blob: 90b756f7117c94443fab0fe3f3939bbaa07dff3a [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 Chinner6ca1c902013-08-12 20:49:26 +100020#include "xfs_format.h"
Nathan Scotta844f452005-11-02 14:38:42 +110021#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110023#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_trans.h"
Nathan Scotta844f452005-11-02 14:38:42 +110025#include "xfs_sb.h"
26#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_mount.h"
28#include "xfs_error.h"
29#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110030#include "xfs_alloc_btree.h"
31#include "xfs_ialloc_btree.h"
Christoph Hellwigee1a47a2013-04-21 14:53:46 -050032#include "xfs_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_dinode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include "xfs_inode.h"
Nathan Scotta844f452005-11-02 14:38:42 +110035#include "xfs_inode_item.h"
Nathan Scotta844f452005-11-02 14:38:42 +110036#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include "xfs_ialloc.h"
38#include "xfs_log_priv.h"
39#include "xfs_buf_item.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include "xfs_log_recover.h"
41#include "xfs_extfree_item.h"
42#include "xfs_trans_priv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include "xfs_quota.h"
Christoph Hellwig0e446be2012-11-12 22:54:24 +110044#include "xfs_cksum.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000045#include "xfs_trace.h"
Dave Chinner33479e02012-10-08 21:56:11 +110046#include "xfs_icache.h"
Dave Chinner28c8e412013-06-27 16:04:55 +100047#include "xfs_icreate_item.h"
Dave Chinnerd75afeb2013-04-03 16:11:29 +110048
49/* Need all the magic numbers and buffer ops structures from these headers */
Dave Chinnerf948dd72013-04-03 16:11:19 +110050#include "xfs_symlink.h"
Dave Chinnerd75afeb2013-04-03 16:11:29 +110051#include "xfs_da_btree.h"
52#include "xfs_dir2_format.h"
Dave Chinner2b9ab5a2013-08-12 20:49:37 +100053#include "xfs_dir2.h"
Dave Chinnerd75afeb2013-04-03 16:11:29 +110054#include "xfs_attr_leaf.h"
55#include "xfs_attr_remote.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070056
Dave Chinnerfc06c6d2013-08-12 20:49:22 +100057#define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
58
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -050059STATIC int
60xlog_find_zeroed(
61 struct xlog *,
62 xfs_daddr_t *);
63STATIC int
64xlog_clear_stale_blocks(
65 struct xlog *,
66 xfs_lsn_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -070067#if defined(DEBUG)
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -050068STATIC void
69xlog_recover_check_summary(
70 struct xlog *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070071#else
72#define xlog_recover_check_summary(log)
Linus Torvalds1da177e2005-04-16 15:20:36 -070073#endif
74
Linus Torvalds1da177e2005-04-16 15:20:36 -070075/*
Christoph Hellwigd5689ea2010-12-01 22:06:22 +000076 * This structure is used during recovery to record the buf log items which
77 * have been canceled and should not be replayed.
78 */
79struct xfs_buf_cancel {
80 xfs_daddr_t bc_blkno;
81 uint bc_len;
82 int bc_refcount;
83 struct list_head bc_list;
84};
85
86/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 * Sector aligned buffer routines for buffer create/read/write/access
88 */
89
Alex Elderff30a622010-04-13 15:22:58 +100090/*
91 * Verify the given count of basic blocks is valid number of blocks
92 * to specify for an operation involving the given XFS log buffer.
93 * Returns nonzero if the count is valid, 0 otherwise.
94 */
95
96static inline int
97xlog_buf_bbcount_valid(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -050098 struct xlog *log,
Alex Elderff30a622010-04-13 15:22:58 +100099 int bbcount)
100{
101 return bbcount > 0 && bbcount <= log->l_logBBsize;
102}
103
Alex Elder36adecf2010-04-13 15:21:13 +1000104/*
105 * Allocate a buffer to hold log data. The buffer needs to be able
106 * to map to a range of nbblks basic blocks at any valid (basic
107 * block) offset within the log.
108 */
Eric Sandeen5d77c0d2009-11-19 15:52:00 +0000109STATIC xfs_buf_t *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110xlog_get_bp(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500111 struct xlog *log,
Dave Chinner32281492009-01-22 15:37:47 +1100112 int nbblks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113{
Christoph Hellwigc8da0fa2011-07-08 14:36:25 +0200114 struct xfs_buf *bp;
115
Alex Elderff30a622010-04-13 15:22:58 +1000116 if (!xlog_buf_bbcount_valid(log, nbblks)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100117 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
Alex Elderff30a622010-04-13 15:22:58 +1000118 nbblks);
119 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
Dave Chinner32281492009-01-22 15:37:47 +1100120 return NULL;
121 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122
Alex Elder36adecf2010-04-13 15:21:13 +1000123 /*
124 * We do log I/O in units of log sectors (a power-of-2
125 * multiple of the basic block size), so we round up the
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300126 * requested size to accommodate the basic blocks required
Alex Elder36adecf2010-04-13 15:21:13 +1000127 * for complete log sectors.
128 *
129 * In addition, the buffer may be used for a non-sector-
130 * aligned block offset, in which case an I/O of the
131 * requested size could extend beyond the end of the
132 * buffer. If the requested size is only 1 basic block it
133 * will never straddle a sector boundary, so this won't be
134 * an issue. Nor will this be a problem if the log I/O is
135 * done in basic blocks (sector size 1). But otherwise we
136 * extend the buffer by one extra log sector to ensure
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300137 * there's space to accommodate this possibility.
Alex Elder36adecf2010-04-13 15:21:13 +1000138 */
Alex Elder69ce58f2010-04-20 17:09:59 +1000139 if (nbblks > 1 && log->l_sectBBsize > 1)
140 nbblks += log->l_sectBBsize;
141 nbblks = round_up(nbblks, log->l_sectBBsize);
Alex Elder36adecf2010-04-13 15:21:13 +1000142
Dave Chinnere70b73f2012-04-23 15:58:49 +1000143 bp = xfs_buf_get_uncached(log->l_mp->m_logdev_targp, nbblks, 0);
Christoph Hellwigc8da0fa2011-07-08 14:36:25 +0200144 if (bp)
145 xfs_buf_unlock(bp);
146 return bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147}
148
Eric Sandeen5d77c0d2009-11-19 15:52:00 +0000149STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150xlog_put_bp(
151 xfs_buf_t *bp)
152{
153 xfs_buf_free(bp);
154}
155
Alex Elder48389ef2010-04-20 17:10:21 +1000156/*
157 * Return the address of the start of the given block number's data
158 * in a log buffer. The buffer covers a log sector-aligned region.
159 */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100160STATIC xfs_caddr_t
161xlog_align(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500162 struct xlog *log,
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100163 xfs_daddr_t blk_no,
164 int nbblks,
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500165 struct xfs_buf *bp)
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100166{
Christoph Hellwigfdc07f42010-05-10 17:28:14 +0000167 xfs_daddr_t offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100168
Dave Chinner4e94b712012-04-23 15:58:51 +1000169 ASSERT(offset + nbblks <= bp->b_length);
Chandra Seetharaman62926042011-07-22 23:40:15 +0000170 return bp->b_addr + BBTOB(offset);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100171}
172
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173
174/*
175 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
176 */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100177STATIC int
178xlog_bread_noalign(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500179 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 xfs_daddr_t blk_no,
181 int nbblks,
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500182 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183{
184 int error;
185
Alex Elderff30a622010-04-13 15:22:58 +1000186 if (!xlog_buf_bbcount_valid(log, nbblks)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100187 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
Alex Elderff30a622010-04-13 15:22:58 +1000188 nbblks);
189 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
Dave Chinner32281492009-01-22 15:37:47 +1100190 return EFSCORRUPTED;
191 }
192
Alex Elder69ce58f2010-04-20 17:09:59 +1000193 blk_no = round_down(blk_no, log->l_sectBBsize);
194 nbblks = round_up(nbblks, log->l_sectBBsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195
196 ASSERT(nbblks > 0);
Dave Chinner4e94b712012-04-23 15:58:51 +1000197 ASSERT(nbblks <= bp->b_length);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198
199 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
200 XFS_BUF_READ(bp);
Dave Chinneraa0e8832012-04-23 15:58:52 +1000201 bp->b_io_length = nbblks;
Dave Chinner0e95f192012-04-23 15:58:46 +1000202 bp->b_error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203
204 xfsbdstrat(log->l_mp, bp);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +0000205 error = xfs_buf_iowait(bp);
David Chinnerd64e31a2008-04-10 12:22:17 +1000206 if (error)
Christoph Hellwig901796a2011-10-10 16:52:49 +0000207 xfs_buf_ioerror_alert(bp, __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 return error;
209}
210
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100211STATIC int
212xlog_bread(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500213 struct xlog *log,
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100214 xfs_daddr_t blk_no,
215 int nbblks,
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500216 struct xfs_buf *bp,
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100217 xfs_caddr_t *offset)
218{
219 int error;
220
221 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
222 if (error)
223 return error;
224
225 *offset = xlog_align(log, blk_no, nbblks, bp);
226 return 0;
227}
228
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229/*
Dave Chinner44396472011-04-21 09:34:27 +0000230 * Read at an offset into the buffer. Returns with the buffer in it's original
231 * state regardless of the result of the read.
232 */
233STATIC int
234xlog_bread_offset(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500235 struct xlog *log,
Dave Chinner44396472011-04-21 09:34:27 +0000236 xfs_daddr_t blk_no, /* block to read from */
237 int nbblks, /* blocks to read */
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500238 struct xfs_buf *bp,
Dave Chinner44396472011-04-21 09:34:27 +0000239 xfs_caddr_t offset)
240{
Chandra Seetharaman62926042011-07-22 23:40:15 +0000241 xfs_caddr_t orig_offset = bp->b_addr;
Dave Chinner4e94b712012-04-23 15:58:51 +1000242 int orig_len = BBTOB(bp->b_length);
Dave Chinner44396472011-04-21 09:34:27 +0000243 int error, error2;
244
Chandra Seetharaman02fe03d2011-07-22 23:40:22 +0000245 error = xfs_buf_associate_memory(bp, offset, BBTOB(nbblks));
Dave Chinner44396472011-04-21 09:34:27 +0000246 if (error)
247 return error;
248
249 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
250
251 /* must reset buffer pointer even on error */
Chandra Seetharaman02fe03d2011-07-22 23:40:22 +0000252 error2 = xfs_buf_associate_memory(bp, orig_offset, orig_len);
Dave Chinner44396472011-04-21 09:34:27 +0000253 if (error)
254 return error;
255 return error2;
256}
257
258/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 * Write out the buffer at the given block for the given number of blocks.
260 * The buffer is kept locked across the write and is returned locked.
261 * This can only be used for synchronous log writes.
262 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000263STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264xlog_bwrite(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500265 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 xfs_daddr_t blk_no,
267 int nbblks,
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500268 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269{
270 int error;
271
Alex Elderff30a622010-04-13 15:22:58 +1000272 if (!xlog_buf_bbcount_valid(log, nbblks)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100273 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
Alex Elderff30a622010-04-13 15:22:58 +1000274 nbblks);
275 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
Dave Chinner32281492009-01-22 15:37:47 +1100276 return EFSCORRUPTED;
277 }
278
Alex Elder69ce58f2010-04-20 17:09:59 +1000279 blk_no = round_down(blk_no, log->l_sectBBsize);
280 nbblks = round_up(nbblks, log->l_sectBBsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281
282 ASSERT(nbblks > 0);
Dave Chinner4e94b712012-04-23 15:58:51 +1000283 ASSERT(nbblks <= bp->b_length);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284
285 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
286 XFS_BUF_ZEROFLAGS(bp);
Chandra Seetharaman72790aa2011-07-22 23:40:04 +0000287 xfs_buf_hold(bp);
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200288 xfs_buf_lock(bp);
Dave Chinneraa0e8832012-04-23 15:58:52 +1000289 bp->b_io_length = nbblks;
Dave Chinner0e95f192012-04-23 15:58:46 +1000290 bp->b_error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291
Christoph Hellwigc2b006c2011-08-23 08:28:07 +0000292 error = xfs_bwrite(bp);
Christoph Hellwig901796a2011-10-10 16:52:49 +0000293 if (error)
294 xfs_buf_ioerror_alert(bp, __func__);
Christoph Hellwigc2b006c2011-08-23 08:28:07 +0000295 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 return error;
297}
298
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299#ifdef DEBUG
300/*
301 * dump debug superblock and log record information
302 */
303STATIC void
304xlog_header_check_dump(
305 xfs_mount_t *mp,
306 xlog_rec_header_t *head)
307{
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100308 xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d\n",
Joe Perches03daa572009-12-14 18:01:10 -0800309 __func__, &mp->m_sb.sb_uuid, XLOG_FMT);
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100310 xfs_debug(mp, " log : uuid = %pU, fmt = %d\n",
Joe Perches03daa572009-12-14 18:01:10 -0800311 &head->h_fs_uuid, be32_to_cpu(head->h_fmt));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312}
313#else
314#define xlog_header_check_dump(mp, head)
315#endif
316
317/*
318 * check log record header for recovery
319 */
320STATIC int
321xlog_header_check_recover(
322 xfs_mount_t *mp,
323 xlog_rec_header_t *head)
324{
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200325 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326
327 /*
328 * IRIX doesn't write the h_fmt field and leaves it zeroed
329 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
330 * a dirty log created in IRIX.
331 */
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200332 if (unlikely(head->h_fmt != cpu_to_be32(XLOG_FMT))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100333 xfs_warn(mp,
334 "dirty log written in incompatible format - can't recover");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 xlog_header_check_dump(mp, head);
336 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
337 XFS_ERRLEVEL_HIGH, mp);
338 return XFS_ERROR(EFSCORRUPTED);
339 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100340 xfs_warn(mp,
341 "dirty log entry has mismatched uuid - can't recover");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 xlog_header_check_dump(mp, head);
343 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
344 XFS_ERRLEVEL_HIGH, mp);
345 return XFS_ERROR(EFSCORRUPTED);
346 }
347 return 0;
348}
349
350/*
351 * read the head block of the log and check the header
352 */
353STATIC int
354xlog_header_check_mount(
355 xfs_mount_t *mp,
356 xlog_rec_header_t *head)
357{
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200358 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359
360 if (uuid_is_nil(&head->h_fs_uuid)) {
361 /*
362 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
363 * h_fs_uuid is nil, we assume this log was last mounted
364 * by IRIX and continue.
365 */
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100366 xfs_warn(mp, "nil uuid in log - IRIX style log");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100368 xfs_warn(mp, "log has mismatched uuid - can't recover");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 xlog_header_check_dump(mp, head);
370 XFS_ERROR_REPORT("xlog_header_check_mount",
371 XFS_ERRLEVEL_HIGH, mp);
372 return XFS_ERROR(EFSCORRUPTED);
373 }
374 return 0;
375}
376
377STATIC void
378xlog_recover_iodone(
379 struct xfs_buf *bp)
380{
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +0000381 if (bp->b_error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 /*
383 * We're not going to bother about retrying
384 * this during recovery. One strike!
385 */
Christoph Hellwig901796a2011-10-10 16:52:49 +0000386 xfs_buf_ioerror_alert(bp, __func__);
Dave Chinnerebad8612010-09-22 10:47:20 +1000387 xfs_force_shutdown(bp->b_target->bt_mount,
388 SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 }
Christoph Hellwigcb669ca2011-07-13 13:43:49 +0200390 bp->b_iodone = NULL;
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +0000391 xfs_buf_ioend(bp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392}
393
394/*
395 * This routine finds (to an approximation) the first block in the physical
396 * log which contains the given cycle. It uses a binary search algorithm.
397 * Note that the algorithm can not be perfect because the disk will not
398 * necessarily be perfect.
399 */
David Chinnera8272ce2007-11-23 16:28:09 +1100400STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401xlog_find_cycle_start(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500402 struct xlog *log,
403 struct xfs_buf *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 xfs_daddr_t first_blk,
405 xfs_daddr_t *last_blk,
406 uint cycle)
407{
408 xfs_caddr_t offset;
409 xfs_daddr_t mid_blk;
Alex Eldere3bb2e32010-04-15 18:17:30 +0000410 xfs_daddr_t end_blk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 uint mid_cycle;
412 int error;
413
Alex Eldere3bb2e32010-04-15 18:17:30 +0000414 end_blk = *last_blk;
415 mid_blk = BLK_AVG(first_blk, end_blk);
416 while (mid_blk != first_blk && mid_blk != end_blk) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100417 error = xlog_bread(log, mid_blk, 1, bp, &offset);
418 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 return error;
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000420 mid_cycle = xlog_get_cycle(offset);
Alex Eldere3bb2e32010-04-15 18:17:30 +0000421 if (mid_cycle == cycle)
422 end_blk = mid_blk; /* last_half_cycle == mid_cycle */
423 else
424 first_blk = mid_blk; /* first_half_cycle == mid_cycle */
425 mid_blk = BLK_AVG(first_blk, end_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 }
Alex Eldere3bb2e32010-04-15 18:17:30 +0000427 ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) ||
428 (mid_blk == end_blk && mid_blk-1 == first_blk));
429
430 *last_blk = end_blk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431
432 return 0;
433}
434
435/*
Alex Elder3f943d82010-04-15 18:17:34 +0000436 * Check that a range of blocks does not contain stop_on_cycle_no.
437 * Fill in *new_blk with the block offset where such a block is
438 * found, or with -1 (an invalid block number) if there is no such
439 * block in the range. The scan needs to occur from front to back
440 * and the pointer into the region must be updated since a later
441 * routine will need to perform another test.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 */
443STATIC int
444xlog_find_verify_cycle(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500445 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 xfs_daddr_t start_blk,
447 int nbblks,
448 uint stop_on_cycle_no,
449 xfs_daddr_t *new_blk)
450{
451 xfs_daddr_t i, j;
452 uint cycle;
453 xfs_buf_t *bp;
454 xfs_daddr_t bufblks;
455 xfs_caddr_t buf = NULL;
456 int error = 0;
457
Alex Elder6881a222010-04-13 15:22:29 +1000458 /*
459 * Greedily allocate a buffer big enough to handle the full
460 * range of basic blocks we'll be examining. If that fails,
461 * try a smaller size. We need to be able to read at least
462 * a log sector, or we're out of luck.
463 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 bufblks = 1 << ffs(nbblks);
Dave Chinner81158e02012-04-27 19:45:22 +1000465 while (bufblks > log->l_logBBsize)
466 bufblks >>= 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 while (!(bp = xlog_get_bp(log, bufblks))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 bufblks >>= 1;
Alex Elder69ce58f2010-04-20 17:09:59 +1000469 if (bufblks < log->l_sectBBsize)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 return ENOMEM;
471 }
472
473 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
474 int bcount;
475
476 bcount = min(bufblks, (start_blk + nbblks - i));
477
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100478 error = xlog_bread(log, i, bcount, bp, &buf);
479 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 goto out;
481
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 for (j = 0; j < bcount; j++) {
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000483 cycle = xlog_get_cycle(buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 if (cycle == stop_on_cycle_no) {
485 *new_blk = i+j;
486 goto out;
487 }
488
489 buf += BBSIZE;
490 }
491 }
492
493 *new_blk = -1;
494
495out:
496 xlog_put_bp(bp);
497 return error;
498}
499
500/*
501 * Potentially backup over partial log record write.
502 *
503 * In the typical case, last_blk is the number of the block directly after
504 * a good log record. Therefore, we subtract one to get the block number
505 * of the last block in the given buffer. extra_bblks contains the number
506 * of blocks we would have read on a previous read. This happens when the
507 * last log record is split over the end of the physical log.
508 *
509 * extra_bblks is the number of blocks potentially verified on a previous
510 * call to this routine.
511 */
512STATIC int
513xlog_find_verify_log_record(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500514 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 xfs_daddr_t start_blk,
516 xfs_daddr_t *last_blk,
517 int extra_bblks)
518{
519 xfs_daddr_t i;
520 xfs_buf_t *bp;
521 xfs_caddr_t offset = NULL;
522 xlog_rec_header_t *head = NULL;
523 int error = 0;
524 int smallmem = 0;
525 int num_blks = *last_blk - start_blk;
526 int xhdrs;
527
528 ASSERT(start_blk != 0 || *last_blk != start_blk);
529
530 if (!(bp = xlog_get_bp(log, num_blks))) {
531 if (!(bp = xlog_get_bp(log, 1)))
532 return ENOMEM;
533 smallmem = 1;
534 } else {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100535 error = xlog_bread(log, start_blk, num_blks, bp, &offset);
536 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 offset += ((num_blks - 1) << BBSHIFT);
539 }
540
541 for (i = (*last_blk) - 1; i >= 0; i--) {
542 if (i < start_blk) {
543 /* valid log record not found */
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100544 xfs_warn(log->l_mp,
545 "Log inconsistent (didn't find previous header)");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 ASSERT(0);
547 error = XFS_ERROR(EIO);
548 goto out;
549 }
550
551 if (smallmem) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100552 error = xlog_bread(log, i, 1, bp, &offset);
553 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 }
556
557 head = (xlog_rec_header_t *)offset;
558
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200559 if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 break;
561
562 if (!smallmem)
563 offset -= BBSIZE;
564 }
565
566 /*
567 * We hit the beginning of the physical log & still no header. Return
568 * to caller. If caller can handle a return of -1, then this routine
569 * will be called again for the end of the physical log.
570 */
571 if (i == -1) {
572 error = -1;
573 goto out;
574 }
575
576 /*
577 * We have the final block of the good log (the first block
578 * of the log record _before_ the head. So we check the uuid.
579 */
580 if ((error = xlog_header_check_mount(log->l_mp, head)))
581 goto out;
582
583 /*
584 * We may have found a log record header before we expected one.
585 * last_blk will be the 1st block # with a given cycle #. We may end
586 * up reading an entire log record. In this case, we don't want to
587 * reset last_blk. Only when last_blk points in the middle of a log
588 * record do we update last_blk.
589 */
Eric Sandeen62118702008-03-06 13:44:28 +1100590 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000591 uint h_size = be32_to_cpu(head->h_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592
593 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
594 if (h_size % XLOG_HEADER_CYCLE_SIZE)
595 xhdrs++;
596 } else {
597 xhdrs = 1;
598 }
599
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000600 if (*last_blk - i + extra_bblks !=
601 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 *last_blk = i;
603
604out:
605 xlog_put_bp(bp);
606 return error;
607}
608
609/*
610 * Head is defined to be the point of the log where the next log write
Zhi Yong Wu0a94da22013-08-07 10:11:08 +0000611 * could go. This means that incomplete LR writes at the end are
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 * eliminated when calculating the head. We aren't guaranteed that previous
613 * LR have complete transactions. We only know that a cycle number of
614 * current cycle number -1 won't be present in the log if we start writing
615 * from our current block number.
616 *
617 * last_blk contains the block number of the first block with a given
618 * cycle number.
619 *
620 * Return: zero if normal, non-zero if error.
621 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000622STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623xlog_find_head(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500624 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 xfs_daddr_t *return_head_blk)
626{
627 xfs_buf_t *bp;
628 xfs_caddr_t offset;
629 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
630 int num_scan_bblks;
631 uint first_half_cycle, last_half_cycle;
632 uint stop_on_cycle;
633 int error, log_bbnum = log->l_logBBsize;
634
635 /* Is the end of the log device zeroed? */
636 if ((error = xlog_find_zeroed(log, &first_blk)) == -1) {
637 *return_head_blk = first_blk;
638
639 /* Is the whole lot zeroed? */
640 if (!first_blk) {
641 /* Linux XFS shouldn't generate totally zeroed logs -
642 * mkfs etc write a dummy unmount record to a fresh
643 * log so we can store the uuid in there
644 */
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100645 xfs_warn(log->l_mp, "totally zeroed log");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 }
647
648 return 0;
649 } else if (error) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100650 xfs_warn(log->l_mp, "empty log check failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 return error;
652 }
653
654 first_blk = 0; /* get cycle # of 1st block */
655 bp = xlog_get_bp(log, 1);
656 if (!bp)
657 return ENOMEM;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100658
659 error = xlog_bread(log, 0, 1, bp, &offset);
660 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 goto bp_err;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100662
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000663 first_half_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664
665 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100666 error = xlog_bread(log, last_blk, 1, bp, &offset);
667 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 goto bp_err;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100669
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000670 last_half_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 ASSERT(last_half_cycle != 0);
672
673 /*
674 * If the 1st half cycle number is equal to the last half cycle number,
675 * then the entire log is stamped with the same cycle number. In this
676 * case, head_blk can't be set to zero (which makes sense). The below
677 * math doesn't work out properly with head_blk equal to zero. Instead,
678 * we set it to log_bbnum which is an invalid block number, but this
679 * value makes the math correct. If head_blk doesn't changed through
680 * all the tests below, *head_blk is set to zero at the very end rather
681 * than log_bbnum. In a sense, log_bbnum and zero are the same block
682 * in a circular file.
683 */
684 if (first_half_cycle == last_half_cycle) {
685 /*
686 * In this case we believe that the entire log should have
687 * cycle number last_half_cycle. We need to scan backwards
688 * from the end verifying that there are no holes still
689 * containing last_half_cycle - 1. If we find such a hole,
690 * then the start of that hole will be the new head. The
691 * simple case looks like
692 * x | x ... | x - 1 | x
693 * Another case that fits this picture would be
694 * x | x + 1 | x ... | x
Nathan Scottc41564b2006-03-29 08:55:14 +1000695 * In this case the head really is somewhere at the end of the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 * log, as one of the latest writes at the beginning was
697 * incomplete.
698 * One more case is
699 * x | x + 1 | x ... | x - 1 | x
700 * This is really the combination of the above two cases, and
701 * the head has to end up at the start of the x-1 hole at the
702 * end of the log.
703 *
704 * In the 256k log case, we will read from the beginning to the
705 * end of the log and search for cycle numbers equal to x-1.
706 * We don't worry about the x+1 blocks that we encounter,
707 * because we know that they cannot be the head since the log
708 * started with x.
709 */
710 head_blk = log_bbnum;
711 stop_on_cycle = last_half_cycle - 1;
712 } else {
713 /*
714 * In this case we want to find the first block with cycle
715 * number matching last_half_cycle. We expect the log to be
716 * some variation on
Alex Elder3f943d82010-04-15 18:17:34 +0000717 * x + 1 ... | x ... | x
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 * The first block with cycle number x (last_half_cycle) will
719 * be where the new head belongs. First we do a binary search
720 * for the first occurrence of last_half_cycle. The binary
721 * search may not be totally accurate, so then we scan back
722 * from there looking for occurrences of last_half_cycle before
723 * us. If that backwards scan wraps around the beginning of
724 * the log, then we look for occurrences of last_half_cycle - 1
725 * at the end of the log. The cases we're looking for look
726 * like
Alex Elder3f943d82010-04-15 18:17:34 +0000727 * v binary search stopped here
728 * x + 1 ... | x | x + 1 | x ... | x
729 * ^ but we want to locate this spot
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 * or
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 * <---------> less than scan distance
Alex Elder3f943d82010-04-15 18:17:34 +0000732 * x + 1 ... | x ... | x - 1 | x
733 * ^ we want to locate this spot
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 */
735 stop_on_cycle = last_half_cycle;
736 if ((error = xlog_find_cycle_start(log, bp, first_blk,
737 &head_blk, last_half_cycle)))
738 goto bp_err;
739 }
740
741 /*
742 * Now validate the answer. Scan back some number of maximum possible
743 * blocks and make sure each one has the expected cycle number. The
744 * maximum is determined by the total possible amount of buffering
745 * in the in-core log. The following number can be made tighter if
746 * we actually look at the block size of the filesystem.
747 */
748 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
749 if (head_blk >= num_scan_bblks) {
750 /*
751 * We are guaranteed that the entire check can be performed
752 * in one buffer.
753 */
754 start_blk = head_blk - num_scan_bblks;
755 if ((error = xlog_find_verify_cycle(log,
756 start_blk, num_scan_bblks,
757 stop_on_cycle, &new_blk)))
758 goto bp_err;
759 if (new_blk != -1)
760 head_blk = new_blk;
761 } else { /* need to read 2 parts of log */
762 /*
763 * We are going to scan backwards in the log in two parts.
764 * First we scan the physical end of the log. In this part
765 * of the log, we are looking for blocks with cycle number
766 * last_half_cycle - 1.
767 * If we find one, then we know that the log starts there, as
768 * we've found a hole that didn't get written in going around
769 * the end of the physical log. The simple case for this is
770 * x + 1 ... | x ... | x - 1 | x
771 * <---------> less than scan distance
772 * If all of the blocks at the end of the log have cycle number
773 * last_half_cycle, then we check the blocks at the start of
774 * the log looking for occurrences of last_half_cycle. If we
775 * find one, then our current estimate for the location of the
776 * first occurrence of last_half_cycle is wrong and we move
777 * back to the hole we've found. This case looks like
778 * x + 1 ... | x | x + 1 | x ...
779 * ^ binary search stopped here
780 * Another case we need to handle that only occurs in 256k
781 * logs is
782 * x + 1 ... | x ... | x+1 | x ...
783 * ^ binary search stops here
784 * In a 256k log, the scan at the end of the log will see the
785 * x + 1 blocks. We need to skip past those since that is
786 * certainly not the head of the log. By searching for
787 * last_half_cycle-1 we accomplish that.
788 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 ASSERT(head_blk <= INT_MAX &&
Alex Elder3f943d82010-04-15 18:17:34 +0000790 (xfs_daddr_t) num_scan_bblks >= head_blk);
791 start_blk = log_bbnum - (num_scan_bblks - head_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 if ((error = xlog_find_verify_cycle(log, start_blk,
793 num_scan_bblks - (int)head_blk,
794 (stop_on_cycle - 1), &new_blk)))
795 goto bp_err;
796 if (new_blk != -1) {
797 head_blk = new_blk;
Alex Elder9db127e2010-04-15 18:17:26 +0000798 goto validate_head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 }
800
801 /*
802 * Scan beginning of log now. The last part of the physical
803 * log is good. This scan needs to verify that it doesn't find
804 * the last_half_cycle.
805 */
806 start_blk = 0;
807 ASSERT(head_blk <= INT_MAX);
808 if ((error = xlog_find_verify_cycle(log,
809 start_blk, (int)head_blk,
810 stop_on_cycle, &new_blk)))
811 goto bp_err;
812 if (new_blk != -1)
813 head_blk = new_blk;
814 }
815
Alex Elder9db127e2010-04-15 18:17:26 +0000816validate_head:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 /*
818 * Now we need to make sure head_blk is not pointing to a block in
819 * the middle of a log record.
820 */
821 num_scan_bblks = XLOG_REC_SHIFT(log);
822 if (head_blk >= num_scan_bblks) {
823 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
824
825 /* start ptr at last block ptr before head_blk */
826 if ((error = xlog_find_verify_log_record(log, start_blk,
827 &head_blk, 0)) == -1) {
828 error = XFS_ERROR(EIO);
829 goto bp_err;
830 } else if (error)
831 goto bp_err;
832 } else {
833 start_blk = 0;
834 ASSERT(head_blk <= INT_MAX);
835 if ((error = xlog_find_verify_log_record(log, start_blk,
836 &head_blk, 0)) == -1) {
837 /* We hit the beginning of the log during our search */
Alex Elder3f943d82010-04-15 18:17:34 +0000838 start_blk = log_bbnum - (num_scan_bblks - head_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 new_blk = log_bbnum;
840 ASSERT(start_blk <= INT_MAX &&
841 (xfs_daddr_t) log_bbnum-start_blk >= 0);
842 ASSERT(head_blk <= INT_MAX);
843 if ((error = xlog_find_verify_log_record(log,
844 start_blk, &new_blk,
845 (int)head_blk)) == -1) {
846 error = XFS_ERROR(EIO);
847 goto bp_err;
848 } else if (error)
849 goto bp_err;
850 if (new_blk != log_bbnum)
851 head_blk = new_blk;
852 } else if (error)
853 goto bp_err;
854 }
855
856 xlog_put_bp(bp);
857 if (head_blk == log_bbnum)
858 *return_head_blk = 0;
859 else
860 *return_head_blk = head_blk;
861 /*
862 * When returning here, we have a good block number. Bad block
863 * means that during a previous crash, we didn't have a clean break
864 * from cycle number N to cycle number N-1. In this case, we need
865 * to find the first block with cycle number N-1.
866 */
867 return 0;
868
869 bp_err:
870 xlog_put_bp(bp);
871
872 if (error)
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100873 xfs_warn(log->l_mp, "failed to find log head");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 return error;
875}
876
877/*
878 * Find the sync block number or the tail of the log.
879 *
880 * This will be the block number of the last record to have its
881 * associated buffers synced to disk. Every log record header has
882 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
883 * to get a sync block number. The only concern is to figure out which
884 * log record header to believe.
885 *
886 * The following algorithm uses the log record header with the largest
887 * lsn. The entire log record does not need to be valid. We only care
888 * that the header is valid.
889 *
890 * We could speed up search by using current head_blk buffer, but it is not
891 * available.
892 */
Eric Sandeen5d77c0d2009-11-19 15:52:00 +0000893STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894xlog_find_tail(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500895 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 xfs_daddr_t *head_blk,
Eric Sandeen65be6052006-01-11 15:34:19 +1100897 xfs_daddr_t *tail_blk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898{
899 xlog_rec_header_t *rhead;
900 xlog_op_header_t *op_head;
901 xfs_caddr_t offset = NULL;
902 xfs_buf_t *bp;
903 int error, i, found;
904 xfs_daddr_t umount_data_blk;
905 xfs_daddr_t after_umount_blk;
906 xfs_lsn_t tail_lsn;
907 int hblks;
908
909 found = 0;
910
911 /*
912 * Find previous log record
913 */
914 if ((error = xlog_find_head(log, head_blk)))
915 return error;
916
917 bp = xlog_get_bp(log, 1);
918 if (!bp)
919 return ENOMEM;
920 if (*head_blk == 0) { /* special case */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100921 error = xlog_bread(log, 0, 1, bp, &offset);
922 if (error)
Alex Elder9db127e2010-04-15 18:17:26 +0000923 goto done;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100924
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000925 if (xlog_get_cycle(offset) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 *tail_blk = 0;
927 /* leave all other log inited values alone */
Alex Elder9db127e2010-04-15 18:17:26 +0000928 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 }
930 }
931
932 /*
933 * Search backwards looking for log record header block
934 */
935 ASSERT(*head_blk < INT_MAX);
936 for (i = (int)(*head_blk) - 1; i >= 0; i--) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100937 error = xlog_bread(log, i, 1, bp, &offset);
938 if (error)
Alex Elder9db127e2010-04-15 18:17:26 +0000939 goto done;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100940
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200941 if (*(__be32 *)offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 found = 1;
943 break;
944 }
945 }
946 /*
947 * If we haven't found the log record header block, start looking
948 * again from the end of the physical log. XXXmiken: There should be
949 * a check here to make sure we didn't search more than N blocks in
950 * the previous code.
951 */
952 if (!found) {
953 for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100954 error = xlog_bread(log, i, 1, bp, &offset);
955 if (error)
Alex Elder9db127e2010-04-15 18:17:26 +0000956 goto done;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100957
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200958 if (*(__be32 *)offset ==
959 cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 found = 2;
961 break;
962 }
963 }
964 }
965 if (!found) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100966 xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__);
Eric Sandeen050a1952013-07-31 20:33:47 -0500967 xlog_put_bp(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 ASSERT(0);
969 return XFS_ERROR(EIO);
970 }
971
972 /* find blk_no of tail of log */
973 rhead = (xlog_rec_header_t *)offset;
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000974 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975
976 /*
977 * Reset log values according to the state of the log when we
978 * crashed. In the case where head_blk == 0, we bump curr_cycle
979 * one because the next write starts a new cycle rather than
980 * continuing the cycle of the last good log record. At this
981 * point we have guaranteed that all partial log records have been
982 * accounted for. Therefore, we know that the last good log record
983 * written was complete and ended exactly on the end boundary
984 * of the physical log.
985 */
986 log->l_prev_block = i;
987 log->l_curr_block = (int)*head_blk;
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000988 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 if (found == 2)
990 log->l_curr_cycle++;
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100991 atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn));
Dave Chinner84f3c682010-12-03 22:11:29 +1100992 atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn));
Christoph Hellwig28496962012-02-20 02:31:25 +0000993 xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle,
Dave Chinnera69ed032010-12-21 12:08:20 +1100994 BBTOB(log->l_curr_block));
Christoph Hellwig28496962012-02-20 02:31:25 +0000995 xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle,
Dave Chinnera69ed032010-12-21 12:08:20 +1100996 BBTOB(log->l_curr_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997
998 /*
999 * Look for unmount record. If we find it, then we know there
1000 * was a clean unmount. Since 'i' could be the last block in
1001 * the physical log, we convert to a log block before comparing
1002 * to the head_blk.
1003 *
1004 * Save the current tail lsn to use to pass to
1005 * xlog_clear_stale_blocks() below. We won't want to clear the
1006 * unmount record if there is one, so we pass the lsn of the
1007 * unmount record rather than the block after it.
1008 */
Eric Sandeen62118702008-03-06 13:44:28 +11001009 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001010 int h_size = be32_to_cpu(rhead->h_size);
1011 int h_version = be32_to_cpu(rhead->h_version);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012
1013 if ((h_version & XLOG_VERSION_2) &&
1014 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
1015 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
1016 if (h_size % XLOG_HEADER_CYCLE_SIZE)
1017 hblks++;
1018 } else {
1019 hblks = 1;
1020 }
1021 } else {
1022 hblks = 1;
1023 }
1024 after_umount_blk = (i + hblks + (int)
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001025 BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize;
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001026 tail_lsn = atomic64_read(&log->l_tail_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 if (*head_blk == after_umount_blk &&
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001028 be32_to_cpu(rhead->h_num_logops) == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 umount_data_blk = (i + hblks) % log->l_logBBsize;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001030 error = xlog_bread(log, umount_data_blk, 1, bp, &offset);
1031 if (error)
Alex Elder9db127e2010-04-15 18:17:26 +00001032 goto done;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001033
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 op_head = (xlog_op_header_t *)offset;
1035 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
1036 /*
1037 * Set tail and last sync so that newly written
1038 * log records will point recovery to after the
1039 * current unmount record.
1040 */
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001041 xlog_assign_atomic_lsn(&log->l_tail_lsn,
1042 log->l_curr_cycle, after_umount_blk);
1043 xlog_assign_atomic_lsn(&log->l_last_sync_lsn,
1044 log->l_curr_cycle, after_umount_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 *tail_blk = after_umount_blk;
David Chinner92821e22007-05-24 15:26:31 +10001046
1047 /*
1048 * Note that the unmount was clean. If the unmount
1049 * was not clean, we need to know this to rebuild the
1050 * superblock counters from the perag headers if we
1051 * have a filesystem using non-persistent counters.
1052 */
1053 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 }
1055 }
1056
1057 /*
1058 * Make sure that there are no blocks in front of the head
1059 * with the same cycle number as the head. This can happen
1060 * because we allow multiple outstanding log writes concurrently,
1061 * and the later writes might make it out before earlier ones.
1062 *
1063 * We use the lsn from before modifying it so that we'll never
1064 * overwrite the unmount record after a clean unmount.
1065 *
1066 * Do this only if we are going to recover the filesystem
1067 *
1068 * NOTE: This used to say "if (!readonly)"
1069 * However on Linux, we can & do recover a read-only filesystem.
1070 * We only skip recovery if NORECOVERY is specified on mount,
1071 * in which case we would not be here.
1072 *
1073 * But... if the -device- itself is readonly, just skip this.
1074 * We can't recover this device anyway, so it won't matter.
1075 */
Alex Elder9db127e2010-04-15 18:17:26 +00001076 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 error = xlog_clear_stale_blocks(log, tail_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078
Alex Elder9db127e2010-04-15 18:17:26 +00001079done:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 xlog_put_bp(bp);
1081
1082 if (error)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001083 xfs_warn(log->l_mp, "failed to locate log tail");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 return error;
1085}
1086
1087/*
1088 * Is the log zeroed at all?
1089 *
1090 * The last binary search should be changed to perform an X block read
1091 * once X becomes small enough. You can then search linearly through
1092 * the X blocks. This will cut down on the number of reads we need to do.
1093 *
1094 * If the log is partially zeroed, this routine will pass back the blkno
1095 * of the first block with cycle number 0. It won't have a complete LR
1096 * preceding it.
1097 *
1098 * Return:
1099 * 0 => the log is completely written to
1100 * -1 => use *blk_no as the first block of the log
1101 * >0 => error has occurred
1102 */
David Chinnera8272ce2007-11-23 16:28:09 +11001103STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104xlog_find_zeroed(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001105 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 xfs_daddr_t *blk_no)
1107{
1108 xfs_buf_t *bp;
1109 xfs_caddr_t offset;
1110 uint first_cycle, last_cycle;
1111 xfs_daddr_t new_blk, last_blk, start_blk;
1112 xfs_daddr_t num_scan_bblks;
1113 int error, log_bbnum = log->l_logBBsize;
1114
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10001115 *blk_no = 0;
1116
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 /* check totally zeroed log */
1118 bp = xlog_get_bp(log, 1);
1119 if (!bp)
1120 return ENOMEM;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001121 error = xlog_bread(log, 0, 1, bp, &offset);
1122 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 goto bp_err;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001124
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10001125 first_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 if (first_cycle == 0) { /* completely zeroed log */
1127 *blk_no = 0;
1128 xlog_put_bp(bp);
1129 return -1;
1130 }
1131
1132 /* check partially zeroed log */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001133 error = xlog_bread(log, log_bbnum-1, 1, bp, &offset);
1134 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 goto bp_err;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001136
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10001137 last_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 if (last_cycle != 0) { /* log completely written to */
1139 xlog_put_bp(bp);
1140 return 0;
1141 } else if (first_cycle != 1) {
1142 /*
1143 * If the cycle of the last block is zero, the cycle of
1144 * the first block must be 1. If it's not, maybe we're
1145 * not looking at a log... Bail out.
1146 */
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001147 xfs_warn(log->l_mp,
1148 "Log inconsistent or not a log (last==0, first!=1)");
Eric Sandeen5d0a6542013-07-31 20:32:30 -05001149 error = XFS_ERROR(EINVAL);
1150 goto bp_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 }
1152
1153 /* we have a partially zeroed log */
1154 last_blk = log_bbnum-1;
1155 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1156 goto bp_err;
1157
1158 /*
1159 * Validate the answer. Because there is no way to guarantee that
1160 * the entire log is made up of log records which are the same size,
1161 * we scan over the defined maximum blocks. At this point, the maximum
1162 * is not chosen to mean anything special. XXXmiken
1163 */
1164 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1165 ASSERT(num_scan_bblks <= INT_MAX);
1166
1167 if (last_blk < num_scan_bblks)
1168 num_scan_bblks = last_blk;
1169 start_blk = last_blk - num_scan_bblks;
1170
1171 /*
1172 * We search for any instances of cycle number 0 that occur before
1173 * our current estimate of the head. What we're trying to detect is
1174 * 1 ... | 0 | 1 | 0...
1175 * ^ binary search ends here
1176 */
1177 if ((error = xlog_find_verify_cycle(log, start_blk,
1178 (int)num_scan_bblks, 0, &new_blk)))
1179 goto bp_err;
1180 if (new_blk != -1)
1181 last_blk = new_blk;
1182
1183 /*
1184 * Potentially backup over partial log record write. We don't need
1185 * to search the end of the log because we know it is zero.
1186 */
1187 if ((error = xlog_find_verify_log_record(log, start_blk,
1188 &last_blk, 0)) == -1) {
1189 error = XFS_ERROR(EIO);
1190 goto bp_err;
1191 } else if (error)
1192 goto bp_err;
1193
1194 *blk_no = last_blk;
1195bp_err:
1196 xlog_put_bp(bp);
1197 if (error)
1198 return error;
1199 return -1;
1200}
1201
1202/*
1203 * These are simple subroutines used by xlog_clear_stale_blocks() below
1204 * to initialize a buffer full of empty log record headers and write
1205 * them into the log.
1206 */
1207STATIC void
1208xlog_add_record(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001209 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 xfs_caddr_t buf,
1211 int cycle,
1212 int block,
1213 int tail_cycle,
1214 int tail_block)
1215{
1216 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1217
1218 memset(buf, 0, BBSIZE);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001219 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1220 recp->h_cycle = cpu_to_be32(cycle);
1221 recp->h_version = cpu_to_be32(
Eric Sandeen62118702008-03-06 13:44:28 +11001222 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001223 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1224 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1225 recp->h_fmt = cpu_to_be32(XLOG_FMT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1227}
1228
1229STATIC int
1230xlog_write_log_records(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001231 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 int cycle,
1233 int start_block,
1234 int blocks,
1235 int tail_cycle,
1236 int tail_block)
1237{
1238 xfs_caddr_t offset;
1239 xfs_buf_t *bp;
1240 int balign, ealign;
Alex Elder69ce58f2010-04-20 17:09:59 +10001241 int sectbb = log->l_sectBBsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 int end_block = start_block + blocks;
1243 int bufblks;
1244 int error = 0;
1245 int i, j = 0;
1246
Alex Elder6881a222010-04-13 15:22:29 +10001247 /*
1248 * Greedily allocate a buffer big enough to handle the full
1249 * range of basic blocks to be written. If that fails, try
1250 * a smaller size. We need to be able to write at least a
1251 * log sector, or we're out of luck.
1252 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 bufblks = 1 << ffs(blocks);
Dave Chinner81158e02012-04-27 19:45:22 +10001254 while (bufblks > log->l_logBBsize)
1255 bufblks >>= 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 while (!(bp = xlog_get_bp(log, bufblks))) {
1257 bufblks >>= 1;
Alex Elder69ce58f2010-04-20 17:09:59 +10001258 if (bufblks < sectbb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 return ENOMEM;
1260 }
1261
1262 /* We may need to do a read at the start to fill in part of
1263 * the buffer in the starting sector not covered by the first
1264 * write below.
1265 */
Alex Elder5c17f532010-04-13 15:22:48 +10001266 balign = round_down(start_block, sectbb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 if (balign != start_block) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001268 error = xlog_bread_noalign(log, start_block, 1, bp);
1269 if (error)
1270 goto out_put_bp;
1271
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 j = start_block - balign;
1273 }
1274
1275 for (i = start_block; i < end_block; i += bufblks) {
1276 int bcount, endcount;
1277
1278 bcount = min(bufblks, end_block - start_block);
1279 endcount = bcount - j;
1280
1281 /* We may need to do a read at the end to fill in part of
1282 * the buffer in the final sector not covered by the write.
1283 * If this is the same sector as the above read, skip it.
1284 */
Alex Elder5c17f532010-04-13 15:22:48 +10001285 ealign = round_down(end_block, sectbb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 if (j == 0 && (start_block + endcount > ealign)) {
Chandra Seetharaman62926042011-07-22 23:40:15 +00001287 offset = bp->b_addr + BBTOB(ealign - start_block);
Dave Chinner44396472011-04-21 09:34:27 +00001288 error = xlog_bread_offset(log, ealign, sectbb,
1289 bp, offset);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001290 if (error)
1291 break;
1292
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 }
1294
1295 offset = xlog_align(log, start_block, endcount, bp);
1296 for (; j < endcount; j++) {
1297 xlog_add_record(log, offset, cycle, i+j,
1298 tail_cycle, tail_block);
1299 offset += BBSIZE;
1300 }
1301 error = xlog_bwrite(log, start_block, endcount, bp);
1302 if (error)
1303 break;
1304 start_block += endcount;
1305 j = 0;
1306 }
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001307
1308 out_put_bp:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 xlog_put_bp(bp);
1310 return error;
1311}
1312
1313/*
1314 * This routine is called to blow away any incomplete log writes out
1315 * in front of the log head. We do this so that we won't become confused
1316 * if we come up, write only a little bit more, and then crash again.
1317 * If we leave the partial log records out there, this situation could
1318 * cause us to think those partial writes are valid blocks since they
1319 * have the current cycle number. We get rid of them by overwriting them
1320 * with empty log records with the old cycle number rather than the
1321 * current one.
1322 *
1323 * The tail lsn is passed in rather than taken from
1324 * the log so that we will not write over the unmount record after a
1325 * clean unmount in a 512 block log. Doing so would leave the log without
1326 * any valid log records in it until a new one was written. If we crashed
1327 * during that time we would not be able to recover.
1328 */
1329STATIC int
1330xlog_clear_stale_blocks(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001331 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 xfs_lsn_t tail_lsn)
1333{
1334 int tail_cycle, head_cycle;
1335 int tail_block, head_block;
1336 int tail_distance, max_distance;
1337 int distance;
1338 int error;
1339
1340 tail_cycle = CYCLE_LSN(tail_lsn);
1341 tail_block = BLOCK_LSN(tail_lsn);
1342 head_cycle = log->l_curr_cycle;
1343 head_block = log->l_curr_block;
1344
1345 /*
1346 * Figure out the distance between the new head of the log
1347 * and the tail. We want to write over any blocks beyond the
1348 * head that we may have written just before the crash, but
1349 * we don't want to overwrite the tail of the log.
1350 */
1351 if (head_cycle == tail_cycle) {
1352 /*
1353 * The tail is behind the head in the physical log,
1354 * so the distance from the head to the tail is the
1355 * distance from the head to the end of the log plus
1356 * the distance from the beginning of the log to the
1357 * tail.
1358 */
1359 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1360 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1361 XFS_ERRLEVEL_LOW, log->l_mp);
1362 return XFS_ERROR(EFSCORRUPTED);
1363 }
1364 tail_distance = tail_block + (log->l_logBBsize - head_block);
1365 } else {
1366 /*
1367 * The head is behind the tail in the physical log,
1368 * so the distance from the head to the tail is just
1369 * the tail block minus the head block.
1370 */
1371 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1372 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1373 XFS_ERRLEVEL_LOW, log->l_mp);
1374 return XFS_ERROR(EFSCORRUPTED);
1375 }
1376 tail_distance = tail_block - head_block;
1377 }
1378
1379 /*
1380 * If the head is right up against the tail, we can't clear
1381 * anything.
1382 */
1383 if (tail_distance <= 0) {
1384 ASSERT(tail_distance == 0);
1385 return 0;
1386 }
1387
1388 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1389 /*
1390 * Take the smaller of the maximum amount of outstanding I/O
1391 * we could have and the distance to the tail to clear out.
1392 * We take the smaller so that we don't overwrite the tail and
1393 * we don't waste all day writing from the head to the tail
1394 * for no reason.
1395 */
1396 max_distance = MIN(max_distance, tail_distance);
1397
1398 if ((head_block + max_distance) <= log->l_logBBsize) {
1399 /*
1400 * We can stomp all the blocks we need to without
1401 * wrapping around the end of the log. Just do it
1402 * in a single write. Use the cycle number of the
1403 * current cycle minus one so that the log will look like:
1404 * n ... | n - 1 ...
1405 */
1406 error = xlog_write_log_records(log, (head_cycle - 1),
1407 head_block, max_distance, tail_cycle,
1408 tail_block);
1409 if (error)
1410 return error;
1411 } else {
1412 /*
1413 * We need to wrap around the end of the physical log in
1414 * order to clear all the blocks. Do it in two separate
1415 * I/Os. The first write should be from the head to the
1416 * end of the physical log, and it should use the current
1417 * cycle number minus one just like above.
1418 */
1419 distance = log->l_logBBsize - head_block;
1420 error = xlog_write_log_records(log, (head_cycle - 1),
1421 head_block, distance, tail_cycle,
1422 tail_block);
1423
1424 if (error)
1425 return error;
1426
1427 /*
1428 * Now write the blocks at the start of the physical log.
1429 * This writes the remainder of the blocks we want to clear.
1430 * It uses the current cycle number since we're now on the
1431 * same cycle as the head so that we get:
1432 * n ... n ... | n - 1 ...
1433 * ^^^^^ blocks we're writing
1434 */
1435 distance = max_distance - (log->l_logBBsize - head_block);
1436 error = xlog_write_log_records(log, head_cycle, 0, distance,
1437 tail_cycle, tail_block);
1438 if (error)
1439 return error;
1440 }
1441
1442 return 0;
1443}
1444
1445/******************************************************************************
1446 *
1447 * Log recover routines
1448 *
1449 ******************************************************************************
1450 */
1451
1452STATIC xlog_recover_t *
1453xlog_recover_find_tid(
Dave Chinnerf0a76952010-01-11 11:49:57 +00001454 struct hlist_head *head,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 xlog_tid_t tid)
1456{
Dave Chinnerf0a76952010-01-11 11:49:57 +00001457 xlog_recover_t *trans;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458
Sasha Levinb67bfe02013-02-27 17:06:00 -08001459 hlist_for_each_entry(trans, head, r_list) {
Dave Chinnerf0a76952010-01-11 11:49:57 +00001460 if (trans->r_log_tid == tid)
1461 return trans;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 }
Dave Chinnerf0a76952010-01-11 11:49:57 +00001463 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464}
1465
1466STATIC void
Dave Chinnerf0a76952010-01-11 11:49:57 +00001467xlog_recover_new_tid(
1468 struct hlist_head *head,
1469 xlog_tid_t tid,
1470 xfs_lsn_t lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471{
Dave Chinnerf0a76952010-01-11 11:49:57 +00001472 xlog_recover_t *trans;
1473
1474 trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
1475 trans->r_log_tid = tid;
1476 trans->r_lsn = lsn;
1477 INIT_LIST_HEAD(&trans->r_itemq);
1478
1479 INIT_HLIST_NODE(&trans->r_list);
1480 hlist_add_head(&trans->r_list, head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481}
1482
1483STATIC void
1484xlog_recover_add_item(
Dave Chinnerf0a76952010-01-11 11:49:57 +00001485 struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486{
1487 xlog_recover_item_t *item;
1488
1489 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
Dave Chinnerf0a76952010-01-11 11:49:57 +00001490 INIT_LIST_HEAD(&item->ri_list);
1491 list_add_tail(&item->ri_list, head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492}
1493
1494STATIC int
1495xlog_recover_add_to_cont_trans(
Mark Tinguelyad223e62012-06-14 09:22:15 -05001496 struct xlog *log,
1497 struct xlog_recover *trans,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 xfs_caddr_t dp,
1499 int len)
1500{
1501 xlog_recover_item_t *item;
1502 xfs_caddr_t ptr, old_ptr;
1503 int old_len;
1504
Dave Chinnerf0a76952010-01-11 11:49:57 +00001505 if (list_empty(&trans->r_itemq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 /* finish copying rest of trans header */
1507 xlog_recover_add_item(&trans->r_itemq);
1508 ptr = (xfs_caddr_t) &trans->r_theader +
1509 sizeof(xfs_trans_header_t) - len;
1510 memcpy(ptr, dp, len); /* d, s, l */
1511 return 0;
1512 }
Dave Chinnerf0a76952010-01-11 11:49:57 +00001513 /* take the tail entry */
1514 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515
1516 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
1517 old_len = item->ri_buf[item->ri_cnt-1].i_len;
1518
Mitsuo Hayasaka45053602012-01-27 06:37:26 +00001519 ptr = kmem_realloc(old_ptr, len+old_len, old_len, KM_SLEEP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 memcpy(&ptr[old_len], dp, len); /* d, s, l */
1521 item->ri_buf[item->ri_cnt-1].i_len += len;
1522 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
Dave Chinner9abbc532010-04-13 15:06:46 +10001523 trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 return 0;
1525}
1526
1527/*
1528 * The next region to add is the start of a new region. It could be
1529 * a whole region or it could be the first part of a new region. Because
1530 * of this, the assumption here is that the type and size fields of all
1531 * format structures fit into the first 32 bits of the structure.
1532 *
1533 * This works because all regions must be 32 bit aligned. Therefore, we
1534 * either have both fields or we have neither field. In the case we have
1535 * neither field, the data part of the region is zero length. We only have
1536 * a log_op_header and can throw away the header since a new one will appear
1537 * later. If we have at least 4 bytes, then we can determine how many regions
1538 * will appear in the current log item.
1539 */
1540STATIC int
1541xlog_recover_add_to_trans(
Mark Tinguelyad223e62012-06-14 09:22:15 -05001542 struct xlog *log,
1543 struct xlog_recover *trans,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 xfs_caddr_t dp,
1545 int len)
1546{
1547 xfs_inode_log_format_t *in_f; /* any will do */
1548 xlog_recover_item_t *item;
1549 xfs_caddr_t ptr;
1550
1551 if (!len)
1552 return 0;
Dave Chinnerf0a76952010-01-11 11:49:57 +00001553 if (list_empty(&trans->r_itemq)) {
David Chinner5a792c42008-10-30 17:40:09 +11001554 /* we need to catch log corruptions here */
1555 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001556 xfs_warn(log->l_mp, "%s: bad header magic number",
1557 __func__);
David Chinner5a792c42008-10-30 17:40:09 +11001558 ASSERT(0);
1559 return XFS_ERROR(EIO);
1560 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 if (len == sizeof(xfs_trans_header_t))
1562 xlog_recover_add_item(&trans->r_itemq);
1563 memcpy(&trans->r_theader, dp, len); /* d, s, l */
1564 return 0;
1565 }
1566
1567 ptr = kmem_alloc(len, KM_SLEEP);
1568 memcpy(ptr, dp, len);
1569 in_f = (xfs_inode_log_format_t *)ptr;
1570
Dave Chinnerf0a76952010-01-11 11:49:57 +00001571 /* take the tail entry */
1572 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
1573 if (item->ri_total != 0 &&
1574 item->ri_total == item->ri_cnt) {
1575 /* tail item is in use, get a new one */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 xlog_recover_add_item(&trans->r_itemq);
Dave Chinnerf0a76952010-01-11 11:49:57 +00001577 item = list_entry(trans->r_itemq.prev,
1578 xlog_recover_item_t, ri_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580
1581 if (item->ri_total == 0) { /* first region to be added */
Christoph Hellwige8fa6b42009-03-03 14:48:36 -05001582 if (in_f->ilf_size == 0 ||
1583 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001584 xfs_warn(log->l_mp,
1585 "bad number of regions (%d) in inode log format",
Christoph Hellwige8fa6b42009-03-03 14:48:36 -05001586 in_f->ilf_size);
1587 ASSERT(0);
1588 return XFS_ERROR(EIO);
1589 }
1590
1591 item->ri_total = in_f->ilf_size;
1592 item->ri_buf =
1593 kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
1594 KM_SLEEP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 }
1596 ASSERT(item->ri_total > item->ri_cnt);
1597 /* Description region is ri_buf[0] */
1598 item->ri_buf[item->ri_cnt].i_addr = ptr;
1599 item->ri_buf[item->ri_cnt].i_len = len;
1600 item->ri_cnt++;
Dave Chinner9abbc532010-04-13 15:06:46 +10001601 trace_xfs_log_recover_item_add(log, trans, item, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602 return 0;
1603}
1604
Dave Chinnerf0a76952010-01-11 11:49:57 +00001605/*
Dave Chinnera775ad72013-06-05 12:09:07 +10001606 * Sort the log items in the transaction.
1607 *
1608 * The ordering constraints are defined by the inode allocation and unlink
1609 * behaviour. The rules are:
1610 *
1611 * 1. Every item is only logged once in a given transaction. Hence it
1612 * represents the last logged state of the item. Hence ordering is
1613 * dependent on the order in which operations need to be performed so
1614 * required initial conditions are always met.
1615 *
1616 * 2. Cancelled buffers are recorded in pass 1 in a separate table and
1617 * there's nothing to replay from them so we can simply cull them
1618 * from the transaction. However, we can't do that until after we've
1619 * replayed all the other items because they may be dependent on the
1620 * cancelled buffer and replaying the cancelled buffer can remove it
1621 * form the cancelled buffer table. Hence they have tobe done last.
1622 *
1623 * 3. Inode allocation buffers must be replayed before inode items that
Dave Chinner28c8e412013-06-27 16:04:55 +10001624 * read the buffer and replay changes into it. For filesystems using the
1625 * ICREATE transactions, this means XFS_LI_ICREATE objects need to get
1626 * treated the same as inode allocation buffers as they create and
1627 * initialise the buffers directly.
Dave Chinnera775ad72013-06-05 12:09:07 +10001628 *
1629 * 4. Inode unlink buffers must be replayed after inode items are replayed.
1630 * This ensures that inodes are completely flushed to the inode buffer
1631 * in a "free" state before we remove the unlinked inode list pointer.
1632 *
1633 * Hence the ordering needs to be inode allocation buffers first, inode items
1634 * second, inode unlink buffers third and cancelled buffers last.
1635 *
1636 * But there's a problem with that - we can't tell an inode allocation buffer
1637 * apart from a regular buffer, so we can't separate them. We can, however,
1638 * tell an inode unlink buffer from the others, and so we can separate them out
1639 * from all the other buffers and move them to last.
1640 *
1641 * Hence, 4 lists, in order from head to tail:
Dave Chinner28c8e412013-06-27 16:04:55 +10001642 * - buffer_list for all buffers except cancelled/inode unlink buffers
1643 * - item_list for all non-buffer items
1644 * - inode_buffer_list for inode unlink buffers
1645 * - cancel_list for the cancelled buffers
1646 *
1647 * Note that we add objects to the tail of the lists so that first-to-last
1648 * ordering is preserved within the lists. Adding objects to the head of the
1649 * list means when we traverse from the head we walk them in last-to-first
1650 * order. For cancelled buffers and inode unlink buffers this doesn't matter,
1651 * but for all other items there may be specific ordering that we need to
1652 * preserve.
Dave Chinnerf0a76952010-01-11 11:49:57 +00001653 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654STATIC int
1655xlog_recover_reorder_trans(
Mark Tinguelyad223e62012-06-14 09:22:15 -05001656 struct xlog *log,
1657 struct xlog_recover *trans,
Dave Chinner9abbc532010-04-13 15:06:46 +10001658 int pass)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659{
Dave Chinnerf0a76952010-01-11 11:49:57 +00001660 xlog_recover_item_t *item, *n;
1661 LIST_HEAD(sort_list);
Dave Chinnera775ad72013-06-05 12:09:07 +10001662 LIST_HEAD(cancel_list);
1663 LIST_HEAD(buffer_list);
1664 LIST_HEAD(inode_buffer_list);
1665 LIST_HEAD(inode_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666
Dave Chinnerf0a76952010-01-11 11:49:57 +00001667 list_splice_init(&trans->r_itemq, &sort_list);
1668 list_for_each_entry_safe(item, n, &sort_list, ri_list) {
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10001669 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
Dave Chinnerf0a76952010-01-11 11:49:57 +00001670
1671 switch (ITEM_TYPE(item)) {
Dave Chinner28c8e412013-06-27 16:04:55 +10001672 case XFS_LI_ICREATE:
1673 list_move_tail(&item->ri_list, &buffer_list);
1674 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 case XFS_LI_BUF:
Dave Chinnera775ad72013-06-05 12:09:07 +10001676 if (buf_f->blf_flags & XFS_BLF_CANCEL) {
Dave Chinner9abbc532010-04-13 15:06:46 +10001677 trace_xfs_log_recover_item_reorder_head(log,
1678 trans, item, pass);
Dave Chinnera775ad72013-06-05 12:09:07 +10001679 list_move(&item->ri_list, &cancel_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 break;
1681 }
Dave Chinnera775ad72013-06-05 12:09:07 +10001682 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
1683 list_move(&item->ri_list, &inode_buffer_list);
1684 break;
1685 }
1686 list_move_tail(&item->ri_list, &buffer_list);
1687 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688 case XFS_LI_INODE:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 case XFS_LI_DQUOT:
1690 case XFS_LI_QUOTAOFF:
1691 case XFS_LI_EFD:
1692 case XFS_LI_EFI:
Dave Chinner9abbc532010-04-13 15:06:46 +10001693 trace_xfs_log_recover_item_reorder_tail(log,
1694 trans, item, pass);
Dave Chinnera775ad72013-06-05 12:09:07 +10001695 list_move_tail(&item->ri_list, &inode_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 break;
1697 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001698 xfs_warn(log->l_mp,
1699 "%s: unrecognized type of log operation",
1700 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 ASSERT(0);
1702 return XFS_ERROR(EIO);
1703 }
Dave Chinnerf0a76952010-01-11 11:49:57 +00001704 }
1705 ASSERT(list_empty(&sort_list));
Dave Chinnera775ad72013-06-05 12:09:07 +10001706 if (!list_empty(&buffer_list))
1707 list_splice(&buffer_list, &trans->r_itemq);
1708 if (!list_empty(&inode_list))
1709 list_splice_tail(&inode_list, &trans->r_itemq);
1710 if (!list_empty(&inode_buffer_list))
1711 list_splice_tail(&inode_buffer_list, &trans->r_itemq);
1712 if (!list_empty(&cancel_list))
1713 list_splice_tail(&cancel_list, &trans->r_itemq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 return 0;
1715}
1716
1717/*
1718 * Build up the table of buf cancel records so that we don't replay
1719 * cancelled data in the second pass. For buffer records that are
1720 * not cancel records, there is nothing to do here so we just return.
1721 *
1722 * If we get a cancel record which is already in the table, this indicates
1723 * that the buffer was cancelled multiple times. In order to ensure
1724 * that during pass 2 we keep the record in the table until we reach its
1725 * last occurrence in the log, we keep a reference count in the cancel
1726 * record in the table to tell us how many times we expect to see this
1727 * record during the second pass.
1728 */
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001729STATIC int
1730xlog_recover_buffer_pass1(
Mark Tinguelyad223e62012-06-14 09:22:15 -05001731 struct xlog *log,
1732 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001734 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001735 struct list_head *bucket;
1736 struct xfs_buf_cancel *bcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737
1738 /*
1739 * If this isn't a cancel buffer item, then just return.
1740 */
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001741 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
Dave Chinner9abbc532010-04-13 15:06:46 +10001742 trace_xfs_log_recover_buf_not_cancel(log, buf_f);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001743 return 0;
Dave Chinner9abbc532010-04-13 15:06:46 +10001744 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745
1746 /*
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001747 * Insert an xfs_buf_cancel record into the hash table of them.
1748 * If there is already an identical record, bump its reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 */
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001750 bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno);
1751 list_for_each_entry(bcp, bucket, bc_list) {
1752 if (bcp->bc_blkno == buf_f->blf_blkno &&
1753 bcp->bc_len == buf_f->blf_len) {
1754 bcp->bc_refcount++;
Dave Chinner9abbc532010-04-13 15:06:46 +10001755 trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001756 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 }
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001759
1760 bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP);
1761 bcp->bc_blkno = buf_f->blf_blkno;
1762 bcp->bc_len = buf_f->blf_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 bcp->bc_refcount = 1;
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001764 list_add_tail(&bcp->bc_list, bucket);
1765
Dave Chinner9abbc532010-04-13 15:06:46 +10001766 trace_xfs_log_recover_buf_cancel_add(log, buf_f);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001767 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768}
1769
1770/*
1771 * Check to see whether the buffer being recovered has a corresponding
Dave Chinner84a5b732013-08-27 08:10:53 +10001772 * entry in the buffer cancel record table. If it is, return the cancel
1773 * buffer structure to the caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 */
Dave Chinner84a5b732013-08-27 08:10:53 +10001775STATIC struct xfs_buf_cancel *
1776xlog_peek_buffer_cancelled(
Mark Tinguelyad223e62012-06-14 09:22:15 -05001777 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 xfs_daddr_t blkno,
1779 uint len,
1780 ushort flags)
1781{
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001782 struct list_head *bucket;
1783 struct xfs_buf_cancel *bcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784
Dave Chinner84a5b732013-08-27 08:10:53 +10001785 if (!log->l_buf_cancel_table) {
1786 /* empty table means no cancelled buffers in the log */
Dave Chinnerc1155412010-05-07 11:05:19 +10001787 ASSERT(!(flags & XFS_BLF_CANCEL));
Dave Chinner84a5b732013-08-27 08:10:53 +10001788 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 }
1790
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001791 bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno);
1792 list_for_each_entry(bcp, bucket, bc_list) {
1793 if (bcp->bc_blkno == blkno && bcp->bc_len == len)
Dave Chinner84a5b732013-08-27 08:10:53 +10001794 return bcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 }
1796
1797 /*
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001798 * We didn't find a corresponding entry in the table, so return 0 so
1799 * that the buffer is NOT cancelled.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 */
Dave Chinnerc1155412010-05-07 11:05:19 +10001801 ASSERT(!(flags & XFS_BLF_CANCEL));
Dave Chinner84a5b732013-08-27 08:10:53 +10001802 return NULL;
1803}
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001804
Dave Chinner84a5b732013-08-27 08:10:53 +10001805/*
1806 * If the buffer is being cancelled then return 1 so that it will be cancelled,
1807 * otherwise return 0. If the buffer is actually a buffer cancel item
1808 * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the
1809 * table and remove it from the table if this is the last reference.
1810 *
1811 * We remove the cancel record from the table when we encounter its last
1812 * occurrence in the log so that if the same buffer is re-used again after its
1813 * last cancellation we actually replay the changes made at that point.
1814 */
1815STATIC int
1816xlog_check_buffer_cancelled(
1817 struct xlog *log,
1818 xfs_daddr_t blkno,
1819 uint len,
1820 ushort flags)
1821{
1822 struct xfs_buf_cancel *bcp;
1823
1824 bcp = xlog_peek_buffer_cancelled(log, blkno, len, flags);
1825 if (!bcp)
1826 return 0;
1827
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001828 /*
1829 * We've go a match, so return 1 so that the recovery of this buffer
1830 * is cancelled. If this buffer is actually a buffer cancel log
1831 * item, then decrement the refcount on the one in the table and
1832 * remove it if this is the last reference.
1833 */
1834 if (flags & XFS_BLF_CANCEL) {
1835 if (--bcp->bc_refcount == 0) {
1836 list_del(&bcp->bc_list);
1837 kmem_free(bcp);
1838 }
1839 }
1840 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841}
1842
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843/*
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001844 * Perform recovery for a buffer full of inodes. In these buffers, the only
1845 * data which should be recovered is that which corresponds to the
1846 * di_next_unlinked pointers in the on disk inode structures. The rest of the
1847 * data for the inodes is always logged through the inodes themselves rather
1848 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 *
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001850 * The only time when buffers full of inodes are fully recovered is when the
1851 * buffer is full of newly allocated inodes. In this case the buffer will
1852 * not be marked as an inode buffer and so will be sent to
1853 * xlog_recover_do_reg_buffer() below during recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 */
1855STATIC int
1856xlog_recover_do_inode_buffer(
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001857 struct xfs_mount *mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 xlog_recover_item_t *item,
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001859 struct xfs_buf *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 xfs_buf_log_format_t *buf_f)
1861{
1862 int i;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001863 int item_index = 0;
1864 int bit = 0;
1865 int nbits = 0;
1866 int reg_buf_offset = 0;
1867 int reg_buf_bytes = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868 int next_unlinked_offset;
1869 int inodes_per_buf;
1870 xfs_agino_t *logged_nextp;
1871 xfs_agino_t *buffer_nextp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872
Dave Chinner9abbc532010-04-13 15:06:46 +10001873 trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f);
Dave Chinner9222a9c2013-06-12 12:19:06 +10001874
1875 /*
1876 * Post recovery validation only works properly on CRC enabled
1877 * filesystems.
1878 */
1879 if (xfs_sb_version_hascrc(&mp->m_sb))
1880 bp->b_ops = &xfs_inode_buf_ops;
Dave Chinner9abbc532010-04-13 15:06:46 +10001881
Dave Chinneraa0e8832012-04-23 15:58:52 +10001882 inodes_per_buf = BBTOB(bp->b_io_length) >> mp->m_sb.sb_inodelog;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 for (i = 0; i < inodes_per_buf; i++) {
1884 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
1885 offsetof(xfs_dinode_t, di_next_unlinked);
1886
1887 while (next_unlinked_offset >=
1888 (reg_buf_offset + reg_buf_bytes)) {
1889 /*
1890 * The next di_next_unlinked field is beyond
1891 * the current logged region. Find the next
1892 * logged region that contains or is beyond
1893 * the current di_next_unlinked field.
1894 */
1895 bit += nbits;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001896 bit = xfs_next_bit(buf_f->blf_data_map,
1897 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898
1899 /*
1900 * If there are no more logged regions in the
1901 * buffer, then we're done.
1902 */
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001903 if (bit == -1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001906 nbits = xfs_contig_bits(buf_f->blf_data_map,
1907 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908 ASSERT(nbits > 0);
Dave Chinnerc1155412010-05-07 11:05:19 +10001909 reg_buf_offset = bit << XFS_BLF_SHIFT;
1910 reg_buf_bytes = nbits << XFS_BLF_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911 item_index++;
1912 }
1913
1914 /*
1915 * If the current logged region starts after the current
1916 * di_next_unlinked field, then move on to the next
1917 * di_next_unlinked field.
1918 */
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001919 if (next_unlinked_offset < reg_buf_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921
1922 ASSERT(item->ri_buf[item_index].i_addr != NULL);
Dave Chinnerc1155412010-05-07 11:05:19 +10001923 ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0);
Dave Chinneraa0e8832012-04-23 15:58:52 +10001924 ASSERT((reg_buf_offset + reg_buf_bytes) <=
1925 BBTOB(bp->b_io_length));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926
1927 /*
1928 * The current logged region contains a copy of the
1929 * current di_next_unlinked field. Extract its value
1930 * and copy it to the buffer copy.
1931 */
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10001932 logged_nextp = item->ri_buf[item_index].i_addr +
1933 next_unlinked_offset - reg_buf_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 if (unlikely(*logged_nextp == 0)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001935 xfs_alert(mp,
1936 "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
1937 "Trying to replay bad (0) inode di_next_unlinked field.",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 item, bp);
1939 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1940 XFS_ERRLEVEL_LOW, mp);
1941 return XFS_ERROR(EFSCORRUPTED);
1942 }
1943
1944 buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
1945 next_unlinked_offset);
Tim Shimmin87c199c2006-06-09 14:56:16 +10001946 *buffer_nextp = *logged_nextp;
Dave Chinner0a32c262013-06-05 12:09:08 +10001947
1948 /*
1949 * If necessary, recalculate the CRC in the on-disk inode. We
1950 * have to leave the inode in a consistent state for whoever
1951 * reads it next....
1952 */
1953 xfs_dinode_calc_crc(mp, (struct xfs_dinode *)
1954 xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize));
1955
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 }
1957
1958 return 0;
1959}
1960
1961/*
Dave Chinnerd75afeb2013-04-03 16:11:29 +11001962 * Validate the recovered buffer is of the correct type and attach the
1963 * appropriate buffer operations to them for writeback. Magic numbers are in a
1964 * few places:
1965 * the first 16 bits of the buffer (inode buffer, dquot buffer),
1966 * the first 32 bits of the buffer (most blocks),
1967 * inside a struct xfs_da_blkinfo at the start of the buffer.
1968 */
1969static void
1970xlog_recovery_validate_buf_type(
1971 struct xfs_mount *mp,
1972 struct xfs_buf *bp,
1973 xfs_buf_log_format_t *buf_f)
1974{
1975 struct xfs_da_blkinfo *info = bp->b_addr;
1976 __uint32_t magic32;
1977 __uint16_t magic16;
1978 __uint16_t magicda;
1979
1980 magic32 = be32_to_cpu(*(__be32 *)bp->b_addr);
1981 magic16 = be16_to_cpu(*(__be16*)bp->b_addr);
1982 magicda = be16_to_cpu(info->magic);
Dave Chinner61fe1352013-04-03 16:11:30 +11001983 switch (xfs_blft_from_flags(buf_f)) {
1984 case XFS_BLFT_BTREE_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11001985 switch (magic32) {
1986 case XFS_ABTB_CRC_MAGIC:
1987 case XFS_ABTC_CRC_MAGIC:
1988 case XFS_ABTB_MAGIC:
1989 case XFS_ABTC_MAGIC:
1990 bp->b_ops = &xfs_allocbt_buf_ops;
1991 break;
1992 case XFS_IBT_CRC_MAGIC:
1993 case XFS_IBT_MAGIC:
1994 bp->b_ops = &xfs_inobt_buf_ops;
1995 break;
1996 case XFS_BMAP_CRC_MAGIC:
1997 case XFS_BMAP_MAGIC:
1998 bp->b_ops = &xfs_bmbt_buf_ops;
1999 break;
2000 default:
2001 xfs_warn(mp, "Bad btree block magic!");
2002 ASSERT(0);
2003 break;
2004 }
2005 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002006 case XFS_BLFT_AGF_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002007 if (magic32 != XFS_AGF_MAGIC) {
2008 xfs_warn(mp, "Bad AGF block magic!");
2009 ASSERT(0);
2010 break;
2011 }
2012 bp->b_ops = &xfs_agf_buf_ops;
2013 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002014 case XFS_BLFT_AGFL_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002015 if (!xfs_sb_version_hascrc(&mp->m_sb))
2016 break;
2017 if (magic32 != XFS_AGFL_MAGIC) {
2018 xfs_warn(mp, "Bad AGFL block magic!");
2019 ASSERT(0);
2020 break;
2021 }
2022 bp->b_ops = &xfs_agfl_buf_ops;
2023 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002024 case XFS_BLFT_AGI_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002025 if (magic32 != XFS_AGI_MAGIC) {
2026 xfs_warn(mp, "Bad AGI block magic!");
2027 ASSERT(0);
2028 break;
2029 }
2030 bp->b_ops = &xfs_agi_buf_ops;
2031 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002032 case XFS_BLFT_UDQUOT_BUF:
2033 case XFS_BLFT_PDQUOT_BUF:
2034 case XFS_BLFT_GDQUOT_BUF:
Dave Chinner123887e2013-04-30 21:39:33 +10002035#ifdef CONFIG_XFS_QUOTA
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002036 if (magic16 != XFS_DQUOT_MAGIC) {
2037 xfs_warn(mp, "Bad DQUOT block magic!");
2038 ASSERT(0);
2039 break;
2040 }
2041 bp->b_ops = &xfs_dquot_buf_ops;
Dave Chinner123887e2013-04-30 21:39:33 +10002042#else
2043 xfs_alert(mp,
2044 "Trying to recover dquots without QUOTA support built in!");
2045 ASSERT(0);
2046#endif
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002047 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002048 case XFS_BLFT_DINO_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002049 /*
2050 * we get here with inode allocation buffers, not buffers that
2051 * track unlinked list changes.
2052 */
2053 if (magic16 != XFS_DINODE_MAGIC) {
2054 xfs_warn(mp, "Bad INODE block magic!");
2055 ASSERT(0);
2056 break;
2057 }
2058 bp->b_ops = &xfs_inode_buf_ops;
2059 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002060 case XFS_BLFT_SYMLINK_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002061 if (magic32 != XFS_SYMLINK_MAGIC) {
2062 xfs_warn(mp, "Bad symlink block magic!");
2063 ASSERT(0);
2064 break;
2065 }
2066 bp->b_ops = &xfs_symlink_buf_ops;
2067 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002068 case XFS_BLFT_DIR_BLOCK_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002069 if (magic32 != XFS_DIR2_BLOCK_MAGIC &&
2070 magic32 != XFS_DIR3_BLOCK_MAGIC) {
2071 xfs_warn(mp, "Bad dir block magic!");
2072 ASSERT(0);
2073 break;
2074 }
2075 bp->b_ops = &xfs_dir3_block_buf_ops;
2076 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002077 case XFS_BLFT_DIR_DATA_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002078 if (magic32 != XFS_DIR2_DATA_MAGIC &&
2079 magic32 != XFS_DIR3_DATA_MAGIC) {
2080 xfs_warn(mp, "Bad dir data magic!");
2081 ASSERT(0);
2082 break;
2083 }
2084 bp->b_ops = &xfs_dir3_data_buf_ops;
2085 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002086 case XFS_BLFT_DIR_FREE_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002087 if (magic32 != XFS_DIR2_FREE_MAGIC &&
2088 magic32 != XFS_DIR3_FREE_MAGIC) {
2089 xfs_warn(mp, "Bad dir3 free magic!");
2090 ASSERT(0);
2091 break;
2092 }
2093 bp->b_ops = &xfs_dir3_free_buf_ops;
2094 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002095 case XFS_BLFT_DIR_LEAF1_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002096 if (magicda != XFS_DIR2_LEAF1_MAGIC &&
2097 magicda != XFS_DIR3_LEAF1_MAGIC) {
2098 xfs_warn(mp, "Bad dir leaf1 magic!");
2099 ASSERT(0);
2100 break;
2101 }
2102 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
2103 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002104 case XFS_BLFT_DIR_LEAFN_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002105 if (magicda != XFS_DIR2_LEAFN_MAGIC &&
2106 magicda != XFS_DIR3_LEAFN_MAGIC) {
2107 xfs_warn(mp, "Bad dir leafn magic!");
2108 ASSERT(0);
2109 break;
2110 }
2111 bp->b_ops = &xfs_dir3_leafn_buf_ops;
2112 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002113 case XFS_BLFT_DA_NODE_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002114 if (magicda != XFS_DA_NODE_MAGIC &&
2115 magicda != XFS_DA3_NODE_MAGIC) {
2116 xfs_warn(mp, "Bad da node magic!");
2117 ASSERT(0);
2118 break;
2119 }
2120 bp->b_ops = &xfs_da3_node_buf_ops;
2121 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002122 case XFS_BLFT_ATTR_LEAF_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002123 if (magicda != XFS_ATTR_LEAF_MAGIC &&
2124 magicda != XFS_ATTR3_LEAF_MAGIC) {
2125 xfs_warn(mp, "Bad attr leaf magic!");
2126 ASSERT(0);
2127 break;
2128 }
2129 bp->b_ops = &xfs_attr3_leaf_buf_ops;
2130 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002131 case XFS_BLFT_ATTR_RMT_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002132 if (!xfs_sb_version_hascrc(&mp->m_sb))
2133 break;
Dave Chinnercab09a82013-04-30 21:39:36 +10002134 if (magic32 != XFS_ATTR3_RMT_MAGIC) {
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002135 xfs_warn(mp, "Bad attr remote magic!");
2136 ASSERT(0);
2137 break;
2138 }
2139 bp->b_ops = &xfs_attr3_rmt_buf_ops;
2140 break;
Dave Chinner04a1e6c2013-04-03 16:11:31 +11002141 case XFS_BLFT_SB_BUF:
2142 if (magic32 != XFS_SB_MAGIC) {
2143 xfs_warn(mp, "Bad SB block magic!");
2144 ASSERT(0);
2145 break;
2146 }
2147 bp->b_ops = &xfs_sb_buf_ops;
2148 break;
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002149 default:
Dave Chinner61fe1352013-04-03 16:11:30 +11002150 xfs_warn(mp, "Unknown buffer type %d!",
2151 xfs_blft_from_flags(buf_f));
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002152 break;
2153 }
2154}
2155
2156/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157 * Perform a 'normal' buffer recovery. Each logged region of the
2158 * buffer should be copied over the corresponding region in the
2159 * given buffer. The bitmap in the buf log format structure indicates
2160 * where to place the logged data.
2161 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162STATIC void
2163xlog_recover_do_reg_buffer(
Dave Chinner9abbc532010-04-13 15:06:46 +10002164 struct xfs_mount *mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 xlog_recover_item_t *item,
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002166 struct xfs_buf *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 xfs_buf_log_format_t *buf_f)
2168{
2169 int i;
2170 int bit;
2171 int nbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172 int error;
2173
Dave Chinner9abbc532010-04-13 15:06:46 +10002174 trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f);
2175
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 bit = 0;
2177 i = 1; /* 0 is the buf format structure */
2178 while (1) {
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002179 bit = xfs_next_bit(buf_f->blf_data_map,
2180 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181 if (bit == -1)
2182 break;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002183 nbits = xfs_contig_bits(buf_f->blf_data_map,
2184 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 ASSERT(nbits > 0);
Christoph Hellwig4b809162007-08-16 15:37:36 +10002186 ASSERT(item->ri_buf[i].i_addr != NULL);
Dave Chinnerc1155412010-05-07 11:05:19 +10002187 ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0);
Dave Chinneraa0e8832012-04-23 15:58:52 +10002188 ASSERT(BBTOB(bp->b_io_length) >=
2189 ((uint)bit << XFS_BLF_SHIFT) + (nbits << XFS_BLF_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190
2191 /*
Dave Chinner709da6a2013-05-27 16:38:23 +10002192 * The dirty regions logged in the buffer, even though
2193 * contiguous, may span multiple chunks. This is because the
2194 * dirty region may span a physical page boundary in a buffer
2195 * and hence be split into two separate vectors for writing into
2196 * the log. Hence we need to trim nbits back to the length of
2197 * the current region being copied out of the log.
2198 */
2199 if (item->ri_buf[i].i_len < (nbits << XFS_BLF_SHIFT))
2200 nbits = item->ri_buf[i].i_len >> XFS_BLF_SHIFT;
2201
2202 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203 * Do a sanity check if this is a dquot buffer. Just checking
2204 * the first dquot in the buffer should do. XXXThis is
2205 * probably a good thing to do for other buf types also.
2206 */
2207 error = 0;
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002208 if (buf_f->blf_flags &
Dave Chinnerc1155412010-05-07 11:05:19 +10002209 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002210 if (item->ri_buf[i].i_addr == NULL) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002211 xfs_alert(mp,
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002212 "XFS: NULL dquot in %s.", __func__);
2213 goto next;
2214 }
Jan Rekorajski8ec6dba2009-11-16 11:57:02 +00002215 if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002216 xfs_alert(mp,
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002217 "XFS: dquot too small (%d) in %s.",
2218 item->ri_buf[i].i_len, __func__);
2219 goto next;
2220 }
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002221 error = xfs_qm_dqcheck(mp, item->ri_buf[i].i_addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 -1, 0, XFS_QMOPT_DOWARN,
2223 "dquot_buf_recover");
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002224 if (error)
2225 goto next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 }
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002227
2228 memcpy(xfs_buf_offset(bp,
Dave Chinnerc1155412010-05-07 11:05:19 +10002229 (uint)bit << XFS_BLF_SHIFT), /* dest */
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002230 item->ri_buf[i].i_addr, /* source */
Dave Chinnerc1155412010-05-07 11:05:19 +10002231 nbits<<XFS_BLF_SHIFT); /* length */
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002232 next:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233 i++;
2234 bit += nbits;
2235 }
2236
2237 /* Shouldn't be any more regions */
2238 ASSERT(i == item->ri_total);
Christoph Hellwigee1a47a2013-04-21 14:53:46 -05002239
Dave Chinner9222a9c2013-06-12 12:19:06 +10002240 /*
2241 * We can only do post recovery validation on items on CRC enabled
2242 * fielsystems as we need to know when the buffer was written to be able
2243 * to determine if we should have replayed the item. If we replay old
2244 * metadata over a newer buffer, then it will enter a temporarily
2245 * inconsistent state resulting in verification failures. Hence for now
2246 * just avoid the verification stage for non-crc filesystems
2247 */
2248 if (xfs_sb_version_hascrc(&mp->m_sb))
2249 xlog_recovery_validate_buf_type(mp, bp, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250}
2251
2252/*
2253 * Do some primitive error checking on ondisk dquot data structures.
2254 */
2255int
2256xfs_qm_dqcheck(
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002257 struct xfs_mount *mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 xfs_disk_dquot_t *ddq,
2259 xfs_dqid_t id,
2260 uint type, /* used only when IO_dorepair is true */
2261 uint flags,
2262 char *str)
2263{
2264 xfs_dqblk_t *d = (xfs_dqblk_t *)ddq;
2265 int errs = 0;
2266
2267 /*
2268 * We can encounter an uninitialized dquot buffer for 2 reasons:
2269 * 1. If we crash while deleting the quotainode(s), and those blks got
2270 * used for user data. This is because we take the path of regular
2271 * file deletion; however, the size field of quotainodes is never
2272 * updated, so all the tricks that we play in itruncate_finish
2273 * don't quite matter.
2274 *
2275 * 2. We don't play the quota buffers when there's a quotaoff logitem.
2276 * But the allocation will be replayed so we'll end up with an
2277 * uninitialized quota block.
2278 *
2279 * This is all fine; things are still consistent, and we haven't lost
2280 * any quota information. Just don't complain about bad dquot blks.
2281 */
Christoph Hellwig69ef9212011-07-08 14:36:05 +02002282 if (ddq->d_magic != cpu_to_be16(XFS_DQUOT_MAGIC)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002284 xfs_alert(mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002286 str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 errs++;
2288 }
Christoph Hellwig1149d962005-11-02 15:01:12 +11002289 if (ddq->d_version != XFS_DQUOT_VERSION) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002291 xfs_alert(mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002293 str, id, ddq->d_version, XFS_DQUOT_VERSION);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 errs++;
2295 }
2296
Christoph Hellwig1149d962005-11-02 15:01:12 +11002297 if (ddq->d_flags != XFS_DQ_USER &&
2298 ddq->d_flags != XFS_DQ_PROJ &&
2299 ddq->d_flags != XFS_DQ_GROUP) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002301 xfs_alert(mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002303 str, id, ddq->d_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 errs++;
2305 }
2306
Christoph Hellwig1149d962005-11-02 15:01:12 +11002307 if (id != -1 && id != be32_to_cpu(ddq->d_id)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002309 xfs_alert(mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310 "%s : ondisk-dquot 0x%p, ID mismatch: "
2311 "0x%x expected, found id 0x%x",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002312 str, ddq, id, be32_to_cpu(ddq->d_id));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313 errs++;
2314 }
2315
2316 if (!errs && ddq->d_id) {
Christoph Hellwig1149d962005-11-02 15:01:12 +11002317 if (ddq->d_blk_softlimit &&
Mitsuo Hayasakad0a3fe62012-02-06 12:50:07 +00002318 be64_to_cpu(ddq->d_bcount) >
Christoph Hellwig1149d962005-11-02 15:01:12 +11002319 be64_to_cpu(ddq->d_blk_softlimit)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 if (!ddq->d_btimer) {
2321 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002322 xfs_alert(mp,
2323 "%s : Dquot ID 0x%x (0x%p) BLK TIMER NOT STARTED",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002324 str, (int)be32_to_cpu(ddq->d_id), ddq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325 errs++;
2326 }
2327 }
Christoph Hellwig1149d962005-11-02 15:01:12 +11002328 if (ddq->d_ino_softlimit &&
Mitsuo Hayasakad0a3fe62012-02-06 12:50:07 +00002329 be64_to_cpu(ddq->d_icount) >
Christoph Hellwig1149d962005-11-02 15:01:12 +11002330 be64_to_cpu(ddq->d_ino_softlimit)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331 if (!ddq->d_itimer) {
2332 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002333 xfs_alert(mp,
2334 "%s : Dquot ID 0x%x (0x%p) INODE TIMER NOT STARTED",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002335 str, (int)be32_to_cpu(ddq->d_id), ddq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 errs++;
2337 }
2338 }
Christoph Hellwig1149d962005-11-02 15:01:12 +11002339 if (ddq->d_rtb_softlimit &&
Mitsuo Hayasakad0a3fe62012-02-06 12:50:07 +00002340 be64_to_cpu(ddq->d_rtbcount) >
Christoph Hellwig1149d962005-11-02 15:01:12 +11002341 be64_to_cpu(ddq->d_rtb_softlimit)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342 if (!ddq->d_rtbtimer) {
2343 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002344 xfs_alert(mp,
2345 "%s : Dquot ID 0x%x (0x%p) RTBLK TIMER NOT STARTED",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002346 str, (int)be32_to_cpu(ddq->d_id), ddq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347 errs++;
2348 }
2349 }
2350 }
2351
2352 if (!errs || !(flags & XFS_QMOPT_DQREPAIR))
2353 return errs;
2354
2355 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002356 xfs_notice(mp, "Re-initializing dquot ID 0x%x", id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357
2358 /*
2359 * Typically, a repair is only requested by quotacheck.
2360 */
2361 ASSERT(id != -1);
2362 ASSERT(flags & XFS_QMOPT_DQREPAIR);
2363 memset(d, 0, sizeof(xfs_dqblk_t));
Christoph Hellwig1149d962005-11-02 15:01:12 +11002364
2365 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
2366 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
2367 d->dd_diskdq.d_flags = type;
2368 d->dd_diskdq.d_id = cpu_to_be32(id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369
Dave Chinner6fcdc592013-06-03 15:28:46 +10002370 if (xfs_sb_version_hascrc(&mp->m_sb)) {
2371 uuid_copy(&d->dd_uuid, &mp->m_sb.sb_uuid);
2372 xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
2373 XFS_DQUOT_CRC_OFF);
2374 }
2375
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376 return errs;
2377}
2378
2379/*
2380 * Perform a dquot buffer recovery.
Zhi Yong Wu8ba701e2013-08-12 03:15:01 +00002381 * Simple algorithm: if we have found a QUOTAOFF log item of the same type
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2383 * Else, treat it as a regular buffer and do recovery.
2384 */
2385STATIC void
2386xlog_recover_do_dquot_buffer(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002387 struct xfs_mount *mp,
2388 struct xlog *log,
2389 struct xlog_recover_item *item,
2390 struct xfs_buf *bp,
2391 struct xfs_buf_log_format *buf_f)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392{
2393 uint type;
2394
Dave Chinner9abbc532010-04-13 15:06:46 +10002395 trace_xfs_log_recover_buf_dquot_buf(log, buf_f);
2396
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397 /*
2398 * Filesystems are required to send in quota flags at mount time.
2399 */
2400 if (mp->m_qflags == 0) {
2401 return;
2402 }
2403
2404 type = 0;
Dave Chinnerc1155412010-05-07 11:05:19 +10002405 if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406 type |= XFS_DQ_USER;
Dave Chinnerc1155412010-05-07 11:05:19 +10002407 if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF)
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002408 type |= XFS_DQ_PROJ;
Dave Chinnerc1155412010-05-07 11:05:19 +10002409 if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410 type |= XFS_DQ_GROUP;
2411 /*
2412 * This type of quotas was turned off, so ignore this buffer
2413 */
2414 if (log->l_quotaoffs_flag & type)
2415 return;
2416
Dave Chinner9abbc532010-04-13 15:06:46 +10002417 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418}
2419
2420/*
2421 * This routine replays a modification made to a buffer at runtime.
2422 * There are actually two types of buffer, regular and inode, which
2423 * are handled differently. Inode buffers are handled differently
2424 * in that we only recover a specific set of data from them, namely
2425 * the inode di_next_unlinked fields. This is because all other inode
2426 * data is actually logged via inode records and any data we replay
2427 * here which overlaps that may be stale.
2428 *
2429 * When meta-data buffers are freed at run time we log a buffer item
Dave Chinnerc1155412010-05-07 11:05:19 +10002430 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431 * of the buffer in the log should not be replayed at recovery time.
2432 * This is so that if the blocks covered by the buffer are reused for
2433 * file data before we crash we don't end up replaying old, freed
2434 * meta-data into a user's file.
2435 *
2436 * To handle the cancellation of buffer log items, we make two passes
2437 * over the log during recovery. During the first we build a table of
2438 * those buffers which have been cancelled, and during the second we
2439 * only replay those buffers which do not have corresponding cancel
Zhi Yong Wu34be5ff2013-08-07 10:11:07 +00002440 * records in the table. See xlog_recover_buffer_pass[1,2] above
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441 * for more details on the implementation of the table of cancel records.
2442 */
2443STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002444xlog_recover_buffer_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002445 struct xlog *log,
2446 struct list_head *buffer_list,
2447 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448{
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002449 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002450 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451 xfs_buf_t *bp;
2452 int error;
Christoph Hellwig6ad112b2009-11-24 18:02:23 +00002453 uint buf_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002455 /*
2456 * In this pass we only want to recover all the buffers which have
2457 * not been cancelled and are not cancellation buffers themselves.
2458 */
2459 if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno,
2460 buf_f->blf_len, buf_f->blf_flags)) {
2461 trace_xfs_log_recover_buf_cancel(log, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463 }
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002464
Dave Chinner9abbc532010-04-13 15:06:46 +10002465 trace_xfs_log_recover_buf_recover(log, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466
Dave Chinnera8acad72012-04-23 15:58:54 +10002467 buf_flags = 0;
Dave Chinner611c9942012-04-23 15:59:07 +10002468 if (buf_f->blf_flags & XFS_BLF_INODE_BUF)
2469 buf_flags |= XBF_UNMAPPED;
Christoph Hellwig6ad112b2009-11-24 18:02:23 +00002470
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002471 bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len,
Dave Chinnerc3f8fc72012-11-12 22:54:01 +11002472 buf_flags, NULL);
Chandra Seetharamanac4d6882011-08-03 02:18:29 +00002473 if (!bp)
2474 return XFS_ERROR(ENOMEM);
Chandra Seetharamane5702802011-08-03 02:18:34 +00002475 error = bp->b_error;
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +00002476 if (error) {
Christoph Hellwig901796a2011-10-10 16:52:49 +00002477 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478 xfs_buf_relse(bp);
2479 return error;
2480 }
2481
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002482 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002484 } else if (buf_f->blf_flags &
Dave Chinnerc1155412010-05-07 11:05:19 +10002485 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002486 xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2487 } else {
Dave Chinner9abbc532010-04-13 15:06:46 +10002488 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489 }
2490 if (error)
2491 return XFS_ERROR(error);
2492
2493 /*
2494 * Perform delayed write on the buffer. Asynchronous writes will be
2495 * slower when taking into account all the buffers to be flushed.
2496 *
2497 * Also make sure that only inode buffers with good sizes stay in
2498 * the buffer cache. The kernel moves inodes in buffers of 1 block
2499 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2500 * buffers in the log can be a different size if the log was generated
2501 * by an older kernel using unclustered inode buffers or a newer kernel
2502 * running with a different inode cluster size. Regardless, if the
2503 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2504 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2505 * the buffer out of the buffer cache so that the buffer won't
2506 * overlap with future reads of those inodes.
2507 */
2508 if (XFS_DINODE_MAGIC ==
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002509 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
Dave Chinneraa0e8832012-04-23 15:58:52 +10002510 (BBTOB(bp->b_io_length) != MAX(log->l_mp->m_sb.sb_blocksize,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511 (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) {
Christoph Hellwigc867cb62011-10-10 16:52:46 +00002512 xfs_buf_stale(bp);
Christoph Hellwigc2b006c2011-08-23 08:28:07 +00002513 error = xfs_bwrite(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514 } else {
Dave Chinnerebad8612010-09-22 10:47:20 +10002515 ASSERT(bp->b_target->bt_mount == mp);
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02002516 bp->b_iodone = xlog_recover_iodone;
Christoph Hellwig43ff2122012-04-23 15:58:39 +10002517 xfs_buf_delwri_queue(bp, buffer_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518 }
2519
Christoph Hellwigc2b006c2011-08-23 08:28:07 +00002520 xfs_buf_relse(bp);
2521 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522}
2523
2524STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002525xlog_recover_inode_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002526 struct xlog *log,
2527 struct list_head *buffer_list,
2528 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529{
2530 xfs_inode_log_format_t *in_f;
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002531 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532 xfs_buf_t *bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533 xfs_dinode_t *dip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534 int len;
2535 xfs_caddr_t src;
2536 xfs_caddr_t dest;
2537 int error;
2538 int attr_index;
2539 uint fields;
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002540 xfs_icdinode_t *dicp;
Christoph Hellwig93848a92013-04-03 16:11:17 +11002541 uint isize;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002542 int need_free = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543
Tim Shimmin6d192a92006-06-09 14:55:38 +10002544 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002545 in_f = item->ri_buf[0].i_addr;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002546 } else {
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002547 in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002548 need_free = 1;
2549 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2550 if (error)
2551 goto error;
2552 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553
2554 /*
2555 * Inode buffers can be freed, look out for it,
2556 * and do not replay the inode.
2557 */
Christoph Hellwiga1941892008-11-28 14:23:40 +11002558 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
2559 in_f->ilf_len, 0)) {
Tim Shimmin6d192a92006-06-09 14:55:38 +10002560 error = 0;
Dave Chinner9abbc532010-04-13 15:06:46 +10002561 trace_xfs_log_recover_inode_cancel(log, in_f);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002562 goto error;
2563 }
Dave Chinner9abbc532010-04-13 15:06:46 +10002564 trace_xfs_log_recover_inode_recover(log, in_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565
Dave Chinnerc3f8fc72012-11-12 22:54:01 +11002566 bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len, 0,
Christoph Hellwig93848a92013-04-03 16:11:17 +11002567 &xfs_inode_buf_ops);
Chandra Seetharamanac4d6882011-08-03 02:18:29 +00002568 if (!bp) {
2569 error = ENOMEM;
2570 goto error;
2571 }
Chandra Seetharamane5702802011-08-03 02:18:34 +00002572 error = bp->b_error;
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +00002573 if (error) {
Christoph Hellwig901796a2011-10-10 16:52:49 +00002574 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575 xfs_buf_relse(bp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002576 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
Christoph Hellwiga1941892008-11-28 14:23:40 +11002579 dip = (xfs_dinode_t *)xfs_buf_offset(bp, in_f->ilf_boffset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580
2581 /*
2582 * Make sure the place we're flushing out to really looks
2583 * like an inode!
2584 */
Christoph Hellwig69ef9212011-07-08 14:36:05 +02002585 if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586 xfs_buf_relse(bp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002587 xfs_alert(mp,
2588 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
2589 __func__, dip, bp, in_f->ilf_ino);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002590 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591 XFS_ERRLEVEL_LOW, mp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002592 error = EFSCORRUPTED;
2593 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594 }
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002595 dicp = item->ri_buf[1].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
2597 xfs_buf_relse(bp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002598 xfs_alert(mp,
2599 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
2600 __func__, item, in_f->ilf_ino);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002601 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602 XFS_ERRLEVEL_LOW, mp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002603 error = EFSCORRUPTED;
2604 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605 }
2606
Dave Chinnere60896d2013-07-24 15:47:30 +10002607 /*
2608 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
2609 * are transactional and if ordering is necessary we can determine that
2610 * more accurately by the LSN field in the V3 inode core. Don't trust
2611 * the inode versions we might be changing them here - use the
2612 * superblock flag to determine whether we need to look at di_flushiter
2613 * to skip replay when the on disk inode is newer than the log one
2614 */
2615 if (!xfs_sb_version_hascrc(&mp->m_sb) &&
2616 dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617 /*
2618 * Deal with the wrap case, DI_MAX_FLUSH is less
2619 * than smaller numbers
2620 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002621 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002622 dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 /* do nothing */
2624 } else {
2625 xfs_buf_relse(bp);
Dave Chinner9abbc532010-04-13 15:06:46 +10002626 trace_xfs_log_recover_inode_skip(log, in_f);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002627 error = 0;
2628 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629 }
2630 }
Dave Chinnere60896d2013-07-24 15:47:30 +10002631
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632 /* Take the opportunity to reset the flush iteration count */
2633 dicp->di_flushiter = 0;
2634
Al Viroabbede12011-07-26 02:31:30 -04002635 if (unlikely(S_ISREG(dicp->di_mode))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2637 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002638 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639 XFS_ERRLEVEL_LOW, mp, dicp);
2640 xfs_buf_relse(bp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002641 xfs_alert(mp,
2642 "%s: Bad regular inode log record, rec ptr 0x%p, "
2643 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2644 __func__, item, dip, bp, in_f->ilf_ino);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002645 error = EFSCORRUPTED;
2646 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647 }
Al Viroabbede12011-07-26 02:31:30 -04002648 } else if (unlikely(S_ISDIR(dicp->di_mode))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2650 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2651 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002652 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653 XFS_ERRLEVEL_LOW, mp, dicp);
2654 xfs_buf_relse(bp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002655 xfs_alert(mp,
2656 "%s: Bad dir inode log record, rec ptr 0x%p, "
2657 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2658 __func__, item, dip, bp, in_f->ilf_ino);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002659 error = EFSCORRUPTED;
2660 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661 }
2662 }
2663 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002664 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665 XFS_ERRLEVEL_LOW, mp, dicp);
2666 xfs_buf_relse(bp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002667 xfs_alert(mp,
2668 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2669 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2670 __func__, item, dip, bp, in_f->ilf_ino,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671 dicp->di_nextents + dicp->di_anextents,
2672 dicp->di_nblocks);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002673 error = EFSCORRUPTED;
2674 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675 }
2676 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002677 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678 XFS_ERRLEVEL_LOW, mp, dicp);
2679 xfs_buf_relse(bp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002680 xfs_alert(mp,
2681 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2682 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__,
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002683 item, dip, bp, in_f->ilf_ino, dicp->di_forkoff);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002684 error = EFSCORRUPTED;
2685 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686 }
Christoph Hellwig93848a92013-04-03 16:11:17 +11002687 isize = xfs_icdinode_size(dicp->di_version);
2688 if (unlikely(item->ri_buf[1].i_len > isize)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002689 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690 XFS_ERRLEVEL_LOW, mp, dicp);
2691 xfs_buf_relse(bp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002692 xfs_alert(mp,
2693 "%s: Bad inode log record length %d, rec ptr 0x%p",
2694 __func__, item->ri_buf[1].i_len, item);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002695 error = EFSCORRUPTED;
2696 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697 }
2698
2699 /* The core is in in-core format */
Christoph Hellwig93848a92013-04-03 16:11:17 +11002700 xfs_dinode_to_disk(dip, dicp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002701
2702 /* the rest is in on-disk format */
Christoph Hellwig93848a92013-04-03 16:11:17 +11002703 if (item->ri_buf[1].i_len > isize) {
2704 memcpy((char *)dip + isize,
2705 item->ri_buf[1].i_addr + isize,
2706 item->ri_buf[1].i_len - isize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707 }
2708
2709 fields = in_f->ilf_fields;
2710 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2711 case XFS_ILOG_DEV:
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002712 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713 break;
2714 case XFS_ILOG_UUID:
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002715 memcpy(XFS_DFORK_DPTR(dip),
2716 &in_f->ilf_u.ilfu_uuid,
2717 sizeof(uuid_t));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718 break;
2719 }
2720
2721 if (in_f->ilf_size == 2)
2722 goto write_inode_buffer;
2723 len = item->ri_buf[2].i_len;
2724 src = item->ri_buf[2].i_addr;
2725 ASSERT(in_f->ilf_size <= 4);
2726 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2727 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2728 (len == in_f->ilf_dsize));
2729
2730 switch (fields & XFS_ILOG_DFORK) {
2731 case XFS_ILOG_DDATA:
2732 case XFS_ILOG_DEXT:
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002733 memcpy(XFS_DFORK_DPTR(dip), src, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002734 break;
2735
2736 case XFS_ILOG_DBROOT:
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002737 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002738 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002739 XFS_DFORK_DSIZE(dip, mp));
2740 break;
2741
2742 default:
2743 /*
2744 * There are no data fork flags set.
2745 */
2746 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2747 break;
2748 }
2749
2750 /*
2751 * If we logged any attribute data, recover it. There may or
2752 * may not have been any other non-core data logged in this
2753 * transaction.
2754 */
2755 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2756 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2757 attr_index = 3;
2758 } else {
2759 attr_index = 2;
2760 }
2761 len = item->ri_buf[attr_index].i_len;
2762 src = item->ri_buf[attr_index].i_addr;
2763 ASSERT(len == in_f->ilf_asize);
2764
2765 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2766 case XFS_ILOG_ADATA:
2767 case XFS_ILOG_AEXT:
2768 dest = XFS_DFORK_APTR(dip);
2769 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2770 memcpy(dest, src, len);
2771 break;
2772
2773 case XFS_ILOG_ABROOT:
2774 dest = XFS_DFORK_APTR(dip);
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002775 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
2776 len, (xfs_bmdr_block_t*)dest,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777 XFS_DFORK_ASIZE(dip, mp));
2778 break;
2779
2780 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002781 xfs_warn(log->l_mp, "%s: Invalid flag", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782 ASSERT(0);
2783 xfs_buf_relse(bp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002784 error = EIO;
2785 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786 }
2787 }
2788
2789write_inode_buffer:
Christoph Hellwig93848a92013-04-03 16:11:17 +11002790 /* re-generate the checksum. */
2791 xfs_dinode_calc_crc(log->l_mp, dip);
2792
Dave Chinnerebad8612010-09-22 10:47:20 +10002793 ASSERT(bp->b_target->bt_mount == mp);
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02002794 bp->b_iodone = xlog_recover_iodone;
Christoph Hellwig43ff2122012-04-23 15:58:39 +10002795 xfs_buf_delwri_queue(bp, buffer_list);
Christoph Hellwig61551f12011-08-23 08:28:06 +00002796 xfs_buf_relse(bp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002797error:
2798 if (need_free)
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002799 kmem_free(in_f);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002800 return XFS_ERROR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002801}
2802
2803/*
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002804 * Recover QUOTAOFF records. We simply make a note of it in the xlog
Linus Torvalds1da177e2005-04-16 15:20:36 -07002805 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2806 * of that type.
2807 */
2808STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002809xlog_recover_quotaoff_pass1(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002810 struct xlog *log,
2811 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002813 xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814 ASSERT(qoff_f);
2815
2816 /*
2817 * The logitem format's flag tells us if this was user quotaoff,
Nathan Scott77a7cce2006-01-11 15:35:57 +11002818 * group/project quotaoff or both.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002819 */
2820 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2821 log->l_quotaoffs_flag |= XFS_DQ_USER;
Nathan Scott77a7cce2006-01-11 15:35:57 +11002822 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2823 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002824 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2825 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2826
2827 return (0);
2828}
2829
2830/*
2831 * Recover a dquot record
2832 */
2833STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002834xlog_recover_dquot_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002835 struct xlog *log,
2836 struct list_head *buffer_list,
2837 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002839 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002840 xfs_buf_t *bp;
2841 struct xfs_disk_dquot *ddq, *recddq;
2842 int error;
2843 xfs_dq_logformat_t *dq_f;
2844 uint type;
2845
Linus Torvalds1da177e2005-04-16 15:20:36 -07002846
2847 /*
2848 * Filesystems are required to send in quota flags at mount time.
2849 */
2850 if (mp->m_qflags == 0)
2851 return (0);
2852
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002853 recddq = item->ri_buf[1].i_addr;
2854 if (recddq == NULL) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002855 xfs_alert(log->l_mp, "NULL dquot in %s.", __func__);
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002856 return XFS_ERROR(EIO);
2857 }
Jan Rekorajski8ec6dba2009-11-16 11:57:02 +00002858 if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002859 xfs_alert(log->l_mp, "dquot too small (%d) in %s.",
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002860 item->ri_buf[1].i_len, __func__);
2861 return XFS_ERROR(EIO);
2862 }
2863
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864 /*
2865 * This type of quotas was turned off, so ignore this record.
2866 */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002867 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868 ASSERT(type);
2869 if (log->l_quotaoffs_flag & type)
2870 return (0);
2871
2872 /*
2873 * At this point we know that quota was _not_ turned off.
2874 * Since the mount flags are not indicating to us otherwise, this
2875 * must mean that quota is on, and the dquot needs to be replayed.
2876 * Remember that we may not have fully recovered the superblock yet,
2877 * so we can't do the usual trick of looking at the SB quota bits.
2878 *
2879 * The other possibility, of course, is that the quota subsystem was
2880 * removed since the last mount - ENOSYS.
2881 */
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002882 dq_f = item->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883 ASSERT(dq_f);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002884 error = xfs_qm_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2885 "xlog_recover_dquot_pass2 (log copy)");
2886 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002887 return XFS_ERROR(EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002888 ASSERT(dq_f->qlf_len == 1);
2889
Dave Chinner7ca790a2012-04-23 15:58:55 +10002890 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dq_f->qlf_blkno,
Dave Chinnerc3f8fc72012-11-12 22:54:01 +11002891 XFS_FSB_TO_BB(mp, dq_f->qlf_len), 0, &bp,
2892 NULL);
Dave Chinner7ca790a2012-04-23 15:58:55 +10002893 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002894 return error;
Dave Chinner7ca790a2012-04-23 15:58:55 +10002895
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896 ASSERT(bp);
2897 ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
2898
2899 /*
2900 * At least the magic num portion should be on disk because this
2901 * was among a chunk of dquots created earlier, and we did some
2902 * minimal initialization then.
2903 */
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002904 error = xfs_qm_dqcheck(mp, ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2905 "xlog_recover_dquot_pass2");
2906 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907 xfs_buf_relse(bp);
2908 return XFS_ERROR(EIO);
2909 }
2910
2911 memcpy(ddq, recddq, item->ri_buf[1].i_len);
Dave Chinner6fcdc592013-06-03 15:28:46 +10002912 if (xfs_sb_version_hascrc(&mp->m_sb)) {
2913 xfs_update_cksum((char *)ddq, sizeof(struct xfs_dqblk),
2914 XFS_DQUOT_CRC_OFF);
2915 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002916
2917 ASSERT(dq_f->qlf_size == 2);
Dave Chinnerebad8612010-09-22 10:47:20 +10002918 ASSERT(bp->b_target->bt_mount == mp);
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02002919 bp->b_iodone = xlog_recover_iodone;
Christoph Hellwig43ff2122012-04-23 15:58:39 +10002920 xfs_buf_delwri_queue(bp, buffer_list);
Christoph Hellwig61551f12011-08-23 08:28:06 +00002921 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002922
2923 return (0);
2924}
2925
2926/*
2927 * This routine is called to create an in-core extent free intent
2928 * item from the efi format structure which was logged on disk.
2929 * It allocates an in-core efi, copies the extents from the format
2930 * structure into it, and adds the efi to the AIL with the given
2931 * LSN.
2932 */
Tim Shimmin6d192a92006-06-09 14:55:38 +10002933STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002934xlog_recover_efi_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002935 struct xlog *log,
2936 struct xlog_recover_item *item,
2937 xfs_lsn_t lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002938{
Tim Shimmin6d192a92006-06-09 14:55:38 +10002939 int error;
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002940 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002941 xfs_efi_log_item_t *efip;
2942 xfs_efi_log_format_t *efi_formatp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002943
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002944 efi_formatp = item->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002945
Linus Torvalds1da177e2005-04-16 15:20:36 -07002946 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002947 if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
2948 &(efip->efi_format)))) {
2949 xfs_efi_item_free(efip);
2950 return error;
2951 }
Dave Chinnerb199c8a2010-12-20 11:59:49 +11002952 atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002953
David Chinnera9c21c12008-10-30 17:39:35 +11002954 spin_lock(&log->l_ailp->xa_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002955 /*
David Chinner783a2f62008-10-30 17:39:58 +11002956 * xfs_trans_ail_update() drops the AIL lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002957 */
Dave Chinnere6059942010-12-20 12:34:26 +11002958 xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002959 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002960}
2961
2962
2963/*
2964 * This routine is called when an efd format structure is found in
2965 * a committed transaction in the log. It's purpose is to cancel
2966 * the corresponding efi if it was still in the log. To do this
2967 * it searches the AIL for the efi with an id equal to that in the
2968 * efd format structure. If we find it, we remove the efi from the
2969 * AIL and free it.
2970 */
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002971STATIC int
2972xlog_recover_efd_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002973 struct xlog *log,
2974 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002976 xfs_efd_log_format_t *efd_formatp;
2977 xfs_efi_log_item_t *efip = NULL;
2978 xfs_log_item_t *lip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002979 __uint64_t efi_id;
David Chinner27d8d5f2008-10-30 17:38:39 +11002980 struct xfs_ail_cursor cur;
David Chinner783a2f62008-10-30 17:39:58 +11002981 struct xfs_ail *ailp = log->l_ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002982
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002983 efd_formatp = item->ri_buf[0].i_addr;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002984 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
2985 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
2986 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
2987 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002988 efi_id = efd_formatp->efd_efi_id;
2989
2990 /*
2991 * Search for the efi with the id in the efd format structure
2992 * in the AIL.
2993 */
David Chinnera9c21c12008-10-30 17:39:35 +11002994 spin_lock(&ailp->xa_lock);
2995 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002996 while (lip != NULL) {
2997 if (lip->li_type == XFS_LI_EFI) {
2998 efip = (xfs_efi_log_item_t *)lip;
2999 if (efip->efi_format.efi_id == efi_id) {
3000 /*
David Chinner783a2f62008-10-30 17:39:58 +11003001 * xfs_trans_ail_delete() drops the
Linus Torvalds1da177e2005-04-16 15:20:36 -07003002 * AIL lock.
3003 */
Dave Chinner04913fd2012-04-23 15:58:41 +10003004 xfs_trans_ail_delete(ailp, lip,
3005 SHUTDOWN_CORRUPT_INCORE);
David Chinner8ae2c0f2007-11-23 16:28:17 +11003006 xfs_efi_item_free(efip);
David Chinnera9c21c12008-10-30 17:39:35 +11003007 spin_lock(&ailp->xa_lock);
David Chinner27d8d5f2008-10-30 17:38:39 +11003008 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003009 }
3010 }
David Chinnera9c21c12008-10-30 17:39:35 +11003011 lip = xfs_trans_ail_cursor_next(ailp, &cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003012 }
David Chinnera9c21c12008-10-30 17:39:35 +11003013 xfs_trans_ail_cursor_done(ailp, &cur);
3014 spin_unlock(&ailp->xa_lock);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003015
3016 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017}
3018
3019/*
Dave Chinner28c8e412013-06-27 16:04:55 +10003020 * This routine is called when an inode create format structure is found in a
3021 * committed transaction in the log. It's purpose is to initialise the inodes
3022 * being allocated on disk. This requires us to get inode cluster buffers that
3023 * match the range to be intialised, stamped with inode templates and written
3024 * by delayed write so that subsequent modifications will hit the cached buffer
3025 * and only need writing out at the end of recovery.
3026 */
3027STATIC int
3028xlog_recover_do_icreate_pass2(
3029 struct xlog *log,
3030 struct list_head *buffer_list,
3031 xlog_recover_item_t *item)
3032{
3033 struct xfs_mount *mp = log->l_mp;
3034 struct xfs_icreate_log *icl;
3035 xfs_agnumber_t agno;
3036 xfs_agblock_t agbno;
3037 unsigned int count;
3038 unsigned int isize;
3039 xfs_agblock_t length;
3040
3041 icl = (struct xfs_icreate_log *)item->ri_buf[0].i_addr;
3042 if (icl->icl_type != XFS_LI_ICREATE) {
3043 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad type");
3044 return EINVAL;
3045 }
3046
3047 if (icl->icl_size != 1) {
3048 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad icl size");
3049 return EINVAL;
3050 }
3051
3052 agno = be32_to_cpu(icl->icl_ag);
3053 if (agno >= mp->m_sb.sb_agcount) {
3054 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agno");
3055 return EINVAL;
3056 }
3057 agbno = be32_to_cpu(icl->icl_agbno);
3058 if (!agbno || agbno == NULLAGBLOCK || agbno >= mp->m_sb.sb_agblocks) {
3059 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agbno");
3060 return EINVAL;
3061 }
3062 isize = be32_to_cpu(icl->icl_isize);
3063 if (isize != mp->m_sb.sb_inodesize) {
3064 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad isize");
3065 return EINVAL;
3066 }
3067 count = be32_to_cpu(icl->icl_count);
3068 if (!count) {
3069 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count");
3070 return EINVAL;
3071 }
3072 length = be32_to_cpu(icl->icl_length);
3073 if (!length || length >= mp->m_sb.sb_agblocks) {
3074 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad length");
3075 return EINVAL;
3076 }
3077
3078 /* existing allocation is fixed value */
3079 ASSERT(count == XFS_IALLOC_INODES(mp));
3080 ASSERT(length == XFS_IALLOC_BLOCKS(mp));
3081 if (count != XFS_IALLOC_INODES(mp) ||
3082 length != XFS_IALLOC_BLOCKS(mp)) {
3083 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count 2");
3084 return EINVAL;
3085 }
3086
3087 /*
3088 * Inode buffers can be freed. Do not replay the inode initialisation as
3089 * we could be overwriting something written after this inode buffer was
3090 * cancelled.
3091 *
3092 * XXX: we need to iterate all buffers and only init those that are not
3093 * cancelled. I think that a more fine grained factoring of
3094 * xfs_ialloc_inode_init may be appropriate here to enable this to be
3095 * done easily.
3096 */
3097 if (xlog_check_buffer_cancelled(log,
3098 XFS_AGB_TO_DADDR(mp, agno, agbno), length, 0))
3099 return 0;
3100
3101 xfs_ialloc_inode_init(mp, NULL, buffer_list, agno, agbno, length,
3102 be32_to_cpu(icl->icl_gen));
3103 return 0;
3104}
3105
3106/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003107 * Free up any resources allocated by the transaction
3108 *
3109 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
3110 */
3111STATIC void
3112xlog_recover_free_trans(
Christoph Hellwigd0450942010-12-01 22:06:23 +00003113 struct xlog_recover *trans)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003114{
Dave Chinnerf0a76952010-01-11 11:49:57 +00003115 xlog_recover_item_t *item, *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003116 int i;
3117
Dave Chinnerf0a76952010-01-11 11:49:57 +00003118 list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
3119 /* Free the regions in the item. */
3120 list_del(&item->ri_list);
3121 for (i = 0; i < item->ri_cnt; i++)
3122 kmem_free(item->ri_buf[i].i_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003123 /* Free the item itself */
Dave Chinnerf0a76952010-01-11 11:49:57 +00003124 kmem_free(item->ri_buf);
3125 kmem_free(item);
3126 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003127 /* Free the transaction recover structure */
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003128 kmem_free(trans);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003129}
3130
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003131STATIC void
3132xlog_recover_buffer_ra_pass2(
3133 struct xlog *log,
3134 struct xlog_recover_item *item)
3135{
3136 struct xfs_buf_log_format *buf_f = item->ri_buf[0].i_addr;
3137 struct xfs_mount *mp = log->l_mp;
3138
Dave Chinner84a5b732013-08-27 08:10:53 +10003139 if (xlog_peek_buffer_cancelled(log, buf_f->blf_blkno,
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003140 buf_f->blf_len, buf_f->blf_flags)) {
3141 return;
3142 }
3143
3144 xfs_buf_readahead(mp->m_ddev_targp, buf_f->blf_blkno,
3145 buf_f->blf_len, NULL);
3146}
3147
3148STATIC void
3149xlog_recover_inode_ra_pass2(
3150 struct xlog *log,
3151 struct xlog_recover_item *item)
3152{
3153 struct xfs_inode_log_format ilf_buf;
3154 struct xfs_inode_log_format *ilfp;
3155 struct xfs_mount *mp = log->l_mp;
3156 int error;
3157
3158 if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) {
3159 ilfp = item->ri_buf[0].i_addr;
3160 } else {
3161 ilfp = &ilf_buf;
3162 memset(ilfp, 0, sizeof(*ilfp));
3163 error = xfs_inode_item_format_convert(&item->ri_buf[0], ilfp);
3164 if (error)
3165 return;
3166 }
3167
Dave Chinner84a5b732013-08-27 08:10:53 +10003168 if (xlog_peek_buffer_cancelled(log, ilfp->ilf_blkno, ilfp->ilf_len, 0))
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003169 return;
3170
3171 xfs_buf_readahead(mp->m_ddev_targp, ilfp->ilf_blkno,
3172 ilfp->ilf_len, &xfs_inode_buf_ops);
3173}
3174
3175STATIC void
3176xlog_recover_dquot_ra_pass2(
3177 struct xlog *log,
3178 struct xlog_recover_item *item)
3179{
3180 struct xfs_mount *mp = log->l_mp;
3181 struct xfs_disk_dquot *recddq;
3182 struct xfs_dq_logformat *dq_f;
3183 uint type;
3184
3185
3186 if (mp->m_qflags == 0)
3187 return;
3188
3189 recddq = item->ri_buf[1].i_addr;
3190 if (recddq == NULL)
3191 return;
3192 if (item->ri_buf[1].i_len < sizeof(struct xfs_disk_dquot))
3193 return;
3194
3195 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
3196 ASSERT(type);
3197 if (log->l_quotaoffs_flag & type)
3198 return;
3199
3200 dq_f = item->ri_buf[0].i_addr;
3201 ASSERT(dq_f);
3202 ASSERT(dq_f->qlf_len == 1);
3203
3204 xfs_buf_readahead(mp->m_ddev_targp, dq_f->qlf_blkno,
3205 dq_f->qlf_len, NULL);
3206}
3207
3208STATIC void
3209xlog_recover_ra_pass2(
3210 struct xlog *log,
3211 struct xlog_recover_item *item)
3212{
3213 switch (ITEM_TYPE(item)) {
3214 case XFS_LI_BUF:
3215 xlog_recover_buffer_ra_pass2(log, item);
3216 break;
3217 case XFS_LI_INODE:
3218 xlog_recover_inode_ra_pass2(log, item);
3219 break;
3220 case XFS_LI_DQUOT:
3221 xlog_recover_dquot_ra_pass2(log, item);
3222 break;
3223 case XFS_LI_EFI:
3224 case XFS_LI_EFD:
3225 case XFS_LI_QUOTAOFF:
3226 default:
3227 break;
3228 }
3229}
3230
Linus Torvalds1da177e2005-04-16 15:20:36 -07003231STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003232xlog_recover_commit_pass1(
Mark Tinguelyad223e62012-06-14 09:22:15 -05003233 struct xlog *log,
3234 struct xlog_recover *trans,
3235 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003236{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003237 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1);
Christoph Hellwigd0450942010-12-01 22:06:23 +00003238
3239 switch (ITEM_TYPE(item)) {
3240 case XFS_LI_BUF:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003241 return xlog_recover_buffer_pass1(log, item);
Christoph Hellwigd0450942010-12-01 22:06:23 +00003242 case XFS_LI_QUOTAOFF:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003243 return xlog_recover_quotaoff_pass1(log, item);
3244 case XFS_LI_INODE:
3245 case XFS_LI_EFI:
3246 case XFS_LI_EFD:
3247 case XFS_LI_DQUOT:
Dave Chinner28c8e412013-06-27 16:04:55 +10003248 case XFS_LI_ICREATE:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003249 /* nothing to do in pass 1 */
3250 return 0;
Christoph Hellwigd0450942010-12-01 22:06:23 +00003251 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003252 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
3253 __func__, ITEM_TYPE(item));
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003254 ASSERT(0);
3255 return XFS_ERROR(EIO);
3256 }
3257}
3258
3259STATIC int
3260xlog_recover_commit_pass2(
Mark Tinguelyad223e62012-06-14 09:22:15 -05003261 struct xlog *log,
3262 struct xlog_recover *trans,
3263 struct list_head *buffer_list,
3264 struct xlog_recover_item *item)
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003265{
3266 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2);
3267
3268 switch (ITEM_TYPE(item)) {
3269 case XFS_LI_BUF:
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003270 return xlog_recover_buffer_pass2(log, buffer_list, item);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003271 case XFS_LI_INODE:
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003272 return xlog_recover_inode_pass2(log, buffer_list, item);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003273 case XFS_LI_EFI:
3274 return xlog_recover_efi_pass2(log, item, trans->r_lsn);
3275 case XFS_LI_EFD:
3276 return xlog_recover_efd_pass2(log, item);
3277 case XFS_LI_DQUOT:
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003278 return xlog_recover_dquot_pass2(log, buffer_list, item);
Dave Chinner28c8e412013-06-27 16:04:55 +10003279 case XFS_LI_ICREATE:
3280 return xlog_recover_do_icreate_pass2(log, buffer_list, item);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003281 case XFS_LI_QUOTAOFF:
3282 /* nothing to do in pass2 */
3283 return 0;
3284 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003285 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
3286 __func__, ITEM_TYPE(item));
Christoph Hellwigd0450942010-12-01 22:06:23 +00003287 ASSERT(0);
3288 return XFS_ERROR(EIO);
3289 }
3290}
3291
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003292STATIC int
3293xlog_recover_items_pass2(
3294 struct xlog *log,
3295 struct xlog_recover *trans,
3296 struct list_head *buffer_list,
3297 struct list_head *item_list)
3298{
3299 struct xlog_recover_item *item;
3300 int error = 0;
3301
3302 list_for_each_entry(item, item_list, ri_list) {
3303 error = xlog_recover_commit_pass2(log, trans,
3304 buffer_list, item);
3305 if (error)
3306 return error;
3307 }
3308
3309 return error;
3310}
3311
Christoph Hellwigd0450942010-12-01 22:06:23 +00003312/*
3313 * Perform the transaction.
3314 *
3315 * If the transaction modifies a buffer or inode, do it now. Otherwise,
3316 * EFIs and EFDs get queued up by adding entries into the AIL for them.
3317 */
3318STATIC int
3319xlog_recover_commit_trans(
Mark Tinguelyad223e62012-06-14 09:22:15 -05003320 struct xlog *log,
Christoph Hellwigd0450942010-12-01 22:06:23 +00003321 struct xlog_recover *trans,
3322 int pass)
3323{
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003324 int error = 0;
3325 int error2;
3326 int items_queued = 0;
3327 struct xlog_recover_item *item;
3328 struct xlog_recover_item *next;
3329 LIST_HEAD (buffer_list);
3330 LIST_HEAD (ra_list);
3331 LIST_HEAD (done_list);
3332
3333 #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100
Linus Torvalds1da177e2005-04-16 15:20:36 -07003334
Dave Chinnerf0a76952010-01-11 11:49:57 +00003335 hlist_del(&trans->r_list);
Christoph Hellwigd0450942010-12-01 22:06:23 +00003336
3337 error = xlog_recover_reorder_trans(log, trans, pass);
3338 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003339 return error;
Christoph Hellwigd0450942010-12-01 22:06:23 +00003340
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003341 list_for_each_entry_safe(item, next, &trans->r_itemq, ri_list) {
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003342 switch (pass) {
3343 case XLOG_RECOVER_PASS1:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003344 error = xlog_recover_commit_pass1(log, trans, item);
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003345 break;
3346 case XLOG_RECOVER_PASS2:
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003347 xlog_recover_ra_pass2(log, item);
3348 list_move_tail(&item->ri_list, &ra_list);
3349 items_queued++;
3350 if (items_queued >= XLOG_RECOVER_COMMIT_QUEUE_MAX) {
3351 error = xlog_recover_items_pass2(log, trans,
3352 &buffer_list, &ra_list);
3353 list_splice_tail_init(&ra_list, &done_list);
3354 items_queued = 0;
3355 }
3356
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003357 break;
3358 default:
3359 ASSERT(0);
3360 }
3361
Christoph Hellwigd0450942010-12-01 22:06:23 +00003362 if (error)
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003363 goto out;
Christoph Hellwigd0450942010-12-01 22:06:23 +00003364 }
3365
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003366out:
3367 if (!list_empty(&ra_list)) {
3368 if (!error)
3369 error = xlog_recover_items_pass2(log, trans,
3370 &buffer_list, &ra_list);
3371 list_splice_tail_init(&ra_list, &done_list);
3372 }
3373
3374 if (!list_empty(&done_list))
3375 list_splice_init(&done_list, &trans->r_itemq);
3376
Christoph Hellwigd0450942010-12-01 22:06:23 +00003377 xlog_recover_free_trans(trans);
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003378
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003379 error2 = xfs_buf_delwri_submit(&buffer_list);
3380 return error ? error : error2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003381}
3382
3383STATIC int
3384xlog_recover_unmount_trans(
Mark Tinguelyad223e62012-06-14 09:22:15 -05003385 struct xlog *log,
3386 struct xlog_recover *trans)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003387{
3388 /* Do nothing now */
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003389 xfs_warn(log->l_mp, "%s: Unmount LR", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003390 return 0;
3391}
3392
3393/*
3394 * There are two valid states of the r_state field. 0 indicates that the
3395 * transaction structure is in a normal state. We have either seen the
3396 * start of the transaction or the last operation we added was not a partial
3397 * operation. If the last operation we added to the transaction was a
3398 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
3399 *
3400 * NOTE: skip LRs with 0 data length.
3401 */
3402STATIC int
3403xlog_recover_process_data(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003404 struct xlog *log,
Dave Chinnerf0a76952010-01-11 11:49:57 +00003405 struct hlist_head rhash[],
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003406 struct xlog_rec_header *rhead,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003407 xfs_caddr_t dp,
3408 int pass)
3409{
3410 xfs_caddr_t lp;
3411 int num_logops;
3412 xlog_op_header_t *ohead;
3413 xlog_recover_t *trans;
3414 xlog_tid_t tid;
3415 int error;
3416 unsigned long hash;
3417 uint flags;
3418
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003419 lp = dp + be32_to_cpu(rhead->h_len);
3420 num_logops = be32_to_cpu(rhead->h_num_logops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003421
3422 /* check the log format matches our own - else we can't recover */
3423 if (xlog_header_check_recover(log->l_mp, rhead))
3424 return (XFS_ERROR(EIO));
3425
3426 while ((dp < lp) && num_logops) {
3427 ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
3428 ohead = (xlog_op_header_t *)dp;
3429 dp += sizeof(xlog_op_header_t);
3430 if (ohead->oh_clientid != XFS_TRANSACTION &&
3431 ohead->oh_clientid != XFS_LOG) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003432 xfs_warn(log->l_mp, "%s: bad clientid 0x%x",
3433 __func__, ohead->oh_clientid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003434 ASSERT(0);
3435 return (XFS_ERROR(EIO));
3436 }
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10003437 tid = be32_to_cpu(ohead->oh_tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003438 hash = XLOG_RHASH(tid);
Dave Chinnerf0a76952010-01-11 11:49:57 +00003439 trans = xlog_recover_find_tid(&rhash[hash], tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003440 if (trans == NULL) { /* not found; add new tid */
3441 if (ohead->oh_flags & XLOG_START_TRANS)
3442 xlog_recover_new_tid(&rhash[hash], tid,
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003443 be64_to_cpu(rhead->h_lsn));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003444 } else {
Lachlan McIlroy9742bb92008-01-10 16:43:36 +11003445 if (dp + be32_to_cpu(ohead->oh_len) > lp) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003446 xfs_warn(log->l_mp, "%s: bad length 0x%x",
3447 __func__, be32_to_cpu(ohead->oh_len));
Lachlan McIlroy9742bb92008-01-10 16:43:36 +11003448 WARN_ON(1);
3449 return (XFS_ERROR(EIO));
3450 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003451 flags = ohead->oh_flags & ~XLOG_END_TRANS;
3452 if (flags & XLOG_WAS_CONT_TRANS)
3453 flags &= ~XLOG_CONTINUE_TRANS;
3454 switch (flags) {
3455 case XLOG_COMMIT_TRANS:
3456 error = xlog_recover_commit_trans(log,
Dave Chinnerf0a76952010-01-11 11:49:57 +00003457 trans, pass);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003458 break;
3459 case XLOG_UNMOUNT_TRANS:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003460 error = xlog_recover_unmount_trans(log, trans);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003461 break;
3462 case XLOG_WAS_CONT_TRANS:
Dave Chinner9abbc532010-04-13 15:06:46 +10003463 error = xlog_recover_add_to_cont_trans(log,
3464 trans, dp,
3465 be32_to_cpu(ohead->oh_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003466 break;
3467 case XLOG_START_TRANS:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003468 xfs_warn(log->l_mp, "%s: bad transaction",
3469 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003470 ASSERT(0);
3471 error = XFS_ERROR(EIO);
3472 break;
3473 case 0:
3474 case XLOG_CONTINUE_TRANS:
Dave Chinner9abbc532010-04-13 15:06:46 +10003475 error = xlog_recover_add_to_trans(log, trans,
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10003476 dp, be32_to_cpu(ohead->oh_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003477 break;
3478 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003479 xfs_warn(log->l_mp, "%s: bad flag 0x%x",
3480 __func__, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003481 ASSERT(0);
3482 error = XFS_ERROR(EIO);
3483 break;
3484 }
3485 if (error)
3486 return error;
3487 }
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10003488 dp += be32_to_cpu(ohead->oh_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003489 num_logops--;
3490 }
3491 return 0;
3492}
3493
3494/*
3495 * Process an extent free intent item that was recovered from
3496 * the log. We need to free the extents that it describes.
3497 */
David Chinner3c1e2bb2008-04-10 12:21:11 +10003498STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003499xlog_recover_process_efi(
3500 xfs_mount_t *mp,
3501 xfs_efi_log_item_t *efip)
3502{
3503 xfs_efd_log_item_t *efdp;
3504 xfs_trans_t *tp;
3505 int i;
David Chinner3c1e2bb2008-04-10 12:21:11 +10003506 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003507 xfs_extent_t *extp;
3508 xfs_fsblock_t startblock_fsb;
3509
Dave Chinnerb199c8a2010-12-20 11:59:49 +11003510 ASSERT(!test_bit(XFS_EFI_RECOVERED, &efip->efi_flags));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003511
3512 /*
3513 * First check the validity of the extents described by the
3514 * EFI. If any are bad, then assume that all are bad and
3515 * just toss the EFI.
3516 */
3517 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3518 extp = &(efip->efi_format.efi_extents[i]);
3519 startblock_fsb = XFS_BB_TO_FSB(mp,
3520 XFS_FSB_TO_DADDR(mp, extp->ext_start));
3521 if ((startblock_fsb == 0) ||
3522 (extp->ext_len == 0) ||
3523 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
3524 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
3525 /*
3526 * This will pull the EFI from the AIL and
3527 * free the memory associated with it.
3528 */
Dave Chinner666d6442013-04-03 14:09:21 +11003529 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003530 xfs_efi_release(efip, efip->efi_format.efi_nextents);
David Chinner3c1e2bb2008-04-10 12:21:11 +10003531 return XFS_ERROR(EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003532 }
3533 }
3534
3535 tp = xfs_trans_alloc(mp, 0);
Jie Liu3d3c8b52013-08-12 20:49:59 +10003536 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
David Chinnerfc6149d2008-04-10 12:21:53 +10003537 if (error)
3538 goto abort_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003539 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
3540
3541 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3542 extp = &(efip->efi_format.efi_extents[i]);
David Chinnerfc6149d2008-04-10 12:21:53 +10003543 error = xfs_free_extent(tp, extp->ext_start, extp->ext_len);
3544 if (error)
3545 goto abort_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003546 xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
3547 extp->ext_len);
3548 }
3549
Dave Chinnerb199c8a2010-12-20 11:59:49 +11003550 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
David Chinnere5720ee2008-04-10 12:21:18 +10003551 error = xfs_trans_commit(tp, 0);
David Chinner3c1e2bb2008-04-10 12:21:11 +10003552 return error;
David Chinnerfc6149d2008-04-10 12:21:53 +10003553
3554abort_error:
3555 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3556 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003557}
3558
3559/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003560 * When this is called, all of the EFIs which did not have
3561 * corresponding EFDs should be in the AIL. What we do now
3562 * is free the extents associated with each one.
3563 *
3564 * Since we process the EFIs in normal transactions, they
3565 * will be removed at some point after the commit. This prevents
3566 * us from just walking down the list processing each one.
3567 * We'll use a flag in the EFI to skip those that we've already
3568 * processed and use the AIL iteration mechanism's generation
3569 * count to try to speed this up at least a bit.
3570 *
3571 * When we start, we know that the EFIs are the only things in
3572 * the AIL. As we process them, however, other items are added
3573 * to the AIL. Since everything added to the AIL must come after
3574 * everything already in the AIL, we stop processing as soon as
3575 * we see something other than an EFI in the AIL.
3576 */
David Chinner3c1e2bb2008-04-10 12:21:11 +10003577STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003578xlog_recover_process_efis(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003579 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003580{
3581 xfs_log_item_t *lip;
3582 xfs_efi_log_item_t *efip;
David Chinner3c1e2bb2008-04-10 12:21:11 +10003583 int error = 0;
David Chinner27d8d5f2008-10-30 17:38:39 +11003584 struct xfs_ail_cursor cur;
David Chinnera9c21c12008-10-30 17:39:35 +11003585 struct xfs_ail *ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003586
David Chinnera9c21c12008-10-30 17:39:35 +11003587 ailp = log->l_ailp;
3588 spin_lock(&ailp->xa_lock);
3589 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003590 while (lip != NULL) {
3591 /*
3592 * We're done when we see something other than an EFI.
David Chinner27d8d5f2008-10-30 17:38:39 +11003593 * There should be no EFIs left in the AIL now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003594 */
3595 if (lip->li_type != XFS_LI_EFI) {
David Chinner27d8d5f2008-10-30 17:38:39 +11003596#ifdef DEBUG
David Chinnera9c21c12008-10-30 17:39:35 +11003597 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
David Chinner27d8d5f2008-10-30 17:38:39 +11003598 ASSERT(lip->li_type != XFS_LI_EFI);
3599#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003600 break;
3601 }
3602
3603 /*
3604 * Skip EFIs that we've already processed.
3605 */
3606 efip = (xfs_efi_log_item_t *)lip;
Dave Chinnerb199c8a2010-12-20 11:59:49 +11003607 if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) {
David Chinnera9c21c12008-10-30 17:39:35 +11003608 lip = xfs_trans_ail_cursor_next(ailp, &cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003609 continue;
3610 }
3611
David Chinnera9c21c12008-10-30 17:39:35 +11003612 spin_unlock(&ailp->xa_lock);
3613 error = xlog_recover_process_efi(log->l_mp, efip);
3614 spin_lock(&ailp->xa_lock);
David Chinner27d8d5f2008-10-30 17:38:39 +11003615 if (error)
3616 goto out;
David Chinnera9c21c12008-10-30 17:39:35 +11003617 lip = xfs_trans_ail_cursor_next(ailp, &cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003618 }
David Chinner27d8d5f2008-10-30 17:38:39 +11003619out:
David Chinnera9c21c12008-10-30 17:39:35 +11003620 xfs_trans_ail_cursor_done(ailp, &cur);
3621 spin_unlock(&ailp->xa_lock);
David Chinner3c1e2bb2008-04-10 12:21:11 +10003622 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003623}
3624
3625/*
3626 * This routine performs a transaction to null out a bad inode pointer
3627 * in an agi unlinked inode hash bucket.
3628 */
3629STATIC void
3630xlog_recover_clear_agi_bucket(
3631 xfs_mount_t *mp,
3632 xfs_agnumber_t agno,
3633 int bucket)
3634{
3635 xfs_trans_t *tp;
3636 xfs_agi_t *agi;
3637 xfs_buf_t *agibp;
3638 int offset;
3639 int error;
3640
3641 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
Jie Liu3d3c8b52013-08-12 20:49:59 +10003642 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_clearagi, 0, 0);
David Chinnere5720ee2008-04-10 12:21:18 +10003643 if (error)
3644 goto out_abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003645
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003646 error = xfs_read_agi(mp, tp, agno, &agibp);
3647 if (error)
David Chinnere5720ee2008-04-10 12:21:18 +10003648 goto out_abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003649
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003650 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11003651 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003652 offset = offsetof(xfs_agi_t, agi_unlinked) +
3653 (sizeof(xfs_agino_t) * bucket);
3654 xfs_trans_log_buf(tp, agibp, offset,
3655 (offset + sizeof(xfs_agino_t) - 1));
3656
David Chinnere5720ee2008-04-10 12:21:18 +10003657 error = xfs_trans_commit(tp, 0);
3658 if (error)
3659 goto out_error;
3660 return;
3661
3662out_abort:
3663 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3664out_error:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003665 xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno);
David Chinnere5720ee2008-04-10 12:21:18 +10003666 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003667}
3668
Christoph Hellwig23fac502008-11-28 14:23:40 +11003669STATIC xfs_agino_t
3670xlog_recover_process_one_iunlink(
3671 struct xfs_mount *mp,
3672 xfs_agnumber_t agno,
3673 xfs_agino_t agino,
3674 int bucket)
3675{
3676 struct xfs_buf *ibp;
3677 struct xfs_dinode *dip;
3678 struct xfs_inode *ip;
3679 xfs_ino_t ino;
3680 int error;
3681
3682 ino = XFS_AGINO_TO_INO(mp, agno, agino);
Dave Chinner7b6259e2010-06-24 11:35:17 +10003683 error = xfs_iget(mp, NULL, ino, 0, 0, &ip);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003684 if (error)
3685 goto fail;
3686
3687 /*
3688 * Get the on disk inode to find the next inode in the bucket.
3689 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04003690 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &ibp, 0, 0);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003691 if (error)
Christoph Hellwig0e446672008-11-28 14:23:42 +11003692 goto fail_iput;
Christoph Hellwig23fac502008-11-28 14:23:40 +11003693
Christoph Hellwig23fac502008-11-28 14:23:40 +11003694 ASSERT(ip->i_d.di_nlink == 0);
Christoph Hellwig0e446672008-11-28 14:23:42 +11003695 ASSERT(ip->i_d.di_mode != 0);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003696
3697 /* setup for the next pass */
3698 agino = be32_to_cpu(dip->di_next_unlinked);
3699 xfs_buf_relse(ibp);
3700
3701 /*
3702 * Prevent any DMAPI event from being sent when the reference on
3703 * the inode is dropped.
3704 */
3705 ip->i_d.di_dmevmask = 0;
3706
Christoph Hellwig0e446672008-11-28 14:23:42 +11003707 IRELE(ip);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003708 return agino;
3709
Christoph Hellwig0e446672008-11-28 14:23:42 +11003710 fail_iput:
3711 IRELE(ip);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003712 fail:
3713 /*
3714 * We can't read in the inode this bucket points to, or this inode
3715 * is messed up. Just ditch this bucket of inodes. We will lose
3716 * some inodes and space, but at least we won't hang.
3717 *
3718 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3719 * clear the inode pointer in the bucket.
3720 */
3721 xlog_recover_clear_agi_bucket(mp, agno, bucket);
3722 return NULLAGINO;
3723}
3724
Linus Torvalds1da177e2005-04-16 15:20:36 -07003725/*
3726 * xlog_iunlink_recover
3727 *
3728 * This is called during recovery to process any inodes which
3729 * we unlinked but not freed when the system crashed. These
3730 * inodes will be on the lists in the AGI blocks. What we do
3731 * here is scan all the AGIs and fully truncate and free any
3732 * inodes found on the lists. Each inode is removed from the
3733 * lists when it has been fully truncated and is freed. The
3734 * freeing of the inode and its removal from the list must be
3735 * atomic.
3736 */
Eric Sandeend96f8f82009-07-02 00:09:33 -05003737STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07003738xlog_recover_process_iunlinks(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003739 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003740{
3741 xfs_mount_t *mp;
3742 xfs_agnumber_t agno;
3743 xfs_agi_t *agi;
3744 xfs_buf_t *agibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003745 xfs_agino_t agino;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003746 int bucket;
3747 int error;
3748 uint mp_dmevmask;
3749
3750 mp = log->l_mp;
3751
3752 /*
3753 * Prevent any DMAPI event from being sent while in this function.
3754 */
3755 mp_dmevmask = mp->m_dmevmask;
3756 mp->m_dmevmask = 0;
3757
3758 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3759 /*
3760 * Find the agi for this ag.
3761 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003762 error = xfs_read_agi(mp, NULL, agno, &agibp);
3763 if (error) {
3764 /*
3765 * AGI is b0rked. Don't process it.
3766 *
3767 * We should probably mark the filesystem as corrupt
3768 * after we've recovered all the ag's we can....
3769 */
3770 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003771 }
Jan Karad97d32e2012-03-15 09:34:02 +00003772 /*
3773 * Unlock the buffer so that it can be acquired in the normal
3774 * course of the transaction to truncate and free each inode.
3775 * Because we are not racing with anyone else here for the AGI
3776 * buffer, we don't even need to hold it locked to read the
3777 * initial unlinked bucket entries out of the buffer. We keep
3778 * buffer reference though, so that it stays pinned in memory
3779 * while we need the buffer.
3780 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003781 agi = XFS_BUF_TO_AGI(agibp);
Jan Karad97d32e2012-03-15 09:34:02 +00003782 xfs_buf_unlock(agibp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003783
3784 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
Christoph Hellwig16259e72005-11-02 15:11:25 +11003785 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003786 while (agino != NULLAGINO) {
Christoph Hellwig23fac502008-11-28 14:23:40 +11003787 agino = xlog_recover_process_one_iunlink(mp,
3788 agno, agino, bucket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003789 }
3790 }
Jan Karad97d32e2012-03-15 09:34:02 +00003791 xfs_buf_rele(agibp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003792 }
3793
3794 mp->m_dmevmask = mp_dmevmask;
3795}
3796
Linus Torvalds1da177e2005-04-16 15:20:36 -07003797/*
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003798 * Upack the log buffer data and crc check it. If the check fails, issue a
3799 * warning if and only if the CRC in the header is non-zero. This makes the
3800 * check an advisory warning, and the zero CRC check will prevent failure
3801 * warnings from being emitted when upgrading the kernel from one that does not
3802 * add CRCs by default.
3803 *
3804 * When filesystems are CRC enabled, this CRC mismatch becomes a fatal log
3805 * corruption failure
Linus Torvalds1da177e2005-04-16 15:20:36 -07003806 */
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003807STATIC int
3808xlog_unpack_data_crc(
3809 struct xlog_rec_header *rhead,
3810 xfs_caddr_t dp,
3811 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003812{
Dave Chinnerf9668a02012-11-28 13:01:03 +11003813 __le32 crc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003814
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003815 crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len));
3816 if (crc != rhead->h_crc) {
3817 if (rhead->h_crc || xfs_sb_version_hascrc(&log->l_mp->m_sb)) {
3818 xfs_alert(log->l_mp,
3819 "log record CRC mismatch: found 0x%x, expected 0x%x.\n",
Dave Chinnerf9668a02012-11-28 13:01:03 +11003820 le32_to_cpu(rhead->h_crc),
3821 le32_to_cpu(crc));
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003822 xfs_hex_dump(dp, 32);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003823 }
3824
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003825 /*
3826 * If we've detected a log record corruption, then we can't
3827 * recover past this point. Abort recovery if we are enforcing
3828 * CRC protection by punting an error back up the stack.
3829 */
3830 if (xfs_sb_version_hascrc(&log->l_mp->m_sb))
3831 return EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003832 }
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003833
3834 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003835}
3836
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003837STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003838xlog_unpack_data(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003839 struct xlog_rec_header *rhead,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003840 xfs_caddr_t dp,
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003841 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003842{
3843 int i, j, k;
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003844 int error;
3845
3846 error = xlog_unpack_data_crc(rhead, dp, log);
3847 if (error)
3848 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003849
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003850 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07003851 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003852 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003853 dp += BBSIZE;
3854 }
3855
Eric Sandeen62118702008-03-06 13:44:28 +11003856 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Christoph Hellwigb28708d2008-11-28 14:23:38 +11003857 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003858 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003859 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3860 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003861 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003862 dp += BBSIZE;
3863 }
3864 }
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003865
3866 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003867}
3868
3869STATIC int
3870xlog_valid_rec_header(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003871 struct xlog *log,
3872 struct xlog_rec_header *rhead,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003873 xfs_daddr_t blkno)
3874{
3875 int hlen;
3876
Christoph Hellwig69ef9212011-07-08 14:36:05 +02003877 if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003878 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3879 XFS_ERRLEVEL_LOW, log->l_mp);
3880 return XFS_ERROR(EFSCORRUPTED);
3881 }
3882 if (unlikely(
3883 (!rhead->h_version ||
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003884 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003885 xfs_warn(log->l_mp, "%s: unrecognised log version (%d).",
Harvey Harrison34a622b2008-04-10 12:19:21 +10003886 __func__, be32_to_cpu(rhead->h_version));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003887 return XFS_ERROR(EIO);
3888 }
3889
3890 /* LR body must have data or it wouldn't have been written */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003891 hlen = be32_to_cpu(rhead->h_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003892 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
3893 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3894 XFS_ERRLEVEL_LOW, log->l_mp);
3895 return XFS_ERROR(EFSCORRUPTED);
3896 }
3897 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
3898 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3899 XFS_ERRLEVEL_LOW, log->l_mp);
3900 return XFS_ERROR(EFSCORRUPTED);
3901 }
3902 return 0;
3903}
3904
3905/*
3906 * Read the log from tail to head and process the log records found.
3907 * Handle the two cases where the tail and head are in the same cycle
3908 * and where the active portion of the log wraps around the end of
3909 * the physical log separately. The pass parameter is passed through
3910 * to the routines called to process the data and is not looked at
3911 * here.
3912 */
3913STATIC int
3914xlog_do_recovery_pass(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003915 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003916 xfs_daddr_t head_blk,
3917 xfs_daddr_t tail_blk,
3918 int pass)
3919{
3920 xlog_rec_header_t *rhead;
3921 xfs_daddr_t blk_no;
Andy Polingfc5bc4c2009-11-03 17:26:47 +00003922 xfs_caddr_t offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003923 xfs_buf_t *hbp, *dbp;
3924 int error = 0, h_size;
3925 int bblks, split_bblks;
3926 int hblks, split_hblks, wrapped_hblks;
Dave Chinnerf0a76952010-01-11 11:49:57 +00003927 struct hlist_head rhash[XLOG_RHASH_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003928
3929 ASSERT(head_blk != tail_blk);
3930
3931 /*
3932 * Read the header of the tail block and get the iclog buffer size from
3933 * h_size. Use this to tell how many sectors make up the log header.
3934 */
Eric Sandeen62118702008-03-06 13:44:28 +11003935 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003936 /*
3937 * When using variable length iclogs, read first sector of
3938 * iclog header and extract the header size from it. Get a
3939 * new hbp that is the correct size.
3940 */
3941 hbp = xlog_get_bp(log, 1);
3942 if (!hbp)
3943 return ENOMEM;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003944
3945 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
3946 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003947 goto bread_err1;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003948
Linus Torvalds1da177e2005-04-16 15:20:36 -07003949 rhead = (xlog_rec_header_t *)offset;
3950 error = xlog_valid_rec_header(log, rhead, tail_blk);
3951 if (error)
3952 goto bread_err1;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003953 h_size = be32_to_cpu(rhead->h_size);
3954 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07003955 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
3956 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
3957 if (h_size % XLOG_HEADER_CYCLE_SIZE)
3958 hblks++;
3959 xlog_put_bp(hbp);
3960 hbp = xlog_get_bp(log, hblks);
3961 } else {
3962 hblks = 1;
3963 }
3964 } else {
Alex Elder69ce58f2010-04-20 17:09:59 +10003965 ASSERT(log->l_sectBBsize == 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003966 hblks = 1;
3967 hbp = xlog_get_bp(log, 1);
3968 h_size = XLOG_BIG_RECORD_BSIZE;
3969 }
3970
3971 if (!hbp)
3972 return ENOMEM;
3973 dbp = xlog_get_bp(log, BTOBB(h_size));
3974 if (!dbp) {
3975 xlog_put_bp(hbp);
3976 return ENOMEM;
3977 }
3978
3979 memset(rhash, 0, sizeof(rhash));
3980 if (tail_blk <= head_blk) {
3981 for (blk_no = tail_blk; blk_no < head_blk; ) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003982 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
3983 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003984 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003985
Linus Torvalds1da177e2005-04-16 15:20:36 -07003986 rhead = (xlog_rec_header_t *)offset;
3987 error = xlog_valid_rec_header(log, rhead, blk_no);
3988 if (error)
3989 goto bread_err2;
3990
3991 /* blocks in data section */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003992 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003993 error = xlog_bread(log, blk_no + hblks, bblks, dbp,
3994 &offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003995 if (error)
3996 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003997
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003998 error = xlog_unpack_data(rhead, offset, log);
3999 if (error)
4000 goto bread_err2;
4001
4002 error = xlog_recover_process_data(log,
4003 rhash, rhead, offset, pass);
4004 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004005 goto bread_err2;
4006 blk_no += bblks + hblks;
4007 }
4008 } else {
4009 /*
4010 * Perform recovery around the end of the physical log.
4011 * When the head is not on the same cycle number as the tail,
4012 * we can't do a sequential recovery as above.
4013 */
4014 blk_no = tail_blk;
4015 while (blk_no < log->l_logBBsize) {
4016 /*
4017 * Check for header wrapping around physical end-of-log
4018 */
Chandra Seetharaman62926042011-07-22 23:40:15 +00004019 offset = hbp->b_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004020 split_hblks = 0;
4021 wrapped_hblks = 0;
4022 if (blk_no + hblks <= log->l_logBBsize) {
4023 /* Read header in one read */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004024 error = xlog_bread(log, blk_no, hblks, hbp,
4025 &offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004026 if (error)
4027 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004028 } else {
4029 /* This LR is split across physical log end */
4030 if (blk_no != log->l_logBBsize) {
4031 /* some data before physical log end */
4032 ASSERT(blk_no <= INT_MAX);
4033 split_hblks = log->l_logBBsize - (int)blk_no;
4034 ASSERT(split_hblks > 0);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004035 error = xlog_bread(log, blk_no,
4036 split_hblks, hbp,
4037 &offset);
4038 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004039 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004040 }
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004041
Linus Torvalds1da177e2005-04-16 15:20:36 -07004042 /*
4043 * Note: this black magic still works with
4044 * large sector sizes (non-512) only because:
4045 * - we increased the buffer size originally
4046 * by 1 sector giving us enough extra space
4047 * for the second read;
4048 * - the log start is guaranteed to be sector
4049 * aligned;
4050 * - we read the log end (LR header start)
4051 * _first_, then the log start (LR header end)
4052 * - order is important.
4053 */
David Chinner234f56a2008-04-10 12:24:24 +10004054 wrapped_hblks = hblks - split_hblks;
Dave Chinner44396472011-04-21 09:34:27 +00004055 error = xlog_bread_offset(log, 0,
4056 wrapped_hblks, hbp,
4057 offset + BBTOB(split_hblks));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004058 if (error)
4059 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004060 }
4061 rhead = (xlog_rec_header_t *)offset;
4062 error = xlog_valid_rec_header(log, rhead,
4063 split_hblks ? blk_no : 0);
4064 if (error)
4065 goto bread_err2;
4066
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004067 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004068 blk_no += hblks;
4069
4070 /* Read in data for log record */
4071 if (blk_no + bblks <= log->l_logBBsize) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004072 error = xlog_bread(log, blk_no, bblks, dbp,
4073 &offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004074 if (error)
4075 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004076 } else {
4077 /* This log record is split across the
4078 * physical end of log */
Chandra Seetharaman62926042011-07-22 23:40:15 +00004079 offset = dbp->b_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004080 split_bblks = 0;
4081 if (blk_no != log->l_logBBsize) {
4082 /* some data is before the physical
4083 * end of log */
4084 ASSERT(!wrapped_hblks);
4085 ASSERT(blk_no <= INT_MAX);
4086 split_bblks =
4087 log->l_logBBsize - (int)blk_no;
4088 ASSERT(split_bblks > 0);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004089 error = xlog_bread(log, blk_no,
4090 split_bblks, dbp,
4091 &offset);
4092 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004093 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004094 }
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004095
Linus Torvalds1da177e2005-04-16 15:20:36 -07004096 /*
4097 * Note: this black magic still works with
4098 * large sector sizes (non-512) only because:
4099 * - we increased the buffer size originally
4100 * by 1 sector giving us enough extra space
4101 * for the second read;
4102 * - the log start is guaranteed to be sector
4103 * aligned;
4104 * - we read the log end (LR header start)
4105 * _first_, then the log start (LR header end)
4106 * - order is important.
4107 */
Dave Chinner44396472011-04-21 09:34:27 +00004108 error = xlog_bread_offset(log, 0,
Dave Chinner009507b2012-11-02 11:38:44 +11004109 bblks - split_bblks, dbp,
Dave Chinner44396472011-04-21 09:34:27 +00004110 offset + BBTOB(split_bblks));
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004111 if (error)
4112 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004113 }
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004114
4115 error = xlog_unpack_data(rhead, offset, log);
4116 if (error)
4117 goto bread_err2;
4118
4119 error = xlog_recover_process_data(log, rhash,
4120 rhead, offset, pass);
4121 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004122 goto bread_err2;
4123 blk_no += bblks;
4124 }
4125
4126 ASSERT(blk_no >= log->l_logBBsize);
4127 blk_no -= log->l_logBBsize;
4128
4129 /* read first part of physical log */
4130 while (blk_no < head_blk) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004131 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
4132 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004133 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004134
Linus Torvalds1da177e2005-04-16 15:20:36 -07004135 rhead = (xlog_rec_header_t *)offset;
4136 error = xlog_valid_rec_header(log, rhead, blk_no);
4137 if (error)
4138 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004139
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004140 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004141 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
4142 &offset);
4143 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004144 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004145
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004146 error = xlog_unpack_data(rhead, offset, log);
4147 if (error)
4148 goto bread_err2;
4149
4150 error = xlog_recover_process_data(log, rhash,
4151 rhead, offset, pass);
4152 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004153 goto bread_err2;
4154 blk_no += bblks + hblks;
4155 }
4156 }
4157
4158 bread_err2:
4159 xlog_put_bp(dbp);
4160 bread_err1:
4161 xlog_put_bp(hbp);
4162 return error;
4163}
4164
4165/*
4166 * Do the recovery of the log. We actually do this in two phases.
4167 * The two passes are necessary in order to implement the function
4168 * of cancelling a record written into the log. The first pass
4169 * determines those things which have been cancelled, and the
4170 * second pass replays log items normally except for those which
4171 * have been cancelled. The handling of the replay and cancellations
4172 * takes place in the log item type specific routines.
4173 *
4174 * The table of items which have cancel records in the log is allocated
4175 * and freed at this level, since only here do we know when all of
4176 * the log recovery has been completed.
4177 */
4178STATIC int
4179xlog_do_log_recovery(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004180 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004181 xfs_daddr_t head_blk,
4182 xfs_daddr_t tail_blk)
4183{
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00004184 int error, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004185
4186 ASSERT(head_blk != tail_blk);
4187
4188 /*
4189 * First do a pass to find all of the cancelled buf log items.
4190 * Store them in the buf_cancel_table for use in the second pass.
4191 */
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00004192 log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE *
4193 sizeof(struct list_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07004194 KM_SLEEP);
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00004195 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
4196 INIT_LIST_HEAD(&log->l_buf_cancel_table[i]);
4197
Linus Torvalds1da177e2005-04-16 15:20:36 -07004198 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
4199 XLOG_RECOVER_PASS1);
4200 if (error != 0) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10004201 kmem_free(log->l_buf_cancel_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004202 log->l_buf_cancel_table = NULL;
4203 return error;
4204 }
4205 /*
4206 * Then do a second pass to actually recover the items in the log.
4207 * When it is complete free the table of buf cancel items.
4208 */
4209 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
4210 XLOG_RECOVER_PASS2);
4211#ifdef DEBUG
Tim Shimmin6d192a92006-06-09 14:55:38 +10004212 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004213 int i;
4214
4215 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00004216 ASSERT(list_empty(&log->l_buf_cancel_table[i]));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004217 }
4218#endif /* DEBUG */
4219
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10004220 kmem_free(log->l_buf_cancel_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004221 log->l_buf_cancel_table = NULL;
4222
4223 return error;
4224}
4225
4226/*
4227 * Do the actual recovery
4228 */
4229STATIC int
4230xlog_do_recover(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004231 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004232 xfs_daddr_t head_blk,
4233 xfs_daddr_t tail_blk)
4234{
4235 int error;
4236 xfs_buf_t *bp;
4237 xfs_sb_t *sbp;
4238
4239 /*
4240 * First replay the images in the log.
4241 */
4242 error = xlog_do_log_recovery(log, head_blk, tail_blk);
Christoph Hellwig43ff2122012-04-23 15:58:39 +10004243 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004244 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004245
4246 /*
4247 * If IO errors happened during recovery, bail out.
4248 */
4249 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
4250 return (EIO);
4251 }
4252
4253 /*
4254 * We now update the tail_lsn since much of the recovery has completed
4255 * and there may be space available to use. If there were no extent
4256 * or iunlinks, we can free up the entire log and set the tail_lsn to
4257 * be the last_sync_lsn. This was set in xlog_find_tail to be the
4258 * lsn of the last known good LR on disk. If there are extent frees
4259 * or iunlinks they will have some entries in the AIL; so we look at
4260 * the AIL to determine how to set the tail_lsn.
4261 */
4262 xlog_assign_tail_lsn(log->l_mp);
4263
4264 /*
4265 * Now that we've finished replaying all buffer and inode
Dave Chinner98021822012-11-12 22:54:03 +11004266 * updates, re-read in the superblock and reverify it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004267 */
4268 bp = xfs_getsb(log->l_mp, 0);
4269 XFS_BUF_UNDONE(bp);
Lachlan McIlroybebf9632007-10-15 13:18:02 +10004270 ASSERT(!(XFS_BUF_ISWRITE(bp)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004271 XFS_BUF_READ(bp);
Lachlan McIlroybebf9632007-10-15 13:18:02 +10004272 XFS_BUF_UNASYNC(bp);
Dave Chinner1813dd62012-11-14 17:54:40 +11004273 bp->b_ops = &xfs_sb_buf_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004274 xfsbdstrat(log->l_mp, bp);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +00004275 error = xfs_buf_iowait(bp);
David Chinnerd64e31a2008-04-10 12:22:17 +10004276 if (error) {
Christoph Hellwig901796a2011-10-10 16:52:49 +00004277 xfs_buf_ioerror_alert(bp, __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004278 ASSERT(0);
4279 xfs_buf_relse(bp);
4280 return error;
4281 }
4282
4283 /* Convert superblock from on-disk format */
4284 sbp = &log->l_mp->m_sb;
Dave Chinner98021822012-11-12 22:54:03 +11004285 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004286 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
Eric Sandeen62118702008-03-06 13:44:28 +11004287 ASSERT(xfs_sb_good_version(sbp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004288 xfs_buf_relse(bp);
4289
Lachlan McIlroy5478eea2007-02-10 18:36:29 +11004290 /* We've re-read the superblock so re-initialize per-cpu counters */
4291 xfs_icsb_reinit_counters(log->l_mp);
4292
Linus Torvalds1da177e2005-04-16 15:20:36 -07004293 xlog_recover_check_summary(log);
4294
4295 /* Normal transactions can now occur */
4296 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
4297 return 0;
4298}
4299
4300/*
4301 * Perform recovery and re-initialize some log variables in xlog_find_tail.
4302 *
4303 * Return error or zero.
4304 */
4305int
4306xlog_recover(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004307 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004308{
4309 xfs_daddr_t head_blk, tail_blk;
4310 int error;
4311
4312 /* find the tail of the log */
Eric Sandeen65be6052006-01-11 15:34:19 +11004313 if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004314 return error;
4315
4316 if (tail_blk != head_blk) {
4317 /* There used to be a comment here:
4318 *
4319 * disallow recovery on read-only mounts. note -- mount
4320 * checks for ENOSPC and turns it into an intelligent
4321 * error message.
4322 * ...but this is no longer true. Now, unless you specify
4323 * NORECOVERY (in which case this function would never be
4324 * called), we just go ahead and recover. We do this all
4325 * under the vfs layer, so we can get away with it unless
4326 * the device itself is read-only, in which case we fail.
4327 */
Utako Kusaka3a02ee12007-05-08 13:50:06 +10004328 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004329 return error;
4330 }
4331
Dave Chinnere721f502013-04-03 16:11:32 +11004332 /*
4333 * Version 5 superblock log feature mask validation. We know the
4334 * log is dirty so check if there are any unknown log features
4335 * in what we need to recover. If there are unknown features
4336 * (e.g. unsupported transactions, then simply reject the
4337 * attempt at recovery before touching anything.
4338 */
4339 if (XFS_SB_VERSION_NUM(&log->l_mp->m_sb) == XFS_SB_VERSION_5 &&
4340 xfs_sb_has_incompat_log_feature(&log->l_mp->m_sb,
4341 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
4342 xfs_warn(log->l_mp,
4343"Superblock has unknown incompatible log features (0x%x) enabled.\n"
4344"The log can not be fully and/or safely recovered by this kernel.\n"
4345"Please recover the log on a kernel that supports the unknown features.",
4346 (log->l_mp->m_sb.sb_features_log_incompat &
4347 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
4348 return EINVAL;
4349 }
4350
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004351 xfs_notice(log->l_mp, "Starting recovery (logdev: %s)",
4352 log->l_mp->m_logname ? log->l_mp->m_logname
4353 : "internal");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004354
4355 error = xlog_do_recover(log, head_blk, tail_blk);
4356 log->l_flags |= XLOG_RECOVERY_NEEDED;
4357 }
4358 return error;
4359}
4360
4361/*
4362 * In the first part of recovery we replay inodes and buffers and build
4363 * up the list of extent free items which need to be processed. Here
4364 * we process the extent free items and clean up the on disk unlinked
4365 * inode lists. This is separated from the first part of recovery so
4366 * that the root and real-time bitmap inodes can be read in from disk in
4367 * between the two stages. This is necessary so that we can free space
4368 * in the real-time portion of the file system.
4369 */
4370int
4371xlog_recover_finish(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004372 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004373{
4374 /*
4375 * Now we're ready to do the transactions needed for the
4376 * rest of recovery. Start with completing all the extent
4377 * free intent records and then process the unlinked inode
4378 * lists. At this point, we essentially run in normal mode
4379 * except that we're still performing recovery actions
4380 * rather than accepting new requests.
4381 */
4382 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
David Chinner3c1e2bb2008-04-10 12:21:11 +10004383 int error;
4384 error = xlog_recover_process_efis(log);
4385 if (error) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004386 xfs_alert(log->l_mp, "Failed to recover EFIs");
David Chinner3c1e2bb2008-04-10 12:21:11 +10004387 return error;
4388 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004389 /*
4390 * Sync the log to get all the EFIs out of the AIL.
4391 * This isn't absolutely necessary, but it helps in
4392 * case the unlink transactions would have problems
4393 * pushing the EFIs out of the way.
4394 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +00004395 xfs_log_force(log->l_mp, XFS_LOG_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004396
Christoph Hellwig42490232008-08-13 16:49:32 +10004397 xlog_recover_process_iunlinks(log);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004398
4399 xlog_recover_check_summary(log);
4400
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004401 xfs_notice(log->l_mp, "Ending recovery (logdev: %s)",
4402 log->l_mp->m_logname ? log->l_mp->m_logname
4403 : "internal");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004404 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
4405 } else {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004406 xfs_info(log->l_mp, "Ending clean mount");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004407 }
4408 return 0;
4409}
4410
4411
4412#if defined(DEBUG)
4413/*
4414 * Read all of the agf and agi counters and check that they
4415 * are consistent with the superblock counters.
4416 */
4417void
4418xlog_recover_check_summary(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004419 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004420{
4421 xfs_mount_t *mp;
4422 xfs_agf_t *agfp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004423 xfs_buf_t *agfbp;
4424 xfs_buf_t *agibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004425 xfs_agnumber_t agno;
4426 __uint64_t freeblks;
4427 __uint64_t itotal;
4428 __uint64_t ifree;
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11004429 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004430
4431 mp = log->l_mp;
4432
4433 freeblks = 0LL;
4434 itotal = 0LL;
4435 ifree = 0LL;
4436 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
From: Christoph Hellwig48056212008-11-28 14:23:38 +11004437 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
4438 if (error) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004439 xfs_alert(mp, "%s agf read failed agno %d error %d",
4440 __func__, agno, error);
From: Christoph Hellwig48056212008-11-28 14:23:38 +11004441 } else {
4442 agfp = XFS_BUF_TO_AGF(agfbp);
4443 freeblks += be32_to_cpu(agfp->agf_freeblks) +
4444 be32_to_cpu(agfp->agf_flcount);
4445 xfs_buf_relse(agfbp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004446 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004447
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11004448 error = xfs_read_agi(mp, NULL, agno, &agibp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004449 if (error) {
4450 xfs_alert(mp, "%s agi read failed agno %d error %d",
4451 __func__, agno, error);
4452 } else {
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11004453 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004454
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11004455 itotal += be32_to_cpu(agi->agi_count);
4456 ifree += be32_to_cpu(agi->agi_freecount);
4457 xfs_buf_relse(agibp);
4458 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004459 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004460}
4461#endif /* DEBUG */