_check_dump(mp, head)
307#endif
308
309/*
310 * check log record header for recovery
311 */
312STATIC int
313xlog_header_check_recover(
314 xfs_mount_t *mp,
315 xlog_rec_header_t *head)
316{
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200317 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318
319 /*
320 * IRIX doesn't write the h_fmt field and leaves it zeroed
321 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
322 * a dirty log created in IRIX.
323 */
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200324 if (unlikely(head->h_fmt != cpu_to_be32(XLOG_FMT))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100325 xfs_warn(mp,
326 "dirty log written in incompatible format - can't recover");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 xlog_header_check_dump(mp, head);
328 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
329 XFS_ERRLEVEL_HIGH, mp);
Dave Chinner24513372014-06-25 14:58:08 +1000330 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100332 xfs_warn(mp,
333 "dirty log entry has mismatched uuid - can't recover");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 xlog_header_check_dump(mp, head);
335 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
336 XFS_ERRLEVEL_HIGH, mp);
Dave Chinner24513372014-06-25 14:58:08 +1000337 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 }
339 return 0;
340}
341
342/*
343 * read the head block of the log and check the header
344 */
345STATIC int
346xlog_header_check_mount(
347 xfs_mount_t *mp,
348 xlog_rec_header_t *head)
349{
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200350 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351
352 if (uuid_is_nil(&head->h_fs_uuid)) {
353 /*
354 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
355 * h_fs_uuid is nil, we assume this log was last mounted
356 * by IRIX and continue.
357 */
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100358 xfs_warn(mp, "nil uuid in log - IRIX style log");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100360 xfs_warn(mp, "log has mismatched uuid - can't recover");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 xlog_header_check_dump(mp, head);
362 XFS_ERROR_REPORT("xlog_header_check_mount",
363 XFS_ERRLEVEL_HIGH, mp);
Dave Chinner24513372014-06-25 14:58:08 +1000364 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 }
366 return 0;
367}
368
369STATIC void
370xlog_recover_iodone(
371 struct xfs_buf *bp)
372{
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +0000373 if (bp->b_error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 /*
375 * We're not going to bother about retrying
376 * this during recovery. One strike!
377 */
Dave Chinner595bff72014-10-02 09:05:14 +1000378 if (!XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) {
379 xfs_buf_ioerror_alert(bp, __func__);
380 xfs_force_shutdown(bp->b_target->bt_mount,
381 SHUTDOWN_META_IO_ERROR);
382 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 }
Christoph Hellwigcb669ca2011-07-13 13:43:49 +0200384 bp->b_iodone = NULL;
Dave Chinnere8aaba92014-10-02 09:04:22 +1000385 xfs_buf_ioend(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386}
387
388/*
389 * This routine finds (to an approximation) the first block in the physical
390 * log which contains the given cycle. It uses a binary search algorithm.
391 * Note that the algorithm can not be perfect because the disk will not
392 * necessarily be perfect.
393 */
David Chinnera8272ce2007-11-23 16:28:09 +1100394STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395xlog_find_cycle_start(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500396 struct xlog *log,
397 struct xfs_buf *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 xfs_daddr_t first_blk,
399 xfs_daddr_t *last_blk,
400 uint cycle)
401{
Christoph Hellwigb2a922c2015-06-22 09:45:10 +1000402 char *offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 xfs_daddr_t mid_blk;
Alex Eldere3bb2e32010-04-15 18:17:30 +0000404 xfs_daddr_t end_blk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 uint mid_cycle;
406 int error;
407
Alex Eldere3bb2e32010-04-15 18:17:30 +0000408 end_blk = *last_blk;
409 mid_blk = BLK_AVG(first_blk, end_blk);
410 while (mid_blk != first_blk && mid_blk != end_blk) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100411 error = xlog_bread(log, mid_blk, 1, bp, &offset);
412 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 return error;
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000414 mid_cycle = xlog_get_cycle(offset);
Alex Eldere3bb2e32010-04-15 18:17:30 +0000415 if (mid_cycle == cycle)
416 end_blk = mid_blk; /* last_half_cycle == mid_cycle */
417 else
418 first_blk = mid_blk; /* first_half_cycle == mid_cycle */
419 mid_blk = BLK_AVG(first_blk, end_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 }
Alex Eldere3bb2e32010-04-15 18:17:30 +0000421 ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) ||
422 (mid_blk == end_blk && mid_blk-1 == first_blk));
423
424 *last_blk = end_blk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425
426 return 0;
427}
428
429/*
Alex Elder3f943d82010-04-15 18:17:34 +0000430 * Check that a range of blocks does not contain stop_on_cycle_no.
431 * Fill in *new_blk with the block offset where such a block is
432 * found, or with -1 (an invalid block number) if there is no such
433 * block in the range. The scan needs to occur from front to back
434 * and the pointer into the region must be updated since a later
435 * routine will need to perform another test.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 */
437STATIC int
438xlog_find_verify_cycle(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500439 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 xfs_daddr_t start_blk,
441 int nbblks,
442 uint stop_on_cycle_no,
443 xfs_daddr_t *new_blk)
444{
445 xfs_daddr_t i, j;
446 uint cycle;
447 xfs_buf_t *bp;
448 xfs_daddr_t bufblks;
Christoph Hellwigb2a922c2015-06-22 09:45:10 +1000449 char *buf = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 int error = 0;
451
Alex Elder6881a222010-04-13 15:22:29 +1000452 /*
453 * Greedily allocate a buffer big enough to handle the full
454 * range of basic blocks we'll be examining. If that fails,
455 * try a smaller size. We need to be able to read at least
456 * a log sector, or we're out of luck.
457 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 bufblks = 1 << ffs(nbblks);
Dave Chinner81158e02012-04-27 19:45:22 +1000459 while (bufblks > log->l_logBBsize)
460 bufblks >>= 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 while (!(bp = xlog_get_bp(log, bufblks))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 bufblks >>= 1;
Alex Elder69ce58f2010-04-20 17:09:59 +1000463 if (bufblks < log->l_sectBBsize)
Dave Chinner24513372014-06-25 14:58:08 +1000464 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 }
466
467 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
468 int bcount;
469
470 bcount = min(bufblks, (start_blk + nbblks - i));
471
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100472 error = xlog_bread(log, i, bcount, bp, &buf);
473 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 goto out;
475
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 for (j = 0; j < bcount; j++) {
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000477 cycle = xlog_get_cycle(buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 if (cycle == stop_on_cycle_no) {
479 *new_blk = i+j;
480 goto out;
481 }
482
483 buf += BBSIZE;
484 }
485 }
486
487 *new_blk = -1;
488
489out:
490 xlog_put_bp(bp);
491 return error;
492}
493
494/*
495 * Potentially backup over partial log record write.
496 *
497 * In the typical case, last_blk is the number of the block directly after
498 * a good log record. Therefore, we subtract one to get the block number
499 * of the last block in the given buffer. extra_bblks contains the number
500 * of blocks we would have read on a previous read. This happens when the
501 * last log record is split over the end of the physical log.
502 *
503 * extra_bblks is the number of blocks potentially verified on a previous
504 * call to this routine.
505 */
506STATIC int
507xlog_find_verify_log_record(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500508 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 xfs_daddr_t start_blk,
510 xfs_daddr_t *last_blk,
511 int extra_bblks)
512{
513 xfs_daddr_t i;
514 xfs_buf_t *bp;
Christoph Hellwigb2a922c2015-06-22 09:45:10 +1000515 char *offset = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 xlog_rec_header_t *head = NULL;
517 int error = 0;
518 int smallmem = 0;
519 int num_blks = *last_blk - start_blk;
520 int xhdrs;
521
522 ASSERT(start_blk != 0 || *last_blk != start_blk);
523
524 if (!(bp = xlog_get_bp(log, num_blks))) {
525 if (!(bp = xlog_get_bp(log, 1)))
Dave Chinner24513372014-06-25 14:58:08 +1000526 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 smallmem = 1;
528 } else {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100529 error = xlog_bread(log, start_blk, num_blks, bp, &offset);
530 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 offset += ((num_blks - 1) << BBSHIFT);
533 }
534
535 for (i = (*last_blk) - 1; i >= 0; i--) {
536 if (i < start_blk) {
537 /* valid log record not found */
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100538 xfs_warn(log->l_mp,
539 "Log inconsistent (didn't find previous header)");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 ASSERT(0);
Dave Chinner24513372014-06-25 14:58:08 +1000541 error = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 goto out;
543 }
544
545 if (smallmem) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100546 error = xlog_bread(log, i, 1, bp, &offset);
547 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 }
550
551 head = (xlog_rec_header_t *)offset;
552
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200553 if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 break;
555
556 if (!smallmem)
557 offset -= BBSIZE;
558 }
559
560 /*
561 * We hit the beginning of the physical log & still no header. Return
562 * to caller. If caller can handle a return of -1, then this routine
563 * will be called again for the end of the physical log.
564 */
565 if (i == -1) {
Dave Chinner24513372014-06-25 14:58:08 +1000566 error = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 goto out;
568 }
569
570 /*
571 * We have the final block of the good log (the first block
572 * of the log record _before_ the head. So we check the uuid.
573 */
574 if ((error = xlog_header_check_mount(log->l_mp, head)))
575 goto out;
576
577 /*
578 * We may have found a log record header before we expected one.
579 * last_blk will be the 1st block # with a given cycle #. We may end
580 * up reading an entire log record. In this case, we don't want to
581 * reset last_blk. Only when last_blk points in the middle of a log
582 * record do we update last_blk.
583 */
Eric Sandeen62118702008-03-06 13:44:28 +1100584 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000585 uint h_size = be32_to_cpu(head->h_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586
587 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
588 if (h_size % XLOG_HEADER_CYCLE_SIZE)
589 xhdrs++;
590 } else {
591 xhdrs = 1;
592 }
593
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000594 if (*last_blk - i + extra_bblks !=
595 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 *last_blk = i;
597
598out:
599 xlog_put_bp(bp);
600 return error;
601}
602
603/*
604 * Head is defined to be the point of the log where the next log write
Zhi Yong Wu0a94da22013-08-07 10:11:08 +0000605 * could go. This means that incomplete LR writes at the end are
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 * eliminated when calculating the head. We aren't guaranteed that previous
607 * LR have complete transactions. We only know that a cycle number of
608 * current cycle number -1 won't be present in the log if we start writing
609 * from our current block number.
610 *
611 * last_blk contains the block number of the first block with a given
612 * cycle number.
613 *
614 * Return: zero if normal, non-zero if error.
615 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000616STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617xlog_find_head(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500618 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 xfs_daddr_t *return_head_blk)
620{
621 xfs_buf_t *bp;
Christoph Hellwigb2a922c2015-06-22 09:45:10 +1000622 char *offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
624 int num_scan_bblks;
625 uint first_half_cycle, last_half_cycle;
626 uint stop_on_cycle;
627 int error, log_bbnum = log->l_logBBsize;
628
629 /* Is the end of the log device zeroed? */
Dave Chinner24513372014-06-25 14:58:08 +1000630 error = xlog_find_zeroed(log, &first_blk);
631 if (error < 0) {
632 xfs_warn(log->l_mp, "empty log check failed");
633 return error;
634 }
635 if (error == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 *return_head_blk = first_blk;
637
638 /* Is the whole lot zeroed? */
639 if (!first_blk) {
640 /* Linux XFS shouldn't generate totally zeroed logs -
641 * mkfs etc write a dummy unmount record to a fresh
642 * log so we can store the uuid in there
643 */
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100644 xfs_warn(log->l_mp, "totally zeroed log");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 }
646
647 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 }
649
650 first_blk = 0; /* get cycle # of 1st block */
651 bp = xlog_get_bp(log, 1);
652 if (!bp)
Dave Chinner24513372014-06-25 14:58:08 +1000653 return -ENOMEM;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100654
655 error = xlog_bread(log, 0, 1, bp, &offset);
656 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 goto bp_err;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100658
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000659 first_half_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660
661 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100662 error = xlog_bread(log, last_blk, 1, bp, &offset);
663 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 goto bp_err;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100665
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000666 last_half_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 ASSERT(last_half_cycle != 0);
668
669 /*
670 * If the 1st half cycle number is equal to the last half cycle number,
671 * then the entire log is stamped with the same cycle number. In this
672 * case, head_blk can't be set to zero (which makes sense). The below
673 * math doesn't work out properly with head_blk equal to zero. Instead,
674 * we set it to log_bbnum which is an invalid block number, but this
675 * value makes the math correct. If head_blk doesn't changed through
676 * all the tests below, *head_blk is set to zero at the very end rather
677 * than log_bbnum. In a sense, log_bbnum and zero are the same block
678 * in a circular file.
679 */
680 if (first_half_cycle == last_half_cycle) {
681 /*
682 * In this case we believe that the entire log should have
683 * cycle number last_half_cycle. We need to scan backwards
684 * from the end verifying that there are no holes still
685 * containing last_half_cycle - 1. If we find such a hole,
686 * then the start of that hole will be the new head. The
687 * simple case looks like
688 * x | x ... | x - 1 | x
689 * Another case that fits this picture would be
690 * x | x + 1 | x ... | x
Nathan Scottc41564b2006-03-29 08:55:14 +1000691 * In this case the head really is somewhere at the end of the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 * log, as one of the latest writes at the beginning was
693 * incomplete.
694 * One more case is
695 * x | x + 1 | x ... | x - 1 | x
696 * This is really the combination of the above two cases, and
697 * the head has to end up at the start of the x-1 hole at the
698 * end of the log.
699 *
700 * In the 256k log case, we will read from the beginning to the
701 * end of the log and search for cycle numbers equal to x-1.
702 * We don't worry about the x+1 blocks that we encounter,
703 * because we know that they cannot be the head since the log
704 * started with x.
705 */
706 head_blk = log_bbnum;
707 stop_on_cycle = last_half_cycle - 1;
708 } else {
709 /*
710 * In this case we want to find the first block with cycle
711 * number matching last_half_cycle. We expect the log to be
712 * some variation on
Alex Elder3f943d82010-04-15 18:17:34 +0000713 * x + 1 ... | x ... | x
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 * The first block with cycle number x (last_half_cycle) will
715 * be where the new head belongs. First we do a binary search
716 * for the first occurrence of last_half_cycle. The binary
717 * search may not be totally accurate, so then we scan back
718 * from there looking for occurrences of last_half_cycle before
719 * us. If that backwards scan wraps around the beginning of
720 * the log, then we look for occurrences of last_half_cycle - 1
721 * at the end of the log. The cases we're looking for look
722 * like
Alex Elder3f943d82010-04-15 18:17:34 +0000723 * v binary search stopped here
724 * x + 1 ... | x | x + 1 | x ... | x
725 * ^ but we want to locate this spot
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 * or
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 * <---------> less than scan distance
Alex Elder3f943d82010-04-15 18:17:34 +0000728 * x + 1 ... | x ... | x - 1 | x
729 * ^ we want to locate this spot
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 */
731 stop_on_cycle = last_half_cycle;
732 if ((error = xlog_find_cycle_start(log, bp, first_blk,
733 &head_blk, last_half_cycle)))
734 goto bp_err;
735 }
736
737 /*
738 * Now validate the answer. Scan back some number of maximum possible
739 * blocks and make sure each one has the expected cycle number. The
740 * maximum is determined by the total possible amount of buffering
741 * in the in-core log. The following number can be made tighter if
742 * we actually look at the block size of the filesystem.
743 */
744 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
745 if (head_blk >= num_scan_bblks) {
746 /*
747 * We are guaranteed that the entire check can be performed
748 * in one buffer.
749 */
750 start_blk = head_blk - num_scan_bblks;
751 if ((error = xlog_find_verify_cycle(log,
752 start_blk, num_scan_bblks,
753 stop_on_cycle, &new_blk)))
754 goto bp_err;
755 if (new_blk != -1)
756 head_blk = new_blk;
757 } else { /* need to read 2 parts of log */
758 /*
759 * We are going to scan backwards in the log in two parts.
760 * First we scan the physical end of the log. In this part
761 * of the log, we are looking for blocks with cycle number
762 * last_half_cycle - 1.
763 * If we find one, then we know that the log starts there, as
764 * we've found a hole that didn't get written in going around
765 * the end of the physical log. The simple case for this is
766 * x + 1 ... | x ... | x - 1 | x
767 * <---------> less than scan distance
768 * If all of the blocks at the end of the log have cycle number
769 * last_half_cycle, then we check the blocks at the start of
770 * the log looking for occurrences of last_half_cycle. If we
771 * find one, then our current estimate for the location of the
772 * first occurrence of last_half_cycle is wrong and we move
773 * back to the hole we've found. This case looks like
774 * x + 1 ... | x | x + 1 | x ...
775 * ^ binary search stopped here
776 * Another case we need to handle that only occurs in 256k
777 * logs is
778 * x + 1 ... | x ... | x+1 | x ...
779 * ^ binary search stops here
780 * In a 256k log, the scan at the end of the log will see the
781 * x + 1 blocks. We need to skip past those since that is
782 * certainly not the head of the log. By searching for
783 * last_half_cycle-1 we accomplish that.
784 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 ASSERT(head_blk <= INT_MAX &&
Alex Elder3f943d82010-04-15 18:17:34 +0000786 (xfs_daddr_t) num_scan_bblks >= head_blk);
787 start_blk = log_bbnum - (num_scan_bblks - head_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 if ((error = xlog_find_verify_cycle(log, start_blk,
789 num_scan_bblks - (int)head_blk,
790 (stop_on_cycle - 1), &new_blk)))
791 goto bp_err;
792 if (new_blk != -1) {
793 head_blk = new_blk;
Alex Elder9db127e2010-04-15 18:17:26 +0000794 goto validate_head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 }
796
797 /*
798 * Scan beginning of log now. The last part of the physical
799 * log is good. This scan needs to verify that it doesn't find
800 * the last_half_cycle.
801 */
802 start_blk = 0;
803 ASSERT(head_blk <= INT_MAX);
804 if ((error = xlog_find_verify_cycle(log,
805 start_blk, (int)head_blk,
806 stop_on_cycle, &new_blk)))
807 goto bp_err;
808 if (new_blk != -1)
809 head_blk = new_blk;
810 }
811
Alex Elder9db127e2010-04-15 18:17:26 +0000812validate_head:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 /*
814 * Now we need to make sure head_blk is not pointing to a block in
815 * the middle of a log record.
816 */
817 num_scan_bblks = XLOG_REC_SHIFT(log);
818 if (head_blk >= num_scan_bblks) {
819 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
820
821 /* start ptr at last block ptr before head_blk */
Dave Chinner24513372014-06-25 14:58:08 +1000822 error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0);
823 if (error == 1)
824 error = -EIO;
825 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 goto bp_err;
827 } else {
828 start_blk = 0;
829 ASSERT(head_blk <= INT_MAX);
Dave Chinner24513372014-06-25 14:58:08 +1000830 error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0);
831 if (error < 0)
832 goto bp_err;
833 if (error == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 /* We hit the beginning of the log during our search */
Alex Elder3f943d82010-04-15 18:17:34 +0000835 start_blk = log_bbnum - (num_scan_bblks - head_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 new_blk = log_bbnum;
837 ASSERT(start_blk <= INT_MAX &&
838 (xfs_daddr_t) log_bbnum-start_blk >= 0);
839 ASSERT(head_blk <= INT_MAX);
Dave Chinner24513372014-06-25 14:58:08 +1000840 error = xlog_find_verify_log_record(log, start_blk,
841 &new_blk, (int)head_blk);
842 if (error == 1)
843 error = -EIO;
844 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 goto bp_err;
846 if (new_blk != log_bbnum)
847 head_blk = new_blk;
848 } else if (error)
849 goto bp_err;
850 }
851
852 xlog_put_bp(bp);
853 if (head_blk == log_bbnum)
854 *return_head_blk = 0;
855 else
856 *return_head_blk = head_blk;
857 /*
858 * When returning here, we have a good block number. Bad block
859 * means that during a previous crash, we didn't have a clean break
860 * from cycle number N to cycle number N-1. In this case, we need
861 * to find the first block with cycle number N-1.
862 */
863 return 0;
864
865 bp_err:
866 xlog_put_bp(bp);
867
868 if (error)
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100869 xfs_warn(log->l_mp, "failed to find log head");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 return error;
871}
872
873/*
Brian Fostereed6b462016-01-04 15:55:10 +1100874 * Seek backwards in the log for log record headers.
875 *
876 * Given a starting log block, walk backwards until we find the provided number
877 * of records or hit the provided tail block. The return value is the number of
878 * records encountered or a negative error code. The log block and buffer
879 * pointer of the last record seen are returned in rblk and rhead respectively.
880 */
881STATIC int
882xlog_rseek_logrec_hdr(
883 struct xlog *log,
884 xfs_daddr_t head_blk,
885 xfs_daddr_t tail_blk,
886 int count,
887 struct xfs_buf *bp,
888 xfs_daddr_t *rblk,
889 struct xlog_rec_header **rhead,
890 bool *wrapped)
891{
892 int i;
893 int error;
894 int found = 0;
895 char *offset = NULL;
896 xfs_daddr_t end_blk;
897
898 *wrapped = false;
899
900 /*
901 * Walk backwards from the head block until we hit the tail or the first
902 * block in the log.
903 */
904 end_blk = head_blk > tail_blk ? tail_blk : 0;
905 for (i = (int) head_blk - 1; i >= end_blk; i--) {
906 error = xlog_bread(log, i, 1, bp, &offset);
907 if (error)
908 goto out_error;
909
910 if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
911 *rblk = i;
912 *rhead = (struct xlog_rec_header *) offset;
913 if (++found == count)
914 break;
915 }
916 }
917
918 /*
919 * If we haven't hit the tail block or the log record header count,
920 * start looking again from the end of the physical log. Note that
921 * callers can pass head == tail if the tail is not yet known.
922 */
923 if (tail_blk >= head_blk && found != count) {
924 for (i = log->l_logBBsize - 1; i >= (int) tail_blk; i--) {
925 error = xlog_bread(log, i, 1, bp, &offset);
926 if (error)
927 goto out_error;
928
929 if (*(__be32 *)offset ==
930 cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
931 *wrapped = true;
932 *rblk = i;
933 *rhead = (struct xlog_rec_header *) offset;
934 if (++found == count)
935 break;
936 }
937 }
938 }
939
940 return found;
941
942out_error:
943 return error;
944}
945
946/*
Brian Foster7088c412016-01-05 07:40:16 +1100947 * Seek forward in the log for log record headers.
948 *
949 * Given head and tail blocks, walk forward from the tail block until we find
950 * the provided number of records or hit the head block. The return value is the
951 * number of records encountered or a negative error code. The log block and
952 * buffer pointer of the last record seen are returned in rblk and rhead
953 * respectively.
954 */
955STATIC int
956xlog_seek_logrec_hdr(
957 struct xlog *log,
958 xfs_daddr_t head_blk,
959 xfs_daddr_t tail_blk,
960 int count,
961 struct xfs_buf *bp,
962 xfs_daddr_t *rblk,
963 struct xlog_rec_header **rhead,
964 bool *wrapped)
965{
966 int i;
967 int error;
968 int found = 0;
969 char *offset = NULL;
970 xfs_daddr_t end_blk;
971
972 *wrapped = false;
973
974 /*
975 * Walk forward from the tail block until we hit the head or the last
976 * block in the log.
977 */
978 end_blk = head_blk > tail_blk ? head_blk : log->l_logBBsize - 1;
979 for (i = (int) tail_blk; i <= end_blk; i++) {
980 error = xlog_bread(log, i, 1, bp, &offset);
981 if (error)
982 goto out_error;
983
984 if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
985 *rblk = i;
986 *rhead = (struct xlog_rec_header *) offset;
987 if (++found == count)
988 break;
989 }
990 }
991
992 /*
993 * If we haven't hit the head block or the log record header count,
994 * start looking again from the start of the physical log.
995 */
996 if (tail_blk > head_blk && found != count) {
997 for (i = 0; i < (int) head_blk; i++) {
998 error = xlog_bread(log, i, 1, bp, &offset);
999 if (error)
1000 goto out_error;
1001
1002 if (*(__be32 *)offset ==
1003 cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
1004 *wrapped = true;
1005 *rblk = i;
1006 *rhead = (struct xlog_rec_header *) offset;
1007 if (++found == count)
1008 break;
1009 }
1010 }
1011 }
1012
1013 return found;
1014
1015out_error:
1016 return error;
1017}
1018
1019/*
1020 * Check the log tail for torn writes. This is required when torn writes are
1021 * detected at the head and the head had to be walked back to a previous record.
1022 * The tail of the previous record must now be verified to ensure the torn
1023 * writes didn't corrupt the previous tail.
1024 *
1025 * Return an error if CRC verification fails as recovery cannot proceed.
1026 */
1027STATIC int
1028xlog_verify_tail(
1029 struct xlog *log,
1030 xfs_daddr_t head_blk,
1031 xfs_daddr_t tail_blk)
1032{
1033 struct xlog_rec_header *thead;
1034 struct xfs_buf *bp;
1035 xfs_daddr_t first_bad;
1036 int count;
1037 int error = 0;
1038 bool wrapped;
1039 xfs_daddr_t tmp_head;
1040
1041 bp = xlog_get_bp(log, 1);
1042 if (!bp)
1043 return -ENOMEM;
1044
1045 /*
1046 * Seek XLOG_MAX_ICLOGS + 1 records past the current tail record to get
1047 * a temporary head block that points after the last possible
1048 * concurrently written record of the tail.
1049 */
1050 count = xlog_seek_logrec_hdr(log, head_blk, tail_blk,
1051 XLOG_MAX_ICLOGS + 1, bp, &tmp_head, &thead,
1052 &wrapped);
1053 if (count < 0) {
1054 error = count;
1055 goto out;
1056 }
1057
1058 /*
1059 * If the call above didn't find XLOG_MAX_ICLOGS + 1 records, we ran
1060 * into the actual log head. tmp_head points to the start of the record
1061 * so update it to the actual head block.
1062 */
1063 if (count < XLOG_MAX_ICLOGS + 1)
1064 tmp_head = head_blk;
1065
1066 /*
1067 * We now have a tail and temporary head block that covers at least
1068 * XLOG_MAX_ICLOGS records from the tail. We need to verify that these
1069 * records were completely written. Run a CRC verification pass from
1070 * tail to head and return the result.
1071 */
1072 error = xlog_do_recovery_pass(log, tmp_head, tail_blk,
1073 XLOG_RECOVER_CRCPASS, &first_bad);
1074
1075out:
1076 xlog_put_bp(bp);
1077 return error;
1078}
1079
1080/*
1081 * Detect and trim torn writes from the head of the log.
1082 *
1083 * Storage without sector atomicity guarantees can result in torn writes in the
1084 * log in the event of a crash. Our only means to detect this scenario is via
1085 * CRC verification. While we can't always be certain that CRC verification
1086 * failure is due to a torn write vs. an unrelated corruption, we do know that
1087 * only a certain number (XLOG_MAX_ICLOGS) of log records can be written out at
1088 * one time. Therefore, CRC verify up to XLOG_MAX_ICLOGS records at the head of
1089 * the log and treat failures in this range as torn writes as a matter of
1090 * policy. In the event of CRC failure, the head is walked back to the last good
1091 * record in the log and the tail is updated from that record and verified.
1092 */
1093STATIC int
1094xlog_verify_head(
1095 struct xlog *log,
1096 xfs_daddr_t *head_blk, /* in/out: unverified head */
1097 xfs_daddr_t *tail_blk, /* out: tail block */
1098 struct xfs_buf *bp,
1099 xfs_daddr_t *rhead_blk, /* start blk of last record */
1100 struct xlog_rec_header **rhead, /* ptr to last record */
1101 bool *wrapped) /* last rec. wraps phys. log */
1102{
1103 struct xlog_rec_header *tmp_rhead;
1104 struct xfs_buf *tmp_bp;
1105 xfs_daddr_t first_bad;
1106 xfs_daddr_t tmp_rhead_blk;
1107 int found;
1108 int error;
1109 bool tmp_wrapped;
1110
1111 /*
1112 * Search backwards through the log looking for the log record header
1113 * block. This wraps all the way back around to the head so something is
1114 * seriously wrong if we can't find it.
1115 */
1116 found = xlog_rseek_logrec_hdr(log, *head_blk, *head_blk, 1, bp, rhead_blk,
1117 rhead, wrapped);
1118 if (found < 0)
1119 return found;
1120 if (!found) {
1121 xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__);
1122 return -EIO;
1123 }
1124
1125 *tail_blk = BLOCK_LSN(be64_to_cpu((*rhead)->h_tail_lsn));
1126
1127 /*
1128 * Now that we have a tail block, check the head of the log for torn
1129 * writes. Search again until we hit the tail or the maximum number of
1130 * log record I/Os that could have been in flight at one time. Use a
1131 * temporary buffer so we don't trash the rhead/bp pointer from the
1132 * call above.
1133 */
1134 tmp_bp = xlog_get_bp(log, 1);
1135 if (!tmp_bp)
1136 return -ENOMEM;
1137 error = xlog_rseek_logrec_hdr(log, *head_blk, *tail_blk,
1138 XLOG_MAX_ICLOGS, tmp_bp, &tmp_rhead_blk,
1139 &tmp_rhead, &tmp_wrapped);
1140 xlog_put_bp(tmp_bp);
1141 if (error < 0)
1142 return error;
1143
1144 /*
1145 * Now run a CRC verification pass over the records starting at the
1146 * block found above to the current head. If a CRC failure occurs, the
1147 * log block of the first bad record is saved in first_bad.
1148 */
1149 error = xlog_do_recovery_pass(log, *head_blk, tmp_rhead_blk,
1150 XLOG_RECOVER_CRCPASS, &first_bad);
1151 if (error == -EFSBADCRC) {
1152 /*
1153 * We've hit a potential torn write. Reset the error and warn
1154 * about it.
1155 */
1156 error = 0;
1157 xfs_warn(log->l_mp,
1158"Torn write (CRC failure) detected at log block 0x%llx. Truncating head block from 0x%llx.",
1159 first_bad, *head_blk);
1160
1161 /*
1162 * Get the header block and buffer pointer for the last good
1163 * record before the bad record.
1164 *
1165 * Note that xlog_find_tail() clears the blocks at the new head
1166 * (i.e., the records with invalid CRC) if the cycle number
1167 * matches the the current cycle.
1168 */
1169 found = xlog_rseek_logrec_hdr(log, first_bad, *tail_blk, 1, bp,
1170 rhead_blk, rhead, wrapped);
1171 if (found < 0)
1172 return found;
1173 if (found == 0) /* XXX: right thing to do here? */
1174 return -EIO;
1175
1176 /*
1177 * Reset the head block to the starting block of the first bad
1178 * log record and set the tail block based on the last good
1179 * record.
1180 *
1181 * Bail out if the updated head/tail match as this indicates
1182 * possible corruption outside of the acceptable
1183 * (XLOG_MAX_ICLOGS) range. This is a job for xfs_repair...
1184 */
1185 *head_blk = first_bad;
1186 *tail_blk = BLOCK_LSN(be64_to_cpu((*rhead)->h_tail_lsn));
1187 if (*head_blk == *tail_blk) {
1188 ASSERT(0);
1189 return 0;
1190 }
1191
1192 /*
1193 * Now verify the tail based on the updated head. This is
1194 * required because the torn writes trimmed from the head could
1195 * have been written over the tail of a previous record. Return
1196 * any errors since recovery cannot proceed if the tail is
1197 * corrupt.
1198 *
1199 * XXX: This leaves a gap in truly robust protection from torn
1200 * writes in the log. If the head is behind the tail, the tail
1201 * pushes forward to create some space and then a crash occurs
1202 * causing the writes into the previous record's tail region to
1203 * tear, log recovery isn't able to recover.
1204 *
1205 * How likely is this to occur? If possible, can we do something
1206 * more intelligent here? Is it safe to push the tail forward if
1207 * we can determine that the tail is within the range of the
1208 * torn write (e.g., the kernel can only overwrite the tail if
1209 * it has actually been pushed forward)? Alternatively, could we
1210 * somehow prevent this condition at runtime?
1211 */
1212 error = xlog_verify_tail(log, *head_blk, *tail_blk);
1213 }
1214
1215 return error;
1216}
1217
1218/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 * Find the sync block number or the tail of the log.
1220 *
1221 * This will be the block number of the last record to have its
1222 * associated buffers synced to disk. Every log record header has
1223 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
1224 * to get a sync block number. The only concern is to figure out which
1225 * log record header to believe.
1226 *
1227 * The following algorithm uses the log record header with the largest
1228 * lsn. The entire log record does not need to be valid. We only care
1229 * that the header is valid.
1230 *
1231 * We could speed up search by using current head_blk buffer, but it is not
1232 * available.
1233 */
Eric Sandeen5d77c0d2009-11-19 15:52:00 +00001234STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235xlog_find_tail(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001236 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 xfs_daddr_t *head_blk,
Eric Sandeen65be6052006-01-11 15:34:19 +11001238 xfs_daddr_t *tail_blk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239{
1240 xlog_rec_header_t *rhead;
1241 xlog_op_header_t *op_head;
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10001242 char *offset = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 xfs_buf_t *bp;
Brian Foster7088c412016-01-05 07:40:16 +11001244 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 xfs_daddr_t umount_data_blk;
1246 xfs_daddr_t after_umount_blk;
Brian Foster7088c412016-01-05 07:40:16 +11001247 xfs_daddr_t rhead_blk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 xfs_lsn_t tail_lsn;
1249 int hblks;
Brian Fostereed6b462016-01-04 15:55:10 +11001250 bool wrapped = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251
1252 /*
1253 * Find previous log record
1254 */
1255 if ((error = xlog_find_head(log, head_blk)))
1256 return error;
1257
1258 bp = xlog_get_bp(log, 1);
1259 if (!bp)
Dave Chinner24513372014-06-25 14:58:08 +10001260 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 if (*head_blk == 0) { /* special case */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001262 error = xlog_bread(log, 0, 1, bp, &offset);
1263 if (error)
Alex Elder9db127e2010-04-15 18:17:26 +00001264 goto done;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001265
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10001266 if (xlog_get_cycle(offset) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 *tail_blk = 0;
1268 /* leave all other log inited values alone */
Alex Elder9db127e2010-04-15 18:17:26 +00001269 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 }
1271 }
1272
1273 /*
Brian Foster7088c412016-01-05 07:40:16 +11001274 * Trim the head block back to skip over torn records. We can have
1275 * multiple log I/Os in flight at any time, so we assume CRC failures
1276 * back through the previous several records are torn writes and skip
1277 * them.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 */
1279 ASSERT(*head_blk < INT_MAX);
Brian Foster7088c412016-01-05 07:40:16 +11001280 error = xlog_verify_head(log, head_blk, tail_blk, bp, &rhead_blk,
1281 &rhead, &wrapped);
1282 if (error)
Brian Fostereed6b462016-01-04 15:55:10 +11001283 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284
1285 /*
1286 * Reset log values according to the state of the log when we
1287 * crashed. In the case where head_blk == 0, we bump curr_cycle
1288 * one because the next write starts a new cycle rather than
1289 * continuing the cycle of the last good log record. At this
1290 * point we have guaranteed that all partial log records have been
1291 * accounted for. Therefore, we know that the last good log record
1292 * written was complete and ended exactly on the end boundary
1293 * of the physical log.
1294 */
Brian Foster7088c412016-01-05 07:40:16 +11001295 log->l_prev_block = rhead_blk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 log->l_curr_block = (int)*head_blk;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001297 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
Brian Fostereed6b462016-01-04 15:55:10 +11001298 if (wrapped)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 log->l_curr_cycle++;
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001300 atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn));
Dave Chinner84f3c682010-12-03 22:11:29 +11001301 atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn));
Christoph Hellwig28496962012-02-20 02:31:25 +00001302 xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle,
Dave Chinnera69ed032010-12-21 12:08:20 +11001303 BBTOB(log->l_curr_block));
Christoph Hellwig28496962012-02-20 02:31:25 +00001304 xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle,
Dave Chinnera69ed032010-12-21 12:08:20 +11001305 BBTOB(log->l_curr_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306
1307 /*
1308 * Look for unmount record. If we find it, then we know there
1309 * was a clean unmount. Since 'i' could be the last block in
1310 * the physical log, we convert to a log block before comparing
1311 * to the head_blk.
1312 *
1313 * Save the current tail lsn to use to pass to
1314 * xlog_clear_stale_blocks() below. We won't want to clear the
1315 * unmount record if there is one, so we pass the lsn of the
1316 * unmount record rather than the block after it.
1317 */
Eric Sandeen62118702008-03-06 13:44:28 +11001318 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001319 int h_size = be32_to_cpu(rhead->h_size);
1320 int h_version = be32_to_cpu(rhead->h_version);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321
1322 if ((h_version & XLOG_VERSION_2) &&
1323 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
1324 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
1325 if (h_size % XLOG_HEADER_CYCLE_SIZE)
1326 hblks++;
1327 } else {
1328 hblks = 1;
1329 }
1330 } else {
1331 hblks = 1;
1332 }
Brian Foster7088c412016-01-05 07:40:16 +11001333 after_umount_blk = rhead_blk + hblks + BTOBB(be32_to_cpu(rhead->h_len));
1334 after_umount_blk = do_mod(after_umount_blk, log->l_logBBsize);
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001335 tail_lsn = atomic64_read(&log->l_tail_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 if (*head_blk == after_umount_blk &&
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001337 be32_to_cpu(rhead->h_num_logops) == 1) {
Brian Foster7088c412016-01-05 07:40:16 +11001338 umount_data_blk = rhead_blk + hblks;
1339 umount_data_blk = do_mod(umount_data_blk, log->l_logBBsize);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001340 error = xlog_bread(log, umount_data_blk, 1, bp, &offset);
1341 if (error)
Alex Elder9db127e2010-04-15 18:17:26 +00001342 goto done;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001343
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 op_head = (xlog_op_header_t *)offset;
1345 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
1346 /*
1347 * Set tail and last sync so that newly written
1348 * log records will point recovery to after the
1349 * current unmount record.
1350 */
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001351 xlog_assign_atomic_lsn(&log->l_tail_lsn,
1352 log->l_curr_cycle, after_umount_blk);
1353 xlog_assign_atomic_lsn(&log->l_last_sync_lsn,
1354 log->l_curr_cycle, after_umount_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 *tail_blk = after_umount_blk;
David Chinner92821e22007-05-24 15:26:31 +10001356
1357 /*
1358 * Note that the unmount was clean. If the unmount
1359 * was not clean, we need to know this to rebuild the
1360 * superblock counters from the perag headers if we
1361 * have a filesystem using non-persistent counters.
1362 */
1363 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 }
1365 }
1366
1367 /*
1368 * Make sure that there are no blocks in front of the head
1369 * with the same cycle number as the head. This can happen
1370 * because we allow multiple outstanding log writes concurrently,
1371 * and the later writes might make it out before earlier ones.
1372 *
1373 * We use the lsn from before modifying it so that we'll never
1374 * overwrite the unmount record after a clean unmount.
1375 *
1376 * Do this only if we are going to recover the filesystem
1377 *
1378 * NOTE: This used to say "if (!readonly)"
1379 * However on Linux, we can & do recover a read-only filesystem.
1380 * We only skip recovery if NORECOVERY is specified on mount,
1381 * in which case we would not be here.
1382 *
1383 * But... if the -device- itself is readonly, just skip this.
1384 * We can't recover this device anyway, so it won't matter.
1385 */
Alex Elder9db127e2010-04-15 18:17:26 +00001386 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 error = xlog_clear_stale_blocks(log, tail_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388
Alex Elder9db127e2010-04-15 18:17:26 +00001389done:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 xlog_put_bp(bp);
1391
1392 if (error)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001393 xfs_warn(log->l_mp, "failed to locate log tail");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 return error;
1395}
1396
1397/*
1398 * Is the log zeroed at all?
1399 *
1400 * The last binary search should be changed to perform an X block read
1401 * once X becomes small enough. You can then search linearly through
1402 * the X blocks. This will cut down on the number of reads we need to do.
1403 *
1404 * If the log is partially zeroed, this routine will pass back the blkno
1405 * of the first block with cycle number 0. It won't have a complete LR
1406 * preceding it.
1407 *
1408 * Return:
1409 * 0 => the log is completely written to
Dave Chinner24513372014-06-25 14:58:08 +10001410 * 1 => use *blk_no as the first block of the log
1411 * <0 => error has occurred
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 */
David Chinnera8272ce2007-11-23 16:28:09 +11001413STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414xlog_find_zeroed(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001415 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 xfs_daddr_t *blk_no)
1417{
1418 xfs_buf_t *bp;
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10001419 char *offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 uint first_cycle, last_cycle;
1421 xfs_daddr_t new_blk, last_blk, start_blk;
1422 xfs_daddr_t num_scan_bblks;
1423 int error, log_bbnum = log->l_logBBsize;
1424
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10001425 *blk_no = 0;
1426
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 /* check totally zeroed log */
1428 bp = xlog_get_bp(log, 1);
1429 if (!bp)
Dave Chinner24513372014-06-25 14:58:08 +10001430 return -ENOMEM;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001431 error = xlog_bread(log, 0, 1, bp, &offset);
1432 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 goto bp_err;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001434
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10001435 first_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 if (first_cycle == 0) { /* completely zeroed log */
1437 *blk_no = 0;
1438 xlog_put_bp(bp);
Dave Chinner24513372014-06-25 14:58:08 +10001439 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 }
1441
1442 /* check partially zeroed log */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001443 error = xlog_bread(log, log_bbnum-1, 1, bp, &offset);
1444 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 goto bp_err;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001446
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10001447 last_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 if (last_cycle != 0) { /* log completely written to */
1449 xlog_put_bp(bp);
1450 return 0;
1451 } else if (first_cycle != 1) {
1452 /*
1453 * If the cycle of the last block is zero, the cycle of
1454 * the first block must be 1. If it's not, maybe we're
1455 * not looking at a log... Bail out.
1456 */
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001457 xfs_warn(log->l_mp,
1458 "Log inconsistent or not a log (last==0, first!=1)");
Dave Chinner24513372014-06-25 14:58:08 +10001459 error = -EINVAL;
Eric Sandeen5d0a6542013-07-31 20:32:30 -05001460 goto bp_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 }
1462
1463 /* we have a partially zeroed log */
1464 last_blk = log_bbnum-1;
1465 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1466 goto bp_err;
1467
1468 /*
1469 * Validate the answer. Because there is no way to guarantee that
1470 * the entire log is made up of log records which are the same size,
1471 * we scan over the defined maximum blocks. At this point, the maximum
1472 * is not chosen to mean anything special. XXXmiken
1473 */
1474 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1475 ASSERT(num_scan_bblks <= INT_MAX);
1476
1477 if (last_blk < num_scan_bblks)
1478 num_scan_bblks = last_blk;
1479 start_blk = last_blk - num_scan_bblks;
1480
1481 /*
1482 * We search for any instances of cycle number 0 that occur before
1483 * our current estimate of the head. What we're trying to detect is
1484 * 1 ... | 0 | 1 | 0...
1485 * ^ binary search ends here
1486 */
1487 if ((error = xlog_find_verify_cycle(log, start_blk,
1488 (int)num_scan_bblks, 0, &new_blk)))
1489 goto bp_err;
1490 if (new_blk != -1)
1491 last_blk = new_blk;
1492
1493 /*
1494 * Potentially backup over partial log record write. We don't need
1495 * to search the end of the log because we know it is zero.
1496 */
Dave Chinner24513372014-06-25 14:58:08 +10001497 error = xlog_find_verify_log_record(log, start_blk, &last_blk, 0);
1498 if (error == 1)
1499 error = -EIO;
1500 if (error)
1501 goto bp_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502
1503 *blk_no = last_blk;
1504bp_err:
1505 xlog_put_bp(bp);
1506 if (error)
1507 return error;
Dave Chinner24513372014-06-25 14:58:08 +10001508 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509}
1510
1511/*
1512 * These are simple subroutines used by xlog_clear_stale_blocks() below
1513 * to initialize a buffer full of empty log record headers and write
1514 * them into the log.
1515 */
1516STATIC void
1517xlog_add_record(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001518 struct xlog *log,
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10001519 char *buf,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 int cycle,
1521 int block,
1522 int tail_cycle,
1523 int tail_block)
1524{
1525 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1526
1527 memset(buf, 0, BBSIZE);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001528 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1529 recp->h_cycle = cpu_to_be32(cycle);
1530 recp->h_version = cpu_to_be32(
Eric Sandeen62118702008-03-06 13:44:28 +11001531 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001532 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1533 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1534 recp->h_fmt = cpu_to_be32(XLOG_FMT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1536}
1537
1538STATIC int
1539xlog_write_log_records(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001540 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 int cycle,
1542 int start_block,
1543 int blocks,
1544 int tail_cycle,
1545 int tail_block)
1546{
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10001547 char *offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 xfs_buf_t *bp;
1549 int balign, ealign;
Alex Elder69ce58f2010-04-20 17:09:59 +10001550 int sectbb = log->l_sectBBsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 int end_block = start_block + blocks;
1552 int bufblks;
1553 int error = 0;
1554 int i, j = 0;
1555
Alex Elder6881a222010-04-13 15:22:29 +10001556 /*
1557 * Greedily allocate a buffer big enough to handle the full
1558 * range of basic blocks to be written. If that fails, try
1559 * a smaller size. We need to be able to write at least a
1560 * log sector, or we're out of luck.
1561 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 bufblks = 1 << ffs(blocks);
Dave Chinner81158e02012-04-27 19:45:22 +10001563 while (bufblks > log->l_logBBsize)
1564 bufblks >>= 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 while (!(bp = xlog_get_bp(log, bufblks))) {
1566 bufblks >>= 1;
Alex Elder69ce58f2010-04-20 17:09:59 +10001567 if (bufblks < sectbb)
Dave Chinner24513372014-06-25 14:58:08 +10001568 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 }
1570
1571 /* We may need to do a read at the start to fill in part of
1572 * the buffer in the starting sector not covered by the first
1573 * write below.
1574 */
Alex Elder5c17f532010-04-13 15:22:48 +10001575 balign = round_down(start_block, sectbb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 if (balign != start_block) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001577 error = xlog_bread_noalign(log, start_block, 1, bp);
1578 if (error)
1579 goto out_put_bp;
1580
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 j = start_block - balign;
1582 }
1583
1584 for (i = start_block; i < end_block; i += bufblks) {
1585 int bcount, endcount;
1586
1587 bcount = min(bufblks, end_block - start_block);
1588 endcount = bcount - j;
1589
1590 /* We may need to do a read at the end to fill in part of
1591 * the buffer in the final sector not covered by the write.
1592 * If this is the same sector as the above read, skip it.
1593 */
Alex Elder5c17f532010-04-13 15:22:48 +10001594 ealign = round_down(end_block, sectbb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 if (j == 0 && (start_block + endcount > ealign)) {
Chandra Seetharaman62926042011-07-22 23:40:15 +00001596 offset = bp->b_addr + BBTOB(ealign - start_block);
Dave Chinner44396472011-04-21 09:34:27 +00001597 error = xlog_bread_offset(log, ealign, sectbb,
1598 bp, offset);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001599 if (error)
1600 break;
1601
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602 }
1603
1604 offset = xlog_align(log, start_block, endcount, bp);
1605 for (; j < endcount; j++) {
1606 xlog_add_record(log, offset, cycle, i+j,
1607 tail_cycle, tail_block);
1608 offset += BBSIZE;
1609 }
1610 error = xlog_bwrite(log, start_block, endcount, bp);
1611 if (error)
1612 break;
1613 start_block += endcount;
1614 j = 0;
1615 }
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001616
1617 out_put_bp:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 xlog_put_bp(bp);
1619 return error;
1620}
1621
1622/*
1623 * This routine is called to blow away any incomplete log writes out
1624 * in front of the log head. We do this so that we won't become confused
1625 * if we come up, write only a little bit more, and then crash again.
1626 * If we leave the partial log records out there, this situation could
1627 * cause us to think those partial writes are valid blocks since they
1628 * have the current cycle number. We get rid of them by overwriting them
1629 * with empty log records with the old cycle number rather than the
1630 * current one.
1631 *
1632 * The tail lsn is passed in rather than taken from
1633 * the log so that we will not write over the unmount record after a
1634 * clean unmount in a 512 block log. Doing so would leave the log without
1635 * any valid log records in it until a new one was written. If we crashed
1636 * during that time we would not be able to recover.
1637 */
1638STATIC int
1639xlog_clear_stale_blocks(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001640 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 xfs_lsn_t tail_lsn)
1642{
1643 int tail_cycle, head_cycle;
1644 int tail_block, head_block;
1645 int tail_distance, max_distance;
1646 int distance;
1647 int error;
1648
1649 tail_cycle = CYCLE_LSN(tail_lsn);
1650 tail_block = BLOCK_LSN(tail_lsn);
1651 head_cycle = log->l_curr_cycle;
1652 head_block = log->l_curr_block;
1653
1654 /*
1655 * Figure out the distance between the new head of the log
1656 * and the tail. We want to write over any blocks beyond the
1657 * head that we may have written just before the crash, but
1658 * we don't want to overwrite the tail of the log.
1659 */
1660 if (head_cycle == tail_cycle) {
1661 /*
1662 * The tail is behind the head in the physical log,
1663 * so the distance from the head to the tail is the
1664 * distance from the head to the end of the log plus
1665 * the distance from the beginning of the log to the
1666 * tail.
1667 */
1668 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1669 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1670 XFS_ERRLEVEL_LOW, log->l_mp);
Dave Chinner24513372014-06-25 14:58:08 +10001671 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 }
1673 tail_distance = tail_block + (log->l_logBBsize - head_block);
1674 } else {
1675 /*
1676 * The head is behind the tail in the physical log,
1677 * so the distance from the head to the tail is just
1678 * the tail block minus the head block.
1679 */
1680 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1681 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1682 XFS_ERRLEVEL_LOW, log->l_mp);
Dave Chinner24513372014-06-25 14:58:08 +10001683 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 }
1685 tail_distance = tail_block - head_block;
1686 }
1687
1688 /*
1689 * If the head is right up against the tail, we can't clear
1690 * anything.
1691 */
1692 if (tail_distance <= 0) {
1693 ASSERT(tail_distance == 0);
1694 return 0;
1695 }
1696
1697 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1698 /*
1699 * Take the smaller of the maximum amount of outstanding I/O
1700 * we could have and the distance to the tail to clear out.
1701 * We take the smaller so that we don't overwrite the tail and
1702 * we don't waste all day writing from the head to the tail
1703 * for no reason.
1704 */
1705 max_distance = MIN(max_distance, tail_distance);
1706
1707 if ((head_block + max_distance) <= log->l_logBBsize) {
1708 /*
1709 * We can stomp all the blocks we need to without
1710 * wrapping around the end of the log. Just do it
1711 * in a single write. Use the cycle number of the
1712 * current cycle minus one so that the log will look like:
1713 * n ... | n - 1 ...
1714 */
1715 error = xlog_write_log_records(log, (head_cycle - 1),
1716 head_block, max_distance, tail_cycle,
1717 tail_block);
1718 if (error)
1719 return error;
1720 } else {
1721 /*
1722 * We need to wrap around the end of the physical log in
1723 * order to clear all the blocks. Do it in two separate
1724 * I/Os. The first write should be from the head to the
1725 * end of the physical log, and it should use the current
1726 * cycle number minus one just like above.
1727 */
1728 distance = log->l_logBBsize - head_block;
1729 error = xlog_write_log_records(log, (head_cycle - 1),
1730 head_block, distance, tail_cycle,
1731 tail_block);
1732
1733 if (error)
1734 return error;
1735
1736 /*
1737 * Now write the blocks at the start of the physical log.
1738 * This writes the remainder of the blocks we want to clear.
1739 * It uses the current cycle number since we're now on the
1740 * same cycle as the head so that we get:
1741 * n ... n ... | n - 1 ...
1742 * ^^^^^ blocks we're writing
1743 */
1744 distance = max_distance - (log->l_logBBsize - head_block);
1745 error = xlog_write_log_records(log, head_cycle, 0, distance,
1746 tail_cycle, tail_block);
1747 if (error)
1748 return error;
1749 }
1750
1751 return 0;
1752}
1753
1754/******************************************************************************
1755 *
1756 * Log recover routines
1757 *
1758 ******************************************************************************
1759 */
1760
Dave Chinnerf0a76952010-01-11 11:49:57 +00001761/*
Dave Chinnera775ad72013-06-05 12:09:07 +10001762 * Sort the log items in the transaction.
1763 *
1764 * The ordering constraints are defined by the inode allocation and unlink
1765 * behaviour. The rules are:
1766 *
1767 * 1. Every item is only logged once in a given transaction. Hence it
1768 * represents the last logged state of the item. Hence ordering is
1769 * dependent on the order in which operations need to be performed so
1770 * required initial conditions are always met.
1771 *
1772 * 2. Cancelled buffers are recorded in pass 1 in a separate table and
1773 * there's nothing to replay from them so we can simply cull them
1774 * from the transaction. However, we can't do that until after we've
1775 * replayed all the other items because they may be dependent on the
1776 * cancelled buffer and replaying the cancelled buffer can remove it
1777 * form the cancelled buffer table. Hence they have tobe done last.
1778 *
1779 * 3. Inode allocation buffers must be replayed before inode items that
Dave Chinner28c8e412013-06-27 16:04:55 +10001780 * read the buffer and replay changes into it. For filesystems using the
1781 * ICREATE transactions, this means XFS_LI_ICREATE objects need to get
1782 * treated the same as inode allocation buffers as they create and
1783 * initialise the buffers directly.
Dave Chinnera775ad72013-06-05 12:09:07 +10001784 *
1785 * 4. Inode unlink buffers must be replayed after inode items are replayed.
1786 * This ensures that inodes are completely flushed to the inode buffer
1787 * in a "free" state before we remove the unlinked inode list pointer.
1788 *
1789 * Hence the ordering needs to be inode allocation buffers first, inode items
1790 * second, inode unlink buffers third and cancelled buffers last.
1791 *
1792 * But there's a problem with that - we can't tell an inode allocation buffer
1793 * apart from a regular buffer, so we can't separate them. We can, however,
1794 * tell an inode unlink buffer from the others, and so we can separate them out
1795 * from all the other buffers and move them to last.
1796 *
1797 * Hence, 4 lists, in order from head to tail:
Dave Chinner28c8e412013-06-27 16:04:55 +10001798 * - buffer_list for all buffers except cancelled/inode unlink buffers
1799 * - item_list for all non-buffer items
1800 * - inode_buffer_list for inode unlink buffers
1801 * - cancel_list for the cancelled buffers
1802 *
1803 * Note that we add objects to the tail of the lists so that first-to-last
1804 * ordering is preserved within the lists. Adding objects to the head of the
1805 * list means when we traverse from the head we walk them in last-to-first
1806 * order. For cancelled buffers and inode unlink buffers this doesn't matter,
1807 * but for all other items there may be specific ordering that we need to
1808 * preserve.
Dave Chinnerf0a76952010-01-11 11:49:57 +00001809 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810STATIC int
1811xlog_recover_reorder_trans(
Mark Tinguelyad223e62012-06-14 09:22:15 -05001812 struct xlog *log,
1813 struct xlog_recover *trans,
Dave Chinner9abbc532010-04-13 15:06:46 +10001814 int pass)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815{
Dave Chinnerf0a76952010-01-11 11:49:57 +00001816 xlog_recover_item_t *item, *n;
Mark Tinguely2a841082013-10-02 07:51:12 -05001817 int error = 0;
Dave Chinnerf0a76952010-01-11 11:49:57 +00001818 LIST_HEAD(sort_list);
Dave Chinnera775ad72013-06-05 12:09:07 +10001819 LIST_HEAD(cancel_list);
1820 LIST_HEAD(buffer_list);
1821 LIST_HEAD(inode_buffer_list);
1822 LIST_HEAD(inode_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823
Dave Chinnerf0a76952010-01-11 11:49:57 +00001824 list_splice_init(&trans->r_itemq, &sort_list);
1825 list_for_each_entry_safe(item, n, &sort_list, ri_list) {
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10001826 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
Dave Chinnerf0a76952010-01-11 11:49:57 +00001827
1828 switch (ITEM_TYPE(item)) {
Dave Chinner28c8e412013-06-27 16:04:55 +10001829 case XFS_LI_ICREATE:
1830 list_move_tail(&item->ri_list, &buffer_list);
1831 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 case XFS_LI_BUF:
Dave Chinnera775ad72013-06-05 12:09:07 +10001833 if (buf_f->blf_flags & XFS_BLF_CANCEL) {
Dave Chinner9abbc532010-04-13 15:06:46 +10001834 trace_xfs_log_recover_item_reorder_head(log,
1835 trans, item, pass);
Dave Chinnera775ad72013-06-05 12:09:07 +10001836 list_move(&item->ri_list, &cancel_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 break;
1838 }
Dave Chinnera775ad72013-06-05 12:09:07 +10001839 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
1840 list_move(&item->ri_list, &inode_buffer_list);
1841 break;
1842 }
1843 list_move_tail(&item->ri_list, &buffer_list);
1844 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 case XFS_LI_INODE:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846 case XFS_LI_DQUOT:
1847 case XFS_LI_QUOTAOFF:
1848 case XFS_LI_EFD:
1849 case XFS_LI_EFI:
Dave Chinner9abbc532010-04-13 15:06:46 +10001850 trace_xfs_log_recover_item_reorder_tail(log,
1851 trans, item, pass);
Dave Chinnera775ad72013-06-05 12:09:07 +10001852 list_move_tail(&item->ri_list, &inode_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 break;
1854 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001855 xfs_warn(log->l_mp,
1856 "%s: unrecognized type of log operation",
1857 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 ASSERT(0);
Mark Tinguely2a841082013-10-02 07:51:12 -05001859 /*
1860 * return the remaining items back to the transaction
1861 * item list so they can be freed in caller.
1862 */
1863 if (!list_empty(&sort_list))
1864 list_splice_init(&sort_list, &trans->r_itemq);
Dave Chinner24513372014-06-25 14:58:08 +10001865 error = -EIO;
Mark Tinguely2a841082013-10-02 07:51:12 -05001866 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 }
Dave Chinnerf0a76952010-01-11 11:49:57 +00001868 }
Mark Tinguely2a841082013-10-02 07:51:12 -05001869out:
Dave Chinnerf0a76952010-01-11 11:49:57 +00001870 ASSERT(list_empty(&sort_list));
Dave Chinnera775ad72013-06-05 12:09:07 +10001871 if (!list_empty(&buffer_list))
1872 list_splice(&buffer_list, &trans->r_itemq);
1873 if (!list_empty(&inode_list))
1874 list_splice_tail(&inode_list, &trans->r_itemq);
1875 if (!list_empty(&inode_buffer_list))
1876 list_splice_tail(&inode_buffer_list, &trans->r_itemq);
1877 if (!list_empty(&cancel_list))
1878 list_splice_tail(&cancel_list, &trans->r_itemq);
Mark Tinguely2a841082013-10-02 07:51:12 -05001879 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880}
1881
1882/*
1883 * Build up the table of buf cancel records so that we don't replay
1884 * cancelled data in the second pass. For buffer records that are
1885 * not cancel records, there is nothing to do here so we just return.
1886 *
1887 * If we get a cancel record which is already in the table, this indicates
1888 * that the buffer was cancelled multiple times. In order to ensure
1889 * that during pass 2 we keep the record in the table until we reach its
1890 * last occurrence in the log, we keep a reference count in the cancel
1891 * record in the table to tell us how many times we expect to see this
1892 * record during the second pass.
1893 */
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001894STATIC int
1895xlog_recover_buffer_pass1(
Mark Tinguelyad223e62012-06-14 09:22:15 -05001896 struct xlog *log,
1897 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001899 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001900 struct list_head *bucket;
1901 struct xfs_buf_cancel *bcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902
1903 /*
1904 * If this isn't a cancel buffer item, then just return.
1905 */
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001906 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
Dave Chinner9abbc532010-04-13 15:06:46 +10001907 trace_xfs_log_recover_buf_not_cancel(log, buf_f);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001908 return 0;
Dave Chinner9abbc532010-04-13 15:06:46 +10001909 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910
1911 /*
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001912 * Insert an xfs_buf_cancel record into the hash table of them.
1913 * If there is already an identical record, bump its reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 */
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001915 bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno);
1916 list_for_each_entry(bcp, bucket, bc_list) {
1917 if (bcp->bc_blkno == buf_f->blf_blkno &&
1918 bcp->bc_len == buf_f->blf_len) {
1919 bcp->bc_refcount++;
Dave Chinner9abbc532010-04-13 15:06:46 +10001920 trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001921 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 }
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001924
1925 bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP);
1926 bcp->bc_blkno = buf_f->blf_blkno;
1927 bcp->bc_len = buf_f->blf_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928 bcp->bc_refcount = 1;
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001929 list_add_tail(&bcp->bc_list, bucket);
1930
Dave Chinner9abbc532010-04-13 15:06:46 +10001931 trace_xfs_log_recover_buf_cancel_add(log, buf_f);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001932 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933}
1934
1935/*
1936 * Check to see whether the buffer being recovered has a corresponding
Dave Chinner84a5b732013-08-27 08:10:53 +10001937 * entry in the buffer cancel record table. If it is, return the cancel
1938 * buffer structure to the caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 */
Dave Chinner84a5b732013-08-27 08:10:53 +10001940STATIC struct xfs_buf_cancel *
1941xlog_peek_buffer_cancelled(
Mark Tinguelyad223e62012-06-14 09:22:15 -05001942 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 xfs_daddr_t blkno,
1944 uint len,
1945 ushort flags)
1946{
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001947 struct list_head *bucket;
1948 struct xfs_buf_cancel *bcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949
Dave Chinner84a5b732013-08-27 08:10:53 +10001950 if (!log->l_buf_cancel_table) {
1951 /* empty table means no cancelled buffers in the log */
Dave Chinnerc1155412010-05-07 11:05:19 +10001952 ASSERT(!(flags & XFS_BLF_CANCEL));
Dave Chinner84a5b732013-08-27 08:10:53 +10001953 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 }
1955
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001956 bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno);
1957 list_for_each_entry(bcp, bucket, bc_list) {
1958 if (bcp->bc_blkno == blkno && bcp->bc_len == len)
Dave Chinner84a5b732013-08-27 08:10:53 +10001959 return bcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 }
1961
1962 /*
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001963 * We didn't find a corresponding entry in the table, so return 0 so
1964 * that the buffer is NOT cancelled.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 */
Dave Chinnerc1155412010-05-07 11:05:19 +10001966 ASSERT(!(flags & XFS_BLF_CANCEL));
Dave Chinner84a5b732013-08-27 08:10:53 +10001967 return NULL;
1968}
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001969
Dave Chinner84a5b732013-08-27 08:10:53 +10001970/*
1971 * If the buffer is being cancelled then return 1 so that it will be cancelled,
1972 * otherwise return 0. If the buffer is actually a buffer cancel item
1973 * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the
1974 * table and remove it from the table if this is the last reference.
1975 *
1976 * We remove the cancel record from the table when we encounter its last
1977 * occurrence in the log so that if the same buffer is re-used again after its
1978 * last cancellation we actually replay the changes made at that point.
1979 */
1980STATIC int
1981xlog_check_buffer_cancelled(
1982 struct xlog *log,
1983 xfs_daddr_t blkno,
1984 uint len,
1985 ushort flags)
1986{
1987 struct xfs_buf_cancel *bcp;
1988
1989 bcp = xlog_peek_buffer_cancelled(log, blkno, len, flags);
1990 if (!bcp)
1991 return 0;
1992
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001993 /*
1994 * We've go a match, so return 1 so that the recovery of this buffer
1995 * is cancelled. If this buffer is actually a buffer cancel log
1996 * item, then decrement the refcount on the one in the table and
1997 * remove it if this is the last reference.
1998 */
1999 if (flags & XFS_BLF_CANCEL) {
2000 if (--bcp->bc_refcount == 0) {
2001 list_del(&bcp->bc_list);
2002 kmem_free(bcp);
2003 }
2004 }
2005 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006}
2007
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008/*
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002009 * Perform recovery for a buffer full of inodes. In these buffers, the only
2010 * data which should be recovered is that which corresponds to the
2011 * di_next_unlinked pointers in the on disk inode structures. The rest of the
2012 * data for the inodes is always logged through the inodes themselves rather
2013 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 *
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002015 * The only time when buffers full of inodes are fully recovered is when the
2016 * buffer is full of newly allocated inodes. In this case the buffer will
2017 * not be marked as an inode buffer and so will be sent to
2018 * xlog_recover_do_reg_buffer() below during recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 */
2020STATIC int
2021xlog_recover_do_inode_buffer(
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002022 struct xfs_mount *mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 xlog_recover_item_t *item,
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002024 struct xfs_buf *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 xfs_buf_log_format_t *buf_f)
2026{
2027 int i;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002028 int item_index = 0;
2029 int bit = 0;
2030 int nbits = 0;
2031 int reg_buf_offset = 0;
2032 int reg_buf_bytes = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 int next_unlinked_offset;
2034 int inodes_per_buf;
2035 xfs_agino_t *logged_nextp;
2036 xfs_agino_t *buffer_nextp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037
Dave Chinner9abbc532010-04-13 15:06:46 +10002038 trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f);
Dave Chinner9222a9c2013-06-12 12:19:06 +10002039
2040 /*
2041 * Post recovery validation only works properly on CRC enabled
2042 * filesystems.
2043 */
2044 if (xfs_sb_version_hascrc(&mp->m_sb))
2045 bp->b_ops = &xfs_inode_buf_ops;
Dave Chinner9abbc532010-04-13 15:06:46 +10002046
Dave Chinneraa0e8832012-04-23 15:58:52 +10002047 inodes_per_buf = BBTOB(bp->b_io_length) >> mp->m_sb.sb_inodelog;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 for (i = 0; i < inodes_per_buf; i++) {
2049 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
2050 offsetof(xfs_dinode_t, di_next_unlinked);
2051
2052 while (next_unlinked_offset >=
2053 (reg_buf_offset + reg_buf_bytes)) {
2054 /*
2055 * The next di_next_unlinked field is beyond
2056 * the current logged region. Find the next
2057 * logged region that contains or is beyond
2058 * the current di_next_unlinked field.
2059 */
2060 bit += nbits;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002061 bit = xfs_next_bit(buf_f->blf_data_map,
2062 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063
2064 /*
2065 * If there are no more logged regions in the
2066 * buffer, then we're done.
2067 */
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002068 if (bit == -1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002071 nbits = xfs_contig_bits(buf_f->blf_data_map,
2072 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073 ASSERT(nbits > 0);
Dave Chinnerc1155412010-05-07 11:05:19 +10002074 reg_buf_offset = bit << XFS_BLF_SHIFT;
2075 reg_buf_bytes = nbits << XFS_BLF_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 item_index++;
2077 }
2078
2079 /*
2080 * If the current logged region starts after the current
2081 * di_next_unlinked field, then move on to the next
2082 * di_next_unlinked field.
2083 */
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002084 if (next_unlinked_offset < reg_buf_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086
2087 ASSERT(item->ri_buf[item_index].i_addr != NULL);
Dave Chinnerc1155412010-05-07 11:05:19 +10002088 ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0);
Dave Chinneraa0e8832012-04-23 15:58:52 +10002089 ASSERT((reg_buf_offset + reg_buf_bytes) <=
2090 BBTOB(bp->b_io_length));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091
2092 /*
2093 * The current logged region contains a copy of the
2094 * current di_next_unlinked field. Extract its value
2095 * and copy it to the buffer copy.
2096 */
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002097 logged_nextp = item->ri_buf[item_index].i_addr +
2098 next_unlinked_offset - reg_buf_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 if (unlikely(*logged_nextp == 0)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002100 xfs_alert(mp,
2101 "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
2102 "Trying to replay bad (0) inode di_next_unlinked field.",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 item, bp);
2104 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
2105 XFS_ERRLEVEL_LOW, mp);
Dave Chinner24513372014-06-25 14:58:08 +10002106 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107 }
2108
Christoph Hellwig88ee2df2015-06-22 09:44:29 +10002109 buffer_nextp = xfs_buf_offset(bp, next_unlinked_offset);
Tim Shimmin87c199c2006-06-09 14:56:16 +10002110 *buffer_nextp = *logged_nextp;
Dave Chinner0a32c262013-06-05 12:09:08 +10002111
2112 /*
2113 * If necessary, recalculate the CRC in the on-disk inode. We
2114 * have to leave the inode in a consistent state for whoever
2115 * reads it next....
2116 */
Christoph Hellwig88ee2df2015-06-22 09:44:29 +10002117 xfs_dinode_calc_crc(mp,
Dave Chinner0a32c262013-06-05 12:09:08 +10002118 xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize));
2119
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120 }
2121
2122 return 0;
2123}
2124
2125/*
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002126 * V5 filesystems know the age of the buffer on disk being recovered. We can
2127 * have newer objects on disk than we are replaying, and so for these cases we
2128 * don't want to replay the current change as that will make the buffer contents
2129 * temporarily invalid on disk.
2130 *
2131 * The magic number might not match the buffer type we are going to recover
2132 * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags. Hence
2133 * extract the LSN of the existing object in the buffer based on it's current
2134 * magic number. If we don't recognise the magic number in the buffer, then
2135 * return a LSN of -1 so that the caller knows it was an unrecognised block and
2136 * so can recover the buffer.
Dave Chinner566055d2013-09-24 16:01:16 +10002137 *
2138 * Note: we cannot rely solely on magic number matches to determine that the
2139 * buffer has a valid LSN - we also need to verify that it belongs to this
2140 * filesystem, so we need to extract the object's LSN and compare it to that
2141 * which we read from the superblock. If the UUIDs don't match, then we've got a
2142 * stale metadata block from an old filesystem instance that we need to recover
2143 * over the top of.
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002144 */
2145static xfs_lsn_t
2146xlog_recover_get_buf_lsn(
2147 struct xfs_mount *mp,
2148 struct xfs_buf *bp)
2149{
2150 __uint32_t magic32;
2151 __uint16_t magic16;
2152 __uint16_t magicda;
2153 void *blk = bp->b_addr;
Dave Chinner566055d2013-09-24 16:01:16 +10002154 uuid_t *uuid;
2155 xfs_lsn_t lsn = -1;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002156
2157 /* v4 filesystems always recover immediately */
2158 if (!xfs_sb_version_hascrc(&mp->m_sb))
2159 goto recover_immediately;
2160
2161 magic32 = be32_to_cpu(*(__be32 *)blk);
2162 switch (magic32) {
2163 case XFS_ABTB_CRC_MAGIC:
2164 case XFS_ABTC_CRC_MAGIC:
2165 case XFS_ABTB_MAGIC:
2166 case XFS_ABTC_MAGIC:
2167 case XFS_IBT_CRC_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10002168 case XFS_IBT_MAGIC: {
2169 struct xfs_btree_block *btb = blk;
2170
2171 lsn = be64_to_cpu(btb->bb_u.s.bb_lsn);
2172 uuid = &btb->bb_u.s.bb_uuid;
2173 break;
2174 }
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002175 case XFS_BMAP_CRC_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10002176 case XFS_BMAP_MAGIC: {
2177 struct xfs_btree_block *btb = blk;
2178
2179 lsn = be64_to_cpu(btb->bb_u.l.bb_lsn);
2180 uuid = &btb->bb_u.l.bb_uuid;
2181 break;
2182 }
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002183 case XFS_AGF_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10002184 lsn = be64_to_cpu(((struct xfs_agf *)blk)->agf_lsn);
2185 uuid = &((struct xfs_agf *)blk)->agf_uuid;
2186 break;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002187 case XFS_AGFL_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10002188 lsn = be64_to_cpu(((struct xfs_agfl *)blk)->agfl_lsn);
2189 uuid = &((struct xfs_agfl *)blk)->agfl_uuid;
2190 break;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002191 case XFS_AGI_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10002192 lsn = be64_to_cpu(((struct xfs_agi *)blk)->agi_lsn);
2193 uuid = &((struct xfs_agi *)blk)->agi_uuid;
2194 break;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002195 case XFS_SYMLINK_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10002196 lsn = be64_to_cpu(((struct xfs_dsymlink_hdr *)blk)->sl_lsn);
2197 uuid = &((struct xfs_dsymlink_hdr *)blk)->sl_uuid;
2198 break;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002199 case XFS_DIR3_BLOCK_MAGIC:
2200 case XFS_DIR3_DATA_MAGIC:
2201 case XFS_DIR3_FREE_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10002202 lsn = be64_to_cpu(((struct xfs_dir3_blk_hdr *)blk)->lsn);
2203 uuid = &((struct xfs_dir3_blk_hdr *)blk)->uuid;
2204 break;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002205 case XFS_ATTR3_RMT_MAGIC:
Dave Chinnere3c32ee2015-07-29 11:48:01 +10002206 /*
2207 * Remote attr blocks are written synchronously, rather than
2208 * being logged. That means they do not contain a valid LSN
2209 * (i.e. transactionally ordered) in them, and hence any time we
2210 * see a buffer to replay over the top of a remote attribute
2211 * block we should simply do so.
2212 */
2213 goto recover_immediately;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002214 case XFS_SB_MAGIC:
Dave Chinnerfcfbe2c2015-08-19 10:31:54 +10002215 /*
2216 * superblock uuids are magic. We may or may not have a
2217 * sb_meta_uuid on disk, but it will be set in the in-core
2218 * superblock. We set the uuid pointer for verification
2219 * according to the superblock feature mask to ensure we check
2220 * the relevant UUID in the superblock.
2221 */
Dave Chinner566055d2013-09-24 16:01:16 +10002222 lsn = be64_to_cpu(((struct xfs_dsb *)blk)->sb_lsn);
Dave Chinnerfcfbe2c2015-08-19 10:31:54 +10002223 if (xfs_sb_version_hasmetauuid(&mp->m_sb))
2224 uuid = &((struct xfs_dsb *)blk)->sb_meta_uuid;
2225 else
2226 uuid = &((struct xfs_dsb *)blk)->sb_uuid;
Dave Chinner566055d2013-09-24 16:01:16 +10002227 break;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002228 default:
2229 break;
2230 }
2231
Dave Chinner566055d2013-09-24 16:01:16 +10002232 if (lsn != (xfs_lsn_t)-1) {
Dave Chinnerfcfbe2c2015-08-19 10:31:54 +10002233 if (!uuid_equal(&mp->m_sb.sb_meta_uuid, uuid))
Dave Chinner566055d2013-09-24 16:01:16 +10002234 goto recover_immediately;
2235 return lsn;
2236 }
2237
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002238 magicda = be16_to_cpu(((struct xfs_da_blkinfo *)blk)->magic);
2239 switch (magicda) {
2240 case XFS_DIR3_LEAF1_MAGIC:
2241 case XFS_DIR3_LEAFN_MAGIC:
2242 case XFS_DA3_NODE_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10002243 lsn = be64_to_cpu(((struct xfs_da3_blkinfo *)blk)->lsn);
2244 uuid = &((struct xfs_da3_blkinfo *)blk)->uuid;
2245 break;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002246 default:
2247 break;
2248 }
2249
Dave Chinner566055d2013-09-24 16:01:16 +10002250 if (lsn != (xfs_lsn_t)-1) {
2251 if (!uuid_equal(&mp->m_sb.sb_uuid, uuid))
2252 goto recover_immediately;
2253 return lsn;
2254 }
2255
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002256 /*
2257 * We do individual object checks on dquot and inode buffers as they
2258 * have their own individual LSN records. Also, we could have a stale
2259 * buffer here, so we have to at least recognise these buffer types.
2260 *
2261 * A notd complexity here is inode unlinked list processing - it logs
2262 * the inode directly in the buffer, but we don't know which inodes have
2263 * been modified, and there is no global buffer LSN. Hence we need to
2264 * recover all inode buffer types immediately. This problem will be
2265 * fixed by logical logging of the unlinked list modifications.
2266 */
2267 magic16 = be16_to_cpu(*(__be16 *)blk);
2268 switch (magic16) {
2269 case XFS_DQUOT_MAGIC:
2270 case XFS_DINODE_MAGIC:
2271 goto recover_immediately;
2272 default:
2273 break;
2274 }
2275
2276 /* unknown buffer contents, recover immediately */
2277
2278recover_immediately:
2279 return (xfs_lsn_t)-1;
2280
2281}
2282
2283/*
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002284 * Validate the recovered buffer is of the correct type and attach the
2285 * appropriate buffer operations to them for writeback. Magic numbers are in a
2286 * few places:
2287 * the first 16 bits of the buffer (inode buffer, dquot buffer),
2288 * the first 32 bits of the buffer (most blocks),
2289 * inside a struct xfs_da_blkinfo at the start of the buffer.
2290 */
2291static void
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002292xlog_recover_validate_buf_type(
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002293 struct xfs_mount *mp,
2294 struct xfs_buf *bp,
2295 xfs_buf_log_format_t *buf_f)
2296{
2297 struct xfs_da_blkinfo *info = bp->b_addr;
2298 __uint32_t magic32;
2299 __uint16_t magic16;
2300 __uint16_t magicda;
2301
Dave Chinner67dc2882014-08-04 12:43:06 +10002302 /*
2303 * We can only do post recovery validation on items on CRC enabled
2304 * fielsystems as we need to know when the buffer was written to be able
2305 * to determine if we should have replayed the item. If we replay old
2306 * metadata over a newer buffer, then it will enter a temporarily
2307 * inconsistent state resulting in verification failures. Hence for now
2308 * just avoid the verification stage for non-crc filesystems
2309 */
2310 if (!xfs_sb_version_hascrc(&mp->m_sb))
2311 return;
2312
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002313 magic32 = be32_to_cpu(*(__be32 *)bp->b_addr);
2314 magic16 = be16_to_cpu(*(__be16*)bp->b_addr);
2315 magicda = be16_to_cpu(info->magic);
Dave Chinner61fe1352013-04-03 16:11:30 +11002316 switch (xfs_blft_from_flags(buf_f)) {
2317 case XFS_BLFT_BTREE_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002318 switch (magic32) {
2319 case XFS_ABTB_CRC_MAGIC:
2320 case XFS_ABTC_CRC_MAGIC:
2321 case XFS_ABTB_MAGIC:
2322 case XFS_ABTC_MAGIC:
2323 bp->b_ops = &xfs_allocbt_buf_ops;
2324 break;
2325 case XFS_IBT_CRC_MAGIC:
Brian Fosteraafc3c22014-04-24 16:00:52 +10002326 case XFS_FIBT_CRC_MAGIC:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002327 case XFS_IBT_MAGIC:
Brian Fosteraafc3c22014-04-24 16:00:52 +10002328 case XFS_FIBT_MAGIC:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002329 bp->b_ops = &xfs_inobt_buf_ops;
2330 break;
2331 case XFS_BMAP_CRC_MAGIC:
2332 case XFS_BMAP_MAGIC:
2333 bp->b_ops = &xfs_bmbt_buf_ops;
2334 break;
2335 default:
2336 xfs_warn(mp, "Bad btree block magic!");
2337 ASSERT(0);
2338 break;
2339 }
2340 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002341 case XFS_BLFT_AGF_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002342 if (magic32 != XFS_AGF_MAGIC) {
2343 xfs_warn(mp, "Bad AGF block magic!");
2344 ASSERT(0);
2345 break;
2346 }
2347 bp->b_ops = &xfs_agf_buf_ops;
2348 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002349 case XFS_BLFT_AGFL_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002350 if (magic32 != XFS_AGFL_MAGIC) {
2351 xfs_warn(mp, "Bad AGFL block magic!");
2352 ASSERT(0);
2353 break;
2354 }
2355 bp->b_ops = &xfs_agfl_buf_ops;
2356 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002357 case XFS_BLFT_AGI_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002358 if (magic32 != XFS_AGI_MAGIC) {
2359 xfs_warn(mp, "Bad AGI block magic!");
2360 ASSERT(0);
2361 break;
2362 }
2363 bp->b_ops = &xfs_agi_buf_ops;
2364 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002365 case XFS_BLFT_UDQUOT_BUF:
2366 case XFS_BLFT_PDQUOT_BUF:
2367 case XFS_BLFT_GDQUOT_BUF:
Dave Chinner123887e2013-04-30 21:39:33 +10002368#ifdef CONFIG_XFS_QUOTA
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002369 if (magic16 != XFS_DQUOT_MAGIC) {
2370 xfs_warn(mp, "Bad DQUOT block magic!");
2371 ASSERT(0);
2372 break;
2373 }
2374 bp->b_ops = &xfs_dquot_buf_ops;
Dave Chinner123887e2013-04-30 21:39:33 +10002375#else
2376 xfs_alert(mp,
2377 "Trying to recover dquots without QUOTA support built in!");
2378 ASSERT(0);
2379#endif
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002380 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002381 case XFS_BLFT_DINO_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002382 if (magic16 != XFS_DINODE_MAGIC) {
2383 xfs_warn(mp, "Bad INODE block magic!");
2384 ASSERT(0);
2385 break;
2386 }
2387 bp->b_ops = &xfs_inode_buf_ops;
2388 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002389 case XFS_BLFT_SYMLINK_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002390 if (magic32 != XFS_SYMLINK_MAGIC) {
2391 xfs_warn(mp, "Bad symlink block magic!");
2392 ASSERT(0);
2393 break;
2394 }
2395 bp->b_ops = &xfs_symlink_buf_ops;
2396 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002397 case XFS_BLFT_DIR_BLOCK_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002398 if (magic32 != XFS_DIR2_BLOCK_MAGIC &&
2399 magic32 != XFS_DIR3_BLOCK_MAGIC) {
2400 xfs_warn(mp, "Bad dir block magic!");
2401 ASSERT(0);
2402 break;
2403 }
2404 bp->b_ops = &xfs_dir3_block_buf_ops;
2405 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002406 case XFS_BLFT_DIR_DATA_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002407 if (magic32 != XFS_DIR2_DATA_MAGIC &&
2408 magic32 != XFS_DIR3_DATA_MAGIC) {
2409 xfs_warn(mp, "Bad dir data magic!");
2410 ASSERT(0);
2411 break;
2412 }
2413 bp->b_ops = &xfs_dir3_data_buf_ops;
2414 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002415 case XFS_BLFT_DIR_FREE_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002416 if (magic32 != XFS_DIR2_FREE_MAGIC &&
2417 magic32 != XFS_DIR3_FREE_MAGIC) {
2418 xfs_warn(mp, "Bad dir3 free magic!");
2419 ASSERT(0);
2420 break;
2421 }
2422 bp->b_ops = &xfs_dir3_free_buf_ops;
2423 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002424 case XFS_BLFT_DIR_LEAF1_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002425 if (magicda != XFS_DIR2_LEAF1_MAGIC &&
2426 magicda != XFS_DIR3_LEAF1_MAGIC) {
2427 xfs_warn(mp, "Bad dir leaf1 magic!");
2428 ASSERT(0);
2429 break;
2430 }
2431 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
2432 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002433 case XFS_BLFT_DIR_LEAFN_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002434 if (magicda != XFS_DIR2_LEAFN_MAGIC &&
2435 magicda != XFS_DIR3_LEAFN_MAGIC) {
2436 xfs_warn(mp, "Bad dir leafn magic!");
2437 ASSERT(0);
2438 break;
2439 }
2440 bp->b_ops = &xfs_dir3_leafn_buf_ops;
2441 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002442 case XFS_BLFT_DA_NODE_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002443 if (magicda != XFS_DA_NODE_MAGIC &&
2444 magicda != XFS_DA3_NODE_MAGIC) {
2445 xfs_warn(mp, "Bad da node magic!");
2446 ASSERT(0);
2447 break;
2448 }
2449 bp->b_ops = &xfs_da3_node_buf_ops;
2450 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002451 case XFS_BLFT_ATTR_LEAF_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002452 if (magicda != XFS_ATTR_LEAF_MAGIC &&
2453 magicda != XFS_ATTR3_LEAF_MAGIC) {
2454 xfs_warn(mp, "Bad attr leaf magic!");
2455 ASSERT(0);
2456 break;
2457 }
2458 bp->b_ops = &xfs_attr3_leaf_buf_ops;
2459 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002460 case XFS_BLFT_ATTR_RMT_BUF:
Dave Chinnercab09a82013-04-30 21:39:36 +10002461 if (magic32 != XFS_ATTR3_RMT_MAGIC) {
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002462 xfs_warn(mp, "Bad attr remote magic!");
2463 ASSERT(0);
2464 break;
2465 }
2466 bp->b_ops = &xfs_attr3_rmt_buf_ops;
2467 break;
Dave Chinner04a1e6c2013-04-03 16:11:31 +11002468 case XFS_BLFT_SB_BUF:
2469 if (magic32 != XFS_SB_MAGIC) {
2470 xfs_warn(mp, "Bad SB block magic!");
2471 ASSERT(0);
2472 break;
2473 }
2474 bp->b_ops = &xfs_sb_buf_ops;
2475 break;
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002476 default:
Dave Chinner61fe1352013-04-03 16:11:30 +11002477 xfs_warn(mp, "Unknown buffer type %d!",
2478 xfs_blft_from_flags(buf_f));
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002479 break;
2480 }
2481}
2482
2483/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484 * Perform a 'normal' buffer recovery. Each logged region of the
2485 * buffer should be copied over the corresponding region in the
2486 * given buffer. The bitmap in the buf log format structure indicates
2487 * where to place the logged data.
2488 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489STATIC void
2490xlog_recover_do_reg_buffer(
Dave Chinner9abbc532010-04-13 15:06:46 +10002491 struct xfs_mount *mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492 xlog_recover_item_t *item,
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002493 struct xfs_buf *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 xfs_buf_log_format_t *buf_f)
2495{
2496 int i;
2497 int bit;
2498 int nbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499 int error;
2500
Dave Chinner9abbc532010-04-13 15:06:46 +10002501 trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f);
2502
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503 bit = 0;
2504 i = 1; /* 0 is the buf format structure */
2505 while (1) {
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002506 bit = xfs_next_bit(buf_f->blf_data_map,
2507 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508 if (bit == -1)
2509 break;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002510 nbits = xfs_contig_bits(buf_f->blf_data_map,
2511 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002512 ASSERT(nbits > 0);
Christoph Hellwig4b809162007-08-16 15:37:36 +10002513 ASSERT(item->ri_buf[i].i_addr != NULL);
Dave Chinnerc1155412010-05-07 11:05:19 +10002514 ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0);
Dave Chinneraa0e8832012-04-23 15:58:52 +10002515 ASSERT(BBTOB(bp->b_io_length) >=
2516 ((uint)bit << XFS_BLF_SHIFT) + (nbits << XFS_BLF_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517
2518 /*
Dave Chinner709da6a2013-05-27 16:38:23 +10002519 * The dirty regions logged in the buffer, even though
2520 * contiguous, may span multiple chunks. This is because the
2521 * dirty region may span a physical page boundary in a buffer
2522 * and hence be split into two separate vectors for writing into
2523 * the log. Hence we need to trim nbits back to the length of
2524 * the current region being copied out of the log.
2525 */
2526 if (item->ri_buf[i].i_len < (nbits << XFS_BLF_SHIFT))
2527 nbits = item->ri_buf[i].i_len >> XFS_BLF_SHIFT;
2528
2529 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530 * Do a sanity check if this is a dquot buffer. Just checking
2531 * the first dquot in the buffer should do. XXXThis is
2532 * probably a good thing to do for other buf types also.
2533 */
2534 error = 0;
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002535 if (buf_f->blf_flags &
Dave Chinnerc1155412010-05-07 11:05:19 +10002536 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002537 if (item->ri_buf[i].i_addr == NULL) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002538 xfs_alert(mp,
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002539 "XFS: NULL dquot in %s.", __func__);
2540 goto next;
2541 }
Jan Rekorajski8ec6dba2009-11-16 11:57:02 +00002542 if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002543 xfs_alert(mp,
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002544 "XFS: dquot too small (%d) in %s.",
2545 item->ri_buf[i].i_len, __func__);
2546 goto next;
2547 }
Dave Chinner9aede1d2013-10-15 09:17:52 +11002548 error = xfs_dqcheck(mp, item->ri_buf[i].i_addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549 -1, 0, XFS_QMOPT_DOWARN,
2550 "dquot_buf_recover");
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002551 if (error)
2552 goto next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553 }
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002554
2555 memcpy(xfs_buf_offset(bp,
Dave Chinnerc1155412010-05-07 11:05:19 +10002556 (uint)bit << XFS_BLF_SHIFT), /* dest */
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002557 item->ri_buf[i].i_addr, /* source */
Dave Chinnerc1155412010-05-07 11:05:19 +10002558 nbits<<XFS_BLF_SHIFT); /* length */
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002559 next:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560 i++;
2561 bit += nbits;
2562 }
2563
2564 /* Shouldn't be any more regions */
2565 ASSERT(i == item->ri_total);
Christoph Hellwigee1a47a2013-04-21 14:53:46 -05002566
Dave Chinner67dc2882014-08-04 12:43:06 +10002567 xlog_recover_validate_buf_type(mp, bp, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568}
2569
2570/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571 * Perform a dquot buffer recovery.
Zhi Yong Wu8ba701e2013-08-12 03:15:01 +00002572 * Simple algorithm: if we have found a QUOTAOFF log item of the same type
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2574 * Else, treat it as a regular buffer and do recovery.
Dave Chinnerad3714b2014-08-04 12:59:31 +10002575 *
2576 * Return false if the buffer was tossed and true if we recovered the buffer to
2577 * indicate to the caller if the buffer needs writing.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578 */
Dave Chinnerad3714b2014-08-04 12:59:31 +10002579STATIC bool
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580xlog_recover_do_dquot_buffer(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002581 struct xfs_mount *mp,
2582 struct xlog *log,
2583 struct xlog_recover_item *item,
2584 struct xfs_buf *bp,
2585 struct xfs_buf_log_format *buf_f)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586{
2587 uint type;
2588
Dave Chinner9abbc532010-04-13 15:06:46 +10002589 trace_xfs_log_recover_buf_dquot_buf(log, buf_f);
2590
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591 /*
2592 * Filesystems are required to send in quota flags at mount time.
2593 */
Dave Chinnerad3714b2014-08-04 12:59:31 +10002594 if (!mp->m_qflags)
2595 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596
2597 type = 0;
Dave Chinnerc1155412010-05-07 11:05:19 +10002598 if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599 type |= XFS_DQ_USER;
Dave Chinnerc1155412010-05-07 11:05:19 +10002600 if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF)
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002601 type |= XFS_DQ_PROJ;
Dave Chinnerc1155412010-05-07 11:05:19 +10002602 if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603 type |= XFS_DQ_GROUP;
2604 /*
2605 * This type of quotas was turned off, so ignore this buffer
2606 */
2607 if (log->l_quotaoffs_flag & type)
Dave Chinnerad3714b2014-08-04 12:59:31 +10002608 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609
Dave Chinner9abbc532010-04-13 15:06:46 +10002610 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
Dave Chinnerad3714b2014-08-04 12:59:31 +10002611 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612}
2613
2614/*
2615 * This routine replays a modification made to a buffer at runtime.
2616 * There are actually two types of buffer, regular and inode, which
2617 * are handled differently. Inode buffers are handled differently
2618 * in that we only recover a specific set of data from them, namely
2619 * the inode di_next_unlinked fields. This is because all other inode
2620 * data is actually logged via inode records and any data we replay
2621 * here which overlaps that may be stale.
2622 *
2623 * When meta-data buffers are freed at run time we log a buffer item
Dave Chinnerc1155412010-05-07 11:05:19 +10002624 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625 * of the buffer in the log should not be replayed at recovery time.
2626 * This is so that if the blocks covered by the buffer are reused for
2627 * file data before we crash we don't end up replaying old, freed
2628 * meta-data into a user's file.
2629 *
2630 * To handle the cancellation of buffer log items, we make two passes
2631 * over the log during recovery. During the first we build a table of
2632 * those buffers which have been cancelled, and during the second we
2633 * only replay those buffers which do not have corresponding cancel
Zhi Yong Wu34be5ff2013-08-07 10:11:07 +00002634 * records in the table. See xlog_recover_buffer_pass[1,2] above
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635 * for more details on the implementation of the table of cancel records.
2636 */
2637STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002638xlog_recover_buffer_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002639 struct xlog *log,
2640 struct list_head *buffer_list,
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002641 struct xlog_recover_item *item,
2642 xfs_lsn_t current_lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643{
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002644 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002645 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646 xfs_buf_t *bp;
2647 int error;
Christoph Hellwig6ad112b2009-11-24 18:02:23 +00002648 uint buf_flags;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002649 xfs_lsn_t lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002651 /*
2652 * In this pass we only want to recover all the buffers which have
2653 * not been cancelled and are not cancellation buffers themselves.
2654 */
2655 if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno,
2656 buf_f->blf_len, buf_f->blf_flags)) {
2657 trace_xfs_log_recover_buf_cancel(log, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659 }
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002660
Dave Chinner9abbc532010-04-13 15:06:46 +10002661 trace_xfs_log_recover_buf_recover(log, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662
Dave Chinnera8acad72012-04-23 15:58:54 +10002663 buf_flags = 0;
Dave Chinner611c9942012-04-23 15:59:07 +10002664 if (buf_f->blf_flags & XFS_BLF_INODE_BUF)
2665 buf_flags |= XBF_UNMAPPED;
Christoph Hellwig6ad112b2009-11-24 18:02:23 +00002666
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002667 bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len,
Dave Chinnerc3f8fc72012-11-12 22:54:01 +11002668 buf_flags, NULL);
Chandra Seetharamanac4d6882011-08-03 02:18:29 +00002669 if (!bp)
Dave Chinner24513372014-06-25 14:58:08 +10002670 return -ENOMEM;
Chandra Seetharamane5702802011-08-03 02:18:34 +00002671 error = bp->b_error;
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +00002672 if (error) {
Christoph Hellwig901796a2011-10-10 16:52:49 +00002673 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)");
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002674 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675 }
2676
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002677 /*
Dave Chinner67dc2882014-08-04 12:43:06 +10002678 * Recover the buffer only if we get an LSN from it and it's less than
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002679 * the lsn of the transaction we are replaying.
Dave Chinner67dc2882014-08-04 12:43:06 +10002680 *
2681 * Note that we have to be extremely careful of readahead here.
2682 * Readahead does not attach verfiers to the buffers so if we don't
2683 * actually do any replay after readahead because of the LSN we found
2684 * in the buffer if more recent than that current transaction then we
2685 * need to attach the verifier directly. Failure to do so can lead to
2686 * future recovery actions (e.g. EFI and unlinked list recovery) can
2687 * operate on the buffers and they won't get the verifier attached. This
2688 * can lead to blocks on disk having the correct content but a stale
2689 * CRC.
2690 *
2691 * It is safe to assume these clean buffers are currently up to date.
2692 * If the buffer is dirtied by a later transaction being replayed, then
2693 * the verifier will be reset to match whatever recover turns that
2694 * buffer into.
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002695 */
2696 lsn = xlog_recover_get_buf_lsn(mp, bp);
Dave Chinner67dc2882014-08-04 12:43:06 +10002697 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
2698 xlog_recover_validate_buf_type(mp, bp, buf_f);
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002699 goto out_release;
Dave Chinner67dc2882014-08-04 12:43:06 +10002700 }
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002701
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002702 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
Dave Chinnerad3714b2014-08-04 12:59:31 +10002704 if (error)
2705 goto out_release;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002706 } else if (buf_f->blf_flags &
Dave Chinnerc1155412010-05-07 11:05:19 +10002707 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
Dave Chinnerad3714b2014-08-04 12:59:31 +10002708 bool dirty;
2709
2710 dirty = xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2711 if (!dirty)
2712 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713 } else {
Dave Chinner9abbc532010-04-13 15:06:46 +10002714 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716
2717 /*
2718 * Perform delayed write on the buffer. Asynchronous writes will be
2719 * slower when taking into account all the buffers to be flushed.
2720 *
2721 * Also make sure that only inode buffers with good sizes stay in
2722 * the buffer cache. The kernel moves inodes in buffers of 1 block
Jie Liu0f49efd2013-12-13 15:51:48 +11002723 * or mp->m_inode_cluster_size bytes, whichever is bigger. The inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724 * buffers in the log can be a different size if the log was generated
2725 * by an older kernel using unclustered inode buffers or a newer kernel
2726 * running with a different inode cluster size. Regardless, if the
Jie Liu0f49efd2013-12-13 15:51:48 +11002727 * the inode buffer size isn't MAX(blocksize, mp->m_inode_cluster_size)
2728 * for *our* value of mp->m_inode_cluster_size, then we need to keep
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729 * the buffer out of the buffer cache so that the buffer won't
2730 * overlap with future reads of those inodes.
2731 */
2732 if (XFS_DINODE_MAGIC ==
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002733 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
Dave Chinneraa0e8832012-04-23 15:58:52 +10002734 (BBTOB(bp->b_io_length) != MAX(log->l_mp->m_sb.sb_blocksize,
Jie Liu0f49efd2013-12-13 15:51:48 +11002735 (__uint32_t)log->l_mp->m_inode_cluster_size))) {
Christoph Hellwigc867cb62011-10-10 16:52:46 +00002736 xfs_buf_stale(bp);
Christoph Hellwigc2b006c2011-08-23 08:28:07 +00002737 error = xfs_bwrite(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002738 } else {
Dave Chinnerebad8612010-09-22 10:47:20 +10002739 ASSERT(bp->b_target->bt_mount == mp);
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02002740 bp->b_iodone = xlog_recover_iodone;
Christoph Hellwig43ff2122012-04-23 15:58:39 +10002741 xfs_buf_delwri_queue(bp, buffer_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742 }
2743
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002744out_release:
Christoph Hellwigc2b006c2011-08-23 08:28:07 +00002745 xfs_buf_relse(bp);
2746 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002747}
2748
Dave Chinner638f44162013-08-30 10:23:45 +10002749/*
2750 * Inode fork owner changes
2751 *
2752 * If we have been told that we have to reparent the inode fork, it's because an
2753 * extent swap operation on a CRC enabled filesystem has been done and we are
2754 * replaying it. We need to walk the BMBT of the appropriate fork and change the
2755 * owners of it.
2756 *
2757 * The complexity here is that we don't have an inode context to work with, so
2758 * after we've replayed the inode we need to instantiate one. This is where the
2759 * fun begins.
2760 *
2761 * We are in the middle of log recovery, so we can't run transactions. That
2762 * means we cannot use cache coherent inode instantiation via xfs_iget(), as
2763 * that will result in the corresponding iput() running the inode through
2764 * xfs_inactive(). If we've just replayed an inode core that changes the link
2765 * count to zero (i.e. it's been unlinked), then xfs_inactive() will run
2766 * transactions (bad!).
2767 *
2768 * So, to avoid this, we instantiate an inode directly from the inode core we've
2769 * just recovered. We have the buffer still locked, and all we really need to
2770 * instantiate is the inode core and the forks being modified. We can do this
2771 * manually, then run the inode btree owner change, and then tear down the
2772 * xfs_inode without having to run any transactions at all.
2773 *
2774 * Also, because we don't have a transaction context available here but need to
2775 * gather all the buffers we modify for writeback so we pass the buffer_list
2776 * instead for the operation to use.
2777 */
2778
2779STATIC int
2780xfs_recover_inode_owner_change(
2781 struct xfs_mount *mp,
2782 struct xfs_dinode *dip,
2783 struct xfs_inode_log_format *in_f,
2784 struct list_head *buffer_list)
2785{
2786 struct xfs_inode *ip;
2787 int error;
2788
2789 ASSERT(in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER));
2790
2791 ip = xfs_inode_alloc(mp, in_f->ilf_ino);
2792 if (!ip)
Dave Chinner24513372014-06-25 14:58:08 +10002793 return -ENOMEM;
Dave Chinner638f44162013-08-30 10:23:45 +10002794
2795 /* instantiate the inode */
2796 xfs_dinode_from_disk(&ip->i_d, dip);
2797 ASSERT(ip->i_d.di_version >= 3);
2798
2799 error = xfs_iformat_fork(ip, dip);
2800 if (error)
2801 goto out_free_ip;
2802
2803
2804 if (in_f->ilf_fields & XFS_ILOG_DOWNER) {
2805 ASSERT(in_f->ilf_fields & XFS_ILOG_DBROOT);
2806 error = xfs_bmbt_change_owner(NULL, ip, XFS_DATA_FORK,
2807 ip->i_ino, buffer_list);
2808 if (error)
2809 goto out_free_ip;
2810 }
2811
2812 if (in_f->ilf_fields & XFS_ILOG_AOWNER) {
2813 ASSERT(in_f->ilf_fields & XFS_ILOG_ABROOT);
2814 error = xfs_bmbt_change_owner(NULL, ip, XFS_ATTR_FORK,
2815 ip->i_ino, buffer_list);
2816 if (error)
2817 goto out_free_ip;
2818 }
2819
2820out_free_ip:
2821 xfs_inode_free(ip);
2822 return error;
2823}
2824
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002826xlog_recover_inode_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002827 struct xlog *log,
2828 struct list_head *buffer_list,
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002829 struct xlog_recover_item *item,
2830 xfs_lsn_t current_lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831{
2832 xfs_inode_log_format_t *in_f;
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002833 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834 xfs_buf_t *bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002835 xfs_dinode_t *dip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836 int len;
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10002837 char *src;
2838 char *dest;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839 int error;
2840 int attr_index;
2841 uint fields;
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002842 xfs_icdinode_t *dicp;
Christoph Hellwig93848a92013-04-03 16:11:17 +11002843 uint isize;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002844 int need_free = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845
Tim Shimmin6d192a92006-06-09 14:55:38 +10002846 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002847 in_f = item->ri_buf[0].i_addr;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002848 } else {
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002849 in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002850 need_free = 1;
2851 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2852 if (error)
2853 goto error;
2854 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855
2856 /*
2857 * Inode buffers can be freed, look out for it,
2858 * and do not replay the inode.
2859 */
Christoph Hellwiga1941892008-11-28 14:23:40 +11002860 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
2861 in_f->ilf_len, 0)) {
Tim Shimmin6d192a92006-06-09 14:55:38 +10002862 error = 0;
Dave Chinner9abbc532010-04-13 15:06:46 +10002863 trace_xfs_log_recover_inode_cancel(log, in_f);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002864 goto error;
2865 }
Dave Chinner9abbc532010-04-13 15:06:46 +10002866 trace_xfs_log_recover_inode_recover(log, in_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867
Dave Chinnerc3f8fc72012-11-12 22:54:01 +11002868 bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len, 0,
Christoph Hellwig93848a92013-04-03 16:11:17 +11002869 &xfs_inode_buf_ops);
Chandra Seetharamanac4d6882011-08-03 02:18:29 +00002870 if (!bp) {
Dave Chinner24513372014-06-25 14:58:08 +10002871 error = -ENOMEM;
Chandra Seetharamanac4d6882011-08-03 02:18:29 +00002872 goto error;
2873 }
Chandra Seetharamane5702802011-08-03 02:18:34 +00002874 error = bp->b_error;
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +00002875 if (error) {
Christoph Hellwig901796a2011-10-10 16:52:49 +00002876 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)");
Dave Chinner638f44162013-08-30 10:23:45 +10002877 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
Christoph Hellwig88ee2df2015-06-22 09:44:29 +10002880 dip = xfs_buf_offset(bp, in_f->ilf_boffset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881
2882 /*
2883 * Make sure the place we're flushing out to really looks
2884 * like an inode!
2885 */
Christoph Hellwig69ef9212011-07-08 14:36:05 +02002886 if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002887 xfs_alert(mp,
2888 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
2889 __func__, dip, bp, in_f->ilf_ino);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002890 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891 XFS_ERRLEVEL_LOW, mp);
Dave Chinner24513372014-06-25 14:58:08 +10002892 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002893 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002894 }
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002895 dicp = item->ri_buf[1].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002897 xfs_alert(mp,
2898 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
2899 __func__, item, in_f->ilf_ino);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002900 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002901 XFS_ERRLEVEL_LOW, mp);
Dave Chinner24513372014-06-25 14:58:08 +10002902 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002903 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002904 }
2905
Dave Chinnere60896d2013-07-24 15:47:30 +10002906 /*
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002907 * If the inode has an LSN in it, recover the inode only if it's less
Dave Chinner638f44162013-08-30 10:23:45 +10002908 * than the lsn of the transaction we are replaying. Note: we still
2909 * need to replay an owner change even though the inode is more recent
2910 * than the transaction as there is no guarantee that all the btree
2911 * blocks are more recent than this transaction, too.
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002912 */
2913 if (dip->di_version >= 3) {
2914 xfs_lsn_t lsn = be64_to_cpu(dip->di_lsn);
2915
2916 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
2917 trace_xfs_log_recover_inode_skip(log, in_f);
2918 error = 0;
Dave Chinner638f44162013-08-30 10:23:45 +10002919 goto out_owner_change;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002920 }
2921 }
2922
2923 /*
Dave Chinnere60896d2013-07-24 15:47:30 +10002924 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
2925 * are transactional and if ordering is necessary we can determine that
2926 * more accurately by the LSN field in the V3 inode core. Don't trust
2927 * the inode versions we might be changing them here - use the
2928 * superblock flag to determine whether we need to look at di_flushiter
2929 * to skip replay when the on disk inode is newer than the log one
2930 */
2931 if (!xfs_sb_version_hascrc(&mp->m_sb) &&
2932 dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002933 /*
2934 * Deal with the wrap case, DI_MAX_FLUSH is less
2935 * than smaller numbers
2936 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002937 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002938 dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939 /* do nothing */
2940 } else {
Dave Chinner9abbc532010-04-13 15:06:46 +10002941 trace_xfs_log_recover_inode_skip(log, in_f);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002942 error = 0;
Dave Chinner638f44162013-08-30 10:23:45 +10002943 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002944 }
2945 }
Dave Chinnere60896d2013-07-24 15:47:30 +10002946
Linus Torvalds1da177e2005-04-16 15:20:36 -07002947 /* Take the opportunity to reset the flush iteration count */
2948 dicp->di_flushiter = 0;
2949
Al Viroabbede12011-07-26 02:31:30 -04002950 if (unlikely(S_ISREG(dicp->di_mode))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002951 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2952 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002953 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002954 XFS_ERRLEVEL_LOW, mp, dicp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002955 xfs_alert(mp,
2956 "%s: Bad regular inode log record, rec ptr 0x%p, "
2957 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2958 __func__, item, dip, bp, in_f->ilf_ino);
Dave Chinner24513372014-06-25 14:58:08 +10002959 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002960 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002961 }
Al Viroabbede12011-07-26 02:31:30 -04002962 } else if (unlikely(S_ISDIR(dicp->di_mode))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002963 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2964 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2965 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002966 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002967 XFS_ERRLEVEL_LOW, mp, dicp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002968 xfs_alert(mp,
2969 "%s: Bad dir inode log record, rec ptr 0x%p, "
2970 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2971 __func__, item, dip, bp, in_f->ilf_ino);
Dave Chinner24513372014-06-25 14:58:08 +10002972 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002973 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002974 }
2975 }
2976 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002977 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002978 XFS_ERRLEVEL_LOW, mp, dicp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002979 xfs_alert(mp,
2980 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2981 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2982 __func__, item, dip, bp, in_f->ilf_ino,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002983 dicp->di_nextents + dicp->di_anextents,
2984 dicp->di_nblocks);
Dave Chinner24513372014-06-25 14:58:08 +10002985 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002986 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987 }
2988 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002989 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002990 XFS_ERRLEVEL_LOW, mp, dicp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002991 xfs_alert(mp,
2992 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2993 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__,
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002994 item, dip, bp, in_f->ilf_ino, dicp->di_forkoff);
Dave Chinner24513372014-06-25 14:58:08 +10002995 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002996 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002997 }
Christoph Hellwig93848a92013-04-03 16:11:17 +11002998 isize = xfs_icdinode_size(dicp->di_version);
2999 if (unlikely(item->ri_buf[1].i_len > isize)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003000 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001 XFS_ERRLEVEL_LOW, mp, dicp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003002 xfs_alert(mp,
3003 "%s: Bad inode log record length %d, rec ptr 0x%p",
3004 __func__, item->ri_buf[1].i_len, item);
Dave Chinner24513372014-06-25 14:58:08 +10003005 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10003006 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003007 }
3008
3009 /* The core is in in-core format */
Christoph Hellwig93848a92013-04-03 16:11:17 +11003010 xfs_dinode_to_disk(dip, dicp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003011
3012 /* the rest is in on-disk format */
Christoph Hellwig93848a92013-04-03 16:11:17 +11003013 if (item->ri_buf[1].i_len > isize) {
3014 memcpy((char *)dip + isize,
3015 item->ri_buf[1].i_addr + isize,
3016 item->ri_buf[1].i_len - isize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017 }
3018
3019 fields = in_f->ilf_fields;
3020 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
3021 case XFS_ILOG_DEV:
Christoph Hellwig81591fe2008-11-28 14:23:39 +11003022 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003023 break;
3024 case XFS_ILOG_UUID:
Christoph Hellwig81591fe2008-11-28 14:23:39 +11003025 memcpy(XFS_DFORK_DPTR(dip),
3026 &in_f->ilf_u.ilfu_uuid,
3027 sizeof(uuid_t));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003028 break;
3029 }
3030
3031 if (in_f->ilf_size == 2)
Dave Chinner638f44162013-08-30 10:23:45 +10003032 goto out_owner_change;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003033 len = item->ri_buf[2].i_len;
3034 src = item->ri_buf[2].i_addr;
3035 ASSERT(in_f->ilf_size <= 4);
3036 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
3037 ASSERT(!(fields & XFS_ILOG_DFORK) ||
3038 (len == in_f->ilf_dsize));
3039
3040 switch (fields & XFS_ILOG_DFORK) {
3041 case XFS_ILOG_DDATA:
3042 case XFS_ILOG_DEXT:
Christoph Hellwig81591fe2008-11-28 14:23:39 +11003043 memcpy(XFS_DFORK_DPTR(dip), src, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044 break;
3045
3046 case XFS_ILOG_DBROOT:
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11003047 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
Christoph Hellwig81591fe2008-11-28 14:23:39 +11003048 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003049 XFS_DFORK_DSIZE(dip, mp));
3050 break;
3051
3052 default:
3053 /*
3054 * There are no data fork flags set.
3055 */
3056 ASSERT((fields & XFS_ILOG_DFORK) == 0);
3057 break;
3058 }
3059
3060 /*
3061 * If we logged any attribute data, recover it. There may or
3062 * may not have been any other non-core data logged in this
3063 * transaction.
3064 */
3065 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
3066 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
3067 attr_index = 3;
3068 } else {
3069 attr_index = 2;
3070 }
3071 len = item->ri_buf[attr_index].i_len;
3072 src = item->ri_buf[attr_index].i_addr;
3073 ASSERT(len == in_f->ilf_asize);
3074
3075 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
3076 case XFS_ILOG_ADATA:
3077 case XFS_ILOG_AEXT:
3078 dest = XFS_DFORK_APTR(dip);
3079 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
3080 memcpy(dest, src, len);
3081 break;
3082
3083 case XFS_ILOG_ABROOT:
3084 dest = XFS_DFORK_APTR(dip);
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11003085 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
3086 len, (xfs_bmdr_block_t*)dest,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003087 XFS_DFORK_ASIZE(dip, mp));
3088 break;
3089
3090 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003091 xfs_warn(log->l_mp, "%s: Invalid flag", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003092 ASSERT(0);
Dave Chinner24513372014-06-25 14:58:08 +10003093 error = -EIO;
Dave Chinner638f44162013-08-30 10:23:45 +10003094 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003095 }
3096 }
3097
Dave Chinner638f44162013-08-30 10:23:45 +10003098out_owner_change:
3099 if (in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER))
3100 error = xfs_recover_inode_owner_change(mp, dip, in_f,
3101 buffer_list);
Christoph Hellwig93848a92013-04-03 16:11:17 +11003102 /* re-generate the checksum. */
3103 xfs_dinode_calc_crc(log->l_mp, dip);
3104
Dave Chinnerebad8612010-09-22 10:47:20 +10003105 ASSERT(bp->b_target->bt_mount == mp);
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02003106 bp->b_iodone = xlog_recover_iodone;
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003107 xfs_buf_delwri_queue(bp, buffer_list);
Dave Chinner50d5c8d2013-08-28 21:22:47 +10003108
3109out_release:
Christoph Hellwig61551f12011-08-23 08:28:06 +00003110 xfs_buf_relse(bp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10003111error:
3112 if (need_free)
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003113 kmem_free(in_f);
Eric Sandeenb474c7a2014-06-22 15:04:54 +10003114 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003115}
3116
3117/*
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003118 * Recover QUOTAOFF records. We simply make a note of it in the xlog
Linus Torvalds1da177e2005-04-16 15:20:36 -07003119 * structure, so that we know not to do any dquot item or dquot buffer recovery,
3120 * of that type.
3121 */
3122STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003123xlog_recover_quotaoff_pass1(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003124 struct xlog *log,
3125 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003126{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003127 xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003128 ASSERT(qoff_f);
3129
3130 /*
3131 * The logitem format's flag tells us if this was user quotaoff,
Nathan Scott77a7cce2006-01-11 15:35:57 +11003132 * group/project quotaoff or both.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003133 */
3134 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
3135 log->l_quotaoffs_flag |= XFS_DQ_USER;
Nathan Scott77a7cce2006-01-11 15:35:57 +11003136 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
3137 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003138 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
3139 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
3140
Eric Sandeend99831f2014-06-22 15:03:54 +10003141 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003142}
3143
3144/*
3145 * Recover a dquot record
3146 */
3147STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003148xlog_recover_dquot_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003149 struct xlog *log,
3150 struct list_head *buffer_list,
Dave Chinner50d5c8d2013-08-28 21:22:47 +10003151 struct xlog_recover_item *item,
3152 xfs_lsn_t current_lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003153{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003154 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003155 xfs_buf_t *bp;
3156 struct xfs_disk_dquot *ddq, *recddq;
3157 int error;
3158 xfs_dq_logformat_t *dq_f;
3159 uint type;
3160
Linus Torvalds1da177e2005-04-16 15:20:36 -07003161
3162 /*
3163 * Filesystems are required to send in quota flags at mount time.
3164 */
3165 if (mp->m_qflags == 0)
Eric Sandeend99831f2014-06-22 15:03:54 +10003166 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003167
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10003168 recddq = item->ri_buf[1].i_addr;
3169 if (recddq == NULL) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003170 xfs_alert(log->l_mp, "NULL dquot in %s.", __func__);
Dave Chinner24513372014-06-25 14:58:08 +10003171 return -EIO;
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02003172 }
Jan Rekorajski8ec6dba2009-11-16 11:57:02 +00003173 if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003174 xfs_alert(log->l_mp, "dquot too small (%d) in %s.",
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02003175 item->ri_buf[1].i_len, __func__);
Dave Chinner24513372014-06-25 14:58:08 +10003176 return -EIO;
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02003177 }
3178
Linus Torvalds1da177e2005-04-16 15:20:36 -07003179 /*
3180 * This type of quotas was turned off, so ignore this record.
3181 */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003182 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003183 ASSERT(type);
3184 if (log->l_quotaoffs_flag & type)
Eric Sandeend99831f2014-06-22 15:03:54 +10003185 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003186
3187 /*
3188 * At this point we know that quota was _not_ turned off.
3189 * Since the mount flags are not indicating to us otherwise, this
3190 * must mean that quota is on, and the dquot needs to be replayed.
3191 * Remember that we may not have fully recovered the superblock yet,
3192 * so we can't do the usual trick of looking at the SB quota bits.
3193 *
3194 * The other possibility, of course, is that the quota subsystem was
3195 * removed since the last mount - ENOSYS.
3196 */
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10003197 dq_f = item->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003198 ASSERT(dq_f);
Dave Chinner9aede1d2013-10-15 09:17:52 +11003199 error = xfs_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003200 "xlog_recover_dquot_pass2 (log copy)");
3201 if (error)
Dave Chinner24513372014-06-25 14:58:08 +10003202 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003203 ASSERT(dq_f->qlf_len == 1);
3204
Dave Chinnerad3714b2014-08-04 12:59:31 +10003205 /*
3206 * At this point we are assuming that the dquots have been allocated
3207 * and hence the buffer has valid dquots stamped in it. It should,
3208 * therefore, pass verifier validation. If the dquot is bad, then the
3209 * we'll return an error here, so we don't need to specifically check
3210 * the dquot in the buffer after the verifier has run.
3211 */
Dave Chinner7ca790a2012-04-23 15:58:55 +10003212 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dq_f->qlf_blkno,
Dave Chinnerc3f8fc72012-11-12 22:54:01 +11003213 XFS_FSB_TO_BB(mp, dq_f->qlf_len), 0, &bp,
Dave Chinnerad3714b2014-08-04 12:59:31 +10003214 &xfs_dquot_buf_ops);
Dave Chinner7ca790a2012-04-23 15:58:55 +10003215 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003216 return error;
Dave Chinner7ca790a2012-04-23 15:58:55 +10003217
Linus Torvalds1da177e2005-04-16 15:20:36 -07003218 ASSERT(bp);
Christoph Hellwig88ee2df2015-06-22 09:44:29 +10003219 ddq = xfs_buf_offset(bp, dq_f->qlf_boffset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003220
3221 /*
Dave Chinner50d5c8d2013-08-28 21:22:47 +10003222 * If the dquot has an LSN in it, recover the dquot only if it's less
3223 * than the lsn of the transaction we are replaying.
3224 */
3225 if (xfs_sb_version_hascrc(&mp->m_sb)) {
3226 struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddq;
3227 xfs_lsn_t lsn = be64_to_cpu(dqb->dd_lsn);
3228
3229 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
3230 goto out_release;
3231 }
3232 }
3233
Linus Torvalds1da177e2005-04-16 15:20:36 -07003234 memcpy(ddq, recddq, item->ri_buf[1].i_len);
Dave Chinner6fcdc592013-06-03 15:28:46 +10003235 if (xfs_sb_version_hascrc(&mp->m_sb)) {
3236 xfs_update_cksum((char *)ddq, sizeof(struct xfs_dqblk),
3237 XFS_DQUOT_CRC_OFF);
3238 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003239
3240 ASSERT(dq_f->qlf_size == 2);
Dave Chinnerebad8612010-09-22 10:47:20 +10003241 ASSERT(bp->b_target->bt_mount == mp);
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02003242 bp->b_iodone = xlog_recover_iodone;
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003243 xfs_buf_delwri_queue(bp, buffer_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003244
Dave Chinner50d5c8d2013-08-28 21:22:47 +10003245out_release:
3246 xfs_buf_relse(bp);
3247 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003248}
3249
3250/*
3251 * This routine is called to create an in-core extent free intent
3252 * item from the efi format structure which was logged on disk.
3253 * It allocates an in-core efi, copies the extents from the format
3254 * structure into it, and adds the efi to the AIL with the given
3255 * LSN.
3256 */
Tim Shimmin6d192a92006-06-09 14:55:38 +10003257STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003258xlog_recover_efi_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003259 struct xlog *log,
3260 struct xlog_recover_item *item,
3261 xfs_lsn_t lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003262{
Brian Fostere32a1d12015-08-19 09:52:21 +10003263 int error;
3264 struct xfs_mount *mp = log->l_mp;
3265 struct xfs_efi_log_item *efip;
3266 struct xfs_efi_log_format *efi_formatp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003267
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10003268 efi_formatp = item->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003269
Linus Torvalds1da177e2005-04-16 15:20:36 -07003270 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
Brian Fostere32a1d12015-08-19 09:52:21 +10003271 error = xfs_efi_copy_format(&item->ri_buf[0], &efip->efi_format);
3272 if (error) {
Tim Shimmin6d192a92006-06-09 14:55:38 +10003273 xfs_efi_item_free(efip);
3274 return error;
3275 }
Dave Chinnerb199c8a2010-12-20 11:59:49 +11003276 atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003277
David Chinnera9c21c12008-10-30 17:39:35 +11003278 spin_lock(&log->l_ailp->xa_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003279 /*
Brian Fostere32a1d12015-08-19 09:52:21 +10003280 * The EFI has two references. One for the EFD and one for EFI to ensure
3281 * it makes it into the AIL. Insert the EFI into the AIL directly and
3282 * drop the EFI reference. Note that xfs_trans_ail_update() drops the
3283 * AIL lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003284 */
Dave Chinnere6059942010-12-20 12:34:26 +11003285 xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn);
Brian Fostere32a1d12015-08-19 09:52:21 +10003286 xfs_efi_release(efip);
Tim Shimmin6d192a92006-06-09 14:55:38 +10003287 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003288}
3289
3290
3291/*
Brian Fostere32a1d12015-08-19 09:52:21 +10003292 * This routine is called when an EFD format structure is found in a committed
3293 * transaction in the log. Its purpose is to cancel the corresponding EFI if it
3294 * was still in the log. To do this it searches the AIL for the EFI with an id
3295 * equal to that in the EFD format structure. If we find it we drop the EFD
3296 * reference, which removes the EFI from the AIL and frees it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003297 */
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003298STATIC int
3299xlog_recover_efd_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003300 struct xlog *log,
3301 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003302{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003303 xfs_efd_log_format_t *efd_formatp;
3304 xfs_efi_log_item_t *efip = NULL;
3305 xfs_log_item_t *lip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003306 __uint64_t efi_id;
David Chinner27d8d5f2008-10-30 17:38:39 +11003307 struct xfs_ail_cursor cur;
David Chinner783a2f62008-10-30 17:39:58 +11003308 struct xfs_ail *ailp = log->l_ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003309
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10003310 efd_formatp = item->ri_buf[0].i_addr;
Tim Shimmin6d192a92006-06-09 14:55:38 +10003311 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
3312 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
3313 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
3314 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003315 efi_id = efd_formatp->efd_efi_id;
3316
3317 /*
Brian Fostere32a1d12015-08-19 09:52:21 +10003318 * Search for the EFI with the id in the EFD format structure in the
3319 * AIL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003320 */
David Chinnera9c21c12008-10-30 17:39:35 +11003321 spin_lock(&ailp->xa_lock);
3322 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003323 while (lip != NULL) {
3324 if (lip->li_type == XFS_LI_EFI) {
3325 efip = (xfs_efi_log_item_t *)lip;
3326 if (efip->efi_format.efi_id == efi_id) {
3327 /*
Brian Fostere32a1d12015-08-19 09:52:21 +10003328 * Drop the EFD reference to the EFI. This
3329 * removes the EFI from the AIL and frees it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003330 */
Brian Fostere32a1d12015-08-19 09:52:21 +10003331 spin_unlock(&ailp->xa_lock);
3332 xfs_efi_release(efip);
David Chinnera9c21c12008-10-30 17:39:35 +11003333 spin_lock(&ailp->xa_lock);
David Chinner27d8d5f2008-10-30 17:38:39 +11003334 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003335 }
3336 }
David Chinnera9c21c12008-10-30 17:39:35 +11003337 lip = xfs_trans_ail_cursor_next(ailp, &cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003338 }
Brian Fostere32a1d12015-08-19 09:52:21 +10003339
Eric Sandeene4a1e292014-04-14 19:06:05 +10003340 xfs_trans_ail_cursor_done(&cur);
David Chinnera9c21c12008-10-30 17:39:35 +11003341 spin_unlock(&ailp->xa_lock);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003342
3343 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003344}
3345
3346/*
Dave Chinner28c8e412013-06-27 16:04:55 +10003347 * This routine is called when an inode create format structure is found in a
3348 * committed transaction in the log. It's purpose is to initialise the inodes
3349 * being allocated on disk. This requires us to get inode cluster buffers that
3350 * match the range to be intialised, stamped with inode templates and written
3351 * by delayed write so that subsequent modifications will hit the cached buffer
3352 * and only need writing out at the end of recovery.
3353 */
3354STATIC int
3355xlog_recover_do_icreate_pass2(
3356 struct xlog *log,
3357 struct list_head *buffer_list,
3358 xlog_recover_item_t *item)
3359{
3360 struct xfs_mount *mp = log->l_mp;
3361 struct xfs_icreate_log *icl;
3362 xfs_agnumber_t agno;
3363 xfs_agblock_t agbno;
3364 unsigned int count;
3365 unsigned int isize;
3366 xfs_agblock_t length;
Brian Fosterfc0d1652015-08-19 09:59:38 +10003367 int blks_per_cluster;
3368 int bb_per_cluster;
3369 int cancel_count;
3370 int nbufs;
3371 int i;
Dave Chinner28c8e412013-06-27 16:04:55 +10003372
3373 icl = (struct xfs_icreate_log *)item->ri_buf[0].i_addr;
3374 if (icl->icl_type != XFS_LI_ICREATE) {
3375 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad type");
Dave Chinner24513372014-06-25 14:58:08 +10003376 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003377 }
3378
3379 if (icl->icl_size != 1) {
3380 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad icl size");
Dave Chinner24513372014-06-25 14:58:08 +10003381 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003382 }
3383
3384 agno = be32_to_cpu(icl->icl_ag);
3385 if (agno >= mp->m_sb.sb_agcount) {
3386 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agno");
Dave Chinner24513372014-06-25 14:58:08 +10003387 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003388 }
3389 agbno = be32_to_cpu(icl->icl_agbno);
3390 if (!agbno || agbno == NULLAGBLOCK || agbno >= mp->m_sb.sb_agblocks) {
3391 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agbno");
Dave Chinner24513372014-06-25 14:58:08 +10003392 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003393 }
3394 isize = be32_to_cpu(icl->icl_isize);
3395 if (isize != mp->m_sb.sb_inodesize) {
3396 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad isize");
Dave Chinner24513372014-06-25 14:58:08 +10003397 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003398 }
3399 count = be32_to_cpu(icl->icl_count);
3400 if (!count) {
3401 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count");
Dave Chinner24513372014-06-25 14:58:08 +10003402 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003403 }
3404 length = be32_to_cpu(icl->icl_length);
3405 if (!length || length >= mp->m_sb.sb_agblocks) {
3406 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad length");
Dave Chinner24513372014-06-25 14:58:08 +10003407 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003408 }
3409
Brian Foster7f43c902015-05-29 09:06:30 +10003410 /*
3411 * The inode chunk is either full or sparse and we only support
3412 * m_ialloc_min_blks sized sparse allocations at this time.
3413 */
3414 if (length != mp->m_ialloc_blks &&
3415 length != mp->m_ialloc_min_blks) {
3416 xfs_warn(log->l_mp,
3417 "%s: unsupported chunk length", __FUNCTION__);
3418 return -EINVAL;
3419 }
3420
3421 /* verify inode count is consistent with extent length */
3422 if ((count >> mp->m_sb.sb_inopblog) != length) {
3423 xfs_warn(log->l_mp,
3424 "%s: inconsistent inode count and chunk length",
3425 __FUNCTION__);
Dave Chinner24513372014-06-25 14:58:08 +10003426 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003427 }
3428
3429 /*
Brian Fosterfc0d1652015-08-19 09:59:38 +10003430 * The icreate transaction can cover multiple cluster buffers and these
3431 * buffers could have been freed and reused. Check the individual
3432 * buffers for cancellation so we don't overwrite anything written after
3433 * a cancellation.
Dave Chinner28c8e412013-06-27 16:04:55 +10003434 */
Brian Fosterfc0d1652015-08-19 09:59:38 +10003435 blks_per_cluster = xfs_icluster_size_fsb(mp);
3436 bb_per_cluster = XFS_FSB_TO_BB(mp, blks_per_cluster);
3437 nbufs = length / blks_per_cluster;
3438 for (i = 0, cancel_count = 0; i < nbufs; i++) {
3439 xfs_daddr_t daddr;
Dave Chinner28c8e412013-06-27 16:04:55 +10003440
Brian Fosterfc0d1652015-08-19 09:59:38 +10003441 daddr = XFS_AGB_TO_DADDR(mp, agno,
3442 agbno + i * blks_per_cluster);
3443 if (xlog_check_buffer_cancelled(log, daddr, bb_per_cluster, 0))
3444 cancel_count++;
3445 }
3446
3447 /*
3448 * We currently only use icreate for a single allocation at a time. This
3449 * means we should expect either all or none of the buffers to be
3450 * cancelled. Be conservative and skip replay if at least one buffer is
3451 * cancelled, but warn the user that something is awry if the buffers
3452 * are not consistent.
3453 *
3454 * XXX: This must be refined to only skip cancelled clusters once we use
3455 * icreate for multiple chunk allocations.
3456 */
3457 ASSERT(!cancel_count || cancel_count == nbufs);
3458 if (cancel_count) {
3459 if (cancel_count != nbufs)
3460 xfs_warn(mp,
3461 "WARNING: partial inode chunk cancellation, skipped icreate.");
Brian Foster78d57e42015-08-19 09:58:48 +10003462 trace_xfs_log_recover_icreate_cancel(log, icl);
Dave Chinner28c8e412013-06-27 16:04:55 +10003463 return 0;
Brian Foster78d57e42015-08-19 09:58:48 +10003464 }
Dave Chinner28c8e412013-06-27 16:04:55 +10003465
Brian Foster78d57e42015-08-19 09:58:48 +10003466 trace_xfs_log_recover_icreate_recover(log, icl);
Brian Fosterfc0d1652015-08-19 09:59:38 +10003467 return xfs_ialloc_inode_init(mp, NULL, buffer_list, count, agno, agbno,
3468 length, be32_to_cpu(icl->icl_gen));
Dave Chinner28c8e412013-06-27 16:04:55 +10003469}
3470
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003471STATIC void
3472xlog_recover_buffer_ra_pass2(
3473 struct xlog *log,
3474 struct xlog_recover_item *item)
3475{
3476 struct xfs_buf_log_format *buf_f = item->ri_buf[0].i_addr;
3477 struct xfs_mount *mp = log->l_mp;
3478
Dave Chinner84a5b732013-08-27 08:10:53 +10003479 if (xlog_peek_buffer_cancelled(log, buf_f->blf_blkno,
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003480 buf_f->blf_len, buf_f->blf_flags)) {
3481 return;
3482 }
3483
3484 xfs_buf_readahead(mp->m_ddev_targp, buf_f->blf_blkno,
3485 buf_f->blf_len, NULL);
3486}
3487
3488STATIC void
3489xlog_recover_inode_ra_pass2(
3490 struct xlog *log,
3491 struct xlog_recover_item *item)
3492{
3493 struct xfs_inode_log_format ilf_buf;
3494 struct xfs_inode_log_format *ilfp;
3495 struct xfs_mount *mp = log->l_mp;
3496 int error;
3497
3498 if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) {
3499 ilfp = item->ri_buf[0].i_addr;
3500 } else {
3501 ilfp = &ilf_buf;
3502 memset(ilfp, 0, sizeof(*ilfp));
3503 error = xfs_inode_item_format_convert(&item->ri_buf[0], ilfp);
3504 if (error)
3505 return;
3506 }
3507
Dave Chinner84a5b732013-08-27 08:10:53 +10003508 if (xlog_peek_buffer_cancelled(log, ilfp->ilf_blkno, ilfp->ilf_len, 0))
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003509 return;
3510
3511 xfs_buf_readahead(mp->m_ddev_targp, ilfp->ilf_blkno,
Dave Chinnerd8914002013-08-27 11:39:37 +10003512 ilfp->ilf_len, &xfs_inode_buf_ra_ops);
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003513}
3514
3515STATIC void
3516xlog_recover_dquot_ra_pass2(
3517 struct xlog *log,
3518 struct xlog_recover_item *item)
3519{
3520 struct xfs_mount *mp = log->l_mp;
3521 struct xfs_disk_dquot *recddq;
3522 struct xfs_dq_logformat *dq_f;
3523 uint type;
Dave Chinner7d6a13f2016-01-12 07:04:01 +11003524 int len;
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003525
3526
3527 if (mp->m_qflags == 0)
3528 return;
3529
3530 recddq = item->ri_buf[1].i_addr;
3531 if (recddq == NULL)
3532 return;
3533 if (item->ri_buf[1].i_len < sizeof(struct xfs_disk_dquot))
3534 return;
3535
3536 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
3537 ASSERT(type);
3538 if (log->l_quotaoffs_flag & type)
3539 return;
3540
3541 dq_f = item->ri_buf[0].i_addr;
3542 ASSERT(dq_f);
3543 ASSERT(dq_f->qlf_len == 1);
3544
Dave Chinner7d6a13f2016-01-12 07:04:01 +11003545 len = XFS_FSB_TO_BB(mp, dq_f->qlf_len);
3546 if (xlog_peek_buffer_cancelled(log, dq_f->qlf_blkno, len, 0))
3547 return;
3548
3549 xfs_buf_readahead(mp->m_ddev_targp, dq_f->qlf_blkno, len,
3550 &xfs_dquot_buf_ra_ops);
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003551}
3552
3553STATIC void
3554xlog_recover_ra_pass2(
3555 struct xlog *log,
3556 struct xlog_recover_item *item)
3557{
3558 switch (ITEM_TYPE(item)) {
3559 case XFS_LI_BUF:
3560 xlog_recover_buffer_ra_pass2(log, item);
3561 break;
3562 case XFS_LI_INODE:
3563 xlog_recover_inode_ra_pass2(log, item);
3564 break;
3565 case XFS_LI_DQUOT:
3566 xlog_recover_dquot_ra_pass2(log, item);
3567 break;
3568 case XFS_LI_EFI:
3569 case XFS_LI_EFD:
3570 case XFS_LI_QUOTAOFF:
3571 default:
3572 break;
3573 }
3574}
3575
Linus Torvalds1da177e2005-04-16 15:20:36 -07003576STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003577xlog_recover_commit_pass1(
Mark Tinguelyad223e62012-06-14 09:22:15 -05003578 struct xlog *log,
3579 struct xlog_recover *trans,
3580 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003581{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003582 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1);
Christoph Hellwigd0450942010-12-01 22:06:23 +00003583
3584 switch (ITEM_TYPE(item)) {
3585 case XFS_LI_BUF:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003586 return xlog_recover_buffer_pass1(log, item);
Christoph Hellwigd0450942010-12-01 22:06:23 +00003587 case XFS_LI_QUOTAOFF:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003588 return xlog_recover_quotaoff_pass1(log, item);
3589 case XFS_LI_INODE:
3590 case XFS_LI_EFI:
3591 case XFS_LI_EFD:
3592 case XFS_LI_DQUOT:
Dave Chinner28c8e412013-06-27 16:04:55 +10003593 case XFS_LI_ICREATE:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003594 /* nothing to do in pass 1 */
3595 return 0;
Christoph Hellwigd0450942010-12-01 22:06:23 +00003596 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003597 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
3598 __func__, ITEM_TYPE(item));
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003599 ASSERT(0);
Dave Chinner24513372014-06-25 14:58:08 +10003600 return -EIO;
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003601 }
3602}
3603
3604STATIC int
3605xlog_recover_commit_pass2(
Mark Tinguelyad223e62012-06-14 09:22:15 -05003606 struct xlog *log,
3607 struct xlog_recover *trans,
3608 struct list_head *buffer_list,
3609 struct xlog_recover_item *item)
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003610{
3611 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2);
3612
3613 switch (ITEM_TYPE(item)) {
3614 case XFS_LI_BUF:
Dave Chinner50d5c8d2013-08-28 21:22:47 +10003615 return xlog_recover_buffer_pass2(log, buffer_list, item,
3616 trans->r_lsn);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003617 case XFS_LI_INODE:
Dave Chinner50d5c8d2013-08-28 21:22:47 +10003618 return xlog_recover_inode_pass2(log, buffer_list, item,
3619 trans->r_lsn);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003620 case XFS_LI_EFI:
3621 return xlog_recover_efi_pass2(log, item, trans->r_lsn);
3622 case XFS_LI_EFD:
3623 return xlog_recover_efd_pass2(log, item);
3624 case XFS_LI_DQUOT:
Dave Chinner50d5c8d2013-08-28 21:22:47 +10003625 return xlog_recover_dquot_pass2(log, buffer_list, item,
3626 trans->r_lsn);
Dave Chinner28c8e412013-06-27 16:04:55 +10003627 case XFS_LI_ICREATE:
3628 return xlog_recover_do_icreate_pass2(log, buffer_list, item);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003629 case XFS_LI_QUOTAOFF:
3630 /* nothing to do in pass2 */
3631 return 0;
3632 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003633 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
3634 __func__, ITEM_TYPE(item));
Christoph Hellwigd0450942010-12-01 22:06:23 +00003635 ASSERT(0);
Dave Chinner24513372014-06-25 14:58:08 +10003636 return -EIO;
Christoph Hellwigd0450942010-12-01 22:06:23 +00003637 }
3638}
3639
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003640STATIC int
3641xlog_recover_items_pass2(
3642 struct xlog *log,
3643 struct xlog_recover *trans,
3644 struct list_head *buffer_list,
3645 struct list_head *item_list)
3646{
3647 struct xlog_recover_item *item;
3648 int error = 0;
3649
3650 list_for_each_entry(item, item_list, ri_list) {
3651 error = xlog_recover_commit_pass2(log, trans,
3652 buffer_list, item);
3653 if (error)
3654 return error;
3655 }
3656
3657 return error;
3658}
3659
Christoph Hellwigd0450942010-12-01 22:06:23 +00003660/*
3661 * Perform the transaction.
3662 *
3663 * If the transaction modifies a buffer or inode, do it now. Otherwise,
3664 * EFIs and EFDs get queued up by adding entries into the AIL for them.
3665 */
3666STATIC int
3667xlog_recover_commit_trans(
Mark Tinguelyad223e62012-06-14 09:22:15 -05003668 struct xlog *log,
Christoph Hellwigd0450942010-12-01 22:06:23 +00003669 struct xlog_recover *trans,
3670 int pass)
3671{
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003672 int error = 0;
3673 int error2;
3674 int items_queued = 0;
3675 struct xlog_recover_item *item;
3676 struct xlog_recover_item *next;
3677 LIST_HEAD (buffer_list);
3678 LIST_HEAD (ra_list);
3679 LIST_HEAD (done_list);
3680
3681 #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100
Linus Torvalds1da177e2005-04-16 15:20:36 -07003682
Dave Chinnerf0a76952010-01-11 11:49:57 +00003683 hlist_del(&trans->r_list);
Christoph Hellwigd0450942010-12-01 22:06:23 +00003684
3685 error = xlog_recover_reorder_trans(log, trans, pass);
3686 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003687 return error;
Christoph Hellwigd0450942010-12-01 22:06:23 +00003688
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003689 list_for_each_entry_safe(item, next, &trans->r_itemq, ri_list) {
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003690 switch (pass) {
3691 case XLOG_RECOVER_PASS1:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003692 error = xlog_recover_commit_pass1(log, trans, item);
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003693 break;
3694 case XLOG_RECOVER_PASS2:
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003695 xlog_recover_ra_pass2(log, item);
3696 list_move_tail(&item->ri_list, &ra_list);
3697 items_queued++;
3698 if (items_queued >= XLOG_RECOVER_COMMIT_QUEUE_MAX) {
3699 error = xlog_recover_items_pass2(log, trans,
3700 &buffer_list, &ra_list);
3701 list_splice_tail_init(&ra_list, &done_list);
3702 items_queued = 0;
3703 }
3704
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003705 break;
3706 default:
3707 ASSERT(0);
3708 }
3709
Christoph Hellwigd0450942010-12-01 22:06:23 +00003710 if (error)
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003711 goto out;
Christoph Hellwigd0450942010-12-01 22:06:23 +00003712 }
3713
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003714out:
3715 if (!list_empty(&ra_list)) {
3716 if (!error)
3717 error = xlog_recover_items_pass2(log, trans,
3718 &buffer_list, &ra_list);
3719 list_splice_tail_init(&ra_list, &done_list);
3720 }
3721
3722 if (!list_empty(&done_list))
3723 list_splice_init(&done_list, &trans->r_itemq);
3724
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003725 error2 = xfs_buf_delwri_submit(&buffer_list);
3726 return error ? error : error2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003727}
3728
Dave Chinner76560662014-09-29 09:45:42 +10003729STATIC void
3730xlog_recover_add_item(
3731 struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003732{
Dave Chinner76560662014-09-29 09:45:42 +10003733 xlog_recover_item_t *item;
3734
3735 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
3736 INIT_LIST_HEAD(&item->ri_list);
3737 list_add_tail(&item->ri_list, head);
3738}
3739
3740STATIC int
3741xlog_recover_add_to_cont_trans(
3742 struct xlog *log,
3743 struct xlog_recover *trans,
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10003744 char *dp,
Dave Chinner76560662014-09-29 09:45:42 +10003745 int len)
3746{
3747 xlog_recover_item_t *item;
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10003748 char *ptr, *old_ptr;
Dave Chinner76560662014-09-29 09:45:42 +10003749 int old_len;
3750
Brian Foster89cebc842015-07-29 11:51:10 +10003751 /*
3752 * If the transaction is empty, the header was split across this and the
3753 * previous record. Copy the rest of the header.
3754 */
Dave Chinner76560662014-09-29 09:45:42 +10003755 if (list_empty(&trans->r_itemq)) {
Brian Foster848ccfc2015-11-10 10:10:33 +11003756 ASSERT(len <= sizeof(struct xfs_trans_header));
Brian Foster89cebc842015-07-29 11:51:10 +10003757 if (len > sizeof(struct xfs_trans_header)) {
3758 xfs_warn(log->l_mp, "%s: bad header length", __func__);
3759 return -EIO;
3760 }
3761
Dave Chinner76560662014-09-29 09:45:42 +10003762 xlog_recover_add_item(&trans->r_itemq);
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10003763 ptr = (char *)&trans->r_theader +
Brian Foster89cebc842015-07-29 11:51:10 +10003764 sizeof(struct xfs_trans_header) - len;
Dave Chinner76560662014-09-29 09:45:42 +10003765 memcpy(ptr, dp, len);
3766 return 0;
3767 }
Brian Foster89cebc842015-07-29 11:51:10 +10003768
Dave Chinner76560662014-09-29 09:45:42 +10003769 /* take the tail entry */
3770 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
3771
3772 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
3773 old_len = item->ri_buf[item->ri_cnt-1].i_len;
3774
3775 ptr = kmem_realloc(old_ptr, len+old_len, old_len, KM_SLEEP);
3776 memcpy(&ptr[old_len], dp, len);
3777 item->ri_buf[item->ri_cnt-1].i_len += len;
3778 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
3779 trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003780 return 0;
3781}
3782
3783/*
Dave Chinner76560662014-09-29 09:45:42 +10003784 * The next region to add is the start of a new region. It could be
3785 * a whole region or it could be the first part of a new region. Because
3786 * of this, the assumption here is that the type and size fields of all
3787 * format structures fit into the first 32 bits of the structure.
3788 *
3789 * This works because all regions must be 32 bit aligned. Therefore, we
3790 * either have both fields or we have neither field. In the case we have
3791 * neither field, the data part of the region is zero length. We only have
3792 * a log_op_header and can throw away the header since a new one will appear
3793 * later. If we have at least 4 bytes, then we can determine how many regions
3794 * will appear in the current log item.
3795 */
3796STATIC int
3797xlog_recover_add_to_trans(
3798 struct xlog *log,
3799 struct xlog_recover *trans,
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10003800 char *dp,
Dave Chinner76560662014-09-29 09:45:42 +10003801 int len)
3802{
3803 xfs_inode_log_format_t *in_f; /* any will do */
3804 xlog_recover_item_t *item;
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10003805 char *ptr;
Dave Chinner76560662014-09-29 09:45:42 +10003806
3807 if (!len)
3808 return 0;
3809 if (list_empty(&trans->r_itemq)) {
3810 /* we need to catch log corruptions here */
3811 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
3812 xfs_warn(log->l_mp, "%s: bad header magic number",
3813 __func__);
3814 ASSERT(0);
3815 return -EIO;
3816 }
Brian Foster89cebc842015-07-29 11:51:10 +10003817
3818 if (len > sizeof(struct xfs_trans_header)) {
3819 xfs_warn(log->l_mp, "%s: bad header length", __func__);
3820 ASSERT(0);
3821 return -EIO;
3822 }
3823
3824 /*
3825 * The transaction header can be arbitrarily split across op
3826 * records. If we don't have the whole thing here, copy what we
3827 * do have and handle the rest in the next record.
3828 */
3829 if (len == sizeof(struct xfs_trans_header))
Dave Chinner76560662014-09-29 09:45:42 +10003830 xlog_recover_add_item(&trans->r_itemq);
3831 memcpy(&trans->r_theader, dp, len);
3832 return 0;
3833 }
3834
3835 ptr = kmem_alloc(len, KM_SLEEP);
3836 memcpy(ptr, dp, len);
3837 in_f = (xfs_inode_log_format_t *)ptr;
3838
3839 /* take the tail entry */
3840 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
3841 if (item->ri_total != 0 &&
3842 item->ri_total == item->ri_cnt) {
3843 /* tail item is in use, get a new one */
3844 xlog_recover_add_item(&trans->r_itemq);
3845 item = list_entry(trans->r_itemq.prev,
3846 xlog_recover_item_t, ri_list);
3847 }
3848
3849 if (item->ri_total == 0) { /* first region to be added */
3850 if (in_f->ilf_size == 0 ||
3851 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
3852 xfs_warn(log->l_mp,
3853 "bad number of regions (%d) in inode log format",
3854 in_f->ilf_size);
3855 ASSERT(0);
3856 kmem_free(ptr);
3857 return -EIO;
3858 }
3859
3860 item->ri_total = in_f->ilf_size;
3861 item->ri_buf =
3862 kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
3863 KM_SLEEP);
3864 }
3865 ASSERT(item->ri_total > item->ri_cnt);
3866 /* Description region is ri_buf[0] */
3867 item->ri_buf[item->ri_cnt].i_addr = ptr;
3868 item->ri_buf[item->ri_cnt].i_len = len;
3869 item->ri_cnt++;
3870 trace_xfs_log_recover_item_add(log, trans, item, 0);
3871 return 0;
3872}
Dave Chinnerb818cca2014-09-29 09:45:54 +10003873
Dave Chinner76560662014-09-29 09:45:42 +10003874/*
3875 * Free up any resources allocated by the transaction
3876 *
3877 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
3878 */
3879STATIC void
3880xlog_recover_free_trans(
3881 struct xlog_recover *trans)
3882{
3883 xlog_recover_item_t *item, *n;
3884 int i;
3885
3886 list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
3887 /* Free the regions in the item. */
3888 list_del(&item->ri_list);
3889 for (i = 0; i < item->ri_cnt; i++)
3890 kmem_free(item->ri_buf[i].i_addr);
3891 /* Free the item itself */
3892 kmem_free(item->ri_buf);
3893 kmem_free(item);
3894 }
3895 /* Free the transaction recover structure */
3896 kmem_free(trans);
3897}
3898
Dave Chinnere9131e52014-09-29 09:45:18 +10003899/*
3900 * On error or completion, trans is freed.
3901 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003902STATIC int
Dave Chinnereeb11682014-09-29 09:45:03 +10003903xlog_recovery_process_trans(
3904 struct xlog *log,
3905 struct xlog_recover *trans,
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10003906 char *dp,
Dave Chinnereeb11682014-09-29 09:45:03 +10003907 unsigned int len,
3908 unsigned int flags,
3909 int pass)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003910{
Dave Chinnere9131e52014-09-29 09:45:18 +10003911 int error = 0;
3912 bool freeit = false;
Dave Chinnereeb11682014-09-29 09:45:03 +10003913
3914 /* mask off ophdr transaction container flags */
3915 flags &= ~XLOG_END_TRANS;
3916 if (flags & XLOG_WAS_CONT_TRANS)
3917 flags &= ~XLOG_CONTINUE_TRANS;
3918
Dave Chinner88b863d2014-09-29 09:45:32 +10003919 /*
3920 * Callees must not free the trans structure. We'll decide if we need to
3921 * free it or not based on the operation being done and it's result.
3922 */
Dave Chinnereeb11682014-09-29 09:45:03 +10003923 switch (flags) {
3924 /* expected flag values */
3925 case 0:
3926 case XLOG_CONTINUE_TRANS:
3927 error = xlog_recover_add_to_trans(log, trans, dp, len);
3928 break;
3929 case XLOG_WAS_CONT_TRANS:
3930 error = xlog_recover_add_to_cont_trans(log, trans, dp, len);
3931 break;
3932 case XLOG_COMMIT_TRANS:
3933 error = xlog_recover_commit_trans(log, trans, pass);
Dave Chinner88b863d2014-09-29 09:45:32 +10003934 /* success or fail, we are now done with this transaction. */
3935 freeit = true;
Dave Chinnereeb11682014-09-29 09:45:03 +10003936 break;
3937
3938 /* unexpected flag values */
3939 case XLOG_UNMOUNT_TRANS:
Dave Chinnere9131e52014-09-29 09:45:18 +10003940 /* just skip trans */
Dave Chinnereeb11682014-09-29 09:45:03 +10003941 xfs_warn(log->l_mp, "%s: Unmount LR", __func__);
Dave Chinnere9131e52014-09-29 09:45:18 +10003942 freeit = true;
Dave Chinnereeb11682014-09-29 09:45:03 +10003943 break;
3944 case XLOG_START_TRANS:
Dave Chinnereeb11682014-09-29 09:45:03 +10003945 default:
3946 xfs_warn(log->l_mp, "%s: bad flag 0x%x", __func__, flags);
3947 ASSERT(0);
Dave Chinnere9131e52014-09-29 09:45:18 +10003948 error = -EIO;
Dave Chinnereeb11682014-09-29 09:45:03 +10003949 break;
3950 }
Dave Chinnere9131e52014-09-29 09:45:18 +10003951 if (error || freeit)
3952 xlog_recover_free_trans(trans);
Dave Chinnereeb11682014-09-29 09:45:03 +10003953 return error;
3954}
3955
Dave Chinnerb818cca2014-09-29 09:45:54 +10003956/*
3957 * Lookup the transaction recovery structure associated with the ID in the
3958 * current ophdr. If the transaction doesn't exist and the start flag is set in
3959 * the ophdr, then allocate a new transaction for future ID matches to find.
3960 * Either way, return what we found during the lookup - an existing transaction
3961 * or nothing.
3962 */
Dave Chinnereeb11682014-09-29 09:45:03 +10003963STATIC struct xlog_recover *
3964xlog_recover_ophdr_to_trans(
3965 struct hlist_head rhash[],
3966 struct xlog_rec_header *rhead,
3967 struct xlog_op_header *ohead)
3968{
3969 struct xlog_recover *trans;
3970 xlog_tid_t tid;
3971 struct hlist_head *rhp;
3972
3973 tid = be32_to_cpu(ohead->oh_tid);
3974 rhp = &rhash[XLOG_RHASH(tid)];
Dave Chinnerb818cca2014-09-29 09:45:54 +10003975 hlist_for_each_entry(trans, rhp, r_list) {
3976 if (trans->r_log_tid == tid)
3977 return trans;
3978 }
Dave Chinnereeb11682014-09-29 09:45:03 +10003979
3980 /*
Dave Chinnerb818cca2014-09-29 09:45:54 +10003981 * skip over non-start transaction headers - we could be
3982 * processing slack space before the next transaction starts
Dave Chinnereeb11682014-09-29 09:45:03 +10003983 */
Dave Chinnerb818cca2014-09-29 09:45:54 +10003984 if (!(ohead->oh_flags & XLOG_START_TRANS))
3985 return NULL;
3986
3987 ASSERT(be32_to_cpu(ohead->oh_len) == 0);
3988
3989 /*
3990 * This is a new transaction so allocate a new recovery container to
3991 * hold the recovery ops that will follow.
3992 */
3993 trans = kmem_zalloc(sizeof(struct xlog_recover), KM_SLEEP);
3994 trans->r_log_tid = tid;
3995 trans->r_lsn = be64_to_cpu(rhead->h_lsn);
3996 INIT_LIST_HEAD(&trans->r_itemq);
3997 INIT_HLIST_NODE(&trans->r_list);
3998 hlist_add_head(&trans->r_list, rhp);
3999
4000 /*
4001 * Nothing more to do for this ophdr. Items to be added to this new
4002 * transaction will be in subsequent ophdr containers.
4003 */
Dave Chinnereeb11682014-09-29 09:45:03 +10004004 return NULL;
4005}
4006
4007STATIC int
4008xlog_recover_process_ophdr(
4009 struct xlog *log,
4010 struct hlist_head rhash[],
4011 struct xlog_rec_header *rhead,
4012 struct xlog_op_header *ohead,
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10004013 char *dp,
4014 char *end,
Dave Chinnereeb11682014-09-29 09:45:03 +10004015 int pass)
4016{
4017 struct xlog_recover *trans;
Dave Chinnereeb11682014-09-29 09:45:03 +10004018 unsigned int len;
4019
4020 /* Do we understand who wrote this op? */
4021 if (ohead->oh_clientid != XFS_TRANSACTION &&
4022 ohead->oh_clientid != XFS_LOG) {
4023 xfs_warn(log->l_mp, "%s: bad clientid 0x%x",
4024 __func__, ohead->oh_clientid);
4025 ASSERT(0);
4026 return -EIO;
4027 }
4028
4029 /*
4030 * Check the ophdr contains all the data it is supposed to contain.
4031 */
4032 len = be32_to_cpu(ohead->oh_len);
4033 if (dp + len > end) {
4034 xfs_warn(log->l_mp, "%s: bad length 0x%x", __func__, len);
4035 WARN_ON(1);
4036 return -EIO;
4037 }
4038
4039 trans = xlog_recover_ophdr_to_trans(rhash, rhead, ohead);
4040 if (!trans) {
4041 /* nothing to do, so skip over this ophdr */
4042 return 0;
4043 }
4044
Dave Chinnere9131e52014-09-29 09:45:18 +10004045 return xlog_recovery_process_trans(log, trans, dp, len,
4046 ohead->oh_flags, pass);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004047}
4048
4049/*
4050 * There are two valid states of the r_state field. 0 indicates that the
4051 * transaction structure is in a normal state. We have either seen the
4052 * start of the transaction or the last operation we added was not a partial
4053 * operation. If the last operation we added to the transaction was a
4054 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
4055 *
4056 * NOTE: skip LRs with 0 data length.
4057 */
4058STATIC int
4059xlog_recover_process_data(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004060 struct xlog *log,
Dave Chinnerf0a76952010-01-11 11:49:57 +00004061 struct hlist_head rhash[],
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004062 struct xlog_rec_header *rhead,
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10004063 char *dp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004064 int pass)
4065{
Dave Chinnereeb11682014-09-29 09:45:03 +10004066 struct xlog_op_header *ohead;
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10004067 char *end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004068 int num_logops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004069 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004070
Dave Chinnereeb11682014-09-29 09:45:03 +10004071 end = dp + be32_to_cpu(rhead->h_len);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004072 num_logops = be32_to_cpu(rhead->h_num_logops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004073
4074 /* check the log format matches our own - else we can't recover */
4075 if (xlog_header_check_recover(log->l_mp, rhead))
Dave Chinner24513372014-06-25 14:58:08 +10004076 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004077
Dave Chinnereeb11682014-09-29 09:45:03 +10004078 while ((dp < end) && num_logops) {
4079
4080 ohead = (struct xlog_op_header *)dp;
4081 dp += sizeof(*ohead);
4082 ASSERT(dp <= end);
4083
4084 /* errors will abort recovery */
4085 error = xlog_recover_process_ophdr(log, rhash, rhead, ohead,
4086 dp, end, pass);
4087 if (error)
4088 return error;
4089
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10004090 dp += be32_to_cpu(ohead->oh_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004091 num_logops--;
4092 }
4093 return 0;
4094}
4095
4096/*
4097 * Process an extent free intent item that was recovered from
4098 * the log. We need to free the extents that it describes.
4099 */
David Chinner3c1e2bb2008-04-10 12:21:11 +10004100STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07004101xlog_recover_process_efi(
4102 xfs_mount_t *mp,
4103 xfs_efi_log_item_t *efip)
4104{
4105 xfs_efd_log_item_t *efdp;
4106 xfs_trans_t *tp;
4107 int i;
David Chinner3c1e2bb2008-04-10 12:21:11 +10004108 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004109 xfs_extent_t *extp;
4110 xfs_fsblock_t startblock_fsb;
4111
Dave Chinnerb199c8a2010-12-20 11:59:49 +11004112 ASSERT(!test_bit(XFS_EFI_RECOVERED, &efip->efi_flags));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004113
4114 /*
4115 * First check the validity of the extents described by the
4116 * EFI. If any are bad, then assume that all are bad and
4117 * just toss the EFI.
4118 */
4119 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
4120 extp = &(efip->efi_format.efi_extents[i]);
4121 startblock_fsb = XFS_BB_TO_FSB(mp,
4122 XFS_FSB_TO_DADDR(mp, extp->ext_start));
4123 if ((startblock_fsb == 0) ||
4124 (extp->ext_len == 0) ||
4125 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
4126 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
4127 /*
4128 * This will pull the EFI from the AIL and
4129 * free the memory associated with it.
4130 */
Dave Chinner666d6442013-04-03 14:09:21 +11004131 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
Brian Foster5e4b5382015-08-19 09:50:12 +10004132 xfs_efi_release(efip);
Dave Chinner24513372014-06-25 14:58:08 +10004133 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004134 }
4135 }
4136
4137 tp = xfs_trans_alloc(mp, 0);
Jie Liu3d3c8b52013-08-12 20:49:59 +10004138 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
David Chinnerfc6149d2008-04-10 12:21:53 +10004139 if (error)
4140 goto abort_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004141 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
4142
4143 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
4144 extp = &(efip->efi_format.efi_extents[i]);
Brian Foster6bc43af2015-08-19 09:51:43 +10004145 error = xfs_trans_free_extent(tp, efdp, extp->ext_start,
4146 extp->ext_len);
David Chinnerfc6149d2008-04-10 12:21:53 +10004147 if (error)
4148 goto abort_error;
Brian Foster6bc43af2015-08-19 09:51:43 +10004149
Linus Torvalds1da177e2005-04-16 15:20:36 -07004150 }
4151
Dave Chinnerb199c8a2010-12-20 11:59:49 +11004152 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
Christoph Hellwig70393312015-06-04 13:48:08 +10004153 error = xfs_trans_commit(tp);
David Chinner3c1e2bb2008-04-10 12:21:11 +10004154 return error;
David Chinnerfc6149d2008-04-10 12:21:53 +10004155
4156abort_error:
Christoph Hellwig4906e212015-06-04 13:47:56 +10004157 xfs_trans_cancel(tp);
David Chinnerfc6149d2008-04-10 12:21:53 +10004158 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004159}
4160
4161/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07004162 * When this is called, all of the EFIs which did not have
4163 * corresponding EFDs should be in the AIL. What we do now
4164 * is free the extents associated with each one.
4165 *
4166 * Since we process the EFIs in normal transactions, they
4167 * will be removed at some point after the commit. This prevents
4168 * us from just walking down the list processing each one.
4169 * We'll use a flag in the EFI to skip those that we've already
4170 * processed and use the AIL iteration mechanism's generation
4171 * count to try to speed this up at least a bit.
4172 *
4173 * When we start, we know that the EFIs are the only things in
4174 * the AIL. As we process them, however, other items are added
4175 * to the AIL. Since everything added to the AIL must come after
4176 * everything already in the AIL, we stop processing as soon as
4177 * we see something other than an EFI in the AIL.
4178 */
David Chinner3c1e2bb2008-04-10 12:21:11 +10004179STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07004180xlog_recover_process_efis(
Brian Fosterf0b2efa2015-08-19 09:58:36 +10004181 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004182{
Brian Fosterf0b2efa2015-08-19 09:58:36 +10004183 struct xfs_log_item *lip;
4184 struct xfs_efi_log_item *efip;
David Chinner3c1e2bb2008-04-10 12:21:11 +10004185 int error = 0;
David Chinner27d8d5f2008-10-30 17:38:39 +11004186 struct xfs_ail_cursor cur;
David Chinnera9c21c12008-10-30 17:39:35 +11004187 struct xfs_ail *ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004188
David Chinnera9c21c12008-10-30 17:39:35 +11004189 ailp = log->l_ailp;
4190 spin_lock(&ailp->xa_lock);
4191 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004192 while (lip != NULL) {
4193 /*
4194 * We're done when we see something other than an EFI.
David Chinner27d8d5f2008-10-30 17:38:39 +11004195 * There should be no EFIs left in the AIL now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004196 */
4197 if (lip->li_type != XFS_LI_EFI) {
David Chinner27d8d5f2008-10-30 17:38:39 +11004198#ifdef DEBUG
David Chinnera9c21c12008-10-30 17:39:35 +11004199 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
David Chinner27d8d5f2008-10-30 17:38:39 +11004200 ASSERT(lip->li_type != XFS_LI_EFI);
4201#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004202 break;
4203 }
4204
4205 /*
4206 * Skip EFIs that we've already processed.
4207 */
Brian Fosterf0b2efa2015-08-19 09:58:36 +10004208 efip = container_of(lip, struct xfs_efi_log_item, efi_item);
Dave Chinnerb199c8a2010-12-20 11:59:49 +11004209 if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) {
David Chinnera9c21c12008-10-30 17:39:35 +11004210 lip = xfs_trans_ail_cursor_next(ailp, &cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004211 continue;
4212 }
4213
David Chinnera9c21c12008-10-30 17:39:35 +11004214 spin_unlock(&ailp->xa_lock);
4215 error = xlog_recover_process_efi(log->l_mp, efip);
4216 spin_lock(&ailp->xa_lock);
David Chinner27d8d5f2008-10-30 17:38:39 +11004217 if (error)
4218 goto out;
David Chinnera9c21c12008-10-30 17:39:35 +11004219 lip = xfs_trans_ail_cursor_next(ailp, &cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004220 }
David Chinner27d8d5f2008-10-30 17:38:39 +11004221out:
Eric Sandeene4a1e292014-04-14 19:06:05 +10004222 xfs_trans_ail_cursor_done(&cur);
David Chinnera9c21c12008-10-30 17:39:35 +11004223 spin_unlock(&ailp->xa_lock);
David Chinner3c1e2bb2008-04-10 12:21:11 +10004224 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004225}
4226
4227/*
Brian Fosterf0b2efa2015-08-19 09:58:36 +10004228 * A cancel occurs when the mount has failed and we're bailing out. Release all
4229 * pending EFIs so they don't pin the AIL.
4230 */
4231STATIC int
4232xlog_recover_cancel_efis(
4233 struct xlog *log)
4234{
4235 struct xfs_log_item *lip;
4236 struct xfs_efi_log_item *efip;
4237 int error = 0;
4238 struct xfs_ail_cursor cur;
4239 struct xfs_ail *ailp;
4240
4241 ailp = log->l_ailp;
4242 spin_lock(&ailp->xa_lock);
4243 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
4244 while (lip != NULL) {
4245 /*
4246 * We're done when we see something other than an EFI.
4247 * There should be no EFIs left in the AIL now.
4248 */
4249 if (lip->li_type != XFS_LI_EFI) {
4250#ifdef DEBUG
4251 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
4252 ASSERT(lip->li_type != XFS_LI_EFI);
4253#endif
4254 break;
4255 }
4256
4257 efip = container_of(lip, struct xfs_efi_log_item, efi_item);
4258
4259 spin_unlock(&ailp->xa_lock);
4260 xfs_efi_release(efip);
4261 spin_lock(&ailp->xa_lock);
4262
4263 lip = xfs_trans_ail_cursor_next(ailp, &cur);
4264 }
4265
4266 xfs_trans_ail_cursor_done(&cur);
4267 spin_unlock(&ailp->xa_lock);
4268 return error;
4269}
4270
4271/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07004272 * This routine performs a transaction to null out a bad inode pointer
4273 * in an agi unlinked inode hash bucket.
4274 */
4275STATIC void
4276xlog_recover_clear_agi_bucket(
4277 xfs_mount_t *mp,
4278 xfs_agnumber_t agno,
4279 int bucket)
4280{
4281 xfs_trans_t *tp;
4282 xfs_agi_t *agi;
4283 xfs_buf_t *agibp;
4284 int offset;
4285 int error;
4286
4287 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
Jie Liu3d3c8b52013-08-12 20:49:59 +10004288 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_clearagi, 0, 0);
David Chinnere5720ee2008-04-10 12:21:18 +10004289 if (error)
4290 goto out_abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004291
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11004292 error = xfs_read_agi(mp, tp, agno, &agibp);
4293 if (error)
David Chinnere5720ee2008-04-10 12:21:18 +10004294 goto out_abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004295
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11004296 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11004297 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004298 offset = offsetof(xfs_agi_t, agi_unlinked) +
4299 (sizeof(xfs_agino_t) * bucket);
4300 xfs_trans_log_buf(tp, agibp, offset,
4301 (offset + sizeof(xfs_agino_t) - 1));
4302
Christoph Hellwig70393312015-06-04 13:48:08 +10004303 error = xfs_trans_commit(tp);
David Chinnere5720ee2008-04-10 12:21:18 +10004304 if (error)
4305 goto out_error;
4306 return;
4307
4308out_abort:
Christoph Hellwig4906e212015-06-04 13:47:56 +10004309 xfs_trans_cancel(tp);
David Chinnere5720ee2008-04-10 12:21:18 +10004310out_error:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004311 xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno);
David Chinnere5720ee2008-04-10 12:21:18 +10004312 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004313}
4314
Christoph Hellwig23fac502008-11-28 14:23:40 +11004315STATIC xfs_agino_t
4316xlog_recover_process_one_iunlink(
4317 struct xfs_mount *mp,
4318 xfs_agnumber_t agno,
4319 xfs_agino_t agino,
4320 int bucket)
4321{
4322 struct xfs_buf *ibp;
4323 struct xfs_dinode *dip;
4324 struct xfs_inode *ip;
4325 xfs_ino_t ino;
4326 int error;
4327
4328 ino = XFS_AGINO_TO_INO(mp, agno, agino);
Dave Chinner7b6259e2010-06-24 11:35:17 +10004329 error = xfs_iget(mp, NULL, ino, 0, 0, &ip);
Christoph Hellwig23fac502008-11-28 14:23:40 +11004330 if (error)
4331 goto fail;
4332
4333 /*
4334 * Get the on disk inode to find the next inode in the bucket.
4335 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04004336 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &ibp, 0, 0);
Christoph Hellwig23fac502008-11-28 14:23:40 +11004337 if (error)
Christoph Hellwig0e446672008-11-28 14:23:42 +11004338 goto fail_iput;
Christoph Hellwig23fac502008-11-28 14:23:40 +11004339
Christoph Hellwig23fac502008-11-28 14:23:40 +11004340 ASSERT(ip->i_d.di_nlink == 0);
Christoph Hellwig0e446672008-11-28 14:23:42 +11004341 ASSERT(ip->i_d.di_mode != 0);
Christoph Hellwig23fac502008-11-28 14:23:40 +11004342
4343 /* setup for the next pass */
4344 agino = be32_to_cpu(dip->di_next_unlinked);
4345 xfs_buf_relse(ibp);
4346
4347 /*
4348 * Prevent any DMAPI event from being sent when the reference on
4349 * the inode is dropped.
4350 */
4351 ip->i_d.di_dmevmask = 0;
4352
Christoph Hellwig0e446672008-11-28 14:23:42 +11004353 IRELE(ip);
Christoph Hellwig23fac502008-11-28 14:23:40 +11004354 return agino;
4355
Christoph Hellwig0e446672008-11-28 14:23:42 +11004356 fail_iput:
4357 IRELE(ip);
Christoph Hellwig23fac502008-11-28 14:23:40 +11004358 fail:
4359 /*
4360 * We can't read in the inode this bucket points to, or this inode
4361 * is messed up. Just ditch this bucket of inodes. We will lose
4362 * some inodes and space, but at least we won't hang.
4363 *
4364 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
4365 * clear the inode pointer in the bucket.
4366 */
4367 xlog_recover_clear_agi_bucket(mp, agno, bucket);
4368 return NULLAGINO;
4369}
4370
Linus Torvalds1da177e2005-04-16 15:20:36 -07004371/*
4372 * xlog_iunlink_recover
4373 *
4374 * This is called during recovery to process any inodes which
4375 * we unlinked but not freed when the system crashed. These
4376 * inodes will be on the lists in the AGI blocks. What we do
4377 * here is scan all the AGIs and fully truncate and free any
4378 * inodes found on the lists. Each inode is removed from the
4379 * lists when it has been fully truncated and is freed. The
4380 * freeing of the inode and its removal from the list must be
4381 * atomic.
4382 */
Eric Sandeend96f8f82009-07-02 00:09:33 -05004383STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07004384xlog_recover_process_iunlinks(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004385 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004386{
4387 xfs_mount_t *mp;
4388 xfs_agnumber_t agno;
4389 xfs_agi_t *agi;
4390 xfs_buf_t *agibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004391 xfs_agino_t agino;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004392 int bucket;
4393 int error;
4394 uint mp_dmevmask;
4395
4396 mp = log->l_mp;
4397
4398 /*
4399 * Prevent any DMAPI event from being sent while in this function.
4400 */
4401 mp_dmevmask = mp->m_dmevmask;
4402 mp->m_dmevmask = 0;
4403
4404 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
4405 /*
4406 * Find the agi for this ag.
4407 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11004408 error = xfs_read_agi(mp, NULL, agno, &agibp);
4409 if (error) {
4410 /*
4411 * AGI is b0rked. Don't process it.
4412 *
4413 * We should probably mark the filesystem as corrupt
4414 * after we've recovered all the ag's we can....
4415 */
4416 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004417 }
Jan Karad97d32e2012-03-15 09:34:02 +00004418 /*
4419 * Unlock the buffer so that it can be acquired in the normal
4420 * course of the transaction to truncate and free each inode.
4421 * Because we are not racing with anyone else here for the AGI
4422 * buffer, we don't even need to hold it locked to read the
4423 * initial unlinked bucket entries out of the buffer. We keep
4424 * buffer reference though, so that it stays pinned in memory
4425 * while we need the buffer.
4426 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004427 agi = XFS_BUF_TO_AGI(agibp);
Jan Karad97d32e2012-03-15 09:34:02 +00004428 xfs_buf_unlock(agibp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004429
4430 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
Christoph Hellwig16259e72005-11-02 15:11:25 +11004431 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004432 while (agino != NULLAGINO) {
Christoph Hellwig23fac502008-11-28 14:23:40 +11004433 agino = xlog_recover_process_one_iunlink(mp,
4434 agno, agino, bucket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004435 }
4436 }
Jan Karad97d32e2012-03-15 09:34:02 +00004437 xfs_buf_rele(agibp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004438 }
4439
4440 mp->m_dmevmask = mp_dmevmask;
4441}
4442
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004443STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07004444xlog_unpack_data(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004445 struct xlog_rec_header *rhead,
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10004446 char *dp,
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004447 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004448{
4449 int i, j, k;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004450
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004451 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07004452 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004453 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07004454 dp += BBSIZE;
4455 }
4456
Eric Sandeen62118702008-03-06 13:44:28 +11004457 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Christoph Hellwigb28708d2008-11-28 14:23:38 +11004458 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004459 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004460 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
4461 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004462 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
Linus Torvalds1da177e2005-04-16 15:20:36 -07004463 dp += BBSIZE;
4464 }
4465 }
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004466
4467 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004468}
4469
Brian Foster9d949012016-01-04 15:55:10 +11004470/*
Brian Fosterb94fb2d2016-01-04 15:55:10 +11004471 * CRC check, unpack and process a log record.
Brian Foster9d949012016-01-04 15:55:10 +11004472 */
4473STATIC int
4474xlog_recover_process(
4475 struct xlog *log,
4476 struct hlist_head rhash[],
4477 struct xlog_rec_header *rhead,
4478 char *dp,
4479 int pass)
4480{
4481 int error;
Brian Fosterb94fb2d2016-01-04 15:55:10 +11004482 __le32 crc;
4483
Brian Fosterb94fb2d2016-01-04 15:55:10 +11004484 crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len));
Brian Foster65282502016-01-04 15:55:10 +11004485
4486 /*
4487 * Nothing else to do if this is a CRC verification pass. Just return
4488 * if this a record with a non-zero crc. Unfortunately, mkfs always
4489 * sets h_crc to 0 so we must consider this valid even on v5 supers.
4490 * Otherwise, return EFSBADCRC on failure so the callers up the stack
4491 * know precisely what failed.
4492 */
4493 if (pass == XLOG_RECOVER_CRCPASS) {
4494 if (rhead->h_crc && crc != le32_to_cpu(rhead->h_crc))
4495 return -EFSBADCRC;
4496 return 0;
4497 }
4498
4499 /*
4500 * We're in the normal recovery path. Issue a warning if and only if the
4501 * CRC in the header is non-zero. This is an advisory warning and the
4502 * zero CRC check prevents warnings from being emitted when upgrading
4503 * the kernel from one that does not add CRCs by default.
4504 */
Brian Fosterb94fb2d2016-01-04 15:55:10 +11004505 if (crc != le32_to_cpu(rhead->h_crc)) {
4506 if (rhead->h_crc || xfs_sb_version_hascrc(&log->l_mp->m_sb)) {
4507 xfs_alert(log->l_mp,
4508 "log record CRC mismatch: found 0x%x, expected 0x%x.",
4509 le32_to_cpu(rhead->h_crc),
4510 le32_to_cpu(crc));
4511 xfs_hex_dump(dp, 32);
4512 }
4513
4514 /*
4515 * If the filesystem is CRC enabled, this mismatch becomes a
4516 * fatal log corruption failure.
4517 */
4518 if (xfs_sb_version_hascrc(&log->l_mp->m_sb))
4519 return -EFSCORRUPTED;
4520 }
Brian Foster9d949012016-01-04 15:55:10 +11004521
4522 error = xlog_unpack_data(rhead, dp, log);
4523 if (error)
4524 return error;
4525
4526 return xlog_recover_process_data(log, rhash, rhead, dp, pass);
4527}
4528
Linus Torvalds1da177e2005-04-16 15:20:36 -07004529STATIC int
4530xlog_valid_rec_header(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004531 struct xlog *log,
4532 struct xlog_rec_header *rhead,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004533 xfs_daddr_t blkno)
4534{
4535 int hlen;
4536
Christoph Hellwig69ef9212011-07-08 14:36:05 +02004537 if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004538 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
4539 XFS_ERRLEVEL_LOW, log->l_mp);
Dave Chinner24513372014-06-25 14:58:08 +10004540 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004541 }
4542 if (unlikely(
4543 (!rhead->h_version ||
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004544 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004545 xfs_warn(log->l_mp, "%s: unrecognised log version (%d).",
Harvey Harrison34a622b2008-04-10 12:19:21 +10004546 __func__, be32_to_cpu(rhead->h_version));
Dave Chinner24513372014-06-25 14:58:08 +10004547 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004548 }
4549
4550 /* LR body must have data or it wouldn't have been written */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004551 hlen = be32_to_cpu(rhead->h_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004552 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
4553 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
4554 XFS_ERRLEVEL_LOW, log->l_mp);
Dave Chinner24513372014-06-25 14:58:08 +10004555 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004556 }
4557 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
4558 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
4559 XFS_ERRLEVEL_LOW, log->l_mp);
Dave Chinner24513372014-06-25 14:58:08 +10004560 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004561 }
4562 return 0;
4563}
4564
4565/*
4566 * Read the log from tail to head and process the log records found.
4567 * Handle the two cases where the tail and head are in the same cycle
4568 * and where the active portion of the log wraps around the end of
4569 * the physical log separately. The pass parameter is passed through
4570 * to the routines called to process the data and is not looked at
4571 * here.
4572 */
4573STATIC int
4574xlog_do_recovery_pass(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004575 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004576 xfs_daddr_t head_blk,
4577 xfs_daddr_t tail_blk,
Brian Fosterd7f37692016-01-04 15:55:10 +11004578 int pass,
4579 xfs_daddr_t *first_bad) /* out: first bad log rec */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004580{
4581 xlog_rec_header_t *rhead;
4582 xfs_daddr_t blk_no;
Brian Fosterd7f37692016-01-04 15:55:10 +11004583 xfs_daddr_t rhead_blk;
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10004584 char *offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004585 xfs_buf_t *hbp, *dbp;
Brian Fostera70f9fe2016-01-04 15:55:10 +11004586 int error = 0, h_size, h_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004587 int bblks, split_bblks;
4588 int hblks, split_hblks, wrapped_hblks;
Dave Chinnerf0a76952010-01-11 11:49:57 +00004589 struct hlist_head rhash[XLOG_RHASH_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -07004590
4591 ASSERT(head_blk != tail_blk);
Brian Fosterd7f37692016-01-04 15:55:10 +11004592 rhead_blk = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004593
4594 /*
4595 * Read the header of the tail block and get the iclog buffer size from
4596 * h_size. Use this to tell how many sectors make up the log header.
4597 */
Eric Sandeen62118702008-03-06 13:44:28 +11004598 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004599 /*
4600 * When using variable length iclogs, read first sector of
4601 * iclog header and extract the header size from it. Get a
4602 * new hbp that is the correct size.
4603 */
4604 hbp = xlog_get_bp(log, 1);
4605 if (!hbp)
Dave Chinner24513372014-06-25 14:58:08 +10004606 return -ENOMEM;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004607
4608 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
4609 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004610 goto bread_err1;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004611
Linus Torvalds1da177e2005-04-16 15:20:36 -07004612 rhead = (xlog_rec_header_t *)offset;
4613 error = xlog_valid_rec_header(log, rhead, tail_blk);
4614 if (error)
4615 goto bread_err1;
Brian Fostera70f9fe2016-01-04 15:55:10 +11004616
4617 /*
4618 * xfsprogs has a bug where record length is based on lsunit but
4619 * h_size (iclog size) is hardcoded to 32k. Now that we
4620 * unconditionally CRC verify the unmount record, this means the
4621 * log buffer can be too small for the record and cause an
4622 * overrun.
4623 *
4624 * Detect this condition here. Use lsunit for the buffer size as
4625 * long as this looks like the mkfs case. Otherwise, return an
4626 * error to avoid a buffer overrun.
4627 */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004628 h_size = be32_to_cpu(rhead->h_size);
Brian Fostera70f9fe2016-01-04 15:55:10 +11004629 h_len = be32_to_cpu(rhead->h_len);
4630 if (h_len > h_size) {
4631 if (h_len <= log->l_mp->m_logbsize &&
4632 be32_to_cpu(rhead->h_num_logops) == 1) {
4633 xfs_warn(log->l_mp,
4634 "invalid iclog size (%d bytes), using lsunit (%d bytes)",
4635 h_size, log->l_mp->m_logbsize);
4636 h_size = log->l_mp->m_logbsize;
4637 } else
4638 return -EFSCORRUPTED;
4639 }
4640
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004641 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07004642 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
4643 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
4644 if (h_size % XLOG_HEADER_CYCLE_SIZE)
4645 hblks++;
4646 xlog_put_bp(hbp);
4647 hbp = xlog_get_bp(log, hblks);
4648 } else {
4649 hblks = 1;
4650 }
4651 } else {
Alex Elder69ce58f2010-04-20 17:09:59 +10004652 ASSERT(log->l_sectBBsize == 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004653 hblks = 1;
4654 hbp = xlog_get_bp(log, 1);
4655 h_size = XLOG_BIG_RECORD_BSIZE;
4656 }
4657
4658 if (!hbp)
Dave Chinner24513372014-06-25 14:58:08 +10004659 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004660 dbp = xlog_get_bp(log, BTOBB(h_size));
4661 if (!dbp) {
4662 xlog_put_bp(hbp);
Dave Chinner24513372014-06-25 14:58:08 +10004663 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004664 }
4665
4666 memset(rhash, 0, sizeof(rhash));
Brian Fosterd7f37692016-01-04 15:55:10 +11004667 blk_no = rhead_blk = tail_blk;
Eric Sandeen970fd3f2014-09-09 11:57:29 +10004668 if (tail_blk > head_blk) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004669 /*
4670 * Perform recovery around the end of the physical log.
4671 * When the head is not on the same cycle number as the tail,
Eric Sandeen970fd3f2014-09-09 11:57:29 +10004672 * we can't do a sequential recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004673 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004674 while (blk_no < log->l_logBBsize) {
4675 /*
4676 * Check for header wrapping around physical end-of-log
4677 */
Chandra Seetharaman62926042011-07-22 23:40:15 +00004678 offset = hbp->b_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004679 split_hblks = 0;
4680 wrapped_hblks = 0;
4681 if (blk_no + hblks <= log->l_logBBsize) {
4682 /* Read header in one read */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004683 error = xlog_bread(log, blk_no, hblks, hbp,
4684 &offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004685 if (error)
4686 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004687 } else {
4688 /* This LR is split across physical log end */
4689 if (blk_no != log->l_logBBsize) {
4690 /* some data before physical log end */
4691 ASSERT(blk_no <= INT_MAX);
4692 split_hblks = log->l_logBBsize - (int)blk_no;
4693 ASSERT(split_hblks > 0);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004694 error = xlog_bread(log, blk_no,
4695 split_hblks, hbp,
4696 &offset);
4697 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004698 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004699 }
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004700
Linus Torvalds1da177e2005-04-16 15:20:36 -07004701 /*
4702 * Note: this black magic still works with
4703 * large sector sizes (non-512) only because:
4704 * - we increased the buffer size originally
4705 * by 1 sector giving us enough extra space
4706 * for the second read;
4707 * - the log start is guaranteed to be sector
4708 * aligned;
4709 * - we read the log end (LR header start)
4710 * _first_, then the log start (LR header end)
4711 * - order is important.
4712 */
David Chinner234f56a2008-04-10 12:24:24 +10004713 wrapped_hblks = hblks - split_hblks;
Dave Chinner44396472011-04-21 09:34:27 +00004714 error = xlog_bread_offset(log, 0,
4715 wrapped_hblks, hbp,
4716 offset + BBTOB(split_hblks));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004717 if (error)
4718 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004719 }
4720 rhead = (xlog_rec_header_t *)offset;
4721 error = xlog_valid_rec_header(log, rhead,
4722 split_hblks ? blk_no : 0);
4723 if (error)
4724 goto bread_err2;
4725
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004726 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004727 blk_no += hblks;
4728
4729 /* Read in data for log record */
4730 if (blk_no + bblks <= log->l_logBBsize) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004731 error = xlog_bread(log, blk_no, bblks, dbp,
4732 &offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004733 if (error)
4734 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004735 } else {
4736 /* This log record is split across the
4737 * physical end of log */
Chandra Seetharaman62926042011-07-22 23:40:15 +00004738 offset = dbp->b_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004739 split_bblks = 0;
4740 if (blk_no != log->l_logBBsize) {
4741 /* some data is before the physical
4742 * end of log */
4743 ASSERT(!wrapped_hblks);
4744 ASSERT(blk_no <= INT_MAX);
4745 split_bblks =
4746 log->l_logBBsize - (int)blk_no;
4747 ASSERT(split_bblks > 0);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004748 error = xlog_bread(log, blk_no,
4749 split_bblks, dbp,
4750 &offset);
4751 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004752 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004753 }
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004754
Linus Torvalds1da177e2005-04-16 15:20:36 -07004755 /*
4756 * Note: this black magic still works with
4757 * large sector sizes (non-512) only because:
4758 * - we increased the buffer size originally
4759 * by 1 sector giving us enough extra space
4760 * for the second read;
4761 * - the log start is guaranteed to be sector
4762 * aligned;
4763 * - we read the log end (LR header start)
4764 * _first_, then the log start (LR header end)
4765 * - order is important.
4766 */
Dave Chinner44396472011-04-21 09:34:27 +00004767 error = xlog_bread_offset(log, 0,
Dave Chinner009507b2012-11-02 11:38:44 +11004768 bblks - split_bblks, dbp,
Dave Chinner44396472011-04-21 09:34:27 +00004769 offset + BBTOB(split_bblks));
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004770 if (error)
4771 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004772 }
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004773
Brian Foster9d949012016-01-04 15:55:10 +11004774 error = xlog_recover_process(log, rhash, rhead, offset,
4775 pass);
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004776 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004777 goto bread_err2;
Brian Fosterd7f37692016-01-04 15:55:10 +11004778
Linus Torvalds1da177e2005-04-16 15:20:36 -07004779 blk_no += bblks;
Brian Fosterd7f37692016-01-04 15:55:10 +11004780 rhead_blk = blk_no;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004781 }
4782
4783 ASSERT(blk_no >= log->l_logBBsize);
4784 blk_no -= log->l_logBBsize;
Brian Fosterd7f37692016-01-04 15:55:10 +11004785 rhead_blk = blk_no;
Eric Sandeen970fd3f2014-09-09 11:57:29 +10004786 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004787
Eric Sandeen970fd3f2014-09-09 11:57:29 +10004788 /* read first part of physical log */
4789 while (blk_no < head_blk) {
4790 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
4791 if (error)
4792 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004793
Eric Sandeen970fd3f2014-09-09 11:57:29 +10004794 rhead = (xlog_rec_header_t *)offset;
4795 error = xlog_valid_rec_header(log, rhead, blk_no);
4796 if (error)
4797 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004798
Eric Sandeen970fd3f2014-09-09 11:57:29 +10004799 /* blocks in data section */
4800 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
4801 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
4802 &offset);
4803 if (error)
4804 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004805
Brian Foster9d949012016-01-04 15:55:10 +11004806 error = xlog_recover_process(log, rhash, rhead, offset, pass);
Eric Sandeen970fd3f2014-09-09 11:57:29 +10004807 if (error)
4808 goto bread_err2;
Brian Fosterd7f37692016-01-04 15:55:10 +11004809
Eric Sandeen970fd3f2014-09-09 11:57:29 +10004810 blk_no += bblks + hblks;
Brian Fosterd7f37692016-01-04 15:55:10 +11004811 rhead_blk = blk_no;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004812 }
4813
4814 bread_err2:
4815 xlog_put_bp(dbp);
4816 bread_err1:
4817 xlog_put_bp(hbp);
Brian Fosterd7f37692016-01-04 15:55:10 +11004818
4819 if (error && first_bad)
4820 *first_bad = rhead_blk;
4821
Linus Torvalds1da177e2005-04-16 15:20:36 -07004822 return error;
4823}
4824
4825/*
4826 * Do the recovery of the log. We actually do this in two phases.
4827 * The two passes are necessary in order to implement the function
4828 * of cancelling a record written into the log. The first pass
4829 * determines those things which have been cancelled, and the
4830 * second pass replays log items normally except for those which
4831 * have been cancelled. The handling of the replay and cancellations
4832 * takes place in the log item type specific routines.
4833 *
4834 * The table of items which have cancel records in the log is allocated
4835 * and freed at this level, since only here do we know when all of
4836 * the log recovery has been completed.
4837 */
4838STATIC int
4839xlog_do_log_recovery(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004840 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004841 xfs_daddr_t head_blk,
4842 xfs_daddr_t tail_blk)
4843{
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00004844 int error, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004845
4846 ASSERT(head_blk != tail_blk);
4847
4848 /*
4849 * First do a pass to find all of the cancelled buf log items.
4850 * Store them in the buf_cancel_table for use in the second pass.
4851 */
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00004852 log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE *
4853 sizeof(struct list_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07004854 KM_SLEEP);
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00004855 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
4856 INIT_LIST_HEAD(&log->l_buf_cancel_table[i]);
4857
Linus Torvalds1da177e2005-04-16 15:20:36 -07004858 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
Brian Fosterd7f37692016-01-04 15:55:10 +11004859 XLOG_RECOVER_PASS1, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004860 if (error != 0) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10004861 kmem_free(log->l_buf_cancel_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004862 log->l_buf_cancel_table = NULL;
4863 return error;
4864 }
4865 /*
4866 * Then do a second pass to actually recover the items in the log.
4867 * When it is complete free the table of buf cancel items.
4868 */
4869 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
Brian Fosterd7f37692016-01-04 15:55:10 +11004870 XLOG_RECOVER_PASS2, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004871#ifdef DEBUG
Tim Shimmin6d192a92006-06-09 14:55:38 +10004872 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004873 int i;
4874
4875 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00004876 ASSERT(list_empty(&log->l_buf_cancel_table[i]));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004877 }
4878#endif /* DEBUG */
4879
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10004880 kmem_free(log->l_buf_cancel_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004881 log->l_buf_cancel_table = NULL;
4882
4883 return error;
4884}
4885
4886/*
4887 * Do the actual recovery
4888 */
4889STATIC int
4890xlog_do_recover(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004891 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004892 xfs_daddr_t head_blk,
4893 xfs_daddr_t tail_blk)
4894{
4895 int error;
4896 xfs_buf_t *bp;
4897 xfs_sb_t *sbp;
4898
4899 /*
4900 * First replay the images in the log.
4901 */
4902 error = xlog_do_log_recovery(log, head_blk, tail_blk);
Christoph Hellwig43ff2122012-04-23 15:58:39 +10004903 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004904 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004905
4906 /*
4907 * If IO errors happened during recovery, bail out.
4908 */
4909 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
Dave Chinner24513372014-06-25 14:58:08 +10004910 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004911 }
4912
4913 /*
4914 * We now update the tail_lsn since much of the recovery has completed
4915 * and there may be space available to use. If there were no extent
4916 * or iunlinks, we can free up the entire log and set the tail_lsn to
4917 * be the last_sync_lsn. This was set in xlog_find_tail to be the
4918 * lsn of the last known good LR on disk. If there are extent frees
4919 * or iunlinks they will have some entries in the AIL; so we look at
4920 * the AIL to determine how to set the tail_lsn.
4921 */
4922 xlog_assign_tail_lsn(log->l_mp);
4923
4924 /*
4925 * Now that we've finished replaying all buffer and inode
Dave Chinner98021822012-11-12 22:54:03 +11004926 * updates, re-read in the superblock and reverify it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004927 */
4928 bp = xfs_getsb(log->l_mp, 0);
4929 XFS_BUF_UNDONE(bp);
Lachlan McIlroybebf9632007-10-15 13:18:02 +10004930 ASSERT(!(XFS_BUF_ISWRITE(bp)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004931 XFS_BUF_READ(bp);
Lachlan McIlroybebf9632007-10-15 13:18:02 +10004932 XFS_BUF_UNASYNC(bp);
Dave Chinner1813dd62012-11-14 17:54:40 +11004933 bp->b_ops = &xfs_sb_buf_ops;
Christoph Hellwig83a0adc2013-12-17 00:03:52 -08004934
Dave Chinner595bff72014-10-02 09:05:14 +10004935 error = xfs_buf_submit_wait(bp);
David Chinnerd64e31a2008-04-10 12:22:17 +10004936 if (error) {
Dave Chinner595bff72014-10-02 09:05:14 +10004937 if (!XFS_FORCED_SHUTDOWN(log->l_mp)) {
4938 xfs_buf_ioerror_alert(bp, __func__);
4939 ASSERT(0);
4940 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004941 xfs_buf_relse(bp);
4942 return error;
4943 }
4944
4945 /* Convert superblock from on-disk format */
4946 sbp = &log->l_mp->m_sb;
Dave Chinner98021822012-11-12 22:54:03 +11004947 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004948 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
Eric Sandeen62118702008-03-06 13:44:28 +11004949 ASSERT(xfs_sb_good_version(sbp));
Dave Chinner5681ca42015-02-23 21:22:31 +11004950 xfs_reinit_percpu_counters(log->l_mp);
4951
Linus Torvalds1da177e2005-04-16 15:20:36 -07004952 xfs_buf_relse(bp);
4953
Lachlan McIlroy5478eea2007-02-10 18:36:29 +11004954
Linus Torvalds1da177e2005-04-16 15:20:36 -07004955 xlog_recover_check_summary(log);
4956
4957 /* Normal transactions can now occur */
4958 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
4959 return 0;
4960}
4961
4962/*
4963 * Perform recovery and re-initialize some log variables in xlog_find_tail.
4964 *
4965 * Return error or zero.
4966 */
4967int
4968xlog_recover(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004969 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004970{
4971 xfs_daddr_t head_blk, tail_blk;
4972 int error;
4973
4974 /* find the tail of the log */
Brian Fostera45086e2015-10-12 15:59:25 +11004975 error = xlog_find_tail(log, &head_blk, &tail_blk);
4976 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004977 return error;
4978
Brian Fostera45086e2015-10-12 15:59:25 +11004979 /*
4980 * The superblock was read before the log was available and thus the LSN
4981 * could not be verified. Check the superblock LSN against the current
4982 * LSN now that it's known.
4983 */
4984 if (xfs_sb_version_hascrc(&log->l_mp->m_sb) &&
4985 !xfs_log_check_lsn(log->l_mp, log->l_mp->m_sb.sb_lsn))
4986 return -EINVAL;
4987
Linus Torvalds1da177e2005-04-16 15:20:36 -07004988 if (tail_blk != head_blk) {
4989 /* There used to be a comment here:
4990 *
4991 * disallow recovery on read-only mounts. note -- mount
4992 * checks for ENOSPC and turns it into an intelligent
4993 * error message.
4994 * ...but this is no longer true. Now, unless you specify
4995 * NORECOVERY (in which case this function would never be
4996 * called), we just go ahead and recover. We do this all
4997 * under the vfs layer, so we can get away with it unless
4998 * the device itself is read-only, in which case we fail.
4999 */
Utako Kusaka3a02ee12007-05-08 13:50:06 +10005000 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005001 return error;
5002 }
5003
Dave Chinnere721f502013-04-03 16:11:32 +11005004 /*
5005 * Version 5 superblock log feature mask validation. We know the
5006 * log is dirty so check if there are any unknown log features
5007 * in what we need to recover. If there are unknown features
5008 * (e.g. unsupported transactions, then simply reject the
5009 * attempt at recovery before touching anything.
5010 */
5011 if (XFS_SB_VERSION_NUM(&log->l_mp->m_sb) == XFS_SB_VERSION_5 &&
5012 xfs_sb_has_incompat_log_feature(&log->l_mp->m_sb,
5013 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
5014 xfs_warn(log->l_mp,
Joe Perchesf41febd2015-07-29 11:52:04 +10005015"Superblock has unknown incompatible log features (0x%x) enabled.",
Dave Chinnere721f502013-04-03 16:11:32 +11005016 (log->l_mp->m_sb.sb_features_log_incompat &
5017 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
Joe Perchesf41febd2015-07-29 11:52:04 +10005018 xfs_warn(log->l_mp,
5019"The log can not be fully and/or safely recovered by this kernel.");
5020 xfs_warn(log->l_mp,
5021"Please recover the log on a kernel that supports the unknown features.");
Dave Chinner24513372014-06-25 14:58:08 +10005022 return -EINVAL;
Dave Chinnere721f502013-04-03 16:11:32 +11005023 }
5024
Brian Foster2e227172014-09-09 11:56:13 +10005025 /*
5026 * Delay log recovery if the debug hook is set. This is debug
5027 * instrumention to coordinate simulation of I/O failures with
5028 * log recovery.
5029 */
5030 if (xfs_globals.log_recovery_delay) {
5031 xfs_notice(log->l_mp,
5032 "Delaying log recovery for %d seconds.",
5033 xfs_globals.log_recovery_delay);
5034 msleep(xfs_globals.log_recovery_delay * 1000);
5035 }
5036
Dave Chinnera0fa2b62011-03-07 10:01:35 +11005037 xfs_notice(log->l_mp, "Starting recovery (logdev: %s)",
5038 log->l_mp->m_logname ? log->l_mp->m_logname
5039 : "internal");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005040
5041 error = xlog_do_recover(log, head_blk, tail_blk);
5042 log->l_flags |= XLOG_RECOVERY_NEEDED;
5043 }
5044 return error;
5045}
5046
5047/*
5048 * In the first part of recovery we replay inodes and buffers and build
5049 * up the list of extent free items which need to be processed. Here
5050 * we process the extent free items and clean up the on disk unlinked
5051 * inode lists. This is separated from the first part of recovery so
5052 * that the root and real-time bitmap inodes can be read in from disk in
5053 * between the two stages. This is necessary so that we can free space
5054 * in the real-time portion of the file system.
5055 */
5056int
5057xlog_recover_finish(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05005058 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005059{
5060 /*
5061 * Now we're ready to do the transactions needed for the
5062 * rest of recovery. Start with completing all the extent
5063 * free intent records and then process the unlinked inode
5064 * lists. At this point, we essentially run in normal mode
5065 * except that we're still performing recovery actions
5066 * rather than accepting new requests.
5067 */
5068 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
David Chinner3c1e2bb2008-04-10 12:21:11 +10005069 int error;
5070 error = xlog_recover_process_efis(log);
5071 if (error) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11005072 xfs_alert(log->l_mp, "Failed to recover EFIs");
David Chinner3c1e2bb2008-04-10 12:21:11 +10005073 return error;
5074 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005075 /*
5076 * Sync the log to get all the EFIs out of the AIL.
5077 * This isn't absolutely necessary, but it helps in
5078 * case the unlink transactions would have problems
5079 * pushing the EFIs out of the way.
5080 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +00005081 xfs_log_force(log->l_mp, XFS_LOG_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005082
Christoph Hellwig42490232008-08-13 16:49:32 +10005083 xlog_recover_process_iunlinks(log);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005084
5085 xlog_recover_check_summary(log);
5086
Dave Chinnera0fa2b62011-03-07 10:01:35 +11005087 xfs_notice(log->l_mp, "Ending recovery (logdev: %s)",
5088 log->l_mp->m_logname ? log->l_mp->m_logname
5089 : "internal");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005090 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
5091 } else {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11005092 xfs_info(log->l_mp, "Ending clean mount");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005093 }
5094 return 0;
5095}
5096
Brian Fosterf0b2efa2015-08-19 09:58:36 +10005097int
5098xlog_recover_cancel(
5099 struct xlog *log)
5100{
5101 int error = 0;
5102
5103 if (log->l_flags & XLOG_RECOVERY_NEEDED)
5104 error = xlog_recover_cancel_efis(log);
5105
5106 return error;
5107}
Linus Torvalds1da177e2005-04-16 15:20:36 -07005108
5109#if defined(DEBUG)
5110/*
5111 * Read all of the agf and agi counters and check that they
5112 * are consistent with the superblock counters.
5113 */
5114void
5115xlog_recover_check_summary(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05005116 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005117{
5118 xfs_mount_t *mp;
5119 xfs_agf_t *agfp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005120 xfs_buf_t *agfbp;
5121 xfs_buf_t *agibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005122 xfs_agnumber_t agno;
5123 __uint64_t freeblks;
5124 __uint64_t itotal;
5125 __uint64_t ifree;
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11005126 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005127
5128 mp = log->l_mp;
5129
5130 freeblks = 0LL;
5131 itotal = 0LL;
5132 ifree = 0LL;
5133 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
From: Christoph Hellwig48056212008-11-28 14:23:38 +11005134 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
5135 if (error) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11005136 xfs_alert(mp, "%s agf read failed agno %d error %d",
5137 __func__, agno, error);
From: Christoph Hellwig48056212008-11-28 14:23:38 +11005138 } else {
5139 agfp = XFS_BUF_TO_AGF(agfbp);
5140 freeblks += be32_to_cpu(agfp->agf_freeblks) +
5141 be32_to_cpu(agfp->agf_flcount);
5142 xfs_buf_relse(agfbp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005143 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005144
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11005145 error = xfs_read_agi(mp, NULL, agno, &agibp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11005146 if (error) {
5147 xfs_alert(mp, "%s agi read failed agno %d error %d",
5148 __func__, agno, error);
5149 } else {
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11005150 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005151
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11005152 itotal += be32_to_cpu(agi->agi_count);
5153 ifree += be32_to_cpu(agi->agi_freecount);
5154 xfs_buf_relse(agibp);
5155 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005156 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005157}
5158#endif /* DEBUG */