blob: fbc2362d13e3b4c0531d310b9aa4c2e7054409ab [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 Chinner70a98832013-10-23 10:36:05 +110020#include "xfs_shared.h"
Dave Chinner239880e2013-10-23 10:50:10 +110021#include "xfs_format.h"
22#include "xfs_log_format.h"
23#include "xfs_trans_resv.h"
Nathan Scotta844f452005-11-02 14:38:42 +110024#include "xfs_bit.h"
Nathan Scotta844f452005-11-02 14:38:42 +110025#include "xfs_inum.h"
Nathan Scotta844f452005-11-02 14:38:42 +110026#include "xfs_sb.h"
27#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include "xfs_mount.h"
Dave Chinner57062782013-10-15 09:17:51 +110029#include "xfs_da_format.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include "xfs_inode.h"
Dave Chinner239880e2013-10-23 10:50:10 +110031#include "xfs_trans.h"
Dave Chinner239880e2013-10-23 10:50:10 +110032#include "xfs_log.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_log_priv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include "xfs_log_recover.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110035#include "xfs_inode_item.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include "xfs_extfree_item.h"
37#include "xfs_trans_priv.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110038#include "xfs_alloc.h"
39#include "xfs_ialloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include "xfs_quota.h"
Christoph Hellwig0e446be2012-11-12 22:54:24 +110041#include "xfs_cksum.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000042#include "xfs_trace.h"
Dave Chinner33479e02012-10-08 21:56:11 +110043#include "xfs_icache.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110044#include "xfs_bmap_btree.h"
45#include "xfs_dinode.h"
46#include "xfs_error.h"
Dave Chinner2b9ab5a2013-08-12 20:49:37 +100047#include "xfs_dir2.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
Dave Chinnerfc06c6d2013-08-12 20:49:22 +100049#define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
50
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -050051STATIC int
52xlog_find_zeroed(
53 struct xlog *,
54 xfs_daddr_t *);
55STATIC int
56xlog_clear_stale_blocks(
57 struct xlog *,
58 xfs_lsn_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -070059#if defined(DEBUG)
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -050060STATIC void
61xlog_recover_check_summary(
62 struct xlog *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070063#else
64#define xlog_recover_check_summary(log)
Linus Torvalds1da177e2005-04-16 15:20:36 -070065#endif
66
Linus Torvalds1da177e2005-04-16 15:20:36 -070067/*
Christoph Hellwigd5689ea2010-12-01 22:06:22 +000068 * This structure is used during recovery to record the buf log items which
69 * have been canceled and should not be replayed.
70 */
71struct xfs_buf_cancel {
72 xfs_daddr_t bc_blkno;
73 uint bc_len;
74 int bc_refcount;
75 struct list_head bc_list;
76};
77
78/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 * Sector aligned buffer routines for buffer create/read/write/access
80 */
81
Alex Elderff30a622010-04-13 15:22:58 +100082/*
83 * Verify the given count of basic blocks is valid number of blocks
84 * to specify for an operation involving the given XFS log buffer.
85 * Returns nonzero if the count is valid, 0 otherwise.
86 */
87
88static inline int
89xlog_buf_bbcount_valid(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -050090 struct xlog *log,
Alex Elderff30a622010-04-13 15:22:58 +100091 int bbcount)
92{
93 return bbcount > 0 && bbcount <= log->l_logBBsize;
94}
95
Alex Elder36adecf2010-04-13 15:21:13 +100096/*
97 * Allocate a buffer to hold log data. The buffer needs to be able
98 * to map to a range of nbblks basic blocks at any valid (basic
99 * block) offset within the log.
100 */
Eric Sandeen5d77c0d2009-11-19 15:52:00 +0000101STATIC xfs_buf_t *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102xlog_get_bp(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500103 struct xlog *log,
Dave Chinner32281492009-01-22 15:37:47 +1100104 int nbblks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105{
Christoph Hellwigc8da0fa2011-07-08 14:36:25 +0200106 struct xfs_buf *bp;
107
Alex Elderff30a622010-04-13 15:22:58 +1000108 if (!xlog_buf_bbcount_valid(log, nbblks)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100109 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
Alex Elderff30a622010-04-13 15:22:58 +1000110 nbblks);
111 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
Dave Chinner32281492009-01-22 15:37:47 +1100112 return NULL;
113 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114
Alex Elder36adecf2010-04-13 15:21:13 +1000115 /*
116 * We do log I/O in units of log sectors (a power-of-2
117 * multiple of the basic block size), so we round up the
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300118 * requested size to accommodate the basic blocks required
Alex Elder36adecf2010-04-13 15:21:13 +1000119 * for complete log sectors.
120 *
121 * In addition, the buffer may be used for a non-sector-
122 * aligned block offset, in which case an I/O of the
123 * requested size could extend beyond the end of the
124 * buffer. If the requested size is only 1 basic block it
125 * will never straddle a sector boundary, so this won't be
126 * an issue. Nor will this be a problem if the log I/O is
127 * done in basic blocks (sector size 1). But otherwise we
128 * extend the buffer by one extra log sector to ensure
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300129 * there's space to accommodate this possibility.
Alex Elder36adecf2010-04-13 15:21:13 +1000130 */
Alex Elder69ce58f2010-04-20 17:09:59 +1000131 if (nbblks > 1 && log->l_sectBBsize > 1)
132 nbblks += log->l_sectBBsize;
133 nbblks = round_up(nbblks, log->l_sectBBsize);
Alex Elder36adecf2010-04-13 15:21:13 +1000134
Dave Chinnere70b73f2012-04-23 15:58:49 +1000135 bp = xfs_buf_get_uncached(log->l_mp->m_logdev_targp, nbblks, 0);
Christoph Hellwigc8da0fa2011-07-08 14:36:25 +0200136 if (bp)
137 xfs_buf_unlock(bp);
138 return bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139}
140
Eric Sandeen5d77c0d2009-11-19 15:52:00 +0000141STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142xlog_put_bp(
143 xfs_buf_t *bp)
144{
145 xfs_buf_free(bp);
146}
147
Alex Elder48389ef2010-04-20 17:10:21 +1000148/*
149 * Return the address of the start of the given block number's data
150 * in a log buffer. The buffer covers a log sector-aligned region.
151 */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100152STATIC xfs_caddr_t
153xlog_align(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500154 struct xlog *log,
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100155 xfs_daddr_t blk_no,
156 int nbblks,
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500157 struct xfs_buf *bp)
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100158{
Christoph Hellwigfdc07f42010-05-10 17:28:14 +0000159 xfs_daddr_t offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100160
Dave Chinner4e94b712012-04-23 15:58:51 +1000161 ASSERT(offset + nbblks <= bp->b_length);
Chandra Seetharaman62926042011-07-22 23:40:15 +0000162 return bp->b_addr + BBTOB(offset);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100163}
164
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165
166/*
167 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
168 */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100169STATIC int
170xlog_bread_noalign(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500171 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 xfs_daddr_t blk_no,
173 int nbblks,
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500174 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175{
176 int error;
177
Alex Elderff30a622010-04-13 15:22:58 +1000178 if (!xlog_buf_bbcount_valid(log, nbblks)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100179 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
Alex Elderff30a622010-04-13 15:22:58 +1000180 nbblks);
181 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
Dave Chinner24513372014-06-25 14:58:08 +1000182 return -EFSCORRUPTED;
Dave Chinner32281492009-01-22 15:37:47 +1100183 }
184
Alex Elder69ce58f2010-04-20 17:09:59 +1000185 blk_no = round_down(blk_no, log->l_sectBBsize);
186 nbblks = round_up(nbblks, log->l_sectBBsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187
188 ASSERT(nbblks > 0);
Dave Chinner4e94b712012-04-23 15:58:51 +1000189 ASSERT(nbblks <= bp->b_length);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190
191 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
192 XFS_BUF_READ(bp);
Dave Chinneraa0e8832012-04-23 15:58:52 +1000193 bp->b_io_length = nbblks;
Dave Chinner0e95f192012-04-23 15:58:46 +1000194 bp->b_error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195
Christoph Hellwig83a0adc2013-12-17 00:03:52 -0800196 if (XFS_FORCED_SHUTDOWN(log->l_mp))
Dave Chinner24513372014-06-25 14:58:08 +1000197 return -EIO;
Christoph Hellwig83a0adc2013-12-17 00:03:52 -0800198
199 xfs_buf_iorequest(bp);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +0000200 error = xfs_buf_iowait(bp);
David Chinnerd64e31a2008-04-10 12:22:17 +1000201 if (error)
Christoph Hellwig901796a2011-10-10 16:52:49 +0000202 xfs_buf_ioerror_alert(bp, __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 return error;
204}
205
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100206STATIC int
207xlog_bread(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500208 struct xlog *log,
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100209 xfs_daddr_t blk_no,
210 int nbblks,
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500211 struct xfs_buf *bp,
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100212 xfs_caddr_t *offset)
213{
214 int error;
215
216 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
217 if (error)
218 return error;
219
220 *offset = xlog_align(log, blk_no, nbblks, bp);
221 return 0;
222}
223
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224/*
Dave Chinner44396472011-04-21 09:34:27 +0000225 * Read at an offset into the buffer. Returns with the buffer in it's original
226 * state regardless of the result of the read.
227 */
228STATIC int
229xlog_bread_offset(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500230 struct xlog *log,
Dave Chinner44396472011-04-21 09:34:27 +0000231 xfs_daddr_t blk_no, /* block to read from */
232 int nbblks, /* blocks to read */
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500233 struct xfs_buf *bp,
Dave Chinner44396472011-04-21 09:34:27 +0000234 xfs_caddr_t offset)
235{
Chandra Seetharaman62926042011-07-22 23:40:15 +0000236 xfs_caddr_t orig_offset = bp->b_addr;
Dave Chinner4e94b712012-04-23 15:58:51 +1000237 int orig_len = BBTOB(bp->b_length);
Dave Chinner44396472011-04-21 09:34:27 +0000238 int error, error2;
239
Chandra Seetharaman02fe03d2011-07-22 23:40:22 +0000240 error = xfs_buf_associate_memory(bp, offset, BBTOB(nbblks));
Dave Chinner44396472011-04-21 09:34:27 +0000241 if (error)
242 return error;
243
244 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
245
246 /* must reset buffer pointer even on error */
Chandra Seetharaman02fe03d2011-07-22 23:40:22 +0000247 error2 = xfs_buf_associate_memory(bp, orig_offset, orig_len);
Dave Chinner44396472011-04-21 09:34:27 +0000248 if (error)
249 return error;
250 return error2;
251}
252
253/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 * Write out the buffer at the given block for the given number of blocks.
255 * The buffer is kept locked across the write and is returned locked.
256 * This can only be used for synchronous log writes.
257 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000258STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259xlog_bwrite(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500260 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 xfs_daddr_t blk_no,
262 int nbblks,
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500263 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264{
265 int error;
266
Alex Elderff30a622010-04-13 15:22:58 +1000267 if (!xlog_buf_bbcount_valid(log, nbblks)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100268 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
Alex Elderff30a622010-04-13 15:22:58 +1000269 nbblks);
270 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
Dave Chinner24513372014-06-25 14:58:08 +1000271 return -EFSCORRUPTED;
Dave Chinner32281492009-01-22 15:37:47 +1100272 }
273
Alex Elder69ce58f2010-04-20 17:09:59 +1000274 blk_no = round_down(blk_no, log->l_sectBBsize);
275 nbblks = round_up(nbblks, log->l_sectBBsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276
277 ASSERT(nbblks > 0);
Dave Chinner4e94b712012-04-23 15:58:51 +1000278 ASSERT(nbblks <= bp->b_length);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279
280 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
281 XFS_BUF_ZEROFLAGS(bp);
Chandra Seetharaman72790aa2011-07-22 23:40:04 +0000282 xfs_buf_hold(bp);
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200283 xfs_buf_lock(bp);
Dave Chinneraa0e8832012-04-23 15:58:52 +1000284 bp->b_io_length = nbblks;
Dave Chinner0e95f192012-04-23 15:58:46 +1000285 bp->b_error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286
Christoph Hellwigc2b006c2011-08-23 08:28:07 +0000287 error = xfs_bwrite(bp);
Christoph Hellwig901796a2011-10-10 16:52:49 +0000288 if (error)
289 xfs_buf_ioerror_alert(bp, __func__);
Christoph Hellwigc2b006c2011-08-23 08:28:07 +0000290 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 return error;
292}
293
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294#ifdef DEBUG
295/*
296 * dump debug superblock and log record information
297 */
298STATIC void
299xlog_header_check_dump(
300 xfs_mount_t *mp,
301 xlog_rec_header_t *head)
302{
Eric Sandeen08e96e12013-10-11 20:59:05 -0500303 xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d",
Joe Perches03daa572009-12-14 18:01:10 -0800304 __func__, &mp->m_sb.sb_uuid, XLOG_FMT);
Eric Sandeen08e96e12013-10-11 20:59:05 -0500305 xfs_debug(mp, " log : uuid = %pU, fmt = %d",
Joe Perches03daa572009-12-14 18:01:10 -0800306 &head->h_fs_uuid, be32_to_cpu(head->h_fmt));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307}
308#else
309#define xlog_header_check_dump(mp, head)
310#endif
311
312/*
313 * check log record header for recovery
314 */
315STATIC int
316xlog_header_check_recover(
317 xfs_mount_t *mp,
318 xlog_rec_header_t *head)
319{
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200320 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321
322 /*
323 * IRIX doesn't write the h_fmt field and leaves it zeroed
324 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
325 * a dirty log created in IRIX.
326 */
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200327 if (unlikely(head->h_fmt != cpu_to_be32(XLOG_FMT))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100328 xfs_warn(mp,
329 "dirty log written in incompatible format - can't recover");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 xlog_header_check_dump(mp, head);
331 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
332 XFS_ERRLEVEL_HIGH, mp);
Dave Chinner24513372014-06-25 14:58:08 +1000333 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100335 xfs_warn(mp,
336 "dirty log entry has mismatched uuid - can't recover");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 xlog_header_check_dump(mp, head);
338 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
339 XFS_ERRLEVEL_HIGH, mp);
Dave Chinner24513372014-06-25 14:58:08 +1000340 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 }
342 return 0;
343}
344
345/*
346 * read the head block of the log and check the header
347 */
348STATIC int
349xlog_header_check_mount(
350 xfs_mount_t *mp,
351 xlog_rec_header_t *head)
352{
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200353 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354
355 if (uuid_is_nil(&head->h_fs_uuid)) {
356 /*
357 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
358 * h_fs_uuid is nil, we assume this log was last mounted
359 * by IRIX and continue.
360 */
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100361 xfs_warn(mp, "nil uuid in log - IRIX style log");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100363 xfs_warn(mp, "log has mismatched uuid - can't recover");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 xlog_header_check_dump(mp, head);
365 XFS_ERROR_REPORT("xlog_header_check_mount",
366 XFS_ERRLEVEL_HIGH, mp);
Dave Chinner24513372014-06-25 14:58:08 +1000367 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 }
369 return 0;
370}
371
372STATIC void
373xlog_recover_iodone(
374 struct xfs_buf *bp)
375{
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +0000376 if (bp->b_error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 /*
378 * We're not going to bother about retrying
379 * this during recovery. One strike!
380 */
Christoph Hellwig901796a2011-10-10 16:52:49 +0000381 xfs_buf_ioerror_alert(bp, __func__);
Dave Chinnerebad8612010-09-22 10:47:20 +1000382 xfs_force_shutdown(bp->b_target->bt_mount,
383 SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 }
Christoph Hellwigcb669ca2011-07-13 13:43:49 +0200385 bp->b_iodone = NULL;
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +0000386 xfs_buf_ioend(bp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387}
388
389/*
390 * This routine finds (to an approximation) the first block in the physical
391 * log which contains the given cycle. It uses a binary search algorithm.
392 * Note that the algorithm can not be perfect because the disk will not
393 * necessarily be perfect.
394 */
David Chinnera8272ce2007-11-23 16:28:09 +1100395STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396xlog_find_cycle_start(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500397 struct xlog *log,
398 struct xfs_buf *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 xfs_daddr_t first_blk,
400 xfs_daddr_t *last_blk,
401 uint cycle)
402{
403 xfs_caddr_t offset;
404 xfs_daddr_t mid_blk;
Alex Eldere3bb2e32010-04-15 18:17:30 +0000405 xfs_daddr_t end_blk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 uint mid_cycle;
407 int error;
408
Alex Eldere3bb2e32010-04-15 18:17:30 +0000409 end_blk = *last_blk;
410 mid_blk = BLK_AVG(first_blk, end_blk);
411 while (mid_blk != first_blk && mid_blk != end_blk) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100412 error = xlog_bread(log, mid_blk, 1, bp, &offset);
413 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 return error;
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000415 mid_cycle = xlog_get_cycle(offset);
Alex Eldere3bb2e32010-04-15 18:17:30 +0000416 if (mid_cycle == cycle)
417 end_blk = mid_blk; /* last_half_cycle == mid_cycle */
418 else
419 first_blk = mid_blk; /* first_half_cycle == mid_cycle */
420 mid_blk = BLK_AVG(first_blk, end_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 }
Alex Eldere3bb2e32010-04-15 18:17:30 +0000422 ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) ||
423 (mid_blk == end_blk && mid_blk-1 == first_blk));
424
425 *last_blk = end_blk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426
427 return 0;
428}
429
430/*
Alex Elder3f943d82010-04-15 18:17:34 +0000431 * Check that a range of blocks does not contain stop_on_cycle_no.
432 * Fill in *new_blk with the block offset where such a block is
433 * found, or with -1 (an invalid block number) if there is no such
434 * block in the range. The scan needs to occur from front to back
435 * and the pointer into the region must be updated since a later
436 * routine will need to perform another test.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 */
438STATIC int
439xlog_find_verify_cycle(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500440 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 xfs_daddr_t start_blk,
442 int nbblks,
443 uint stop_on_cycle_no,
444 xfs_daddr_t *new_blk)
445{
446 xfs_daddr_t i, j;
447 uint cycle;
448 xfs_buf_t *bp;
449 xfs_daddr_t bufblks;
450 xfs_caddr_t buf = NULL;
451 int error = 0;
452
Alex Elder6881a222010-04-13 15:22:29 +1000453 /*
454 * Greedily allocate a buffer big enough to handle the full
455 * range of basic blocks we'll be examining. If that fails,
456 * try a smaller size. We need to be able to read at least
457 * a log sector, or we're out of luck.
458 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 bufblks = 1 << ffs(nbblks);
Dave Chinner81158e02012-04-27 19:45:22 +1000460 while (bufblks > log->l_logBBsize)
461 bufblks >>= 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 while (!(bp = xlog_get_bp(log, bufblks))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 bufblks >>= 1;
Alex Elder69ce58f2010-04-20 17:09:59 +1000464 if (bufblks < log->l_sectBBsize)
Dave Chinner24513372014-06-25 14:58:08 +1000465 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 }
467
468 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
469 int bcount;
470
471 bcount = min(bufblks, (start_blk + nbblks - i));
472
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100473 error = xlog_bread(log, i, bcount, bp, &buf);
474 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 goto out;
476
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 for (j = 0; j < bcount; j++) {
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000478 cycle = xlog_get_cycle(buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 if (cycle == stop_on_cycle_no) {
480 *new_blk = i+j;
481 goto out;
482 }
483
484 buf += BBSIZE;
485 }
486 }
487
488 *new_blk = -1;
489
490out:
491 xlog_put_bp(bp);
492 return error;
493}
494
495/*
496 * Potentially backup over partial log record write.
497 *
498 * In the typical case, last_blk is the number of the block directly after
499 * a good log record. Therefore, we subtract one to get the block number
500 * of the last block in the given buffer. extra_bblks contains the number
501 * of blocks we would have read on a previous read. This happens when the
502 * last log record is split over the end of the physical log.
503 *
504 * extra_bblks is the number of blocks potentially verified on a previous
505 * call to this routine.
506 */
507STATIC int
508xlog_find_verify_log_record(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500509 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 xfs_daddr_t start_blk,
511 xfs_daddr_t *last_blk,
512 int extra_bblks)
513{
514 xfs_daddr_t i;
515 xfs_buf_t *bp;
516 xfs_caddr_t offset = NULL;
517 xlog_rec_header_t *head = NULL;
518 int error = 0;
519 int smallmem = 0;
520 int num_blks = *last_blk - start_blk;
521 int xhdrs;
522
523 ASSERT(start_blk != 0 || *last_blk != start_blk);
524
525 if (!(bp = xlog_get_bp(log, num_blks))) {
526 if (!(bp = xlog_get_bp(log, 1)))
Dave Chinner24513372014-06-25 14:58:08 +1000527 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 smallmem = 1;
529 } else {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100530 error = xlog_bread(log, start_blk, num_blks, bp, &offset);
531 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 offset += ((num_blks - 1) << BBSHIFT);
534 }
535
536 for (i = (*last_blk) - 1; i >= 0; i--) {
537 if (i < start_blk) {
538 /* valid log record not found */
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100539 xfs_warn(log->l_mp,
540 "Log inconsistent (didn't find previous header)");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 ASSERT(0);
Dave Chinner24513372014-06-25 14:58:08 +1000542 error = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 goto out;
544 }
545
546 if (smallmem) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100547 error = xlog_bread(log, i, 1, bp, &offset);
548 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 }
551
552 head = (xlog_rec_header_t *)offset;
553
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200554 if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 break;
556
557 if (!smallmem)
558 offset -= BBSIZE;
559 }
560
561 /*
562 * We hit the beginning of the physical log & still no header. Return
563 * to caller. If caller can handle a return of -1, then this routine
564 * will be called again for the end of the physical log.
565 */
566 if (i == -1) {
Dave Chinner24513372014-06-25 14:58:08 +1000567 error = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 goto out;
569 }
570
571 /*
572 * We have the final block of the good log (the first block
573 * of the log record _before_ the head. So we check the uuid.
574 */
575 if ((error = xlog_header_check_mount(log->l_mp, head)))
576 goto out;
577
578 /*
579 * We may have found a log record header before we expected one.
580 * last_blk will be the 1st block # with a given cycle #. We may end
581 * up reading an entire log record. In this case, we don't want to
582 * reset last_blk. Only when last_blk points in the middle of a log
583 * record do we update last_blk.
584 */
Eric Sandeen62118702008-03-06 13:44:28 +1100585 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000586 uint h_size = be32_to_cpu(head->h_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587
588 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
589 if (h_size % XLOG_HEADER_CYCLE_SIZE)
590 xhdrs++;
591 } else {
592 xhdrs = 1;
593 }
594
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000595 if (*last_blk - i + extra_bblks !=
596 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 *last_blk = i;
598
599out:
600 xlog_put_bp(bp);
601 return error;
602}
603
604/*
605 * Head is defined to be the point of the log where the next log write
Zhi Yong Wu0a94da22013-08-07 10:11:08 +0000606 * could go. This means that incomplete LR writes at the end are
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 * eliminated when calculating the head. We aren't guaranteed that previous
608 * LR have complete transactions. We only know that a cycle number of
609 * current cycle number -1 won't be present in the log if we start writing
610 * from our current block number.
611 *
612 * last_blk contains the block number of the first block with a given
613 * cycle number.
614 *
615 * Return: zero if normal, non-zero if error.
616 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000617STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618xlog_find_head(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500619 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 xfs_daddr_t *return_head_blk)
621{
622 xfs_buf_t *bp;
623 xfs_caddr_t offset;
624 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
625 int num_scan_bblks;
626 uint first_half_cycle, last_half_cycle;
627 uint stop_on_cycle;
628 int error, log_bbnum = log->l_logBBsize;
629
630 /* Is the end of the log device zeroed? */
Dave Chinner24513372014-06-25 14:58:08 +1000631 error = xlog_find_zeroed(log, &first_blk);
632 if (error < 0) {
633 xfs_warn(log->l_mp, "empty log check failed");
634 return error;
635 }
636 if (error == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 *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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 }
650
651 first_blk = 0; /* get cycle # of 1st block */
652 bp = xlog_get_bp(log, 1);
653 if (!bp)
Dave Chinner24513372014-06-25 14:58:08 +1000654 return -ENOMEM;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100655
656 error = xlog_bread(log, 0, 1, bp, &offset);
657 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 goto bp_err;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100659
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000660 first_half_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661
662 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100663 error = xlog_bread(log, last_blk, 1, bp, &offset);
664 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 goto bp_err;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100666
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000667 last_half_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 ASSERT(last_half_cycle != 0);
669
670 /*
671 * If the 1st half cycle number is equal to the last half cycle number,
672 * then the entire log is stamped with the same cycle number. In this
673 * case, head_blk can't be set to zero (which makes sense). The below
674 * math doesn't work out properly with head_blk equal to zero. Instead,
675 * we set it to log_bbnum which is an invalid block number, but this
676 * value makes the math correct. If head_blk doesn't changed through
677 * all the tests below, *head_blk is set to zero at the very end rather
678 * than log_bbnum. In a sense, log_bbnum and zero are the same block
679 * in a circular file.
680 */
681 if (first_half_cycle == last_half_cycle) {
682 /*
683 * In this case we believe that the entire log should have
684 * cycle number last_half_cycle. We need to scan backwards
685 * from the end verifying that there are no holes still
686 * containing last_half_cycle - 1. If we find such a hole,
687 * then the start of that hole will be the new head. The
688 * simple case looks like
689 * x | x ... | x - 1 | x
690 * Another case that fits this picture would be
691 * x | x + 1 | x ... | x
Nathan Scottc41564b2006-03-29 08:55:14 +1000692 * In this case the head really is somewhere at the end of the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 * log, as one of the latest writes at the beginning was
694 * incomplete.
695 * One more case is
696 * x | x + 1 | x ... | x - 1 | x
697 * This is really the combination of the above two cases, and
698 * the head has to end up at the start of the x-1 hole at the
699 * end of the log.
700 *
701 * In the 256k log case, we will read from the beginning to the
702 * end of the log and search for cycle numbers equal to x-1.
703 * We don't worry about the x+1 blocks that we encounter,
704 * because we know that they cannot be the head since the log
705 * started with x.
706 */
707 head_blk = log_bbnum;
708 stop_on_cycle = last_half_cycle - 1;
709 } else {
710 /*
711 * In this case we want to find the first block with cycle
712 * number matching last_half_cycle. We expect the log to be
713 * some variation on
Alex Elder3f943d82010-04-15 18:17:34 +0000714 * x + 1 ... | x ... | x
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 * The first block with cycle number x (last_half_cycle) will
716 * be where the new head belongs. First we do a binary search
717 * for the first occurrence of last_half_cycle. The binary
718 * search may not be totally accurate, so then we scan back
719 * from there looking for occurrences of last_half_cycle before
720 * us. If that backwards scan wraps around the beginning of
721 * the log, then we look for occurrences of last_half_cycle - 1
722 * at the end of the log. The cases we're looking for look
723 * like
Alex Elder3f943d82010-04-15 18:17:34 +0000724 * v binary search stopped here
725 * x + 1 ... | x | x + 1 | x ... | x
726 * ^ but we want to locate this spot
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 * or
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 * <---------> less than scan distance
Alex Elder3f943d82010-04-15 18:17:34 +0000729 * x + 1 ... | x ... | x - 1 | x
730 * ^ we want to locate this spot
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 */
732 stop_on_cycle = last_half_cycle;
733 if ((error = xlog_find_cycle_start(log, bp, first_blk,
734 &head_blk, last_half_cycle)))
735 goto bp_err;
736 }
737
738 /*
739 * Now validate the answer. Scan back some number of maximum possible
740 * blocks and make sure each one has the expected cycle number. The
741 * maximum is determined by the total possible amount of buffering
742 * in the in-core log. The following number can be made tighter if
743 * we actually look at the block size of the filesystem.
744 */
745 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
746 if (head_blk >= num_scan_bblks) {
747 /*
748 * We are guaranteed that the entire check can be performed
749 * in one buffer.
750 */
751 start_blk = head_blk - num_scan_bblks;
752 if ((error = xlog_find_verify_cycle(log,
753 start_blk, num_scan_bblks,
754 stop_on_cycle, &new_blk)))
755 goto bp_err;
756 if (new_blk != -1)
757 head_blk = new_blk;
758 } else { /* need to read 2 parts of log */
759 /*
760 * We are going to scan backwards in the log in two parts.
761 * First we scan the physical end of the log. In this part
762 * of the log, we are looking for blocks with cycle number
763 * last_half_cycle - 1.
764 * If we find one, then we know that the log starts there, as
765 * we've found a hole that didn't get written in going around
766 * the end of the physical log. The simple case for this is
767 * x + 1 ... | x ... | x - 1 | x
768 * <---------> less than scan distance
769 * If all of the blocks at the end of the log have cycle number
770 * last_half_cycle, then we check the blocks at the start of
771 * the log looking for occurrences of last_half_cycle. If we
772 * find one, then our current estimate for the location of the
773 * first occurrence of last_half_cycle is wrong and we move
774 * back to the hole we've found. This case looks like
775 * x + 1 ... | x | x + 1 | x ...
776 * ^ binary search stopped here
777 * Another case we need to handle that only occurs in 256k
778 * logs is
779 * x + 1 ... | x ... | x+1 | x ...
780 * ^ binary search stops here
781 * In a 256k log, the scan at the end of the log will see the
782 * x + 1 blocks. We need to skip past those since that is
783 * certainly not the head of the log. By searching for
784 * last_half_cycle-1 we accomplish that.
785 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 ASSERT(head_blk <= INT_MAX &&
Alex Elder3f943d82010-04-15 18:17:34 +0000787 (xfs_daddr_t) num_scan_bblks >= head_blk);
788 start_blk = log_bbnum - (num_scan_bblks - head_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 if ((error = xlog_find_verify_cycle(log, start_blk,
790 num_scan_bblks - (int)head_blk,
791 (stop_on_cycle - 1), &new_blk)))
792 goto bp_err;
793 if (new_blk != -1) {
794 head_blk = new_blk;
Alex Elder9db127e2010-04-15 18:17:26 +0000795 goto validate_head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 }
797
798 /*
799 * Scan beginning of log now. The last part of the physical
800 * log is good. This scan needs to verify that it doesn't find
801 * the last_half_cycle.
802 */
803 start_blk = 0;
804 ASSERT(head_blk <= INT_MAX);
805 if ((error = xlog_find_verify_cycle(log,
806 start_blk, (int)head_blk,
807 stop_on_cycle, &new_blk)))
808 goto bp_err;
809 if (new_blk != -1)
810 head_blk = new_blk;
811 }
812
Alex Elder9db127e2010-04-15 18:17:26 +0000813validate_head:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 /*
815 * Now we need to make sure head_blk is not pointing to a block in
816 * the middle of a log record.
817 */
818 num_scan_bblks = XLOG_REC_SHIFT(log);
819 if (head_blk >= num_scan_bblks) {
820 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
821
822 /* start ptr at last block ptr before head_blk */
Dave Chinner24513372014-06-25 14:58:08 +1000823 error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0);
824 if (error == 1)
825 error = -EIO;
826 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 goto bp_err;
828 } else {
829 start_blk = 0;
830 ASSERT(head_blk <= INT_MAX);
Dave Chinner24513372014-06-25 14:58:08 +1000831 error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0);
832 if (error < 0)
833 goto bp_err;
834 if (error == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 /* We hit the beginning of the log during our search */
Alex Elder3f943d82010-04-15 18:17:34 +0000836 start_blk = log_bbnum - (num_scan_bblks - head_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 new_blk = log_bbnum;
838 ASSERT(start_blk <= INT_MAX &&
839 (xfs_daddr_t) log_bbnum-start_blk >= 0);
840 ASSERT(head_blk <= INT_MAX);
Dave Chinner24513372014-06-25 14:58:08 +1000841 error = xlog_find_verify_log_record(log, start_blk,
842 &new_blk, (int)head_blk);
843 if (error == 1)
844 error = -EIO;
845 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 goto bp_err;
847 if (new_blk != log_bbnum)
848 head_blk = new_blk;
849 } else if (error)
850 goto bp_err;
851 }
852
853 xlog_put_bp(bp);
854 if (head_blk == log_bbnum)
855 *return_head_blk = 0;
856 else
857 *return_head_blk = head_blk;
858 /*
859 * When returning here, we have a good block number. Bad block
860 * means that during a previous crash, we didn't have a clean break
861 * from cycle number N to cycle number N-1. In this case, we need
862 * to find the first block with cycle number N-1.
863 */
864 return 0;
865
866 bp_err:
867 xlog_put_bp(bp);
868
869 if (error)
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100870 xfs_warn(log->l_mp, "failed to find log head");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 return error;
872}
873
874/*
875 * Find the sync block number or the tail of the log.
876 *
877 * This will be the block number of the last record to have its
878 * associated buffers synced to disk. Every log record header has
879 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
880 * to get a sync block number. The only concern is to figure out which
881 * log record header to believe.
882 *
883 * The following algorithm uses the log record header with the largest
884 * lsn. The entire log record does not need to be valid. We only care
885 * that the header is valid.
886 *
887 * We could speed up search by using current head_blk buffer, but it is not
888 * available.
889 */
Eric Sandeen5d77c0d2009-11-19 15:52:00 +0000890STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891xlog_find_tail(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500892 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 xfs_daddr_t *head_blk,
Eric Sandeen65be6052006-01-11 15:34:19 +1100894 xfs_daddr_t *tail_blk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895{
896 xlog_rec_header_t *rhead;
897 xlog_op_header_t *op_head;
898 xfs_caddr_t offset = NULL;
899 xfs_buf_t *bp;
900 int error, i, found;
901 xfs_daddr_t umount_data_blk;
902 xfs_daddr_t after_umount_blk;
903 xfs_lsn_t tail_lsn;
904 int hblks;
905
906 found = 0;
907
908 /*
909 * Find previous log record
910 */
911 if ((error = xlog_find_head(log, head_blk)))
912 return error;
913
914 bp = xlog_get_bp(log, 1);
915 if (!bp)
Dave Chinner24513372014-06-25 14:58:08 +1000916 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 if (*head_blk == 0) { /* special case */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100918 error = xlog_bread(log, 0, 1, bp, &offset);
919 if (error)
Alex Elder9db127e2010-04-15 18:17:26 +0000920 goto done;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100921
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000922 if (xlog_get_cycle(offset) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 *tail_blk = 0;
924 /* leave all other log inited values alone */
Alex Elder9db127e2010-04-15 18:17:26 +0000925 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 }
927 }
928
929 /*
930 * Search backwards looking for log record header block
931 */
932 ASSERT(*head_blk < INT_MAX);
933 for (i = (int)(*head_blk) - 1; i >= 0; i--) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100934 error = xlog_bread(log, i, 1, bp, &offset);
935 if (error)
Alex Elder9db127e2010-04-15 18:17:26 +0000936 goto done;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100937
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200938 if (*(__be32 *)offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 found = 1;
940 break;
941 }
942 }
943 /*
944 * If we haven't found the log record header block, start looking
945 * again from the end of the physical log. XXXmiken: There should be
946 * a check here to make sure we didn't search more than N blocks in
947 * the previous code.
948 */
949 if (!found) {
950 for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100951 error = xlog_bread(log, i, 1, bp, &offset);
952 if (error)
Alex Elder9db127e2010-04-15 18:17:26 +0000953 goto done;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100954
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200955 if (*(__be32 *)offset ==
956 cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 found = 2;
958 break;
959 }
960 }
961 }
962 if (!found) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100963 xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__);
Eric Sandeen050a1952013-07-31 20:33:47 -0500964 xlog_put_bp(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 ASSERT(0);
Dave Chinner24513372014-06-25 14:58:08 +1000966 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 }
968
969 /* find blk_no of tail of log */
970 rhead = (xlog_rec_header_t *)offset;
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000971 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972
973 /*
974 * Reset log values according to the state of the log when we
975 * crashed. In the case where head_blk == 0, we bump curr_cycle
976 * one because the next write starts a new cycle rather than
977 * continuing the cycle of the last good log record. At this
978 * point we have guaranteed that all partial log records have been
979 * accounted for. Therefore, we know that the last good log record
980 * written was complete and ended exactly on the end boundary
981 * of the physical log.
982 */
983 log->l_prev_block = i;
984 log->l_curr_block = (int)*head_blk;
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000985 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 if (found == 2)
987 log->l_curr_cycle++;
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100988 atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn));
Dave Chinner84f3c682010-12-03 22:11:29 +1100989 atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn));
Christoph Hellwig28496962012-02-20 02:31:25 +0000990 xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle,
Dave Chinnera69ed032010-12-21 12:08:20 +1100991 BBTOB(log->l_curr_block));
Christoph Hellwig28496962012-02-20 02:31:25 +0000992 xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle,
Dave Chinnera69ed032010-12-21 12:08:20 +1100993 BBTOB(log->l_curr_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994
995 /*
996 * Look for unmount record. If we find it, then we know there
997 * was a clean unmount. Since 'i' could be the last block in
998 * the physical log, we convert to a log block before comparing
999 * to the head_blk.
1000 *
1001 * Save the current tail lsn to use to pass to
1002 * xlog_clear_stale_blocks() below. We won't want to clear the
1003 * unmount record if there is one, so we pass the lsn of the
1004 * unmount record rather than the block after it.
1005 */
Eric Sandeen62118702008-03-06 13:44:28 +11001006 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001007 int h_size = be32_to_cpu(rhead->h_size);
1008 int h_version = be32_to_cpu(rhead->h_version);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009
1010 if ((h_version & XLOG_VERSION_2) &&
1011 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
1012 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
1013 if (h_size % XLOG_HEADER_CYCLE_SIZE)
1014 hblks++;
1015 } else {
1016 hblks = 1;
1017 }
1018 } else {
1019 hblks = 1;
1020 }
1021 after_umount_blk = (i + hblks + (int)
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001022 BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize;
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001023 tail_lsn = atomic64_read(&log->l_tail_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 if (*head_blk == after_umount_blk &&
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001025 be32_to_cpu(rhead->h_num_logops) == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 umount_data_blk = (i + hblks) % log->l_logBBsize;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001027 error = xlog_bread(log, umount_data_blk, 1, bp, &offset);
1028 if (error)
Alex Elder9db127e2010-04-15 18:17:26 +00001029 goto done;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001030
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 op_head = (xlog_op_header_t *)offset;
1032 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
1033 /*
1034 * Set tail and last sync so that newly written
1035 * log records will point recovery to after the
1036 * current unmount record.
1037 */
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001038 xlog_assign_atomic_lsn(&log->l_tail_lsn,
1039 log->l_curr_cycle, after_umount_blk);
1040 xlog_assign_atomic_lsn(&log->l_last_sync_lsn,
1041 log->l_curr_cycle, after_umount_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 *tail_blk = after_umount_blk;
David Chinner92821e22007-05-24 15:26:31 +10001043
1044 /*
1045 * Note that the unmount was clean. If the unmount
1046 * was not clean, we need to know this to rebuild the
1047 * superblock counters from the perag headers if we
1048 * have a filesystem using non-persistent counters.
1049 */
1050 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 }
1052 }
1053
1054 /*
1055 * Make sure that there are no blocks in front of the head
1056 * with the same cycle number as the head. This can happen
1057 * because we allow multiple outstanding log writes concurrently,
1058 * and the later writes might make it out before earlier ones.
1059 *
1060 * We use the lsn from before modifying it so that we'll never
1061 * overwrite the unmount record after a clean unmount.
1062 *
1063 * Do this only if we are going to recover the filesystem
1064 *
1065 * NOTE: This used to say "if (!readonly)"
1066 * However on Linux, we can & do recover a read-only filesystem.
1067 * We only skip recovery if NORECOVERY is specified on mount,
1068 * in which case we would not be here.
1069 *
1070 * But... if the -device- itself is readonly, just skip this.
1071 * We can't recover this device anyway, so it won't matter.
1072 */
Alex Elder9db127e2010-04-15 18:17:26 +00001073 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 error = xlog_clear_stale_blocks(log, tail_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075
Alex Elder9db127e2010-04-15 18:17:26 +00001076done:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 xlog_put_bp(bp);
1078
1079 if (error)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001080 xfs_warn(log->l_mp, "failed to locate log tail");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 return error;
1082}
1083
1084/*
1085 * Is the log zeroed at all?
1086 *
1087 * The last binary search should be changed to perform an X block read
1088 * once X becomes small enough. You can then search linearly through
1089 * the X blocks. This will cut down on the number of reads we need to do.
1090 *
1091 * If the log is partially zeroed, this routine will pass back the blkno
1092 * of the first block with cycle number 0. It won't have a complete LR
1093 * preceding it.
1094 *
1095 * Return:
1096 * 0 => the log is completely written to
Dave Chinner24513372014-06-25 14:58:08 +10001097 * 1 => use *blk_no as the first block of the log
1098 * <0 => error has occurred
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 */
David Chinnera8272ce2007-11-23 16:28:09 +11001100STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101xlog_find_zeroed(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001102 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 xfs_daddr_t *blk_no)
1104{
1105 xfs_buf_t *bp;
1106 xfs_caddr_t offset;
1107 uint first_cycle, last_cycle;
1108 xfs_daddr_t new_blk, last_blk, start_blk;
1109 xfs_daddr_t num_scan_bblks;
1110 int error, log_bbnum = log->l_logBBsize;
1111
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10001112 *blk_no = 0;
1113
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 /* check totally zeroed log */
1115 bp = xlog_get_bp(log, 1);
1116 if (!bp)
Dave Chinner24513372014-06-25 14:58:08 +10001117 return -ENOMEM;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001118 error = xlog_bread(log, 0, 1, bp, &offset);
1119 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 goto bp_err;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001121
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10001122 first_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 if (first_cycle == 0) { /* completely zeroed log */
1124 *blk_no = 0;
1125 xlog_put_bp(bp);
Dave Chinner24513372014-06-25 14:58:08 +10001126 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 }
1128
1129 /* check partially zeroed log */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001130 error = xlog_bread(log, log_bbnum-1, 1, bp, &offset);
1131 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 goto bp_err;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001133
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10001134 last_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 if (last_cycle != 0) { /* log completely written to */
1136 xlog_put_bp(bp);
1137 return 0;
1138 } else if (first_cycle != 1) {
1139 /*
1140 * If the cycle of the last block is zero, the cycle of
1141 * the first block must be 1. If it's not, maybe we're
1142 * not looking at a log... Bail out.
1143 */
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001144 xfs_warn(log->l_mp,
1145 "Log inconsistent or not a log (last==0, first!=1)");
Dave Chinner24513372014-06-25 14:58:08 +10001146 error = -EINVAL;
Eric Sandeen5d0a6542013-07-31 20:32:30 -05001147 goto bp_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 }
1149
1150 /* we have a partially zeroed log */
1151 last_blk = log_bbnum-1;
1152 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1153 goto bp_err;
1154
1155 /*
1156 * Validate the answer. Because there is no way to guarantee that
1157 * the entire log is made up of log records which are the same size,
1158 * we scan over the defined maximum blocks. At this point, the maximum
1159 * is not chosen to mean anything special. XXXmiken
1160 */
1161 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1162 ASSERT(num_scan_bblks <= INT_MAX);
1163
1164 if (last_blk < num_scan_bblks)
1165 num_scan_bblks = last_blk;
1166 start_blk = last_blk - num_scan_bblks;
1167
1168 /*
1169 * We search for any instances of cycle number 0 that occur before
1170 * our current estimate of the head. What we're trying to detect is
1171 * 1 ... | 0 | 1 | 0...
1172 * ^ binary search ends here
1173 */
1174 if ((error = xlog_find_verify_cycle(log, start_blk,
1175 (int)num_scan_bblks, 0, &new_blk)))
1176 goto bp_err;
1177 if (new_blk != -1)
1178 last_blk = new_blk;
1179
1180 /*
1181 * Potentially backup over partial log record write. We don't need
1182 * to search the end of the log because we know it is zero.
1183 */
Dave Chinner24513372014-06-25 14:58:08 +10001184 error = xlog_find_verify_log_record(log, start_blk, &last_blk, 0);
1185 if (error == 1)
1186 error = -EIO;
1187 if (error)
1188 goto bp_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189
1190 *blk_no = last_blk;
1191bp_err:
1192 xlog_put_bp(bp);
1193 if (error)
1194 return error;
Dave Chinner24513372014-06-25 14:58:08 +10001195 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196}
1197
1198/*
1199 * These are simple subroutines used by xlog_clear_stale_blocks() below
1200 * to initialize a buffer full of empty log record headers and write
1201 * them into the log.
1202 */
1203STATIC void
1204xlog_add_record(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001205 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 xfs_caddr_t buf,
1207 int cycle,
1208 int block,
1209 int tail_cycle,
1210 int tail_block)
1211{
1212 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1213
1214 memset(buf, 0, BBSIZE);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001215 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1216 recp->h_cycle = cpu_to_be32(cycle);
1217 recp->h_version = cpu_to_be32(
Eric Sandeen62118702008-03-06 13:44:28 +11001218 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001219 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1220 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1221 recp->h_fmt = cpu_to_be32(XLOG_FMT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1223}
1224
1225STATIC int
1226xlog_write_log_records(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001227 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 int cycle,
1229 int start_block,
1230 int blocks,
1231 int tail_cycle,
1232 int tail_block)
1233{
1234 xfs_caddr_t offset;
1235 xfs_buf_t *bp;
1236 int balign, ealign;
Alex Elder69ce58f2010-04-20 17:09:59 +10001237 int sectbb = log->l_sectBBsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 int end_block = start_block + blocks;
1239 int bufblks;
1240 int error = 0;
1241 int i, j = 0;
1242
Alex Elder6881a222010-04-13 15:22:29 +10001243 /*
1244 * Greedily allocate a buffer big enough to handle the full
1245 * range of basic blocks to be written. If that fails, try
1246 * a smaller size. We need to be able to write at least a
1247 * log sector, or we're out of luck.
1248 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 bufblks = 1 << ffs(blocks);
Dave Chinner81158e02012-04-27 19:45:22 +10001250 while (bufblks > log->l_logBBsize)
1251 bufblks >>= 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 while (!(bp = xlog_get_bp(log, bufblks))) {
1253 bufblks >>= 1;
Alex Elder69ce58f2010-04-20 17:09:59 +10001254 if (bufblks < sectbb)
Dave Chinner24513372014-06-25 14:58:08 +10001255 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 }
1257
1258 /* We may need to do a read at the start to fill in part of
1259 * the buffer in the starting sector not covered by the first
1260 * write below.
1261 */
Alex Elder5c17f532010-04-13 15:22:48 +10001262 balign = round_down(start_block, sectbb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 if (balign != start_block) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001264 error = xlog_bread_noalign(log, start_block, 1, bp);
1265 if (error)
1266 goto out_put_bp;
1267
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 j = start_block - balign;
1269 }
1270
1271 for (i = start_block; i < end_block; i += bufblks) {
1272 int bcount, endcount;
1273
1274 bcount = min(bufblks, end_block - start_block);
1275 endcount = bcount - j;
1276
1277 /* We may need to do a read at the end to fill in part of
1278 * the buffer in the final sector not covered by the write.
1279 * If this is the same sector as the above read, skip it.
1280 */
Alex Elder5c17f532010-04-13 15:22:48 +10001281 ealign = round_down(end_block, sectbb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 if (j == 0 && (start_block + endcount > ealign)) {
Chandra Seetharaman62926042011-07-22 23:40:15 +00001283 offset = bp->b_addr + BBTOB(ealign - start_block);
Dave Chinner44396472011-04-21 09:34:27 +00001284 error = xlog_bread_offset(log, ealign, sectbb,
1285 bp, offset);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001286 if (error)
1287 break;
1288
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 }
1290
1291 offset = xlog_align(log, start_block, endcount, bp);
1292 for (; j < endcount; j++) {
1293 xlog_add_record(log, offset, cycle, i+j,
1294 tail_cycle, tail_block);
1295 offset += BBSIZE;
1296 }
1297 error = xlog_bwrite(log, start_block, endcount, bp);
1298 if (error)
1299 break;
1300 start_block += endcount;
1301 j = 0;
1302 }
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001303
1304 out_put_bp:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 xlog_put_bp(bp);
1306 return error;
1307}
1308
1309/*
1310 * This routine is called to blow away any incomplete log writes out
1311 * in front of the log head. We do this so that we won't become confused
1312 * if we come up, write only a little bit more, and then crash again.
1313 * If we leave the partial log records out there, this situation could
1314 * cause us to think those partial writes are valid blocks since they
1315 * have the current cycle number. We get rid of them by overwriting them
1316 * with empty log records with the old cycle number rather than the
1317 * current one.
1318 *
1319 * The tail lsn is passed in rather than taken from
1320 * the log so that we will not write over the unmount record after a
1321 * clean unmount in a 512 block log. Doing so would leave the log without
1322 * any valid log records in it until a new one was written. If we crashed
1323 * during that time we would not be able to recover.
1324 */
1325STATIC int
1326xlog_clear_stale_blocks(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001327 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 xfs_lsn_t tail_lsn)
1329{
1330 int tail_cycle, head_cycle;
1331 int tail_block, head_block;
1332 int tail_distance, max_distance;
1333 int distance;
1334 int error;
1335
1336 tail_cycle = CYCLE_LSN(tail_lsn);
1337 tail_block = BLOCK_LSN(tail_lsn);
1338 head_cycle = log->l_curr_cycle;
1339 head_block = log->l_curr_block;
1340
1341 /*
1342 * Figure out the distance between the new head of the log
1343 * and the tail. We want to write over any blocks beyond the
1344 * head that we may have written just before the crash, but
1345 * we don't want to overwrite the tail of the log.
1346 */
1347 if (head_cycle == tail_cycle) {
1348 /*
1349 * The tail is behind the head in the physical log,
1350 * so the distance from the head to the tail is the
1351 * distance from the head to the end of the log plus
1352 * the distance from the beginning of the log to the
1353 * tail.
1354 */
1355 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1356 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1357 XFS_ERRLEVEL_LOW, log->l_mp);
Dave Chinner24513372014-06-25 14:58:08 +10001358 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 }
1360 tail_distance = tail_block + (log->l_logBBsize - head_block);
1361 } else {
1362 /*
1363 * The head is behind the tail in the physical log,
1364 * so the distance from the head to the tail is just
1365 * the tail block minus the head block.
1366 */
1367 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1368 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1369 XFS_ERRLEVEL_LOW, log->l_mp);
Dave Chinner24513372014-06-25 14:58:08 +10001370 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 }
1372 tail_distance = tail_block - head_block;
1373 }
1374
1375 /*
1376 * If the head is right up against the tail, we can't clear
1377 * anything.
1378 */
1379 if (tail_distance <= 0) {
1380 ASSERT(tail_distance == 0);
1381 return 0;
1382 }
1383
1384 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1385 /*
1386 * Take the smaller of the maximum amount of outstanding I/O
1387 * we could have and the distance to the tail to clear out.
1388 * We take the smaller so that we don't overwrite the tail and
1389 * we don't waste all day writing from the head to the tail
1390 * for no reason.
1391 */
1392 max_distance = MIN(max_distance, tail_distance);
1393
1394 if ((head_block + max_distance) <= log->l_logBBsize) {
1395 /*
1396 * We can stomp all the blocks we need to without
1397 * wrapping around the end of the log. Just do it
1398 * in a single write. Use the cycle number of the
1399 * current cycle minus one so that the log will look like:
1400 * n ... | n - 1 ...
1401 */
1402 error = xlog_write_log_records(log, (head_cycle - 1),
1403 head_block, max_distance, tail_cycle,
1404 tail_block);
1405 if (error)
1406 return error;
1407 } else {
1408 /*
1409 * We need to wrap around the end of the physical log in
1410 * order to clear all the blocks. Do it in two separate
1411 * I/Os. The first write should be from the head to the
1412 * end of the physical log, and it should use the current
1413 * cycle number minus one just like above.
1414 */
1415 distance = log->l_logBBsize - head_block;
1416 error = xlog_write_log_records(log, (head_cycle - 1),
1417 head_block, distance, tail_cycle,
1418 tail_block);
1419
1420 if (error)
1421 return error;
1422
1423 /*
1424 * Now write the blocks at the start of the physical log.
1425 * This writes the remainder of the blocks we want to clear.
1426 * It uses the current cycle number since we're now on the
1427 * same cycle as the head so that we get:
1428 * n ... n ... | n - 1 ...
1429 * ^^^^^ blocks we're writing
1430 */
1431 distance = max_distance - (log->l_logBBsize - head_block);
1432 error = xlog_write_log_records(log, head_cycle, 0, distance,
1433 tail_cycle, tail_block);
1434 if (error)
1435 return error;
1436 }
1437
1438 return 0;
1439}
1440
1441/******************************************************************************
1442 *
1443 * Log recover routines
1444 *
1445 ******************************************************************************
1446 */
1447
1448STATIC xlog_recover_t *
1449xlog_recover_find_tid(
Dave Chinnerf0a76952010-01-11 11:49:57 +00001450 struct hlist_head *head,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 xlog_tid_t tid)
1452{
Dave Chinnerf0a76952010-01-11 11:49:57 +00001453 xlog_recover_t *trans;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454
Sasha Levinb67bfe02013-02-27 17:06:00 -08001455 hlist_for_each_entry(trans, head, r_list) {
Dave Chinnerf0a76952010-01-11 11:49:57 +00001456 if (trans->r_log_tid == tid)
1457 return trans;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 }
Dave Chinnerf0a76952010-01-11 11:49:57 +00001459 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460}
1461
1462STATIC void
Dave Chinnerf0a76952010-01-11 11:49:57 +00001463xlog_recover_new_tid(
1464 struct hlist_head *head,
1465 xlog_tid_t tid,
1466 xfs_lsn_t lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467{
Dave Chinnerf0a76952010-01-11 11:49:57 +00001468 xlog_recover_t *trans;
1469
1470 trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
1471 trans->r_log_tid = tid;
1472 trans->r_lsn = lsn;
1473 INIT_LIST_HEAD(&trans->r_itemq);
1474
1475 INIT_HLIST_NODE(&trans->r_list);
1476 hlist_add_head(&trans->r_list, head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477}
1478
1479STATIC void
1480xlog_recover_add_item(
Dave Chinnerf0a76952010-01-11 11:49:57 +00001481 struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482{
1483 xlog_recover_item_t *item;
1484
1485 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
Dave Chinnerf0a76952010-01-11 11:49:57 +00001486 INIT_LIST_HEAD(&item->ri_list);
1487 list_add_tail(&item->ri_list, head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488}
1489
1490STATIC int
1491xlog_recover_add_to_cont_trans(
Mark Tinguelyad223e62012-06-14 09:22:15 -05001492 struct xlog *log,
1493 struct xlog_recover *trans,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 xfs_caddr_t dp,
1495 int len)
1496{
1497 xlog_recover_item_t *item;
1498 xfs_caddr_t ptr, old_ptr;
1499 int old_len;
1500
Dave Chinnerf0a76952010-01-11 11:49:57 +00001501 if (list_empty(&trans->r_itemq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 /* finish copying rest of trans header */
1503 xlog_recover_add_item(&trans->r_itemq);
1504 ptr = (xfs_caddr_t) &trans->r_theader +
1505 sizeof(xfs_trans_header_t) - len;
1506 memcpy(ptr, dp, len); /* d, s, l */
1507 return 0;
1508 }
Dave Chinnerf0a76952010-01-11 11:49:57 +00001509 /* take the tail entry */
1510 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511
1512 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
1513 old_len = item->ri_buf[item->ri_cnt-1].i_len;
1514
Mitsuo Hayasaka45053602012-01-27 06:37:26 +00001515 ptr = kmem_realloc(old_ptr, len+old_len, old_len, KM_SLEEP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 memcpy(&ptr[old_len], dp, len); /* d, s, l */
1517 item->ri_buf[item->ri_cnt-1].i_len += len;
1518 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
Dave Chinner9abbc532010-04-13 15:06:46 +10001519 trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 return 0;
1521}
1522
1523/*
1524 * The next region to add is the start of a new region. It could be
1525 * a whole region or it could be the first part of a new region. Because
1526 * of this, the assumption here is that the type and size fields of all
1527 * format structures fit into the first 32 bits of the structure.
1528 *
1529 * This works because all regions must be 32 bit aligned. Therefore, we
1530 * either have both fields or we have neither field. In the case we have
1531 * neither field, the data part of the region is zero length. We only have
1532 * a log_op_header and can throw away the header since a new one will appear
1533 * later. If we have at least 4 bytes, then we can determine how many regions
1534 * will appear in the current log item.
1535 */
1536STATIC int
1537xlog_recover_add_to_trans(
Mark Tinguelyad223e62012-06-14 09:22:15 -05001538 struct xlog *log,
1539 struct xlog_recover *trans,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 xfs_caddr_t dp,
1541 int len)
1542{
1543 xfs_inode_log_format_t *in_f; /* any will do */
1544 xlog_recover_item_t *item;
1545 xfs_caddr_t ptr;
1546
1547 if (!len)
1548 return 0;
Dave Chinnerf0a76952010-01-11 11:49:57 +00001549 if (list_empty(&trans->r_itemq)) {
David Chinner5a792c42008-10-30 17:40:09 +11001550 /* we need to catch log corruptions here */
1551 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001552 xfs_warn(log->l_mp, "%s: bad header magic number",
1553 __func__);
David Chinner5a792c42008-10-30 17:40:09 +11001554 ASSERT(0);
Dave Chinner24513372014-06-25 14:58:08 +10001555 return -EIO;
David Chinner5a792c42008-10-30 17:40:09 +11001556 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 if (len == sizeof(xfs_trans_header_t))
1558 xlog_recover_add_item(&trans->r_itemq);
1559 memcpy(&trans->r_theader, dp, len); /* d, s, l */
1560 return 0;
1561 }
1562
1563 ptr = kmem_alloc(len, KM_SLEEP);
1564 memcpy(ptr, dp, len);
1565 in_f = (xfs_inode_log_format_t *)ptr;
1566
Dave Chinnerf0a76952010-01-11 11:49:57 +00001567 /* take the tail entry */
1568 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
1569 if (item->ri_total != 0 &&
1570 item->ri_total == item->ri_cnt) {
1571 /* tail item is in use, get a new one */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 xlog_recover_add_item(&trans->r_itemq);
Dave Chinnerf0a76952010-01-11 11:49:57 +00001573 item = list_entry(trans->r_itemq.prev,
1574 xlog_recover_item_t, ri_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576
1577 if (item->ri_total == 0) { /* first region to be added */
Christoph Hellwige8fa6b42009-03-03 14:48:36 -05001578 if (in_f->ilf_size == 0 ||
1579 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001580 xfs_warn(log->l_mp,
1581 "bad number of regions (%d) in inode log format",
Christoph Hellwige8fa6b42009-03-03 14:48:36 -05001582 in_f->ilf_size);
1583 ASSERT(0);
Thierry Redingaaaae982013-10-01 16:47:53 +02001584 kmem_free(ptr);
Dave Chinner24513372014-06-25 14:58:08 +10001585 return -EIO;
Christoph Hellwige8fa6b42009-03-03 14:48:36 -05001586 }
1587
1588 item->ri_total = in_f->ilf_size;
1589 item->ri_buf =
1590 kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
1591 KM_SLEEP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 }
1593 ASSERT(item->ri_total > item->ri_cnt);
1594 /* Description region is ri_buf[0] */
1595 item->ri_buf[item->ri_cnt].i_addr = ptr;
1596 item->ri_buf[item->ri_cnt].i_len = len;
1597 item->ri_cnt++;
Dave Chinner9abbc532010-04-13 15:06:46 +10001598 trace_xfs_log_recover_item_add(log, trans, item, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 return 0;
1600}
1601
Dave Chinnerf0a76952010-01-11 11:49:57 +00001602/*
Dave Chinnera775ad72013-06-05 12:09:07 +10001603 * Sort the log items in the transaction.
1604 *
1605 * The ordering constraints are defined by the inode allocation and unlink
1606 * behaviour. The rules are:
1607 *
1608 * 1. Every item is only logged once in a given transaction. Hence it
1609 * represents the last logged state of the item. Hence ordering is
1610 * dependent on the order in which operations need to be performed so
1611 * required initial conditions are always met.
1612 *
1613 * 2. Cancelled buffers are recorded in pass 1 in a separate table and
1614 * there's nothing to replay from them so we can simply cull them
1615 * from the transaction. However, we can't do that until after we've
1616 * replayed all the other items because they may be dependent on the
1617 * cancelled buffer and replaying the cancelled buffer can remove it
1618 * form the cancelled buffer table. Hence they have tobe done last.
1619 *
1620 * 3. Inode allocation buffers must be replayed before inode items that
Dave Chinner28c8e412013-06-27 16:04:55 +10001621 * read the buffer and replay changes into it. For filesystems using the
1622 * ICREATE transactions, this means XFS_LI_ICREATE objects need to get
1623 * treated the same as inode allocation buffers as they create and
1624 * initialise the buffers directly.
Dave Chinnera775ad72013-06-05 12:09:07 +10001625 *
1626 * 4. Inode unlink buffers must be replayed after inode items are replayed.
1627 * This ensures that inodes are completely flushed to the inode buffer
1628 * in a "free" state before we remove the unlinked inode list pointer.
1629 *
1630 * Hence the ordering needs to be inode allocation buffers first, inode items
1631 * second, inode unlink buffers third and cancelled buffers last.
1632 *
1633 * But there's a problem with that - we can't tell an inode allocation buffer
1634 * apart from a regular buffer, so we can't separate them. We can, however,
1635 * tell an inode unlink buffer from the others, and so we can separate them out
1636 * from all the other buffers and move them to last.
1637 *
1638 * Hence, 4 lists, in order from head to tail:
Dave Chinner28c8e412013-06-27 16:04:55 +10001639 * - buffer_list for all buffers except cancelled/inode unlink buffers
1640 * - item_list for all non-buffer items
1641 * - inode_buffer_list for inode unlink buffers
1642 * - cancel_list for the cancelled buffers
1643 *
1644 * Note that we add objects to the tail of the lists so that first-to-last
1645 * ordering is preserved within the lists. Adding objects to the head of the
1646 * list means when we traverse from the head we walk them in last-to-first
1647 * order. For cancelled buffers and inode unlink buffers this doesn't matter,
1648 * but for all other items there may be specific ordering that we need to
1649 * preserve.
Dave Chinnerf0a76952010-01-11 11:49:57 +00001650 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651STATIC int
1652xlog_recover_reorder_trans(
Mark Tinguelyad223e62012-06-14 09:22:15 -05001653 struct xlog *log,
1654 struct xlog_recover *trans,
Dave Chinner9abbc532010-04-13 15:06:46 +10001655 int pass)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656{
Dave Chinnerf0a76952010-01-11 11:49:57 +00001657 xlog_recover_item_t *item, *n;
Mark Tinguely2a841082013-10-02 07:51:12 -05001658 int error = 0;
Dave Chinnerf0a76952010-01-11 11:49:57 +00001659 LIST_HEAD(sort_list);
Dave Chinnera775ad72013-06-05 12:09:07 +10001660 LIST_HEAD(cancel_list);
1661 LIST_HEAD(buffer_list);
1662 LIST_HEAD(inode_buffer_list);
1663 LIST_HEAD(inode_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664
Dave Chinnerf0a76952010-01-11 11:49:57 +00001665 list_splice_init(&trans->r_itemq, &sort_list);
1666 list_for_each_entry_safe(item, n, &sort_list, ri_list) {
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10001667 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
Dave Chinnerf0a76952010-01-11 11:49:57 +00001668
1669 switch (ITEM_TYPE(item)) {
Dave Chinner28c8e412013-06-27 16:04:55 +10001670 case XFS_LI_ICREATE:
1671 list_move_tail(&item->ri_list, &buffer_list);
1672 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 case XFS_LI_BUF:
Dave Chinnera775ad72013-06-05 12:09:07 +10001674 if (buf_f->blf_flags & XFS_BLF_CANCEL) {
Dave Chinner9abbc532010-04-13 15:06:46 +10001675 trace_xfs_log_recover_item_reorder_head(log,
1676 trans, item, pass);
Dave Chinnera775ad72013-06-05 12:09:07 +10001677 list_move(&item->ri_list, &cancel_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 break;
1679 }
Dave Chinnera775ad72013-06-05 12:09:07 +10001680 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
1681 list_move(&item->ri_list, &inode_buffer_list);
1682 break;
1683 }
1684 list_move_tail(&item->ri_list, &buffer_list);
1685 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 case XFS_LI_INODE:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 case XFS_LI_DQUOT:
1688 case XFS_LI_QUOTAOFF:
1689 case XFS_LI_EFD:
1690 case XFS_LI_EFI:
Dave Chinner9abbc532010-04-13 15:06:46 +10001691 trace_xfs_log_recover_item_reorder_tail(log,
1692 trans, item, pass);
Dave Chinnera775ad72013-06-05 12:09:07 +10001693 list_move_tail(&item->ri_list, &inode_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 break;
1695 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001696 xfs_warn(log->l_mp,
1697 "%s: unrecognized type of log operation",
1698 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 ASSERT(0);
Mark Tinguely2a841082013-10-02 07:51:12 -05001700 /*
1701 * return the remaining items back to the transaction
1702 * item list so they can be freed in caller.
1703 */
1704 if (!list_empty(&sort_list))
1705 list_splice_init(&sort_list, &trans->r_itemq);
Dave Chinner24513372014-06-25 14:58:08 +10001706 error = -EIO;
Mark Tinguely2a841082013-10-02 07:51:12 -05001707 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 }
Dave Chinnerf0a76952010-01-11 11:49:57 +00001709 }
Mark Tinguely2a841082013-10-02 07:51:12 -05001710out:
Dave Chinnerf0a76952010-01-11 11:49:57 +00001711 ASSERT(list_empty(&sort_list));
Dave Chinnera775ad72013-06-05 12:09:07 +10001712 if (!list_empty(&buffer_list))
1713 list_splice(&buffer_list, &trans->r_itemq);
1714 if (!list_empty(&inode_list))
1715 list_splice_tail(&inode_list, &trans->r_itemq);
1716 if (!list_empty(&inode_buffer_list))
1717 list_splice_tail(&inode_buffer_list, &trans->r_itemq);
1718 if (!list_empty(&cancel_list))
1719 list_splice_tail(&cancel_list, &trans->r_itemq);
Mark Tinguely2a841082013-10-02 07:51:12 -05001720 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721}
1722
1723/*
1724 * Build up the table of buf cancel records so that we don't replay
1725 * cancelled data in the second pass. For buffer records that are
1726 * not cancel records, there is nothing to do here so we just return.
1727 *
1728 * If we get a cancel record which is already in the table, this indicates
1729 * that the buffer was cancelled multiple times. In order to ensure
1730 * that during pass 2 we keep the record in the table until we reach its
1731 * last occurrence in the log, we keep a reference count in the cancel
1732 * record in the table to tell us how many times we expect to see this
1733 * record during the second pass.
1734 */
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001735STATIC int
1736xlog_recover_buffer_pass1(
Mark Tinguelyad223e62012-06-14 09:22:15 -05001737 struct xlog *log,
1738 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001740 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001741 struct list_head *bucket;
1742 struct xfs_buf_cancel *bcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743
1744 /*
1745 * If this isn't a cancel buffer item, then just return.
1746 */
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001747 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
Dave Chinner9abbc532010-04-13 15:06:46 +10001748 trace_xfs_log_recover_buf_not_cancel(log, buf_f);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001749 return 0;
Dave Chinner9abbc532010-04-13 15:06:46 +10001750 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751
1752 /*
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001753 * Insert an xfs_buf_cancel record into the hash table of them.
1754 * If there is already an identical record, bump its reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 */
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001756 bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno);
1757 list_for_each_entry(bcp, bucket, bc_list) {
1758 if (bcp->bc_blkno == buf_f->blf_blkno &&
1759 bcp->bc_len == buf_f->blf_len) {
1760 bcp->bc_refcount++;
Dave Chinner9abbc532010-04-13 15:06:46 +10001761 trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001762 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 }
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001765
1766 bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP);
1767 bcp->bc_blkno = buf_f->blf_blkno;
1768 bcp->bc_len = buf_f->blf_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 bcp->bc_refcount = 1;
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001770 list_add_tail(&bcp->bc_list, bucket);
1771
Dave Chinner9abbc532010-04-13 15:06:46 +10001772 trace_xfs_log_recover_buf_cancel_add(log, buf_f);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001773 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774}
1775
1776/*
1777 * Check to see whether the buffer being recovered has a corresponding
Dave Chinner84a5b732013-08-27 08:10:53 +10001778 * entry in the buffer cancel record table. If it is, return the cancel
1779 * buffer structure to the caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 */
Dave Chinner84a5b732013-08-27 08:10:53 +10001781STATIC struct xfs_buf_cancel *
1782xlog_peek_buffer_cancelled(
Mark Tinguelyad223e62012-06-14 09:22:15 -05001783 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 xfs_daddr_t blkno,
1785 uint len,
1786 ushort flags)
1787{
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001788 struct list_head *bucket;
1789 struct xfs_buf_cancel *bcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790
Dave Chinner84a5b732013-08-27 08:10:53 +10001791 if (!log->l_buf_cancel_table) {
1792 /* empty table means no cancelled buffers in the log */
Dave Chinnerc1155412010-05-07 11:05:19 +10001793 ASSERT(!(flags & XFS_BLF_CANCEL));
Dave Chinner84a5b732013-08-27 08:10:53 +10001794 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 }
1796
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001797 bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno);
1798 list_for_each_entry(bcp, bucket, bc_list) {
1799 if (bcp->bc_blkno == blkno && bcp->bc_len == len)
Dave Chinner84a5b732013-08-27 08:10:53 +10001800 return bcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 }
1802
1803 /*
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001804 * We didn't find a corresponding entry in the table, so return 0 so
1805 * that the buffer is NOT cancelled.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806 */
Dave Chinnerc1155412010-05-07 11:05:19 +10001807 ASSERT(!(flags & XFS_BLF_CANCEL));
Dave Chinner84a5b732013-08-27 08:10:53 +10001808 return NULL;
1809}
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001810
Dave Chinner84a5b732013-08-27 08:10:53 +10001811/*
1812 * If the buffer is being cancelled then return 1 so that it will be cancelled,
1813 * otherwise return 0. If the buffer is actually a buffer cancel item
1814 * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the
1815 * table and remove it from the table if this is the last reference.
1816 *
1817 * We remove the cancel record from the table when we encounter its last
1818 * occurrence in the log so that if the same buffer is re-used again after its
1819 * last cancellation we actually replay the changes made at that point.
1820 */
1821STATIC int
1822xlog_check_buffer_cancelled(
1823 struct xlog *log,
1824 xfs_daddr_t blkno,
1825 uint len,
1826 ushort flags)
1827{
1828 struct xfs_buf_cancel *bcp;
1829
1830 bcp = xlog_peek_buffer_cancelled(log, blkno, len, flags);
1831 if (!bcp)
1832 return 0;
1833
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001834 /*
1835 * We've go a match, so return 1 so that the recovery of this buffer
1836 * is cancelled. If this buffer is actually a buffer cancel log
1837 * item, then decrement the refcount on the one in the table and
1838 * remove it if this is the last reference.
1839 */
1840 if (flags & XFS_BLF_CANCEL) {
1841 if (--bcp->bc_refcount == 0) {
1842 list_del(&bcp->bc_list);
1843 kmem_free(bcp);
1844 }
1845 }
1846 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847}
1848
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849/*
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001850 * Perform recovery for a buffer full of inodes. In these buffers, the only
1851 * data which should be recovered is that which corresponds to the
1852 * di_next_unlinked pointers in the on disk inode structures. The rest of the
1853 * data for the inodes is always logged through the inodes themselves rather
1854 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 *
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001856 * The only time when buffers full of inodes are fully recovered is when the
1857 * buffer is full of newly allocated inodes. In this case the buffer will
1858 * not be marked as an inode buffer and so will be sent to
1859 * xlog_recover_do_reg_buffer() below during recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 */
1861STATIC int
1862xlog_recover_do_inode_buffer(
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001863 struct xfs_mount *mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864 xlog_recover_item_t *item,
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001865 struct xfs_buf *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 xfs_buf_log_format_t *buf_f)
1867{
1868 int i;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001869 int item_index = 0;
1870 int bit = 0;
1871 int nbits = 0;
1872 int reg_buf_offset = 0;
1873 int reg_buf_bytes = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 int next_unlinked_offset;
1875 int inodes_per_buf;
1876 xfs_agino_t *logged_nextp;
1877 xfs_agino_t *buffer_nextp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878
Dave Chinner9abbc532010-04-13 15:06:46 +10001879 trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f);
Dave Chinner9222a9c2013-06-12 12:19:06 +10001880
1881 /*
1882 * Post recovery validation only works properly on CRC enabled
1883 * filesystems.
1884 */
1885 if (xfs_sb_version_hascrc(&mp->m_sb))
1886 bp->b_ops = &xfs_inode_buf_ops;
Dave Chinner9abbc532010-04-13 15:06:46 +10001887
Dave Chinneraa0e8832012-04-23 15:58:52 +10001888 inodes_per_buf = BBTOB(bp->b_io_length) >> mp->m_sb.sb_inodelog;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 for (i = 0; i < inodes_per_buf; i++) {
1890 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
1891 offsetof(xfs_dinode_t, di_next_unlinked);
1892
1893 while (next_unlinked_offset >=
1894 (reg_buf_offset + reg_buf_bytes)) {
1895 /*
1896 * The next di_next_unlinked field is beyond
1897 * the current logged region. Find the next
1898 * logged region that contains or is beyond
1899 * the current di_next_unlinked field.
1900 */
1901 bit += nbits;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001902 bit = xfs_next_bit(buf_f->blf_data_map,
1903 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904
1905 /*
1906 * If there are no more logged regions in the
1907 * buffer, then we're done.
1908 */
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001909 if (bit == -1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001912 nbits = xfs_contig_bits(buf_f->blf_data_map,
1913 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 ASSERT(nbits > 0);
Dave Chinnerc1155412010-05-07 11:05:19 +10001915 reg_buf_offset = bit << XFS_BLF_SHIFT;
1916 reg_buf_bytes = nbits << XFS_BLF_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 item_index++;
1918 }
1919
1920 /*
1921 * If the current logged region starts after the current
1922 * di_next_unlinked field, then move on to the next
1923 * di_next_unlinked field.
1924 */
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001925 if (next_unlinked_offset < reg_buf_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927
1928 ASSERT(item->ri_buf[item_index].i_addr != NULL);
Dave Chinnerc1155412010-05-07 11:05:19 +10001929 ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0);
Dave Chinneraa0e8832012-04-23 15:58:52 +10001930 ASSERT((reg_buf_offset + reg_buf_bytes) <=
1931 BBTOB(bp->b_io_length));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932
1933 /*
1934 * The current logged region contains a copy of the
1935 * current di_next_unlinked field. Extract its value
1936 * and copy it to the buffer copy.
1937 */
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10001938 logged_nextp = item->ri_buf[item_index].i_addr +
1939 next_unlinked_offset - reg_buf_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940 if (unlikely(*logged_nextp == 0)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001941 xfs_alert(mp,
1942 "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
1943 "Trying to replay bad (0) inode di_next_unlinked field.",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 item, bp);
1945 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1946 XFS_ERRLEVEL_LOW, mp);
Dave Chinner24513372014-06-25 14:58:08 +10001947 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948 }
1949
1950 buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
1951 next_unlinked_offset);
Tim Shimmin87c199c2006-06-09 14:56:16 +10001952 *buffer_nextp = *logged_nextp;
Dave Chinner0a32c262013-06-05 12:09:08 +10001953
1954 /*
1955 * If necessary, recalculate the CRC in the on-disk inode. We
1956 * have to leave the inode in a consistent state for whoever
1957 * reads it next....
1958 */
1959 xfs_dinode_calc_crc(mp, (struct xfs_dinode *)
1960 xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize));
1961
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 }
1963
1964 return 0;
1965}
1966
1967/*
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001968 * V5 filesystems know the age of the buffer on disk being recovered. We can
1969 * have newer objects on disk than we are replaying, and so for these cases we
1970 * don't want to replay the current change as that will make the buffer contents
1971 * temporarily invalid on disk.
1972 *
1973 * The magic number might not match the buffer type we are going to recover
1974 * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags. Hence
1975 * extract the LSN of the existing object in the buffer based on it's current
1976 * magic number. If we don't recognise the magic number in the buffer, then
1977 * return a LSN of -1 so that the caller knows it was an unrecognised block and
1978 * so can recover the buffer.
Dave Chinner566055d2013-09-24 16:01:16 +10001979 *
1980 * Note: we cannot rely solely on magic number matches to determine that the
1981 * buffer has a valid LSN - we also need to verify that it belongs to this
1982 * filesystem, so we need to extract the object's LSN and compare it to that
1983 * which we read from the superblock. If the UUIDs don't match, then we've got a
1984 * stale metadata block from an old filesystem instance that we need to recover
1985 * over the top of.
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001986 */
1987static xfs_lsn_t
1988xlog_recover_get_buf_lsn(
1989 struct xfs_mount *mp,
1990 struct xfs_buf *bp)
1991{
1992 __uint32_t magic32;
1993 __uint16_t magic16;
1994 __uint16_t magicda;
1995 void *blk = bp->b_addr;
Dave Chinner566055d2013-09-24 16:01:16 +10001996 uuid_t *uuid;
1997 xfs_lsn_t lsn = -1;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001998
1999 /* v4 filesystems always recover immediately */
2000 if (!xfs_sb_version_hascrc(&mp->m_sb))
2001 goto recover_immediately;
2002
2003 magic32 = be32_to_cpu(*(__be32 *)blk);
2004 switch (magic32) {
2005 case XFS_ABTB_CRC_MAGIC:
2006 case XFS_ABTC_CRC_MAGIC:
2007 case XFS_ABTB_MAGIC:
2008 case XFS_ABTC_MAGIC:
2009 case XFS_IBT_CRC_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10002010 case XFS_IBT_MAGIC: {
2011 struct xfs_btree_block *btb = blk;
2012
2013 lsn = be64_to_cpu(btb->bb_u.s.bb_lsn);
2014 uuid = &btb->bb_u.s.bb_uuid;
2015 break;
2016 }
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002017 case XFS_BMAP_CRC_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10002018 case XFS_BMAP_MAGIC: {
2019 struct xfs_btree_block *btb = blk;
2020
2021 lsn = be64_to_cpu(btb->bb_u.l.bb_lsn);
2022 uuid = &btb->bb_u.l.bb_uuid;
2023 break;
2024 }
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002025 case XFS_AGF_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10002026 lsn = be64_to_cpu(((struct xfs_agf *)blk)->agf_lsn);
2027 uuid = &((struct xfs_agf *)blk)->agf_uuid;
2028 break;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002029 case XFS_AGFL_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10002030 lsn = be64_to_cpu(((struct xfs_agfl *)blk)->agfl_lsn);
2031 uuid = &((struct xfs_agfl *)blk)->agfl_uuid;
2032 break;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002033 case XFS_AGI_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10002034 lsn = be64_to_cpu(((struct xfs_agi *)blk)->agi_lsn);
2035 uuid = &((struct xfs_agi *)blk)->agi_uuid;
2036 break;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002037 case XFS_SYMLINK_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10002038 lsn = be64_to_cpu(((struct xfs_dsymlink_hdr *)blk)->sl_lsn);
2039 uuid = &((struct xfs_dsymlink_hdr *)blk)->sl_uuid;
2040 break;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002041 case XFS_DIR3_BLOCK_MAGIC:
2042 case XFS_DIR3_DATA_MAGIC:
2043 case XFS_DIR3_FREE_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10002044 lsn = be64_to_cpu(((struct xfs_dir3_blk_hdr *)blk)->lsn);
2045 uuid = &((struct xfs_dir3_blk_hdr *)blk)->uuid;
2046 break;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002047 case XFS_ATTR3_RMT_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10002048 lsn = be64_to_cpu(((struct xfs_attr3_rmt_hdr *)blk)->rm_lsn);
2049 uuid = &((struct xfs_attr3_rmt_hdr *)blk)->rm_uuid;
2050 break;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002051 case XFS_SB_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10002052 lsn = be64_to_cpu(((struct xfs_dsb *)blk)->sb_lsn);
2053 uuid = &((struct xfs_dsb *)blk)->sb_uuid;
2054 break;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002055 default:
2056 break;
2057 }
2058
Dave Chinner566055d2013-09-24 16:01:16 +10002059 if (lsn != (xfs_lsn_t)-1) {
2060 if (!uuid_equal(&mp->m_sb.sb_uuid, uuid))
2061 goto recover_immediately;
2062 return lsn;
2063 }
2064
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002065 magicda = be16_to_cpu(((struct xfs_da_blkinfo *)blk)->magic);
2066 switch (magicda) {
2067 case XFS_DIR3_LEAF1_MAGIC:
2068 case XFS_DIR3_LEAFN_MAGIC:
2069 case XFS_DA3_NODE_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10002070 lsn = be64_to_cpu(((struct xfs_da3_blkinfo *)blk)->lsn);
2071 uuid = &((struct xfs_da3_blkinfo *)blk)->uuid;
2072 break;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002073 default:
2074 break;
2075 }
2076
Dave Chinner566055d2013-09-24 16:01:16 +10002077 if (lsn != (xfs_lsn_t)-1) {
2078 if (!uuid_equal(&mp->m_sb.sb_uuid, uuid))
2079 goto recover_immediately;
2080 return lsn;
2081 }
2082
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002083 /*
2084 * We do individual object checks on dquot and inode buffers as they
2085 * have their own individual LSN records. Also, we could have a stale
2086 * buffer here, so we have to at least recognise these buffer types.
2087 *
2088 * A notd complexity here is inode unlinked list processing - it logs
2089 * the inode directly in the buffer, but we don't know which inodes have
2090 * been modified, and there is no global buffer LSN. Hence we need to
2091 * recover all inode buffer types immediately. This problem will be
2092 * fixed by logical logging of the unlinked list modifications.
2093 */
2094 magic16 = be16_to_cpu(*(__be16 *)blk);
2095 switch (magic16) {
2096 case XFS_DQUOT_MAGIC:
2097 case XFS_DINODE_MAGIC:
2098 goto recover_immediately;
2099 default:
2100 break;
2101 }
2102
2103 /* unknown buffer contents, recover immediately */
2104
2105recover_immediately:
2106 return (xfs_lsn_t)-1;
2107
2108}
2109
2110/*
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002111 * Validate the recovered buffer is of the correct type and attach the
2112 * appropriate buffer operations to them for writeback. Magic numbers are in a
2113 * few places:
2114 * the first 16 bits of the buffer (inode buffer, dquot buffer),
2115 * the first 32 bits of the buffer (most blocks),
2116 * inside a struct xfs_da_blkinfo at the start of the buffer.
2117 */
2118static void
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002119xlog_recover_validate_buf_type(
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002120 struct xfs_mount *mp,
2121 struct xfs_buf *bp,
2122 xfs_buf_log_format_t *buf_f)
2123{
2124 struct xfs_da_blkinfo *info = bp->b_addr;
2125 __uint32_t magic32;
2126 __uint16_t magic16;
2127 __uint16_t magicda;
2128
2129 magic32 = be32_to_cpu(*(__be32 *)bp->b_addr);
2130 magic16 = be16_to_cpu(*(__be16*)bp->b_addr);
2131 magicda = be16_to_cpu(info->magic);
Dave Chinner61fe1352013-04-03 16:11:30 +11002132 switch (xfs_blft_from_flags(buf_f)) {
2133 case XFS_BLFT_BTREE_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002134 switch (magic32) {
2135 case XFS_ABTB_CRC_MAGIC:
2136 case XFS_ABTC_CRC_MAGIC:
2137 case XFS_ABTB_MAGIC:
2138 case XFS_ABTC_MAGIC:
2139 bp->b_ops = &xfs_allocbt_buf_ops;
2140 break;
2141 case XFS_IBT_CRC_MAGIC:
Brian Fosteraafc3c22014-04-24 16:00:52 +10002142 case XFS_FIBT_CRC_MAGIC:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002143 case XFS_IBT_MAGIC:
Brian Fosteraafc3c22014-04-24 16:00:52 +10002144 case XFS_FIBT_MAGIC:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002145 bp->b_ops = &xfs_inobt_buf_ops;
2146 break;
2147 case XFS_BMAP_CRC_MAGIC:
2148 case XFS_BMAP_MAGIC:
2149 bp->b_ops = &xfs_bmbt_buf_ops;
2150 break;
2151 default:
2152 xfs_warn(mp, "Bad btree block magic!");
2153 ASSERT(0);
2154 break;
2155 }
2156 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002157 case XFS_BLFT_AGF_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002158 if (magic32 != XFS_AGF_MAGIC) {
2159 xfs_warn(mp, "Bad AGF block magic!");
2160 ASSERT(0);
2161 break;
2162 }
2163 bp->b_ops = &xfs_agf_buf_ops;
2164 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002165 case XFS_BLFT_AGFL_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002166 if (!xfs_sb_version_hascrc(&mp->m_sb))
2167 break;
2168 if (magic32 != XFS_AGFL_MAGIC) {
2169 xfs_warn(mp, "Bad AGFL block magic!");
2170 ASSERT(0);
2171 break;
2172 }
2173 bp->b_ops = &xfs_agfl_buf_ops;
2174 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002175 case XFS_BLFT_AGI_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002176 if (magic32 != XFS_AGI_MAGIC) {
2177 xfs_warn(mp, "Bad AGI block magic!");
2178 ASSERT(0);
2179 break;
2180 }
2181 bp->b_ops = &xfs_agi_buf_ops;
2182 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002183 case XFS_BLFT_UDQUOT_BUF:
2184 case XFS_BLFT_PDQUOT_BUF:
2185 case XFS_BLFT_GDQUOT_BUF:
Dave Chinner123887e2013-04-30 21:39:33 +10002186#ifdef CONFIG_XFS_QUOTA
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002187 if (magic16 != XFS_DQUOT_MAGIC) {
2188 xfs_warn(mp, "Bad DQUOT block magic!");
2189 ASSERT(0);
2190 break;
2191 }
2192 bp->b_ops = &xfs_dquot_buf_ops;
Dave Chinner123887e2013-04-30 21:39:33 +10002193#else
2194 xfs_alert(mp,
2195 "Trying to recover dquots without QUOTA support built in!");
2196 ASSERT(0);
2197#endif
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002198 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002199 case XFS_BLFT_DINO_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002200 /*
2201 * we get here with inode allocation buffers, not buffers that
2202 * track unlinked list changes.
2203 */
2204 if (magic16 != XFS_DINODE_MAGIC) {
2205 xfs_warn(mp, "Bad INODE block magic!");
2206 ASSERT(0);
2207 break;
2208 }
2209 bp->b_ops = &xfs_inode_buf_ops;
2210 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002211 case XFS_BLFT_SYMLINK_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002212 if (magic32 != XFS_SYMLINK_MAGIC) {
2213 xfs_warn(mp, "Bad symlink block magic!");
2214 ASSERT(0);
2215 break;
2216 }
2217 bp->b_ops = &xfs_symlink_buf_ops;
2218 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002219 case XFS_BLFT_DIR_BLOCK_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002220 if (magic32 != XFS_DIR2_BLOCK_MAGIC &&
2221 magic32 != XFS_DIR3_BLOCK_MAGIC) {
2222 xfs_warn(mp, "Bad dir block magic!");
2223 ASSERT(0);
2224 break;
2225 }
2226 bp->b_ops = &xfs_dir3_block_buf_ops;
2227 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002228 case XFS_BLFT_DIR_DATA_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002229 if (magic32 != XFS_DIR2_DATA_MAGIC &&
2230 magic32 != XFS_DIR3_DATA_MAGIC) {
2231 xfs_warn(mp, "Bad dir data magic!");
2232 ASSERT(0);
2233 break;
2234 }
2235 bp->b_ops = &xfs_dir3_data_buf_ops;
2236 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002237 case XFS_BLFT_DIR_FREE_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002238 if (magic32 != XFS_DIR2_FREE_MAGIC &&
2239 magic32 != XFS_DIR3_FREE_MAGIC) {
2240 xfs_warn(mp, "Bad dir3 free magic!");
2241 ASSERT(0);
2242 break;
2243 }
2244 bp->b_ops = &xfs_dir3_free_buf_ops;
2245 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002246 case XFS_BLFT_DIR_LEAF1_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002247 if (magicda != XFS_DIR2_LEAF1_MAGIC &&
2248 magicda != XFS_DIR3_LEAF1_MAGIC) {
2249 xfs_warn(mp, "Bad dir leaf1 magic!");
2250 ASSERT(0);
2251 break;
2252 }
2253 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
2254 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002255 case XFS_BLFT_DIR_LEAFN_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002256 if (magicda != XFS_DIR2_LEAFN_MAGIC &&
2257 magicda != XFS_DIR3_LEAFN_MAGIC) {
2258 xfs_warn(mp, "Bad dir leafn magic!");
2259 ASSERT(0);
2260 break;
2261 }
2262 bp->b_ops = &xfs_dir3_leafn_buf_ops;
2263 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002264 case XFS_BLFT_DA_NODE_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002265 if (magicda != XFS_DA_NODE_MAGIC &&
2266 magicda != XFS_DA3_NODE_MAGIC) {
2267 xfs_warn(mp, "Bad da node magic!");
2268 ASSERT(0);
2269 break;
2270 }
2271 bp->b_ops = &xfs_da3_node_buf_ops;
2272 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002273 case XFS_BLFT_ATTR_LEAF_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002274 if (magicda != XFS_ATTR_LEAF_MAGIC &&
2275 magicda != XFS_ATTR3_LEAF_MAGIC) {
2276 xfs_warn(mp, "Bad attr leaf magic!");
2277 ASSERT(0);
2278 break;
2279 }
2280 bp->b_ops = &xfs_attr3_leaf_buf_ops;
2281 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002282 case XFS_BLFT_ATTR_RMT_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002283 if (!xfs_sb_version_hascrc(&mp->m_sb))
2284 break;
Dave Chinnercab09a82013-04-30 21:39:36 +10002285 if (magic32 != XFS_ATTR3_RMT_MAGIC) {
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002286 xfs_warn(mp, "Bad attr remote magic!");
2287 ASSERT(0);
2288 break;
2289 }
2290 bp->b_ops = &xfs_attr3_rmt_buf_ops;
2291 break;
Dave Chinner04a1e6c2013-04-03 16:11:31 +11002292 case XFS_BLFT_SB_BUF:
2293 if (magic32 != XFS_SB_MAGIC) {
2294 xfs_warn(mp, "Bad SB block magic!");
2295 ASSERT(0);
2296 break;
2297 }
2298 bp->b_ops = &xfs_sb_buf_ops;
2299 break;
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002300 default:
Dave Chinner61fe1352013-04-03 16:11:30 +11002301 xfs_warn(mp, "Unknown buffer type %d!",
2302 xfs_blft_from_flags(buf_f));
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002303 break;
2304 }
2305}
2306
2307/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 * Perform a 'normal' buffer recovery. Each logged region of the
2309 * buffer should be copied over the corresponding region in the
2310 * given buffer. The bitmap in the buf log format structure indicates
2311 * where to place the logged data.
2312 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313STATIC void
2314xlog_recover_do_reg_buffer(
Dave Chinner9abbc532010-04-13 15:06:46 +10002315 struct xfs_mount *mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 xlog_recover_item_t *item,
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002317 struct xfs_buf *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318 xfs_buf_log_format_t *buf_f)
2319{
2320 int i;
2321 int bit;
2322 int nbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 int error;
2324
Dave Chinner9abbc532010-04-13 15:06:46 +10002325 trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f);
2326
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327 bit = 0;
2328 i = 1; /* 0 is the buf format structure */
2329 while (1) {
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002330 bit = xfs_next_bit(buf_f->blf_data_map,
2331 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332 if (bit == -1)
2333 break;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002334 nbits = xfs_contig_bits(buf_f->blf_data_map,
2335 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 ASSERT(nbits > 0);
Christoph Hellwig4b809162007-08-16 15:37:36 +10002337 ASSERT(item->ri_buf[i].i_addr != NULL);
Dave Chinnerc1155412010-05-07 11:05:19 +10002338 ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0);
Dave Chinneraa0e8832012-04-23 15:58:52 +10002339 ASSERT(BBTOB(bp->b_io_length) >=
2340 ((uint)bit << XFS_BLF_SHIFT) + (nbits << XFS_BLF_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341
2342 /*
Dave Chinner709da6a2013-05-27 16:38:23 +10002343 * The dirty regions logged in the buffer, even though
2344 * contiguous, may span multiple chunks. This is because the
2345 * dirty region may span a physical page boundary in a buffer
2346 * and hence be split into two separate vectors for writing into
2347 * the log. Hence we need to trim nbits back to the length of
2348 * the current region being copied out of the log.
2349 */
2350 if (item->ri_buf[i].i_len < (nbits << XFS_BLF_SHIFT))
2351 nbits = item->ri_buf[i].i_len >> XFS_BLF_SHIFT;
2352
2353 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354 * Do a sanity check if this is a dquot buffer. Just checking
2355 * the first dquot in the buffer should do. XXXThis is
2356 * probably a good thing to do for other buf types also.
2357 */
2358 error = 0;
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002359 if (buf_f->blf_flags &
Dave Chinnerc1155412010-05-07 11:05:19 +10002360 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002361 if (item->ri_buf[i].i_addr == NULL) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002362 xfs_alert(mp,
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002363 "XFS: NULL dquot in %s.", __func__);
2364 goto next;
2365 }
Jan Rekorajski8ec6dba2009-11-16 11:57:02 +00002366 if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002367 xfs_alert(mp,
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002368 "XFS: dquot too small (%d) in %s.",
2369 item->ri_buf[i].i_len, __func__);
2370 goto next;
2371 }
Dave Chinner9aede1d2013-10-15 09:17:52 +11002372 error = xfs_dqcheck(mp, item->ri_buf[i].i_addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373 -1, 0, XFS_QMOPT_DOWARN,
2374 "dquot_buf_recover");
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002375 if (error)
2376 goto next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377 }
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002378
2379 memcpy(xfs_buf_offset(bp,
Dave Chinnerc1155412010-05-07 11:05:19 +10002380 (uint)bit << XFS_BLF_SHIFT), /* dest */
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002381 item->ri_buf[i].i_addr, /* source */
Dave Chinnerc1155412010-05-07 11:05:19 +10002382 nbits<<XFS_BLF_SHIFT); /* length */
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002383 next:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384 i++;
2385 bit += nbits;
2386 }
2387
2388 /* Shouldn't be any more regions */
2389 ASSERT(i == item->ri_total);
Christoph Hellwigee1a47a2013-04-21 14:53:46 -05002390
Dave Chinner9222a9c2013-06-12 12:19:06 +10002391 /*
2392 * We can only do post recovery validation on items on CRC enabled
2393 * fielsystems as we need to know when the buffer was written to be able
2394 * to determine if we should have replayed the item. If we replay old
2395 * metadata over a newer buffer, then it will enter a temporarily
2396 * inconsistent state resulting in verification failures. Hence for now
2397 * just avoid the verification stage for non-crc filesystems
2398 */
2399 if (xfs_sb_version_hascrc(&mp->m_sb))
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002400 xlog_recover_validate_buf_type(mp, bp, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401}
2402
2403/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404 * Perform a dquot buffer recovery.
Zhi Yong Wu8ba701e2013-08-12 03:15:01 +00002405 * Simple algorithm: if we have found a QUOTAOFF log item of the same type
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2407 * Else, treat it as a regular buffer and do recovery.
2408 */
2409STATIC void
2410xlog_recover_do_dquot_buffer(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002411 struct xfs_mount *mp,
2412 struct xlog *log,
2413 struct xlog_recover_item *item,
2414 struct xfs_buf *bp,
2415 struct xfs_buf_log_format *buf_f)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416{
2417 uint type;
2418
Dave Chinner9abbc532010-04-13 15:06:46 +10002419 trace_xfs_log_recover_buf_dquot_buf(log, buf_f);
2420
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 /*
2422 * Filesystems are required to send in quota flags at mount time.
2423 */
2424 if (mp->m_qflags == 0) {
2425 return;
2426 }
2427
2428 type = 0;
Dave Chinnerc1155412010-05-07 11:05:19 +10002429 if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430 type |= XFS_DQ_USER;
Dave Chinnerc1155412010-05-07 11:05:19 +10002431 if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF)
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002432 type |= XFS_DQ_PROJ;
Dave Chinnerc1155412010-05-07 11:05:19 +10002433 if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434 type |= XFS_DQ_GROUP;
2435 /*
2436 * This type of quotas was turned off, so ignore this buffer
2437 */
2438 if (log->l_quotaoffs_flag & type)
2439 return;
2440
Dave Chinner9abbc532010-04-13 15:06:46 +10002441 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442}
2443
2444/*
2445 * This routine replays a modification made to a buffer at runtime.
2446 * There are actually two types of buffer, regular and inode, which
2447 * are handled differently. Inode buffers are handled differently
2448 * in that we only recover a specific set of data from them, namely
2449 * the inode di_next_unlinked fields. This is because all other inode
2450 * data is actually logged via inode records and any data we replay
2451 * here which overlaps that may be stale.
2452 *
2453 * When meta-data buffers are freed at run time we log a buffer item
Dave Chinnerc1155412010-05-07 11:05:19 +10002454 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455 * of the buffer in the log should not be replayed at recovery time.
2456 * This is so that if the blocks covered by the buffer are reused for
2457 * file data before we crash we don't end up replaying old, freed
2458 * meta-data into a user's file.
2459 *
2460 * To handle the cancellation of buffer log items, we make two passes
2461 * over the log during recovery. During the first we build a table of
2462 * those buffers which have been cancelled, and during the second we
2463 * only replay those buffers which do not have corresponding cancel
Zhi Yong Wu34be5ff2013-08-07 10:11:07 +00002464 * records in the table. See xlog_recover_buffer_pass[1,2] above
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465 * for more details on the implementation of the table of cancel records.
2466 */
2467STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002468xlog_recover_buffer_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002469 struct xlog *log,
2470 struct list_head *buffer_list,
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002471 struct xlog_recover_item *item,
2472 xfs_lsn_t current_lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473{
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002474 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002475 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476 xfs_buf_t *bp;
2477 int error;
Christoph Hellwig6ad112b2009-11-24 18:02:23 +00002478 uint buf_flags;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002479 xfs_lsn_t lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002481 /*
2482 * In this pass we only want to recover all the buffers which have
2483 * not been cancelled and are not cancellation buffers themselves.
2484 */
2485 if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno,
2486 buf_f->blf_len, buf_f->blf_flags)) {
2487 trace_xfs_log_recover_buf_cancel(log, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489 }
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002490
Dave Chinner9abbc532010-04-13 15:06:46 +10002491 trace_xfs_log_recover_buf_recover(log, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492
Dave Chinnera8acad72012-04-23 15:58:54 +10002493 buf_flags = 0;
Dave Chinner611c9942012-04-23 15:59:07 +10002494 if (buf_f->blf_flags & XFS_BLF_INODE_BUF)
2495 buf_flags |= XBF_UNMAPPED;
Christoph Hellwig6ad112b2009-11-24 18:02:23 +00002496
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002497 bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len,
Dave Chinnerc3f8fc72012-11-12 22:54:01 +11002498 buf_flags, NULL);
Chandra Seetharamanac4d6882011-08-03 02:18:29 +00002499 if (!bp)
Dave Chinner24513372014-06-25 14:58:08 +10002500 return -ENOMEM;
Chandra Seetharamane5702802011-08-03 02:18:34 +00002501 error = bp->b_error;
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +00002502 if (error) {
Christoph Hellwig901796a2011-10-10 16:52:49 +00002503 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)");
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002504 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505 }
2506
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002507 /*
2508 * recover the buffer only if we get an LSN from it and it's less than
2509 * the lsn of the transaction we are replaying.
2510 */
2511 lsn = xlog_recover_get_buf_lsn(mp, bp);
2512 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0)
2513 goto out_release;
2514
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002515 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002517 } else if (buf_f->blf_flags &
Dave Chinnerc1155412010-05-07 11:05:19 +10002518 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519 xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2520 } else {
Dave Chinner9abbc532010-04-13 15:06:46 +10002521 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522 }
2523 if (error)
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002524 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525
2526 /*
2527 * Perform delayed write on the buffer. Asynchronous writes will be
2528 * slower when taking into account all the buffers to be flushed.
2529 *
2530 * Also make sure that only inode buffers with good sizes stay in
2531 * the buffer cache. The kernel moves inodes in buffers of 1 block
Jie Liu0f49efd2013-12-13 15:51:48 +11002532 * or mp->m_inode_cluster_size bytes, whichever is bigger. The inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533 * buffers in the log can be a different size if the log was generated
2534 * by an older kernel using unclustered inode buffers or a newer kernel
2535 * running with a different inode cluster size. Regardless, if the
Jie Liu0f49efd2013-12-13 15:51:48 +11002536 * the inode buffer size isn't MAX(blocksize, mp->m_inode_cluster_size)
2537 * for *our* value of mp->m_inode_cluster_size, then we need to keep
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 * the buffer out of the buffer cache so that the buffer won't
2539 * overlap with future reads of those inodes.
2540 */
2541 if (XFS_DINODE_MAGIC ==
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002542 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
Dave Chinneraa0e8832012-04-23 15:58:52 +10002543 (BBTOB(bp->b_io_length) != MAX(log->l_mp->m_sb.sb_blocksize,
Jie Liu0f49efd2013-12-13 15:51:48 +11002544 (__uint32_t)log->l_mp->m_inode_cluster_size))) {
Christoph Hellwigc867cb62011-10-10 16:52:46 +00002545 xfs_buf_stale(bp);
Christoph Hellwigc2b006c2011-08-23 08:28:07 +00002546 error = xfs_bwrite(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547 } else {
Dave Chinnerebad8612010-09-22 10:47:20 +10002548 ASSERT(bp->b_target->bt_mount == mp);
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02002549 bp->b_iodone = xlog_recover_iodone;
Christoph Hellwig43ff2122012-04-23 15:58:39 +10002550 xfs_buf_delwri_queue(bp, buffer_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 }
2552
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002553out_release:
Christoph Hellwigc2b006c2011-08-23 08:28:07 +00002554 xfs_buf_relse(bp);
2555 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556}
2557
Dave Chinner638f44162013-08-30 10:23:45 +10002558/*
2559 * Inode fork owner changes
2560 *
2561 * If we have been told that we have to reparent the inode fork, it's because an
2562 * extent swap operation on a CRC enabled filesystem has been done and we are
2563 * replaying it. We need to walk the BMBT of the appropriate fork and change the
2564 * owners of it.
2565 *
2566 * The complexity here is that we don't have an inode context to work with, so
2567 * after we've replayed the inode we need to instantiate one. This is where the
2568 * fun begins.
2569 *
2570 * We are in the middle of log recovery, so we can't run transactions. That
2571 * means we cannot use cache coherent inode instantiation via xfs_iget(), as
2572 * that will result in the corresponding iput() running the inode through
2573 * xfs_inactive(). If we've just replayed an inode core that changes the link
2574 * count to zero (i.e. it's been unlinked), then xfs_inactive() will run
2575 * transactions (bad!).
2576 *
2577 * So, to avoid this, we instantiate an inode directly from the inode core we've
2578 * just recovered. We have the buffer still locked, and all we really need to
2579 * instantiate is the inode core and the forks being modified. We can do this
2580 * manually, then run the inode btree owner change, and then tear down the
2581 * xfs_inode without having to run any transactions at all.
2582 *
2583 * Also, because we don't have a transaction context available here but need to
2584 * gather all the buffers we modify for writeback so we pass the buffer_list
2585 * instead for the operation to use.
2586 */
2587
2588STATIC int
2589xfs_recover_inode_owner_change(
2590 struct xfs_mount *mp,
2591 struct xfs_dinode *dip,
2592 struct xfs_inode_log_format *in_f,
2593 struct list_head *buffer_list)
2594{
2595 struct xfs_inode *ip;
2596 int error;
2597
2598 ASSERT(in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER));
2599
2600 ip = xfs_inode_alloc(mp, in_f->ilf_ino);
2601 if (!ip)
Dave Chinner24513372014-06-25 14:58:08 +10002602 return -ENOMEM;
Dave Chinner638f44162013-08-30 10:23:45 +10002603
2604 /* instantiate the inode */
2605 xfs_dinode_from_disk(&ip->i_d, dip);
2606 ASSERT(ip->i_d.di_version >= 3);
2607
2608 error = xfs_iformat_fork(ip, dip);
2609 if (error)
2610 goto out_free_ip;
2611
2612
2613 if (in_f->ilf_fields & XFS_ILOG_DOWNER) {
2614 ASSERT(in_f->ilf_fields & XFS_ILOG_DBROOT);
2615 error = xfs_bmbt_change_owner(NULL, ip, XFS_DATA_FORK,
2616 ip->i_ino, buffer_list);
2617 if (error)
2618 goto out_free_ip;
2619 }
2620
2621 if (in_f->ilf_fields & XFS_ILOG_AOWNER) {
2622 ASSERT(in_f->ilf_fields & XFS_ILOG_ABROOT);
2623 error = xfs_bmbt_change_owner(NULL, ip, XFS_ATTR_FORK,
2624 ip->i_ino, buffer_list);
2625 if (error)
2626 goto out_free_ip;
2627 }
2628
2629out_free_ip:
2630 xfs_inode_free(ip);
2631 return error;
2632}
2633
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002635xlog_recover_inode_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002636 struct xlog *log,
2637 struct list_head *buffer_list,
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002638 struct xlog_recover_item *item,
2639 xfs_lsn_t current_lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640{
2641 xfs_inode_log_format_t *in_f;
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002642 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643 xfs_buf_t *bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644 xfs_dinode_t *dip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645 int len;
2646 xfs_caddr_t src;
2647 xfs_caddr_t dest;
2648 int error;
2649 int attr_index;
2650 uint fields;
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002651 xfs_icdinode_t *dicp;
Christoph Hellwig93848a92013-04-03 16:11:17 +11002652 uint isize;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002653 int need_free = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654
Tim Shimmin6d192a92006-06-09 14:55:38 +10002655 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002656 in_f = item->ri_buf[0].i_addr;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002657 } else {
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002658 in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002659 need_free = 1;
2660 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2661 if (error)
2662 goto error;
2663 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002664
2665 /*
2666 * Inode buffers can be freed, look out for it,
2667 * and do not replay the inode.
2668 */
Christoph Hellwiga1941892008-11-28 14:23:40 +11002669 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
2670 in_f->ilf_len, 0)) {
Tim Shimmin6d192a92006-06-09 14:55:38 +10002671 error = 0;
Dave Chinner9abbc532010-04-13 15:06:46 +10002672 trace_xfs_log_recover_inode_cancel(log, in_f);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002673 goto error;
2674 }
Dave Chinner9abbc532010-04-13 15:06:46 +10002675 trace_xfs_log_recover_inode_recover(log, in_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676
Dave Chinnerc3f8fc72012-11-12 22:54:01 +11002677 bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len, 0,
Christoph Hellwig93848a92013-04-03 16:11:17 +11002678 &xfs_inode_buf_ops);
Chandra Seetharamanac4d6882011-08-03 02:18:29 +00002679 if (!bp) {
Dave Chinner24513372014-06-25 14:58:08 +10002680 error = -ENOMEM;
Chandra Seetharamanac4d6882011-08-03 02:18:29 +00002681 goto error;
2682 }
Chandra Seetharamane5702802011-08-03 02:18:34 +00002683 error = bp->b_error;
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +00002684 if (error) {
Christoph Hellwig901796a2011-10-10 16:52:49 +00002685 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)");
Dave Chinner638f44162013-08-30 10:23:45 +10002686 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
Christoph Hellwiga1941892008-11-28 14:23:40 +11002689 dip = (xfs_dinode_t *)xfs_buf_offset(bp, in_f->ilf_boffset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690
2691 /*
2692 * Make sure the place we're flushing out to really looks
2693 * like an inode!
2694 */
Christoph Hellwig69ef9212011-07-08 14:36:05 +02002695 if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002696 xfs_alert(mp,
2697 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
2698 __func__, dip, bp, in_f->ilf_ino);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002699 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700 XFS_ERRLEVEL_LOW, mp);
Dave Chinner24513372014-06-25 14:58:08 +10002701 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002702 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703 }
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002704 dicp = item->ri_buf[1].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002706 xfs_alert(mp,
2707 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
2708 __func__, item, in_f->ilf_ino);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002709 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710 XFS_ERRLEVEL_LOW, mp);
Dave Chinner24513372014-06-25 14:58:08 +10002711 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002712 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713 }
2714
Dave Chinnere60896d2013-07-24 15:47:30 +10002715 /*
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002716 * If the inode has an LSN in it, recover the inode only if it's less
Dave Chinner638f44162013-08-30 10:23:45 +10002717 * than the lsn of the transaction we are replaying. Note: we still
2718 * need to replay an owner change even though the inode is more recent
2719 * than the transaction as there is no guarantee that all the btree
2720 * blocks are more recent than this transaction, too.
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002721 */
2722 if (dip->di_version >= 3) {
2723 xfs_lsn_t lsn = be64_to_cpu(dip->di_lsn);
2724
2725 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
2726 trace_xfs_log_recover_inode_skip(log, in_f);
2727 error = 0;
Dave Chinner638f44162013-08-30 10:23:45 +10002728 goto out_owner_change;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002729 }
2730 }
2731
2732 /*
Dave Chinnere60896d2013-07-24 15:47:30 +10002733 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
2734 * are transactional and if ordering is necessary we can determine that
2735 * more accurately by the LSN field in the V3 inode core. Don't trust
2736 * the inode versions we might be changing them here - use the
2737 * superblock flag to determine whether we need to look at di_flushiter
2738 * to skip replay when the on disk inode is newer than the log one
2739 */
2740 if (!xfs_sb_version_hascrc(&mp->m_sb) &&
2741 dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742 /*
2743 * Deal with the wrap case, DI_MAX_FLUSH is less
2744 * than smaller numbers
2745 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002746 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002747 dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748 /* do nothing */
2749 } else {
Dave Chinner9abbc532010-04-13 15:06:46 +10002750 trace_xfs_log_recover_inode_skip(log, in_f);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002751 error = 0;
Dave Chinner638f44162013-08-30 10:23:45 +10002752 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002753 }
2754 }
Dave Chinnere60896d2013-07-24 15:47:30 +10002755
Linus Torvalds1da177e2005-04-16 15:20:36 -07002756 /* Take the opportunity to reset the flush iteration count */
2757 dicp->di_flushiter = 0;
2758
Al Viroabbede12011-07-26 02:31:30 -04002759 if (unlikely(S_ISREG(dicp->di_mode))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2761 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002762 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763 XFS_ERRLEVEL_LOW, mp, dicp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002764 xfs_alert(mp,
2765 "%s: Bad regular inode log record, rec ptr 0x%p, "
2766 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2767 __func__, item, dip, bp, in_f->ilf_ino);
Dave Chinner24513372014-06-25 14:58:08 +10002768 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002769 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770 }
Al Viroabbede12011-07-26 02:31:30 -04002771 } else if (unlikely(S_ISDIR(dicp->di_mode))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002772 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2773 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2774 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002775 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776 XFS_ERRLEVEL_LOW, mp, dicp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002777 xfs_alert(mp,
2778 "%s: Bad dir inode log record, rec ptr 0x%p, "
2779 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2780 __func__, item, dip, bp, in_f->ilf_ino);
Dave Chinner24513372014-06-25 14:58:08 +10002781 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002782 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002783 }
2784 }
2785 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002786 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787 XFS_ERRLEVEL_LOW, mp, dicp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002788 xfs_alert(mp,
2789 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2790 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2791 __func__, item, dip, bp, in_f->ilf_ino,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002792 dicp->di_nextents + dicp->di_anextents,
2793 dicp->di_nblocks);
Dave Chinner24513372014-06-25 14:58:08 +10002794 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002795 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002796 }
2797 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002798 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002799 XFS_ERRLEVEL_LOW, mp, dicp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002800 xfs_alert(mp,
2801 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2802 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__,
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002803 item, dip, bp, in_f->ilf_ino, dicp->di_forkoff);
Dave Chinner24513372014-06-25 14:58:08 +10002804 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002805 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002806 }
Christoph Hellwig93848a92013-04-03 16:11:17 +11002807 isize = xfs_icdinode_size(dicp->di_version);
2808 if (unlikely(item->ri_buf[1].i_len > isize)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002809 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810 XFS_ERRLEVEL_LOW, mp, dicp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002811 xfs_alert(mp,
2812 "%s: Bad inode log record length %d, rec ptr 0x%p",
2813 __func__, item->ri_buf[1].i_len, item);
Dave Chinner24513372014-06-25 14:58:08 +10002814 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002815 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816 }
2817
2818 /* The core is in in-core format */
Christoph Hellwig93848a92013-04-03 16:11:17 +11002819 xfs_dinode_to_disk(dip, dicp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820
2821 /* the rest is in on-disk format */
Christoph Hellwig93848a92013-04-03 16:11:17 +11002822 if (item->ri_buf[1].i_len > isize) {
2823 memcpy((char *)dip + isize,
2824 item->ri_buf[1].i_addr + isize,
2825 item->ri_buf[1].i_len - isize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826 }
2827
2828 fields = in_f->ilf_fields;
2829 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2830 case XFS_ILOG_DEV:
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002831 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002832 break;
2833 case XFS_ILOG_UUID:
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002834 memcpy(XFS_DFORK_DPTR(dip),
2835 &in_f->ilf_u.ilfu_uuid,
2836 sizeof(uuid_t));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837 break;
2838 }
2839
2840 if (in_f->ilf_size == 2)
Dave Chinner638f44162013-08-30 10:23:45 +10002841 goto out_owner_change;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002842 len = item->ri_buf[2].i_len;
2843 src = item->ri_buf[2].i_addr;
2844 ASSERT(in_f->ilf_size <= 4);
2845 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2846 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2847 (len == in_f->ilf_dsize));
2848
2849 switch (fields & XFS_ILOG_DFORK) {
2850 case XFS_ILOG_DDATA:
2851 case XFS_ILOG_DEXT:
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002852 memcpy(XFS_DFORK_DPTR(dip), src, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002853 break;
2854
2855 case XFS_ILOG_DBROOT:
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002856 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002857 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002858 XFS_DFORK_DSIZE(dip, mp));
2859 break;
2860
2861 default:
2862 /*
2863 * There are no data fork flags set.
2864 */
2865 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2866 break;
2867 }
2868
2869 /*
2870 * If we logged any attribute data, recover it. There may or
2871 * may not have been any other non-core data logged in this
2872 * transaction.
2873 */
2874 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2875 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2876 attr_index = 3;
2877 } else {
2878 attr_index = 2;
2879 }
2880 len = item->ri_buf[attr_index].i_len;
2881 src = item->ri_buf[attr_index].i_addr;
2882 ASSERT(len == in_f->ilf_asize);
2883
2884 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2885 case XFS_ILOG_ADATA:
2886 case XFS_ILOG_AEXT:
2887 dest = XFS_DFORK_APTR(dip);
2888 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2889 memcpy(dest, src, len);
2890 break;
2891
2892 case XFS_ILOG_ABROOT:
2893 dest = XFS_DFORK_APTR(dip);
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002894 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
2895 len, (xfs_bmdr_block_t*)dest,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896 XFS_DFORK_ASIZE(dip, mp));
2897 break;
2898
2899 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002900 xfs_warn(log->l_mp, "%s: Invalid flag", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002901 ASSERT(0);
Dave Chinner24513372014-06-25 14:58:08 +10002902 error = -EIO;
Dave Chinner638f44162013-08-30 10:23:45 +10002903 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002904 }
2905 }
2906
Dave Chinner638f44162013-08-30 10:23:45 +10002907out_owner_change:
2908 if (in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER))
2909 error = xfs_recover_inode_owner_change(mp, dip, in_f,
2910 buffer_list);
Christoph Hellwig93848a92013-04-03 16:11:17 +11002911 /* re-generate the checksum. */
2912 xfs_dinode_calc_crc(log->l_mp, dip);
2913
Dave Chinnerebad8612010-09-22 10:47:20 +10002914 ASSERT(bp->b_target->bt_mount == mp);
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02002915 bp->b_iodone = xlog_recover_iodone;
Christoph Hellwig43ff2122012-04-23 15:58:39 +10002916 xfs_buf_delwri_queue(bp, buffer_list);
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002917
2918out_release:
Christoph Hellwig61551f12011-08-23 08:28:06 +00002919 xfs_buf_relse(bp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002920error:
2921 if (need_free)
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002922 kmem_free(in_f);
Eric Sandeenb474c7a2014-06-22 15:04:54 +10002923 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924}
2925
2926/*
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002927 * Recover QUOTAOFF records. We simply make a note of it in the xlog
Linus Torvalds1da177e2005-04-16 15:20:36 -07002928 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2929 * of that type.
2930 */
2931STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002932xlog_recover_quotaoff_pass1(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002933 struct xlog *log,
2934 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002935{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002936 xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002937 ASSERT(qoff_f);
2938
2939 /*
2940 * The logitem format's flag tells us if this was user quotaoff,
Nathan Scott77a7cce2006-01-11 15:35:57 +11002941 * group/project quotaoff or both.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002942 */
2943 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2944 log->l_quotaoffs_flag |= XFS_DQ_USER;
Nathan Scott77a7cce2006-01-11 15:35:57 +11002945 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2946 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002947 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2948 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2949
Eric Sandeend99831f2014-06-22 15:03:54 +10002950 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002951}
2952
2953/*
2954 * Recover a dquot record
2955 */
2956STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002957xlog_recover_dquot_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002958 struct xlog *log,
2959 struct list_head *buffer_list,
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002960 struct xlog_recover_item *item,
2961 xfs_lsn_t current_lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002962{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002963 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002964 xfs_buf_t *bp;
2965 struct xfs_disk_dquot *ddq, *recddq;
2966 int error;
2967 xfs_dq_logformat_t *dq_f;
2968 uint type;
2969
Linus Torvalds1da177e2005-04-16 15:20:36 -07002970
2971 /*
2972 * Filesystems are required to send in quota flags at mount time.
2973 */
2974 if (mp->m_qflags == 0)
Eric Sandeend99831f2014-06-22 15:03:54 +10002975 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002976
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002977 recddq = item->ri_buf[1].i_addr;
2978 if (recddq == NULL) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002979 xfs_alert(log->l_mp, "NULL dquot in %s.", __func__);
Dave Chinner24513372014-06-25 14:58:08 +10002980 return -EIO;
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002981 }
Jan Rekorajski8ec6dba2009-11-16 11:57:02 +00002982 if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002983 xfs_alert(log->l_mp, "dquot too small (%d) in %s.",
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002984 item->ri_buf[1].i_len, __func__);
Dave Chinner24513372014-06-25 14:58:08 +10002985 return -EIO;
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002986 }
2987
Linus Torvalds1da177e2005-04-16 15:20:36 -07002988 /*
2989 * This type of quotas was turned off, so ignore this record.
2990 */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002991 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002992 ASSERT(type);
2993 if (log->l_quotaoffs_flag & type)
Eric Sandeend99831f2014-06-22 15:03:54 +10002994 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995
2996 /*
2997 * At this point we know that quota was _not_ turned off.
2998 * Since the mount flags are not indicating to us otherwise, this
2999 * must mean that quota is on, and the dquot needs to be replayed.
3000 * Remember that we may not have fully recovered the superblock yet,
3001 * so we can't do the usual trick of looking at the SB quota bits.
3002 *
3003 * The other possibility, of course, is that the quota subsystem was
3004 * removed since the last mount - ENOSYS.
3005 */
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10003006 dq_f = item->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003007 ASSERT(dq_f);
Dave Chinner9aede1d2013-10-15 09:17:52 +11003008 error = xfs_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003009 "xlog_recover_dquot_pass2 (log copy)");
3010 if (error)
Dave Chinner24513372014-06-25 14:58:08 +10003011 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003012 ASSERT(dq_f->qlf_len == 1);
3013
Dave Chinner7ca790a2012-04-23 15:58:55 +10003014 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dq_f->qlf_blkno,
Dave Chinnerc3f8fc72012-11-12 22:54:01 +11003015 XFS_FSB_TO_BB(mp, dq_f->qlf_len), 0, &bp,
3016 NULL);
Dave Chinner7ca790a2012-04-23 15:58:55 +10003017 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003018 return error;
Dave Chinner7ca790a2012-04-23 15:58:55 +10003019
Linus Torvalds1da177e2005-04-16 15:20:36 -07003020 ASSERT(bp);
3021 ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
3022
3023 /*
3024 * At least the magic num portion should be on disk because this
3025 * was among a chunk of dquots created earlier, and we did some
3026 * minimal initialization then.
3027 */
Dave Chinner9aede1d2013-10-15 09:17:52 +11003028 error = xfs_dqcheck(mp, ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003029 "xlog_recover_dquot_pass2");
3030 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003031 xfs_buf_relse(bp);
Dave Chinner24513372014-06-25 14:58:08 +10003032 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003033 }
3034
Dave Chinner50d5c8d2013-08-28 21:22:47 +10003035 /*
3036 * If the dquot has an LSN in it, recover the dquot only if it's less
3037 * than the lsn of the transaction we are replaying.
3038 */
3039 if (xfs_sb_version_hascrc(&mp->m_sb)) {
3040 struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddq;
3041 xfs_lsn_t lsn = be64_to_cpu(dqb->dd_lsn);
3042
3043 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
3044 goto out_release;
3045 }
3046 }
3047
Linus Torvalds1da177e2005-04-16 15:20:36 -07003048 memcpy(ddq, recddq, item->ri_buf[1].i_len);
Dave Chinner6fcdc592013-06-03 15:28:46 +10003049 if (xfs_sb_version_hascrc(&mp->m_sb)) {
3050 xfs_update_cksum((char *)ddq, sizeof(struct xfs_dqblk),
3051 XFS_DQUOT_CRC_OFF);
3052 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003053
3054 ASSERT(dq_f->qlf_size == 2);
Dave Chinnerebad8612010-09-22 10:47:20 +10003055 ASSERT(bp->b_target->bt_mount == mp);
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02003056 bp->b_iodone = xlog_recover_iodone;
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003057 xfs_buf_delwri_queue(bp, buffer_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058
Dave Chinner50d5c8d2013-08-28 21:22:47 +10003059out_release:
3060 xfs_buf_relse(bp);
3061 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003062}
3063
3064/*
3065 * This routine is called to create an in-core extent free intent
3066 * item from the efi format structure which was logged on disk.
3067 * It allocates an in-core efi, copies the extents from the format
3068 * structure into it, and adds the efi to the AIL with the given
3069 * LSN.
3070 */
Tim Shimmin6d192a92006-06-09 14:55:38 +10003071STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003072xlog_recover_efi_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003073 struct xlog *log,
3074 struct xlog_recover_item *item,
3075 xfs_lsn_t lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003076{
Tim Shimmin6d192a92006-06-09 14:55:38 +10003077 int error;
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003078 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003079 xfs_efi_log_item_t *efip;
3080 xfs_efi_log_format_t *efi_formatp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003081
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10003082 efi_formatp = item->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003083
Linus Torvalds1da177e2005-04-16 15:20:36 -07003084 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
Tim Shimmin6d192a92006-06-09 14:55:38 +10003085 if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
3086 &(efip->efi_format)))) {
3087 xfs_efi_item_free(efip);
3088 return error;
3089 }
Dave Chinnerb199c8a2010-12-20 11:59:49 +11003090 atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003091
David Chinnera9c21c12008-10-30 17:39:35 +11003092 spin_lock(&log->l_ailp->xa_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003093 /*
David Chinner783a2f62008-10-30 17:39:58 +11003094 * xfs_trans_ail_update() drops the AIL lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003095 */
Dave Chinnere6059942010-12-20 12:34:26 +11003096 xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn);
Tim Shimmin6d192a92006-06-09 14:55:38 +10003097 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003098}
3099
3100
3101/*
3102 * This routine is called when an efd format structure is found in
3103 * a committed transaction in the log. It's purpose is to cancel
3104 * the corresponding efi if it was still in the log. To do this
3105 * it searches the AIL for the efi with an id equal to that in the
3106 * efd format structure. If we find it, we remove the efi from the
3107 * AIL and free it.
3108 */
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003109STATIC int
3110xlog_recover_efd_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003111 struct xlog *log,
3112 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003113{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003114 xfs_efd_log_format_t *efd_formatp;
3115 xfs_efi_log_item_t *efip = NULL;
3116 xfs_log_item_t *lip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003117 __uint64_t efi_id;
David Chinner27d8d5f2008-10-30 17:38:39 +11003118 struct xfs_ail_cursor cur;
David Chinner783a2f62008-10-30 17:39:58 +11003119 struct xfs_ail *ailp = log->l_ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003120
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10003121 efd_formatp = item->ri_buf[0].i_addr;
Tim Shimmin6d192a92006-06-09 14:55:38 +10003122 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
3123 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
3124 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
3125 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003126 efi_id = efd_formatp->efd_efi_id;
3127
3128 /*
3129 * Search for the efi with the id in the efd format structure
3130 * in the AIL.
3131 */
David Chinnera9c21c12008-10-30 17:39:35 +11003132 spin_lock(&ailp->xa_lock);
3133 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003134 while (lip != NULL) {
3135 if (lip->li_type == XFS_LI_EFI) {
3136 efip = (xfs_efi_log_item_t *)lip;
3137 if (efip->efi_format.efi_id == efi_id) {
3138 /*
David Chinner783a2f62008-10-30 17:39:58 +11003139 * xfs_trans_ail_delete() drops the
Linus Torvalds1da177e2005-04-16 15:20:36 -07003140 * AIL lock.
3141 */
Dave Chinner04913fd2012-04-23 15:58:41 +10003142 xfs_trans_ail_delete(ailp, lip,
3143 SHUTDOWN_CORRUPT_INCORE);
David Chinner8ae2c0f2007-11-23 16:28:17 +11003144 xfs_efi_item_free(efip);
David Chinnera9c21c12008-10-30 17:39:35 +11003145 spin_lock(&ailp->xa_lock);
David Chinner27d8d5f2008-10-30 17:38:39 +11003146 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003147 }
3148 }
David Chinnera9c21c12008-10-30 17:39:35 +11003149 lip = xfs_trans_ail_cursor_next(ailp, &cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003150 }
Eric Sandeene4a1e292014-04-14 19:06:05 +10003151 xfs_trans_ail_cursor_done(&cur);
David Chinnera9c21c12008-10-30 17:39:35 +11003152 spin_unlock(&ailp->xa_lock);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003153
3154 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003155}
3156
3157/*
Dave Chinner28c8e412013-06-27 16:04:55 +10003158 * This routine is called when an inode create format structure is found in a
3159 * committed transaction in the log. It's purpose is to initialise the inodes
3160 * being allocated on disk. This requires us to get inode cluster buffers that
3161 * match the range to be intialised, stamped with inode templates and written
3162 * by delayed write so that subsequent modifications will hit the cached buffer
3163 * and only need writing out at the end of recovery.
3164 */
3165STATIC int
3166xlog_recover_do_icreate_pass2(
3167 struct xlog *log,
3168 struct list_head *buffer_list,
3169 xlog_recover_item_t *item)
3170{
3171 struct xfs_mount *mp = log->l_mp;
3172 struct xfs_icreate_log *icl;
3173 xfs_agnumber_t agno;
3174 xfs_agblock_t agbno;
3175 unsigned int count;
3176 unsigned int isize;
3177 xfs_agblock_t length;
3178
3179 icl = (struct xfs_icreate_log *)item->ri_buf[0].i_addr;
3180 if (icl->icl_type != XFS_LI_ICREATE) {
3181 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad type");
Dave Chinner24513372014-06-25 14:58:08 +10003182 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003183 }
3184
3185 if (icl->icl_size != 1) {
3186 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad icl size");
Dave Chinner24513372014-06-25 14:58:08 +10003187 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003188 }
3189
3190 agno = be32_to_cpu(icl->icl_ag);
3191 if (agno >= mp->m_sb.sb_agcount) {
3192 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agno");
Dave Chinner24513372014-06-25 14:58:08 +10003193 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003194 }
3195 agbno = be32_to_cpu(icl->icl_agbno);
3196 if (!agbno || agbno == NULLAGBLOCK || agbno >= mp->m_sb.sb_agblocks) {
3197 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agbno");
Dave Chinner24513372014-06-25 14:58:08 +10003198 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003199 }
3200 isize = be32_to_cpu(icl->icl_isize);
3201 if (isize != mp->m_sb.sb_inodesize) {
3202 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad isize");
Dave Chinner24513372014-06-25 14:58:08 +10003203 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003204 }
3205 count = be32_to_cpu(icl->icl_count);
3206 if (!count) {
3207 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count");
Dave Chinner24513372014-06-25 14:58:08 +10003208 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003209 }
3210 length = be32_to_cpu(icl->icl_length);
3211 if (!length || length >= mp->m_sb.sb_agblocks) {
3212 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad length");
Dave Chinner24513372014-06-25 14:58:08 +10003213 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003214 }
3215
3216 /* existing allocation is fixed value */
Jie Liu71783432013-12-13 15:51:46 +11003217 ASSERT(count == mp->m_ialloc_inos);
Jie Liu126cd102013-12-13 15:51:48 +11003218 ASSERT(length == mp->m_ialloc_blks);
Jie Liu71783432013-12-13 15:51:46 +11003219 if (count != mp->m_ialloc_inos ||
Jie Liu126cd102013-12-13 15:51:48 +11003220 length != mp->m_ialloc_blks) {
Dave Chinner28c8e412013-06-27 16:04:55 +10003221 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count 2");
Dave Chinner24513372014-06-25 14:58:08 +10003222 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003223 }
3224
3225 /*
3226 * Inode buffers can be freed. Do not replay the inode initialisation as
3227 * we could be overwriting something written after this inode buffer was
3228 * cancelled.
3229 *
3230 * XXX: we need to iterate all buffers and only init those that are not
3231 * cancelled. I think that a more fine grained factoring of
3232 * xfs_ialloc_inode_init may be appropriate here to enable this to be
3233 * done easily.
3234 */
3235 if (xlog_check_buffer_cancelled(log,
3236 XFS_AGB_TO_DADDR(mp, agno, agbno), length, 0))
3237 return 0;
3238
3239 xfs_ialloc_inode_init(mp, NULL, buffer_list, agno, agbno, length,
3240 be32_to_cpu(icl->icl_gen));
3241 return 0;
3242}
3243
3244/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003245 * Free up any resources allocated by the transaction
3246 *
3247 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
3248 */
3249STATIC void
3250xlog_recover_free_trans(
Christoph Hellwigd0450942010-12-01 22:06:23 +00003251 struct xlog_recover *trans)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003252{
Dave Chinnerf0a76952010-01-11 11:49:57 +00003253 xlog_recover_item_t *item, *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003254 int i;
3255
Dave Chinnerf0a76952010-01-11 11:49:57 +00003256 list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
3257 /* Free the regions in the item. */
3258 list_del(&item->ri_list);
3259 for (i = 0; i < item->ri_cnt; i++)
3260 kmem_free(item->ri_buf[i].i_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003261 /* Free the item itself */
Dave Chinnerf0a76952010-01-11 11:49:57 +00003262 kmem_free(item->ri_buf);
3263 kmem_free(item);
3264 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003265 /* Free the transaction recover structure */
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003266 kmem_free(trans);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003267}
3268
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003269STATIC void
3270xlog_recover_buffer_ra_pass2(
3271 struct xlog *log,
3272 struct xlog_recover_item *item)
3273{
3274 struct xfs_buf_log_format *buf_f = item->ri_buf[0].i_addr;
3275 struct xfs_mount *mp = log->l_mp;
3276
Dave Chinner84a5b732013-08-27 08:10:53 +10003277 if (xlog_peek_buffer_cancelled(log, buf_f->blf_blkno,
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003278 buf_f->blf_len, buf_f->blf_flags)) {
3279 return;
3280 }
3281
3282 xfs_buf_readahead(mp->m_ddev_targp, buf_f->blf_blkno,
3283 buf_f->blf_len, NULL);
3284}
3285
3286STATIC void
3287xlog_recover_inode_ra_pass2(
3288 struct xlog *log,
3289 struct xlog_recover_item *item)
3290{
3291 struct xfs_inode_log_format ilf_buf;
3292 struct xfs_inode_log_format *ilfp;
3293 struct xfs_mount *mp = log->l_mp;
3294 int error;
3295
3296 if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) {
3297 ilfp = item->ri_buf[0].i_addr;
3298 } else {
3299 ilfp = &ilf_buf;
3300 memset(ilfp, 0, sizeof(*ilfp));
3301 error = xfs_inode_item_format_convert(&item->ri_buf[0], ilfp);
3302 if (error)
3303 return;
3304 }
3305
Dave Chinner84a5b732013-08-27 08:10:53 +10003306 if (xlog_peek_buffer_cancelled(log, ilfp->ilf_blkno, ilfp->ilf_len, 0))
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003307 return;
3308
3309 xfs_buf_readahead(mp->m_ddev_targp, ilfp->ilf_blkno,
Dave Chinnerd8914002013-08-27 11:39:37 +10003310 ilfp->ilf_len, &xfs_inode_buf_ra_ops);
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003311}
3312
3313STATIC void
3314xlog_recover_dquot_ra_pass2(
3315 struct xlog *log,
3316 struct xlog_recover_item *item)
3317{
3318 struct xfs_mount *mp = log->l_mp;
3319 struct xfs_disk_dquot *recddq;
3320 struct xfs_dq_logformat *dq_f;
3321 uint type;
3322
3323
3324 if (mp->m_qflags == 0)
3325 return;
3326
3327 recddq = item->ri_buf[1].i_addr;
3328 if (recddq == NULL)
3329 return;
3330 if (item->ri_buf[1].i_len < sizeof(struct xfs_disk_dquot))
3331 return;
3332
3333 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
3334 ASSERT(type);
3335 if (log->l_quotaoffs_flag & type)
3336 return;
3337
3338 dq_f = item->ri_buf[0].i_addr;
3339 ASSERT(dq_f);
3340 ASSERT(dq_f->qlf_len == 1);
3341
3342 xfs_buf_readahead(mp->m_ddev_targp, dq_f->qlf_blkno,
Dave Chinner0f0d3342013-08-27 13:25:43 +10003343 XFS_FSB_TO_BB(mp, dq_f->qlf_len), NULL);
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003344}
3345
3346STATIC void
3347xlog_recover_ra_pass2(
3348 struct xlog *log,
3349 struct xlog_recover_item *item)
3350{
3351 switch (ITEM_TYPE(item)) {
3352 case XFS_LI_BUF:
3353 xlog_recover_buffer_ra_pass2(log, item);
3354 break;
3355 case XFS_LI_INODE:
3356 xlog_recover_inode_ra_pass2(log, item);
3357 break;
3358 case XFS_LI_DQUOT:
3359 xlog_recover_dquot_ra_pass2(log, item);
3360 break;
3361 case XFS_LI_EFI:
3362 case XFS_LI_EFD:
3363 case XFS_LI_QUOTAOFF:
3364 default:
3365 break;
3366 }
3367}
3368
Linus Torvalds1da177e2005-04-16 15:20:36 -07003369STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003370xlog_recover_commit_pass1(
Mark Tinguelyad223e62012-06-14 09:22:15 -05003371 struct xlog *log,
3372 struct xlog_recover *trans,
3373 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003374{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003375 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1);
Christoph Hellwigd0450942010-12-01 22:06:23 +00003376
3377 switch (ITEM_TYPE(item)) {
3378 case XFS_LI_BUF:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003379 return xlog_recover_buffer_pass1(log, item);
Christoph Hellwigd0450942010-12-01 22:06:23 +00003380 case XFS_LI_QUOTAOFF:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003381 return xlog_recover_quotaoff_pass1(log, item);
3382 case XFS_LI_INODE:
3383 case XFS_LI_EFI:
3384 case XFS_LI_EFD:
3385 case XFS_LI_DQUOT:
Dave Chinner28c8e412013-06-27 16:04:55 +10003386 case XFS_LI_ICREATE:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003387 /* nothing to do in pass 1 */
3388 return 0;
Christoph Hellwigd0450942010-12-01 22:06:23 +00003389 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003390 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
3391 __func__, ITEM_TYPE(item));
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003392 ASSERT(0);
Dave Chinner24513372014-06-25 14:58:08 +10003393 return -EIO;
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003394 }
3395}
3396
3397STATIC int
3398xlog_recover_commit_pass2(
Mark Tinguelyad223e62012-06-14 09:22:15 -05003399 struct xlog *log,
3400 struct xlog_recover *trans,
3401 struct list_head *buffer_list,
3402 struct xlog_recover_item *item)
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003403{
3404 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2);
3405
3406 switch (ITEM_TYPE(item)) {
3407 case XFS_LI_BUF:
Dave Chinner50d5c8d2013-08-28 21:22:47 +10003408 return xlog_recover_buffer_pass2(log, buffer_list, item,
3409 trans->r_lsn);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003410 case XFS_LI_INODE:
Dave Chinner50d5c8d2013-08-28 21:22:47 +10003411 return xlog_recover_inode_pass2(log, buffer_list, item,
3412 trans->r_lsn);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003413 case XFS_LI_EFI:
3414 return xlog_recover_efi_pass2(log, item, trans->r_lsn);
3415 case XFS_LI_EFD:
3416 return xlog_recover_efd_pass2(log, item);
3417 case XFS_LI_DQUOT:
Dave Chinner50d5c8d2013-08-28 21:22:47 +10003418 return xlog_recover_dquot_pass2(log, buffer_list, item,
3419 trans->r_lsn);
Dave Chinner28c8e412013-06-27 16:04:55 +10003420 case XFS_LI_ICREATE:
3421 return xlog_recover_do_icreate_pass2(log, buffer_list, item);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003422 case XFS_LI_QUOTAOFF:
3423 /* nothing to do in pass2 */
3424 return 0;
3425 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003426 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
3427 __func__, ITEM_TYPE(item));
Christoph Hellwigd0450942010-12-01 22:06:23 +00003428 ASSERT(0);
Dave Chinner24513372014-06-25 14:58:08 +10003429 return -EIO;
Christoph Hellwigd0450942010-12-01 22:06:23 +00003430 }
3431}
3432
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003433STATIC int
3434xlog_recover_items_pass2(
3435 struct xlog *log,
3436 struct xlog_recover *trans,
3437 struct list_head *buffer_list,
3438 struct list_head *item_list)
3439{
3440 struct xlog_recover_item *item;
3441 int error = 0;
3442
3443 list_for_each_entry(item, item_list, ri_list) {
3444 error = xlog_recover_commit_pass2(log, trans,
3445 buffer_list, item);
3446 if (error)
3447 return error;
3448 }
3449
3450 return error;
3451}
3452
Christoph Hellwigd0450942010-12-01 22:06:23 +00003453/*
3454 * Perform the transaction.
3455 *
3456 * If the transaction modifies a buffer or inode, do it now. Otherwise,
3457 * EFIs and EFDs get queued up by adding entries into the AIL for them.
3458 */
3459STATIC int
3460xlog_recover_commit_trans(
Mark Tinguelyad223e62012-06-14 09:22:15 -05003461 struct xlog *log,
Christoph Hellwigd0450942010-12-01 22:06:23 +00003462 struct xlog_recover *trans,
3463 int pass)
3464{
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003465 int error = 0;
3466 int error2;
3467 int items_queued = 0;
3468 struct xlog_recover_item *item;
3469 struct xlog_recover_item *next;
3470 LIST_HEAD (buffer_list);
3471 LIST_HEAD (ra_list);
3472 LIST_HEAD (done_list);
3473
3474 #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100
Linus Torvalds1da177e2005-04-16 15:20:36 -07003475
Dave Chinnerf0a76952010-01-11 11:49:57 +00003476 hlist_del(&trans->r_list);
Christoph Hellwigd0450942010-12-01 22:06:23 +00003477
3478 error = xlog_recover_reorder_trans(log, trans, pass);
3479 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003480 return error;
Christoph Hellwigd0450942010-12-01 22:06:23 +00003481
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003482 list_for_each_entry_safe(item, next, &trans->r_itemq, ri_list) {
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003483 switch (pass) {
3484 case XLOG_RECOVER_PASS1:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003485 error = xlog_recover_commit_pass1(log, trans, item);
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003486 break;
3487 case XLOG_RECOVER_PASS2:
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003488 xlog_recover_ra_pass2(log, item);
3489 list_move_tail(&item->ri_list, &ra_list);
3490 items_queued++;
3491 if (items_queued >= XLOG_RECOVER_COMMIT_QUEUE_MAX) {
3492 error = xlog_recover_items_pass2(log, trans,
3493 &buffer_list, &ra_list);
3494 list_splice_tail_init(&ra_list, &done_list);
3495 items_queued = 0;
3496 }
3497
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003498 break;
3499 default:
3500 ASSERT(0);
3501 }
3502
Christoph Hellwigd0450942010-12-01 22:06:23 +00003503 if (error)
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003504 goto out;
Christoph Hellwigd0450942010-12-01 22:06:23 +00003505 }
3506
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003507out:
3508 if (!list_empty(&ra_list)) {
3509 if (!error)
3510 error = xlog_recover_items_pass2(log, trans,
3511 &buffer_list, &ra_list);
3512 list_splice_tail_init(&ra_list, &done_list);
3513 }
3514
3515 if (!list_empty(&done_list))
3516 list_splice_init(&done_list, &trans->r_itemq);
3517
Christoph Hellwigd0450942010-12-01 22:06:23 +00003518 xlog_recover_free_trans(trans);
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003519
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003520 error2 = xfs_buf_delwri_submit(&buffer_list);
3521 return error ? error : error2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003522}
3523
3524STATIC int
3525xlog_recover_unmount_trans(
Eric Sandeenbbe4c662014-04-14 19:06:25 +10003526 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003527{
3528 /* Do nothing now */
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003529 xfs_warn(log->l_mp, "%s: Unmount LR", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003530 return 0;
3531}
3532
3533/*
3534 * There are two valid states of the r_state field. 0 indicates that the
3535 * transaction structure is in a normal state. We have either seen the
3536 * start of the transaction or the last operation we added was not a partial
3537 * operation. If the last operation we added to the transaction was a
3538 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
3539 *
3540 * NOTE: skip LRs with 0 data length.
3541 */
3542STATIC int
3543xlog_recover_process_data(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003544 struct xlog *log,
Dave Chinnerf0a76952010-01-11 11:49:57 +00003545 struct hlist_head rhash[],
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003546 struct xlog_rec_header *rhead,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003547 xfs_caddr_t dp,
3548 int pass)
3549{
3550 xfs_caddr_t lp;
3551 int num_logops;
3552 xlog_op_header_t *ohead;
3553 xlog_recover_t *trans;
3554 xlog_tid_t tid;
3555 int error;
3556 unsigned long hash;
3557 uint flags;
3558
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003559 lp = dp + be32_to_cpu(rhead->h_len);
3560 num_logops = be32_to_cpu(rhead->h_num_logops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003561
3562 /* check the log format matches our own - else we can't recover */
3563 if (xlog_header_check_recover(log->l_mp, rhead))
Dave Chinner24513372014-06-25 14:58:08 +10003564 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003565
3566 while ((dp < lp) && num_logops) {
3567 ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
3568 ohead = (xlog_op_header_t *)dp;
3569 dp += sizeof(xlog_op_header_t);
3570 if (ohead->oh_clientid != XFS_TRANSACTION &&
3571 ohead->oh_clientid != XFS_LOG) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003572 xfs_warn(log->l_mp, "%s: bad clientid 0x%x",
3573 __func__, ohead->oh_clientid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003574 ASSERT(0);
Dave Chinner24513372014-06-25 14:58:08 +10003575 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003576 }
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10003577 tid = be32_to_cpu(ohead->oh_tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003578 hash = XLOG_RHASH(tid);
Dave Chinnerf0a76952010-01-11 11:49:57 +00003579 trans = xlog_recover_find_tid(&rhash[hash], tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003580 if (trans == NULL) { /* not found; add new tid */
3581 if (ohead->oh_flags & XLOG_START_TRANS)
3582 xlog_recover_new_tid(&rhash[hash], tid,
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003583 be64_to_cpu(rhead->h_lsn));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003584 } else {
Lachlan McIlroy9742bb92008-01-10 16:43:36 +11003585 if (dp + be32_to_cpu(ohead->oh_len) > lp) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003586 xfs_warn(log->l_mp, "%s: bad length 0x%x",
3587 __func__, be32_to_cpu(ohead->oh_len));
Lachlan McIlroy9742bb92008-01-10 16:43:36 +11003588 WARN_ON(1);
Dave Chinner24513372014-06-25 14:58:08 +10003589 return -EIO;
Lachlan McIlroy9742bb92008-01-10 16:43:36 +11003590 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003591 flags = ohead->oh_flags & ~XLOG_END_TRANS;
3592 if (flags & XLOG_WAS_CONT_TRANS)
3593 flags &= ~XLOG_CONTINUE_TRANS;
3594 switch (flags) {
3595 case XLOG_COMMIT_TRANS:
3596 error = xlog_recover_commit_trans(log,
Dave Chinnerf0a76952010-01-11 11:49:57 +00003597 trans, pass);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003598 break;
3599 case XLOG_UNMOUNT_TRANS:
Eric Sandeenbbe4c662014-04-14 19:06:25 +10003600 error = xlog_recover_unmount_trans(log);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003601 break;
3602 case XLOG_WAS_CONT_TRANS:
Dave Chinner9abbc532010-04-13 15:06:46 +10003603 error = xlog_recover_add_to_cont_trans(log,
3604 trans, dp,
3605 be32_to_cpu(ohead->oh_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003606 break;
3607 case XLOG_START_TRANS:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003608 xfs_warn(log->l_mp, "%s: bad transaction",
3609 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003610 ASSERT(0);
Dave Chinner24513372014-06-25 14:58:08 +10003611 error = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003612 break;
3613 case 0:
3614 case XLOG_CONTINUE_TRANS:
Dave Chinner9abbc532010-04-13 15:06:46 +10003615 error = xlog_recover_add_to_trans(log, trans,
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10003616 dp, be32_to_cpu(ohead->oh_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003617 break;
3618 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003619 xfs_warn(log->l_mp, "%s: bad flag 0x%x",
3620 __func__, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003621 ASSERT(0);
Dave Chinner24513372014-06-25 14:58:08 +10003622 error = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003623 break;
3624 }
Mark Tinguely2a841082013-10-02 07:51:12 -05003625 if (error) {
3626 xlog_recover_free_trans(trans);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003627 return error;
Mark Tinguely2a841082013-10-02 07:51:12 -05003628 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003629 }
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10003630 dp += be32_to_cpu(ohead->oh_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003631 num_logops--;
3632 }
3633 return 0;
3634}
3635
3636/*
3637 * Process an extent free intent item that was recovered from
3638 * the log. We need to free the extents that it describes.
3639 */
David Chinner3c1e2bb2008-04-10 12:21:11 +10003640STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003641xlog_recover_process_efi(
3642 xfs_mount_t *mp,
3643 xfs_efi_log_item_t *efip)
3644{
3645 xfs_efd_log_item_t *efdp;
3646 xfs_trans_t *tp;
3647 int i;
David Chinner3c1e2bb2008-04-10 12:21:11 +10003648 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003649 xfs_extent_t *extp;
3650 xfs_fsblock_t startblock_fsb;
3651
Dave Chinnerb199c8a2010-12-20 11:59:49 +11003652 ASSERT(!test_bit(XFS_EFI_RECOVERED, &efip->efi_flags));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003653
3654 /*
3655 * First check the validity of the extents described by the
3656 * EFI. If any are bad, then assume that all are bad and
3657 * just toss the EFI.
3658 */
3659 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3660 extp = &(efip->efi_format.efi_extents[i]);
3661 startblock_fsb = XFS_BB_TO_FSB(mp,
3662 XFS_FSB_TO_DADDR(mp, extp->ext_start));
3663 if ((startblock_fsb == 0) ||
3664 (extp->ext_len == 0) ||
3665 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
3666 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
3667 /*
3668 * This will pull the EFI from the AIL and
3669 * free the memory associated with it.
3670 */
Dave Chinner666d6442013-04-03 14:09:21 +11003671 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003672 xfs_efi_release(efip, efip->efi_format.efi_nextents);
Dave Chinner24513372014-06-25 14:58:08 +10003673 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003674 }
3675 }
3676
3677 tp = xfs_trans_alloc(mp, 0);
Jie Liu3d3c8b52013-08-12 20:49:59 +10003678 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
David Chinnerfc6149d2008-04-10 12:21:53 +10003679 if (error)
3680 goto abort_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003681 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
3682
3683 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3684 extp = &(efip->efi_format.efi_extents[i]);
David Chinnerfc6149d2008-04-10 12:21:53 +10003685 error = xfs_free_extent(tp, extp->ext_start, extp->ext_len);
3686 if (error)
3687 goto abort_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003688 xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
3689 extp->ext_len);
3690 }
3691
Dave Chinnerb199c8a2010-12-20 11:59:49 +11003692 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
David Chinnere5720ee2008-04-10 12:21:18 +10003693 error = xfs_trans_commit(tp, 0);
David Chinner3c1e2bb2008-04-10 12:21:11 +10003694 return error;
David Chinnerfc6149d2008-04-10 12:21:53 +10003695
3696abort_error:
3697 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3698 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003699}
3700
3701/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003702 * When this is called, all of the EFIs which did not have
3703 * corresponding EFDs should be in the AIL. What we do now
3704 * is free the extents associated with each one.
3705 *
3706 * Since we process the EFIs in normal transactions, they
3707 * will be removed at some point after the commit. This prevents
3708 * us from just walking down the list processing each one.
3709 * We'll use a flag in the EFI to skip those that we've already
3710 * processed and use the AIL iteration mechanism's generation
3711 * count to try to speed this up at least a bit.
3712 *
3713 * When we start, we know that the EFIs are the only things in
3714 * the AIL. As we process them, however, other items are added
3715 * to the AIL. Since everything added to the AIL must come after
3716 * everything already in the AIL, we stop processing as soon as
3717 * we see something other than an EFI in the AIL.
3718 */
David Chinner3c1e2bb2008-04-10 12:21:11 +10003719STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003720xlog_recover_process_efis(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003721 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003722{
3723 xfs_log_item_t *lip;
3724 xfs_efi_log_item_t *efip;
David Chinner3c1e2bb2008-04-10 12:21:11 +10003725 int error = 0;
David Chinner27d8d5f2008-10-30 17:38:39 +11003726 struct xfs_ail_cursor cur;
David Chinnera9c21c12008-10-30 17:39:35 +11003727 struct xfs_ail *ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003728
David Chinnera9c21c12008-10-30 17:39:35 +11003729 ailp = log->l_ailp;
3730 spin_lock(&ailp->xa_lock);
3731 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003732 while (lip != NULL) {
3733 /*
3734 * We're done when we see something other than an EFI.
David Chinner27d8d5f2008-10-30 17:38:39 +11003735 * There should be no EFIs left in the AIL now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003736 */
3737 if (lip->li_type != XFS_LI_EFI) {
David Chinner27d8d5f2008-10-30 17:38:39 +11003738#ifdef DEBUG
David Chinnera9c21c12008-10-30 17:39:35 +11003739 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
David Chinner27d8d5f2008-10-30 17:38:39 +11003740 ASSERT(lip->li_type != XFS_LI_EFI);
3741#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003742 break;
3743 }
3744
3745 /*
3746 * Skip EFIs that we've already processed.
3747 */
3748 efip = (xfs_efi_log_item_t *)lip;
Dave Chinnerb199c8a2010-12-20 11:59:49 +11003749 if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) {
David Chinnera9c21c12008-10-30 17:39:35 +11003750 lip = xfs_trans_ail_cursor_next(ailp, &cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003751 continue;
3752 }
3753
David Chinnera9c21c12008-10-30 17:39:35 +11003754 spin_unlock(&ailp->xa_lock);
3755 error = xlog_recover_process_efi(log->l_mp, efip);
3756 spin_lock(&ailp->xa_lock);
David Chinner27d8d5f2008-10-30 17:38:39 +11003757 if (error)
3758 goto out;
David Chinnera9c21c12008-10-30 17:39:35 +11003759 lip = xfs_trans_ail_cursor_next(ailp, &cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003760 }
David Chinner27d8d5f2008-10-30 17:38:39 +11003761out:
Eric Sandeene4a1e292014-04-14 19:06:05 +10003762 xfs_trans_ail_cursor_done(&cur);
David Chinnera9c21c12008-10-30 17:39:35 +11003763 spin_unlock(&ailp->xa_lock);
David Chinner3c1e2bb2008-04-10 12:21:11 +10003764 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003765}
3766
3767/*
3768 * This routine performs a transaction to null out a bad inode pointer
3769 * in an agi unlinked inode hash bucket.
3770 */
3771STATIC void
3772xlog_recover_clear_agi_bucket(
3773 xfs_mount_t *mp,
3774 xfs_agnumber_t agno,
3775 int bucket)
3776{
3777 xfs_trans_t *tp;
3778 xfs_agi_t *agi;
3779 xfs_buf_t *agibp;
3780 int offset;
3781 int error;
3782
3783 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
Jie Liu3d3c8b52013-08-12 20:49:59 +10003784 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_clearagi, 0, 0);
David Chinnere5720ee2008-04-10 12:21:18 +10003785 if (error)
3786 goto out_abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003787
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003788 error = xfs_read_agi(mp, tp, agno, &agibp);
3789 if (error)
David Chinnere5720ee2008-04-10 12:21:18 +10003790 goto out_abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003791
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003792 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11003793 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003794 offset = offsetof(xfs_agi_t, agi_unlinked) +
3795 (sizeof(xfs_agino_t) * bucket);
3796 xfs_trans_log_buf(tp, agibp, offset,
3797 (offset + sizeof(xfs_agino_t) - 1));
3798
David Chinnere5720ee2008-04-10 12:21:18 +10003799 error = xfs_trans_commit(tp, 0);
3800 if (error)
3801 goto out_error;
3802 return;
3803
3804out_abort:
3805 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3806out_error:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003807 xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno);
David Chinnere5720ee2008-04-10 12:21:18 +10003808 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003809}
3810
Christoph Hellwig23fac502008-11-28 14:23:40 +11003811STATIC xfs_agino_t
3812xlog_recover_process_one_iunlink(
3813 struct xfs_mount *mp,
3814 xfs_agnumber_t agno,
3815 xfs_agino_t agino,
3816 int bucket)
3817{
3818 struct xfs_buf *ibp;
3819 struct xfs_dinode *dip;
3820 struct xfs_inode *ip;
3821 xfs_ino_t ino;
3822 int error;
3823
3824 ino = XFS_AGINO_TO_INO(mp, agno, agino);
Dave Chinner7b6259e2010-06-24 11:35:17 +10003825 error = xfs_iget(mp, NULL, ino, 0, 0, &ip);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003826 if (error)
3827 goto fail;
3828
3829 /*
3830 * Get the on disk inode to find the next inode in the bucket.
3831 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04003832 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &ibp, 0, 0);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003833 if (error)
Christoph Hellwig0e446672008-11-28 14:23:42 +11003834 goto fail_iput;
Christoph Hellwig23fac502008-11-28 14:23:40 +11003835
Christoph Hellwig23fac502008-11-28 14:23:40 +11003836 ASSERT(ip->i_d.di_nlink == 0);
Christoph Hellwig0e446672008-11-28 14:23:42 +11003837 ASSERT(ip->i_d.di_mode != 0);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003838
3839 /* setup for the next pass */
3840 agino = be32_to_cpu(dip->di_next_unlinked);
3841 xfs_buf_relse(ibp);
3842
3843 /*
3844 * Prevent any DMAPI event from being sent when the reference on
3845 * the inode is dropped.
3846 */
3847 ip->i_d.di_dmevmask = 0;
3848
Christoph Hellwig0e446672008-11-28 14:23:42 +11003849 IRELE(ip);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003850 return agino;
3851
Christoph Hellwig0e446672008-11-28 14:23:42 +11003852 fail_iput:
3853 IRELE(ip);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003854 fail:
3855 /*
3856 * We can't read in the inode this bucket points to, or this inode
3857 * is messed up. Just ditch this bucket of inodes. We will lose
3858 * some inodes and space, but at least we won't hang.
3859 *
3860 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3861 * clear the inode pointer in the bucket.
3862 */
3863 xlog_recover_clear_agi_bucket(mp, agno, bucket);
3864 return NULLAGINO;
3865}
3866
Linus Torvalds1da177e2005-04-16 15:20:36 -07003867/*
3868 * xlog_iunlink_recover
3869 *
3870 * This is called during recovery to process any inodes which
3871 * we unlinked but not freed when the system crashed. These
3872 * inodes will be on the lists in the AGI blocks. What we do
3873 * here is scan all the AGIs and fully truncate and free any
3874 * inodes found on the lists. Each inode is removed from the
3875 * lists when it has been fully truncated and is freed. The
3876 * freeing of the inode and its removal from the list must be
3877 * atomic.
3878 */
Eric Sandeend96f8f82009-07-02 00:09:33 -05003879STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07003880xlog_recover_process_iunlinks(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003881 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003882{
3883 xfs_mount_t *mp;
3884 xfs_agnumber_t agno;
3885 xfs_agi_t *agi;
3886 xfs_buf_t *agibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003887 xfs_agino_t agino;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003888 int bucket;
3889 int error;
3890 uint mp_dmevmask;
3891
3892 mp = log->l_mp;
3893
3894 /*
3895 * Prevent any DMAPI event from being sent while in this function.
3896 */
3897 mp_dmevmask = mp->m_dmevmask;
3898 mp->m_dmevmask = 0;
3899
3900 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3901 /*
3902 * Find the agi for this ag.
3903 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003904 error = xfs_read_agi(mp, NULL, agno, &agibp);
3905 if (error) {
3906 /*
3907 * AGI is b0rked. Don't process it.
3908 *
3909 * We should probably mark the filesystem as corrupt
3910 * after we've recovered all the ag's we can....
3911 */
3912 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003913 }
Jan Karad97d32e2012-03-15 09:34:02 +00003914 /*
3915 * Unlock the buffer so that it can be acquired in the normal
3916 * course of the transaction to truncate and free each inode.
3917 * Because we are not racing with anyone else here for the AGI
3918 * buffer, we don't even need to hold it locked to read the
3919 * initial unlinked bucket entries out of the buffer. We keep
3920 * buffer reference though, so that it stays pinned in memory
3921 * while we need the buffer.
3922 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003923 agi = XFS_BUF_TO_AGI(agibp);
Jan Karad97d32e2012-03-15 09:34:02 +00003924 xfs_buf_unlock(agibp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003925
3926 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
Christoph Hellwig16259e72005-11-02 15:11:25 +11003927 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003928 while (agino != NULLAGINO) {
Christoph Hellwig23fac502008-11-28 14:23:40 +11003929 agino = xlog_recover_process_one_iunlink(mp,
3930 agno, agino, bucket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003931 }
3932 }
Jan Karad97d32e2012-03-15 09:34:02 +00003933 xfs_buf_rele(agibp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003934 }
3935
3936 mp->m_dmevmask = mp_dmevmask;
3937}
3938
Linus Torvalds1da177e2005-04-16 15:20:36 -07003939/*
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003940 * Upack the log buffer data and crc check it. If the check fails, issue a
3941 * warning if and only if the CRC in the header is non-zero. This makes the
3942 * check an advisory warning, and the zero CRC check will prevent failure
3943 * warnings from being emitted when upgrading the kernel from one that does not
3944 * add CRCs by default.
3945 *
3946 * When filesystems are CRC enabled, this CRC mismatch becomes a fatal log
3947 * corruption failure
Linus Torvalds1da177e2005-04-16 15:20:36 -07003948 */
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003949STATIC int
3950xlog_unpack_data_crc(
3951 struct xlog_rec_header *rhead,
3952 xfs_caddr_t dp,
3953 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003954{
Dave Chinnerf9668a02012-11-28 13:01:03 +11003955 __le32 crc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003956
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003957 crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len));
3958 if (crc != rhead->h_crc) {
3959 if (rhead->h_crc || xfs_sb_version_hascrc(&log->l_mp->m_sb)) {
3960 xfs_alert(log->l_mp,
Eric Sandeen08e96e12013-10-11 20:59:05 -05003961 "log record CRC mismatch: found 0x%x, expected 0x%x.",
Dave Chinnerf9668a02012-11-28 13:01:03 +11003962 le32_to_cpu(rhead->h_crc),
3963 le32_to_cpu(crc));
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003964 xfs_hex_dump(dp, 32);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003965 }
3966
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003967 /*
3968 * If we've detected a log record corruption, then we can't
3969 * recover past this point. Abort recovery if we are enforcing
3970 * CRC protection by punting an error back up the stack.
3971 */
3972 if (xfs_sb_version_hascrc(&log->l_mp->m_sb))
Dave Chinner24513372014-06-25 14:58:08 +10003973 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003974 }
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003975
3976 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003977}
3978
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003979STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003980xlog_unpack_data(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003981 struct xlog_rec_header *rhead,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003982 xfs_caddr_t dp,
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003983 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003984{
3985 int i, j, k;
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003986 int error;
3987
3988 error = xlog_unpack_data_crc(rhead, dp, log);
3989 if (error)
3990 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003991
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003992 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07003993 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003994 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003995 dp += BBSIZE;
3996 }
3997
Eric Sandeen62118702008-03-06 13:44:28 +11003998 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Christoph Hellwigb28708d2008-11-28 14:23:38 +11003999 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004000 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004001 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
4002 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004003 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
Linus Torvalds1da177e2005-04-16 15:20:36 -07004004 dp += BBSIZE;
4005 }
4006 }
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004007
4008 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004009}
4010
4011STATIC int
4012xlog_valid_rec_header(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004013 struct xlog *log,
4014 struct xlog_rec_header *rhead,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004015 xfs_daddr_t blkno)
4016{
4017 int hlen;
4018
Christoph Hellwig69ef9212011-07-08 14:36:05 +02004019 if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004020 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
4021 XFS_ERRLEVEL_LOW, log->l_mp);
Dave Chinner24513372014-06-25 14:58:08 +10004022 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004023 }
4024 if (unlikely(
4025 (!rhead->h_version ||
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004026 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004027 xfs_warn(log->l_mp, "%s: unrecognised log version (%d).",
Harvey Harrison34a622b2008-04-10 12:19:21 +10004028 __func__, be32_to_cpu(rhead->h_version));
Dave Chinner24513372014-06-25 14:58:08 +10004029 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004030 }
4031
4032 /* LR body must have data or it wouldn't have been written */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004033 hlen = be32_to_cpu(rhead->h_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004034 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
4035 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
4036 XFS_ERRLEVEL_LOW, log->l_mp);
Dave Chinner24513372014-06-25 14:58:08 +10004037 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004038 }
4039 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
4040 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
4041 XFS_ERRLEVEL_LOW, log->l_mp);
Dave Chinner24513372014-06-25 14:58:08 +10004042 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004043 }
4044 return 0;
4045}
4046
4047/*
4048 * Read the log from tail to head and process the log records found.
4049 * Handle the two cases where the tail and head are in the same cycle
4050 * and where the active portion of the log wraps around the end of
4051 * the physical log separately. The pass parameter is passed through
4052 * to the routines called to process the data and is not looked at
4053 * here.
4054 */
4055STATIC int
4056xlog_do_recovery_pass(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004057 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004058 xfs_daddr_t head_blk,
4059 xfs_daddr_t tail_blk,
4060 int pass)
4061{
4062 xlog_rec_header_t *rhead;
4063 xfs_daddr_t blk_no;
Andy Polingfc5bc4c2009-11-03 17:26:47 +00004064 xfs_caddr_t offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004065 xfs_buf_t *hbp, *dbp;
4066 int error = 0, h_size;
4067 int bblks, split_bblks;
4068 int hblks, split_hblks, wrapped_hblks;
Dave Chinnerf0a76952010-01-11 11:49:57 +00004069 struct hlist_head rhash[XLOG_RHASH_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -07004070
4071 ASSERT(head_blk != tail_blk);
4072
4073 /*
4074 * Read the header of the tail block and get the iclog buffer size from
4075 * h_size. Use this to tell how many sectors make up the log header.
4076 */
Eric Sandeen62118702008-03-06 13:44:28 +11004077 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004078 /*
4079 * When using variable length iclogs, read first sector of
4080 * iclog header and extract the header size from it. Get a
4081 * new hbp that is the correct size.
4082 */
4083 hbp = xlog_get_bp(log, 1);
4084 if (!hbp)
Dave Chinner24513372014-06-25 14:58:08 +10004085 return -ENOMEM;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004086
4087 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
4088 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004089 goto bread_err1;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004090
Linus Torvalds1da177e2005-04-16 15:20:36 -07004091 rhead = (xlog_rec_header_t *)offset;
4092 error = xlog_valid_rec_header(log, rhead, tail_blk);
4093 if (error)
4094 goto bread_err1;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004095 h_size = be32_to_cpu(rhead->h_size);
4096 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07004097 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
4098 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
4099 if (h_size % XLOG_HEADER_CYCLE_SIZE)
4100 hblks++;
4101 xlog_put_bp(hbp);
4102 hbp = xlog_get_bp(log, hblks);
4103 } else {
4104 hblks = 1;
4105 }
4106 } else {
Alex Elder69ce58f2010-04-20 17:09:59 +10004107 ASSERT(log->l_sectBBsize == 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004108 hblks = 1;
4109 hbp = xlog_get_bp(log, 1);
4110 h_size = XLOG_BIG_RECORD_BSIZE;
4111 }
4112
4113 if (!hbp)
Dave Chinner24513372014-06-25 14:58:08 +10004114 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004115 dbp = xlog_get_bp(log, BTOBB(h_size));
4116 if (!dbp) {
4117 xlog_put_bp(hbp);
Dave Chinner24513372014-06-25 14:58:08 +10004118 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004119 }
4120
4121 memset(rhash, 0, sizeof(rhash));
4122 if (tail_blk <= head_blk) {
4123 for (blk_no = tail_blk; blk_no < head_blk; ) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004124 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
4125 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004126 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004127
Linus Torvalds1da177e2005-04-16 15:20:36 -07004128 rhead = (xlog_rec_header_t *)offset;
4129 error = xlog_valid_rec_header(log, rhead, blk_no);
4130 if (error)
4131 goto bread_err2;
4132
4133 /* blocks in data section */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004134 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004135 error = xlog_bread(log, blk_no + hblks, bblks, dbp,
4136 &offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004137 if (error)
4138 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004139
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004140 error = xlog_unpack_data(rhead, offset, log);
4141 if (error)
4142 goto bread_err2;
4143
4144 error = xlog_recover_process_data(log,
4145 rhash, rhead, offset, pass);
4146 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004147 goto bread_err2;
4148 blk_no += bblks + hblks;
4149 }
4150 } else {
4151 /*
4152 * Perform recovery around the end of the physical log.
4153 * When the head is not on the same cycle number as the tail,
4154 * we can't do a sequential recovery as above.
4155 */
4156 blk_no = tail_blk;
4157 while (blk_no < log->l_logBBsize) {
4158 /*
4159 * Check for header wrapping around physical end-of-log
4160 */
Chandra Seetharaman62926042011-07-22 23:40:15 +00004161 offset = hbp->b_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004162 split_hblks = 0;
4163 wrapped_hblks = 0;
4164 if (blk_no + hblks <= log->l_logBBsize) {
4165 /* Read header in one read */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004166 error = xlog_bread(log, blk_no, hblks, hbp,
4167 &offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004168 if (error)
4169 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004170 } else {
4171 /* This LR is split across physical log end */
4172 if (blk_no != log->l_logBBsize) {
4173 /* some data before physical log end */
4174 ASSERT(blk_no <= INT_MAX);
4175 split_hblks = log->l_logBBsize - (int)blk_no;
4176 ASSERT(split_hblks > 0);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004177 error = xlog_bread(log, blk_no,
4178 split_hblks, hbp,
4179 &offset);
4180 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004181 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004182 }
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004183
Linus Torvalds1da177e2005-04-16 15:20:36 -07004184 /*
4185 * Note: this black magic still works with
4186 * large sector sizes (non-512) only because:
4187 * - we increased the buffer size originally
4188 * by 1 sector giving us enough extra space
4189 * for the second read;
4190 * - the log start is guaranteed to be sector
4191 * aligned;
4192 * - we read the log end (LR header start)
4193 * _first_, then the log start (LR header end)
4194 * - order is important.
4195 */
David Chinner234f56a2008-04-10 12:24:24 +10004196 wrapped_hblks = hblks - split_hblks;
Dave Chinner44396472011-04-21 09:34:27 +00004197 error = xlog_bread_offset(log, 0,
4198 wrapped_hblks, hbp,
4199 offset + BBTOB(split_hblks));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004200 if (error)
4201 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004202 }
4203 rhead = (xlog_rec_header_t *)offset;
4204 error = xlog_valid_rec_header(log, rhead,
4205 split_hblks ? blk_no : 0);
4206 if (error)
4207 goto bread_err2;
4208
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004209 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004210 blk_no += hblks;
4211
4212 /* Read in data for log record */
4213 if (blk_no + bblks <= log->l_logBBsize) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004214 error = xlog_bread(log, blk_no, bblks, dbp,
4215 &offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004216 if (error)
4217 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004218 } else {
4219 /* This log record is split across the
4220 * physical end of log */
Chandra Seetharaman62926042011-07-22 23:40:15 +00004221 offset = dbp->b_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004222 split_bblks = 0;
4223 if (blk_no != log->l_logBBsize) {
4224 /* some data is before the physical
4225 * end of log */
4226 ASSERT(!wrapped_hblks);
4227 ASSERT(blk_no <= INT_MAX);
4228 split_bblks =
4229 log->l_logBBsize - (int)blk_no;
4230 ASSERT(split_bblks > 0);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004231 error = xlog_bread(log, blk_no,
4232 split_bblks, dbp,
4233 &offset);
4234 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004235 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004236 }
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004237
Linus Torvalds1da177e2005-04-16 15:20:36 -07004238 /*
4239 * Note: this black magic still works with
4240 * large sector sizes (non-512) only because:
4241 * - we increased the buffer size originally
4242 * by 1 sector giving us enough extra space
4243 * for the second read;
4244 * - the log start is guaranteed to be sector
4245 * aligned;
4246 * - we read the log end (LR header start)
4247 * _first_, then the log start (LR header end)
4248 * - order is important.
4249 */
Dave Chinner44396472011-04-21 09:34:27 +00004250 error = xlog_bread_offset(log, 0,
Dave Chinner009507b2012-11-02 11:38:44 +11004251 bblks - split_bblks, dbp,
Dave Chinner44396472011-04-21 09:34:27 +00004252 offset + BBTOB(split_bblks));
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004253 if (error)
4254 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004255 }
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004256
4257 error = xlog_unpack_data(rhead, offset, log);
4258 if (error)
4259 goto bread_err2;
4260
4261 error = xlog_recover_process_data(log, rhash,
4262 rhead, offset, pass);
4263 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004264 goto bread_err2;
4265 blk_no += bblks;
4266 }
4267
4268 ASSERT(blk_no >= log->l_logBBsize);
4269 blk_no -= log->l_logBBsize;
4270
4271 /* read first part of physical log */
4272 while (blk_no < head_blk) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004273 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
4274 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004275 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004276
Linus Torvalds1da177e2005-04-16 15:20:36 -07004277 rhead = (xlog_rec_header_t *)offset;
4278 error = xlog_valid_rec_header(log, rhead, blk_no);
4279 if (error)
4280 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004281
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004282 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004283 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
4284 &offset);
4285 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004286 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004287
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004288 error = xlog_unpack_data(rhead, offset, log);
4289 if (error)
4290 goto bread_err2;
4291
4292 error = xlog_recover_process_data(log, rhash,
4293 rhead, offset, pass);
4294 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004295 goto bread_err2;
4296 blk_no += bblks + hblks;
4297 }
4298 }
4299
4300 bread_err2:
4301 xlog_put_bp(dbp);
4302 bread_err1:
4303 xlog_put_bp(hbp);
4304 return error;
4305}
4306
4307/*
4308 * Do the recovery of the log. We actually do this in two phases.
4309 * The two passes are necessary in order to implement the function
4310 * of cancelling a record written into the log. The first pass
4311 * determines those things which have been cancelled, and the
4312 * second pass replays log items normally except for those which
4313 * have been cancelled. The handling of the replay and cancellations
4314 * takes place in the log item type specific routines.
4315 *
4316 * The table of items which have cancel records in the log is allocated
4317 * and freed at this level, since only here do we know when all of
4318 * the log recovery has been completed.
4319 */
4320STATIC int
4321xlog_do_log_recovery(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004322 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004323 xfs_daddr_t head_blk,
4324 xfs_daddr_t tail_blk)
4325{
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00004326 int error, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004327
4328 ASSERT(head_blk != tail_blk);
4329
4330 /*
4331 * First do a pass to find all of the cancelled buf log items.
4332 * Store them in the buf_cancel_table for use in the second pass.
4333 */
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00004334 log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE *
4335 sizeof(struct list_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07004336 KM_SLEEP);
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00004337 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
4338 INIT_LIST_HEAD(&log->l_buf_cancel_table[i]);
4339
Linus Torvalds1da177e2005-04-16 15:20:36 -07004340 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
4341 XLOG_RECOVER_PASS1);
4342 if (error != 0) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10004343 kmem_free(log->l_buf_cancel_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004344 log->l_buf_cancel_table = NULL;
4345 return error;
4346 }
4347 /*
4348 * Then do a second pass to actually recover the items in the log.
4349 * When it is complete free the table of buf cancel items.
4350 */
4351 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
4352 XLOG_RECOVER_PASS2);
4353#ifdef DEBUG
Tim Shimmin6d192a92006-06-09 14:55:38 +10004354 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004355 int i;
4356
4357 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00004358 ASSERT(list_empty(&log->l_buf_cancel_table[i]));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004359 }
4360#endif /* DEBUG */
4361
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10004362 kmem_free(log->l_buf_cancel_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004363 log->l_buf_cancel_table = NULL;
4364
4365 return error;
4366}
4367
4368/*
4369 * Do the actual recovery
4370 */
4371STATIC int
4372xlog_do_recover(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004373 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004374 xfs_daddr_t head_blk,
4375 xfs_daddr_t tail_blk)
4376{
4377 int error;
4378 xfs_buf_t *bp;
4379 xfs_sb_t *sbp;
4380
4381 /*
4382 * First replay the images in the log.
4383 */
4384 error = xlog_do_log_recovery(log, head_blk, tail_blk);
Christoph Hellwig43ff2122012-04-23 15:58:39 +10004385 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004386 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004387
4388 /*
4389 * If IO errors happened during recovery, bail out.
4390 */
4391 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
Dave Chinner24513372014-06-25 14:58:08 +10004392 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004393 }
4394
4395 /*
4396 * We now update the tail_lsn since much of the recovery has completed
4397 * and there may be space available to use. If there were no extent
4398 * or iunlinks, we can free up the entire log and set the tail_lsn to
4399 * be the last_sync_lsn. This was set in xlog_find_tail to be the
4400 * lsn of the last known good LR on disk. If there are extent frees
4401 * or iunlinks they will have some entries in the AIL; so we look at
4402 * the AIL to determine how to set the tail_lsn.
4403 */
4404 xlog_assign_tail_lsn(log->l_mp);
4405
4406 /*
4407 * Now that we've finished replaying all buffer and inode
Dave Chinner98021822012-11-12 22:54:03 +11004408 * updates, re-read in the superblock and reverify it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004409 */
4410 bp = xfs_getsb(log->l_mp, 0);
4411 XFS_BUF_UNDONE(bp);
Lachlan McIlroybebf9632007-10-15 13:18:02 +10004412 ASSERT(!(XFS_BUF_ISWRITE(bp)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004413 XFS_BUF_READ(bp);
Lachlan McIlroybebf9632007-10-15 13:18:02 +10004414 XFS_BUF_UNASYNC(bp);
Dave Chinner1813dd62012-11-14 17:54:40 +11004415 bp->b_ops = &xfs_sb_buf_ops;
Christoph Hellwig83a0adc2013-12-17 00:03:52 -08004416
4417 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
4418 xfs_buf_relse(bp);
Dave Chinner24513372014-06-25 14:58:08 +10004419 return -EIO;
Christoph Hellwig83a0adc2013-12-17 00:03:52 -08004420 }
4421
4422 xfs_buf_iorequest(bp);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +00004423 error = xfs_buf_iowait(bp);
David Chinnerd64e31a2008-04-10 12:22:17 +10004424 if (error) {
Christoph Hellwig901796a2011-10-10 16:52:49 +00004425 xfs_buf_ioerror_alert(bp, __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004426 ASSERT(0);
4427 xfs_buf_relse(bp);
4428 return error;
4429 }
4430
4431 /* Convert superblock from on-disk format */
4432 sbp = &log->l_mp->m_sb;
Dave Chinner98021822012-11-12 22:54:03 +11004433 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004434 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
Eric Sandeen62118702008-03-06 13:44:28 +11004435 ASSERT(xfs_sb_good_version(sbp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004436 xfs_buf_relse(bp);
4437
Lachlan McIlroy5478eea2007-02-10 18:36:29 +11004438 /* We've re-read the superblock so re-initialize per-cpu counters */
4439 xfs_icsb_reinit_counters(log->l_mp);
4440
Linus Torvalds1da177e2005-04-16 15:20:36 -07004441 xlog_recover_check_summary(log);
4442
4443 /* Normal transactions can now occur */
4444 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
4445 return 0;
4446}
4447
4448/*
4449 * Perform recovery and re-initialize some log variables in xlog_find_tail.
4450 *
4451 * Return error or zero.
4452 */
4453int
4454xlog_recover(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004455 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004456{
4457 xfs_daddr_t head_blk, tail_blk;
4458 int error;
4459
4460 /* find the tail of the log */
Eric Sandeen65be6052006-01-11 15:34:19 +11004461 if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004462 return error;
4463
4464 if (tail_blk != head_blk) {
4465 /* There used to be a comment here:
4466 *
4467 * disallow recovery on read-only mounts. note -- mount
4468 * checks for ENOSPC and turns it into an intelligent
4469 * error message.
4470 * ...but this is no longer true. Now, unless you specify
4471 * NORECOVERY (in which case this function would never be
4472 * called), we just go ahead and recover. We do this all
4473 * under the vfs layer, so we can get away with it unless
4474 * the device itself is read-only, in which case we fail.
4475 */
Utako Kusaka3a02ee12007-05-08 13:50:06 +10004476 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004477 return error;
4478 }
4479
Dave Chinnere721f502013-04-03 16:11:32 +11004480 /*
4481 * Version 5 superblock log feature mask validation. We know the
4482 * log is dirty so check if there are any unknown log features
4483 * in what we need to recover. If there are unknown features
4484 * (e.g. unsupported transactions, then simply reject the
4485 * attempt at recovery before touching anything.
4486 */
4487 if (XFS_SB_VERSION_NUM(&log->l_mp->m_sb) == XFS_SB_VERSION_5 &&
4488 xfs_sb_has_incompat_log_feature(&log->l_mp->m_sb,
4489 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
4490 xfs_warn(log->l_mp,
4491"Superblock has unknown incompatible log features (0x%x) enabled.\n"
4492"The log can not be fully and/or safely recovered by this kernel.\n"
4493"Please recover the log on a kernel that supports the unknown features.",
4494 (log->l_mp->m_sb.sb_features_log_incompat &
4495 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
Dave Chinner24513372014-06-25 14:58:08 +10004496 return -EINVAL;
Dave Chinnere721f502013-04-03 16:11:32 +11004497 }
4498
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004499 xfs_notice(log->l_mp, "Starting recovery (logdev: %s)",
4500 log->l_mp->m_logname ? log->l_mp->m_logname
4501 : "internal");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004502
4503 error = xlog_do_recover(log, head_blk, tail_blk);
4504 log->l_flags |= XLOG_RECOVERY_NEEDED;
4505 }
4506 return error;
4507}
4508
4509/*
4510 * In the first part of recovery we replay inodes and buffers and build
4511 * up the list of extent free items which need to be processed. Here
4512 * we process the extent free items and clean up the on disk unlinked
4513 * inode lists. This is separated from the first part of recovery so
4514 * that the root and real-time bitmap inodes can be read in from disk in
4515 * between the two stages. This is necessary so that we can free space
4516 * in the real-time portion of the file system.
4517 */
4518int
4519xlog_recover_finish(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004520 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004521{
4522 /*
4523 * Now we're ready to do the transactions needed for the
4524 * rest of recovery. Start with completing all the extent
4525 * free intent records and then process the unlinked inode
4526 * lists. At this point, we essentially run in normal mode
4527 * except that we're still performing recovery actions
4528 * rather than accepting new requests.
4529 */
4530 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
David Chinner3c1e2bb2008-04-10 12:21:11 +10004531 int error;
4532 error = xlog_recover_process_efis(log);
4533 if (error) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004534 xfs_alert(log->l_mp, "Failed to recover EFIs");
David Chinner3c1e2bb2008-04-10 12:21:11 +10004535 return error;
4536 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004537 /*
4538 * Sync the log to get all the EFIs out of the AIL.
4539 * This isn't absolutely necessary, but it helps in
4540 * case the unlink transactions would have problems
4541 * pushing the EFIs out of the way.
4542 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +00004543 xfs_log_force(log->l_mp, XFS_LOG_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004544
Christoph Hellwig42490232008-08-13 16:49:32 +10004545 xlog_recover_process_iunlinks(log);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004546
4547 xlog_recover_check_summary(log);
4548
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004549 xfs_notice(log->l_mp, "Ending recovery (logdev: %s)",
4550 log->l_mp->m_logname ? log->l_mp->m_logname
4551 : "internal");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004552 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
4553 } else {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004554 xfs_info(log->l_mp, "Ending clean mount");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004555 }
4556 return 0;
4557}
4558
4559
4560#if defined(DEBUG)
4561/*
4562 * Read all of the agf and agi counters and check that they
4563 * are consistent with the superblock counters.
4564 */
4565void
4566xlog_recover_check_summary(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004567 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004568{
4569 xfs_mount_t *mp;
4570 xfs_agf_t *agfp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004571 xfs_buf_t *agfbp;
4572 xfs_buf_t *agibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004573 xfs_agnumber_t agno;
4574 __uint64_t freeblks;
4575 __uint64_t itotal;
4576 __uint64_t ifree;
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11004577 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004578
4579 mp = log->l_mp;
4580
4581 freeblks = 0LL;
4582 itotal = 0LL;
4583 ifree = 0LL;
4584 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
From: Christoph Hellwig48056212008-11-28 14:23:38 +11004585 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
4586 if (error) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004587 xfs_alert(mp, "%s agf read failed agno %d error %d",
4588 __func__, agno, error);
From: Christoph Hellwig48056212008-11-28 14:23:38 +11004589 } else {
4590 agfp = XFS_BUF_TO_AGF(agfbp);
4591 freeblks += be32_to_cpu(agfp->agf_freeblks) +
4592 be32_to_cpu(agfp->agf_flcount);
4593 xfs_buf_relse(agfbp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004594 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004595
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11004596 error = xfs_read_agi(mp, NULL, agno, &agibp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004597 if (error) {
4598 xfs_alert(mp, "%s agi read failed agno %d error %d",
4599 __func__, agno, error);
4600 } else {
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11004601 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004602
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11004603 itotal += be32_to_cpu(agi->agi_count);
4604 ifree += be32_to_cpu(agi->agi_freecount);
4605 xfs_buf_relse(agibp);
4606 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004607 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004608}
4609#endif /* DEBUG */