blob: 1fd5787add9924d7ac726329f2ee88c13a90422d [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Tim Shimmin87c199c2006-06-09 14:56:16 +10002 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
Nathan Scott7b718762005-11-02 14:58:39 +11003 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110019#include "xfs_fs.h"
Dave Chinner70a9883c2013-10-23 10:36:05 +110020#include "xfs_shared.h"
Dave Chinner239880e2013-10-23 10:50:10 +110021#include "xfs_format.h"
22#include "xfs_log_format.h"
23#include "xfs_trans_resv.h"
Nathan Scotta844f452005-11-02 14:38:42 +110024#include "xfs_bit.h"
Nathan Scotta844f452005-11-02 14:38:42 +110025#include "xfs_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
Dave Chinner67dc2882014-08-04 12:43:06 +10002129 /*
2130 * We can only do post recovery validation on items on CRC enabled
2131 * fielsystems as we need to know when the buffer was written to be able
2132 * to determine if we should have replayed the item. If we replay old
2133 * metadata over a newer buffer, then it will enter a temporarily
2134 * inconsistent state resulting in verification failures. Hence for now
2135 * just avoid the verification stage for non-crc filesystems
2136 */
2137 if (!xfs_sb_version_hascrc(&mp->m_sb))
2138 return;
2139
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002140 magic32 = be32_to_cpu(*(__be32 *)bp->b_addr);
2141 magic16 = be16_to_cpu(*(__be16*)bp->b_addr);
2142 magicda = be16_to_cpu(info->magic);
Dave Chinner61fe1352013-04-03 16:11:30 +11002143 switch (xfs_blft_from_flags(buf_f)) {
2144 case XFS_BLFT_BTREE_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002145 switch (magic32) {
2146 case XFS_ABTB_CRC_MAGIC:
2147 case XFS_ABTC_CRC_MAGIC:
2148 case XFS_ABTB_MAGIC:
2149 case XFS_ABTC_MAGIC:
2150 bp->b_ops = &xfs_allocbt_buf_ops;
2151 break;
2152 case XFS_IBT_CRC_MAGIC:
Brian Fosteraafc3c22014-04-24 16:00:52 +10002153 case XFS_FIBT_CRC_MAGIC:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002154 case XFS_IBT_MAGIC:
Brian Fosteraafc3c22014-04-24 16:00:52 +10002155 case XFS_FIBT_MAGIC:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002156 bp->b_ops = &xfs_inobt_buf_ops;
2157 break;
2158 case XFS_BMAP_CRC_MAGIC:
2159 case XFS_BMAP_MAGIC:
2160 bp->b_ops = &xfs_bmbt_buf_ops;
2161 break;
2162 default:
2163 xfs_warn(mp, "Bad btree block magic!");
2164 ASSERT(0);
2165 break;
2166 }
2167 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002168 case XFS_BLFT_AGF_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002169 if (magic32 != XFS_AGF_MAGIC) {
2170 xfs_warn(mp, "Bad AGF block magic!");
2171 ASSERT(0);
2172 break;
2173 }
2174 bp->b_ops = &xfs_agf_buf_ops;
2175 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002176 case XFS_BLFT_AGFL_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002177 if (magic32 != XFS_AGFL_MAGIC) {
2178 xfs_warn(mp, "Bad AGFL block magic!");
2179 ASSERT(0);
2180 break;
2181 }
2182 bp->b_ops = &xfs_agfl_buf_ops;
2183 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002184 case XFS_BLFT_AGI_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002185 if (magic32 != XFS_AGI_MAGIC) {
2186 xfs_warn(mp, "Bad AGI block magic!");
2187 ASSERT(0);
2188 break;
2189 }
2190 bp->b_ops = &xfs_agi_buf_ops;
2191 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002192 case XFS_BLFT_UDQUOT_BUF:
2193 case XFS_BLFT_PDQUOT_BUF:
2194 case XFS_BLFT_GDQUOT_BUF:
Dave Chinner123887e2013-04-30 21:39:33 +10002195#ifdef CONFIG_XFS_QUOTA
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002196 if (magic16 != XFS_DQUOT_MAGIC) {
2197 xfs_warn(mp, "Bad DQUOT block magic!");
2198 ASSERT(0);
2199 break;
2200 }
2201 bp->b_ops = &xfs_dquot_buf_ops;
Dave Chinner123887e2013-04-30 21:39:33 +10002202#else
2203 xfs_alert(mp,
2204 "Trying to recover dquots without QUOTA support built in!");
2205 ASSERT(0);
2206#endif
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002207 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002208 case XFS_BLFT_DINO_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002209 if (magic16 != XFS_DINODE_MAGIC) {
2210 xfs_warn(mp, "Bad INODE block magic!");
2211 ASSERT(0);
2212 break;
2213 }
2214 bp->b_ops = &xfs_inode_buf_ops;
2215 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002216 case XFS_BLFT_SYMLINK_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002217 if (magic32 != XFS_SYMLINK_MAGIC) {
2218 xfs_warn(mp, "Bad symlink block magic!");
2219 ASSERT(0);
2220 break;
2221 }
2222 bp->b_ops = &xfs_symlink_buf_ops;
2223 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002224 case XFS_BLFT_DIR_BLOCK_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002225 if (magic32 != XFS_DIR2_BLOCK_MAGIC &&
2226 magic32 != XFS_DIR3_BLOCK_MAGIC) {
2227 xfs_warn(mp, "Bad dir block magic!");
2228 ASSERT(0);
2229 break;
2230 }
2231 bp->b_ops = &xfs_dir3_block_buf_ops;
2232 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002233 case XFS_BLFT_DIR_DATA_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002234 if (magic32 != XFS_DIR2_DATA_MAGIC &&
2235 magic32 != XFS_DIR3_DATA_MAGIC) {
2236 xfs_warn(mp, "Bad dir data magic!");
2237 ASSERT(0);
2238 break;
2239 }
2240 bp->b_ops = &xfs_dir3_data_buf_ops;
2241 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002242 case XFS_BLFT_DIR_FREE_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002243 if (magic32 != XFS_DIR2_FREE_MAGIC &&
2244 magic32 != XFS_DIR3_FREE_MAGIC) {
2245 xfs_warn(mp, "Bad dir3 free magic!");
2246 ASSERT(0);
2247 break;
2248 }
2249 bp->b_ops = &xfs_dir3_free_buf_ops;
2250 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002251 case XFS_BLFT_DIR_LEAF1_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002252 if (magicda != XFS_DIR2_LEAF1_MAGIC &&
2253 magicda != XFS_DIR3_LEAF1_MAGIC) {
2254 xfs_warn(mp, "Bad dir leaf1 magic!");
2255 ASSERT(0);
2256 break;
2257 }
2258 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
2259 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002260 case XFS_BLFT_DIR_LEAFN_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002261 if (magicda != XFS_DIR2_LEAFN_MAGIC &&
2262 magicda != XFS_DIR3_LEAFN_MAGIC) {
2263 xfs_warn(mp, "Bad dir leafn magic!");
2264 ASSERT(0);
2265 break;
2266 }
2267 bp->b_ops = &xfs_dir3_leafn_buf_ops;
2268 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002269 case XFS_BLFT_DA_NODE_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002270 if (magicda != XFS_DA_NODE_MAGIC &&
2271 magicda != XFS_DA3_NODE_MAGIC) {
2272 xfs_warn(mp, "Bad da node magic!");
2273 ASSERT(0);
2274 break;
2275 }
2276 bp->b_ops = &xfs_da3_node_buf_ops;
2277 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002278 case XFS_BLFT_ATTR_LEAF_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002279 if (magicda != XFS_ATTR_LEAF_MAGIC &&
2280 magicda != XFS_ATTR3_LEAF_MAGIC) {
2281 xfs_warn(mp, "Bad attr leaf magic!");
2282 ASSERT(0);
2283 break;
2284 }
2285 bp->b_ops = &xfs_attr3_leaf_buf_ops;
2286 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002287 case XFS_BLFT_ATTR_RMT_BUF:
Dave Chinnercab09a82013-04-30 21:39:36 +10002288 if (magic32 != XFS_ATTR3_RMT_MAGIC) {
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002289 xfs_warn(mp, "Bad attr remote magic!");
2290 ASSERT(0);
2291 break;
2292 }
2293 bp->b_ops = &xfs_attr3_rmt_buf_ops;
2294 break;
Dave Chinner04a1e6c2013-04-03 16:11:31 +11002295 case XFS_BLFT_SB_BUF:
2296 if (magic32 != XFS_SB_MAGIC) {
2297 xfs_warn(mp, "Bad SB block magic!");
2298 ASSERT(0);
2299 break;
2300 }
2301 bp->b_ops = &xfs_sb_buf_ops;
2302 break;
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002303 default:
Dave Chinner61fe1352013-04-03 16:11:30 +11002304 xfs_warn(mp, "Unknown buffer type %d!",
2305 xfs_blft_from_flags(buf_f));
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002306 break;
2307 }
2308}
2309
2310/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 * Perform a 'normal' buffer recovery. Each logged region of the
2312 * buffer should be copied over the corresponding region in the
2313 * given buffer. The bitmap in the buf log format structure indicates
2314 * where to place the logged data.
2315 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316STATIC void
2317xlog_recover_do_reg_buffer(
Dave Chinner9abbc532010-04-13 15:06:46 +10002318 struct xfs_mount *mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319 xlog_recover_item_t *item,
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002320 struct xfs_buf *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321 xfs_buf_log_format_t *buf_f)
2322{
2323 int i;
2324 int bit;
2325 int nbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 int error;
2327
Dave Chinner9abbc532010-04-13 15:06:46 +10002328 trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f);
2329
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 bit = 0;
2331 i = 1; /* 0 is the buf format structure */
2332 while (1) {
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002333 bit = xfs_next_bit(buf_f->blf_data_map,
2334 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 if (bit == -1)
2336 break;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002337 nbits = xfs_contig_bits(buf_f->blf_data_map,
2338 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339 ASSERT(nbits > 0);
Christoph Hellwig4b809162007-08-16 15:37:36 +10002340 ASSERT(item->ri_buf[i].i_addr != NULL);
Dave Chinnerc1155412010-05-07 11:05:19 +10002341 ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0);
Dave Chinneraa0e8832012-04-23 15:58:52 +10002342 ASSERT(BBTOB(bp->b_io_length) >=
2343 ((uint)bit << XFS_BLF_SHIFT) + (nbits << XFS_BLF_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344
2345 /*
Dave Chinner709da6a2013-05-27 16:38:23 +10002346 * The dirty regions logged in the buffer, even though
2347 * contiguous, may span multiple chunks. This is because the
2348 * dirty region may span a physical page boundary in a buffer
2349 * and hence be split into two separate vectors for writing into
2350 * the log. Hence we need to trim nbits back to the length of
2351 * the current region being copied out of the log.
2352 */
2353 if (item->ri_buf[i].i_len < (nbits << XFS_BLF_SHIFT))
2354 nbits = item->ri_buf[i].i_len >> XFS_BLF_SHIFT;
2355
2356 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357 * Do a sanity check if this is a dquot buffer. Just checking
2358 * the first dquot in the buffer should do. XXXThis is
2359 * probably a good thing to do for other buf types also.
2360 */
2361 error = 0;
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002362 if (buf_f->blf_flags &
Dave Chinnerc1155412010-05-07 11:05:19 +10002363 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002364 if (item->ri_buf[i].i_addr == NULL) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002365 xfs_alert(mp,
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002366 "XFS: NULL dquot in %s.", __func__);
2367 goto next;
2368 }
Jan Rekorajski8ec6dba2009-11-16 11:57:02 +00002369 if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002370 xfs_alert(mp,
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002371 "XFS: dquot too small (%d) in %s.",
2372 item->ri_buf[i].i_len, __func__);
2373 goto next;
2374 }
Dave Chinner9aede1d2013-10-15 09:17:52 +11002375 error = xfs_dqcheck(mp, item->ri_buf[i].i_addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376 -1, 0, XFS_QMOPT_DOWARN,
2377 "dquot_buf_recover");
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002378 if (error)
2379 goto next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380 }
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002381
2382 memcpy(xfs_buf_offset(bp,
Dave Chinnerc1155412010-05-07 11:05:19 +10002383 (uint)bit << XFS_BLF_SHIFT), /* dest */
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002384 item->ri_buf[i].i_addr, /* source */
Dave Chinnerc1155412010-05-07 11:05:19 +10002385 nbits<<XFS_BLF_SHIFT); /* length */
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002386 next:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387 i++;
2388 bit += nbits;
2389 }
2390
2391 /* Shouldn't be any more regions */
2392 ASSERT(i == item->ri_total);
Christoph Hellwigee1a47a2013-04-21 14:53:46 -05002393
Dave Chinner67dc2882014-08-04 12:43:06 +10002394 xlog_recover_validate_buf_type(mp, bp, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395}
2396
2397/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398 * Perform a dquot buffer recovery.
Zhi Yong Wu8ba701e2013-08-12 03:15:01 +00002399 * Simple algorithm: if we have found a QUOTAOFF log item of the same type
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2401 * Else, treat it as a regular buffer and do recovery.
Dave Chinnerad3714b2014-08-04 12:59:31 +10002402 *
2403 * Return false if the buffer was tossed and true if we recovered the buffer to
2404 * indicate to the caller if the buffer needs writing.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405 */
Dave Chinnerad3714b2014-08-04 12:59:31 +10002406STATIC bool
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407xlog_recover_do_dquot_buffer(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002408 struct xfs_mount *mp,
2409 struct xlog *log,
2410 struct xlog_recover_item *item,
2411 struct xfs_buf *bp,
2412 struct xfs_buf_log_format *buf_f)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413{
2414 uint type;
2415
Dave Chinner9abbc532010-04-13 15:06:46 +10002416 trace_xfs_log_recover_buf_dquot_buf(log, buf_f);
2417
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418 /*
2419 * Filesystems are required to send in quota flags at mount time.
2420 */
Dave Chinnerad3714b2014-08-04 12:59:31 +10002421 if (!mp->m_qflags)
2422 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423
2424 type = 0;
Dave Chinnerc1155412010-05-07 11:05:19 +10002425 if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426 type |= XFS_DQ_USER;
Dave Chinnerc1155412010-05-07 11:05:19 +10002427 if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF)
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002428 type |= XFS_DQ_PROJ;
Dave Chinnerc1155412010-05-07 11:05:19 +10002429 if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430 type |= XFS_DQ_GROUP;
2431 /*
2432 * This type of quotas was turned off, so ignore this buffer
2433 */
2434 if (log->l_quotaoffs_flag & type)
Dave Chinnerad3714b2014-08-04 12:59:31 +10002435 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436
Dave Chinner9abbc532010-04-13 15:06:46 +10002437 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
Dave Chinnerad3714b2014-08-04 12:59:31 +10002438 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439}
2440
2441/*
2442 * This routine replays a modification made to a buffer at runtime.
2443 * There are actually two types of buffer, regular and inode, which
2444 * are handled differently. Inode buffers are handled differently
2445 * in that we only recover a specific set of data from them, namely
2446 * the inode di_next_unlinked fields. This is because all other inode
2447 * data is actually logged via inode records and any data we replay
2448 * here which overlaps that may be stale.
2449 *
2450 * When meta-data buffers are freed at run time we log a buffer item
Dave Chinnerc1155412010-05-07 11:05:19 +10002451 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452 * of the buffer in the log should not be replayed at recovery time.
2453 * This is so that if the blocks covered by the buffer are reused for
2454 * file data before we crash we don't end up replaying old, freed
2455 * meta-data into a user's file.
2456 *
2457 * To handle the cancellation of buffer log items, we make two passes
2458 * over the log during recovery. During the first we build a table of
2459 * those buffers which have been cancelled, and during the second we
2460 * only replay those buffers which do not have corresponding cancel
Zhi Yong Wu34be5ff2013-08-07 10:11:07 +00002461 * records in the table. See xlog_recover_buffer_pass[1,2] above
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462 * for more details on the implementation of the table of cancel records.
2463 */
2464STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002465xlog_recover_buffer_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002466 struct xlog *log,
2467 struct list_head *buffer_list,
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002468 struct xlog_recover_item *item,
2469 xfs_lsn_t current_lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470{
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002471 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002472 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473 xfs_buf_t *bp;
2474 int error;
Christoph Hellwig6ad112b2009-11-24 18:02:23 +00002475 uint buf_flags;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002476 xfs_lsn_t lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002478 /*
2479 * In this pass we only want to recover all the buffers which have
2480 * not been cancelled and are not cancellation buffers themselves.
2481 */
2482 if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno,
2483 buf_f->blf_len, buf_f->blf_flags)) {
2484 trace_xfs_log_recover_buf_cancel(log, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002486 }
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002487
Dave Chinner9abbc532010-04-13 15:06:46 +10002488 trace_xfs_log_recover_buf_recover(log, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489
Dave Chinnera8acad72012-04-23 15:58:54 +10002490 buf_flags = 0;
Dave Chinner611c9942012-04-23 15:59:07 +10002491 if (buf_f->blf_flags & XFS_BLF_INODE_BUF)
2492 buf_flags |= XBF_UNMAPPED;
Christoph Hellwig6ad112b2009-11-24 18:02:23 +00002493
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002494 bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len,
Dave Chinnerc3f8fc72012-11-12 22:54:01 +11002495 buf_flags, NULL);
Chandra Seetharamanac4d6882011-08-03 02:18:29 +00002496 if (!bp)
Dave Chinner24513372014-06-25 14:58:08 +10002497 return -ENOMEM;
Chandra Seetharamane5702802011-08-03 02:18:34 +00002498 error = bp->b_error;
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +00002499 if (error) {
Christoph Hellwig901796a2011-10-10 16:52:49 +00002500 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)");
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002501 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502 }
2503
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002504 /*
Dave Chinner67dc2882014-08-04 12:43:06 +10002505 * Recover the buffer only if we get an LSN from it and it's less than
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002506 * the lsn of the transaction we are replaying.
Dave Chinner67dc2882014-08-04 12:43:06 +10002507 *
2508 * Note that we have to be extremely careful of readahead here.
2509 * Readahead does not attach verfiers to the buffers so if we don't
2510 * actually do any replay after readahead because of the LSN we found
2511 * in the buffer if more recent than that current transaction then we
2512 * need to attach the verifier directly. Failure to do so can lead to
2513 * future recovery actions (e.g. EFI and unlinked list recovery) can
2514 * operate on the buffers and they won't get the verifier attached. This
2515 * can lead to blocks on disk having the correct content but a stale
2516 * CRC.
2517 *
2518 * It is safe to assume these clean buffers are currently up to date.
2519 * If the buffer is dirtied by a later transaction being replayed, then
2520 * the verifier will be reset to match whatever recover turns that
2521 * buffer into.
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002522 */
2523 lsn = xlog_recover_get_buf_lsn(mp, bp);
Dave Chinner67dc2882014-08-04 12:43:06 +10002524 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
2525 xlog_recover_validate_buf_type(mp, bp, buf_f);
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002526 goto out_release;
Dave Chinner67dc2882014-08-04 12:43:06 +10002527 }
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002528
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002529 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
Dave Chinnerad3714b2014-08-04 12:59:31 +10002531 if (error)
2532 goto out_release;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002533 } else if (buf_f->blf_flags &
Dave Chinnerc1155412010-05-07 11:05:19 +10002534 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
Dave Chinnerad3714b2014-08-04 12:59:31 +10002535 bool dirty;
2536
2537 dirty = xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2538 if (!dirty)
2539 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540 } else {
Dave Chinner9abbc532010-04-13 15:06:46 +10002541 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543
2544 /*
2545 * Perform delayed write on the buffer. Asynchronous writes will be
2546 * slower when taking into account all the buffers to be flushed.
2547 *
2548 * Also make sure that only inode buffers with good sizes stay in
2549 * the buffer cache. The kernel moves inodes in buffers of 1 block
Jie Liu0f49efd2013-12-13 15:51:48 +11002550 * or mp->m_inode_cluster_size bytes, whichever is bigger. The inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 * buffers in the log can be a different size if the log was generated
2552 * by an older kernel using unclustered inode buffers or a newer kernel
2553 * running with a different inode cluster size. Regardless, if the
Jie Liu0f49efd2013-12-13 15:51:48 +11002554 * the inode buffer size isn't MAX(blocksize, mp->m_inode_cluster_size)
2555 * for *our* value of mp->m_inode_cluster_size, then we need to keep
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556 * the buffer out of the buffer cache so that the buffer won't
2557 * overlap with future reads of those inodes.
2558 */
2559 if (XFS_DINODE_MAGIC ==
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002560 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
Dave Chinneraa0e8832012-04-23 15:58:52 +10002561 (BBTOB(bp->b_io_length) != MAX(log->l_mp->m_sb.sb_blocksize,
Jie Liu0f49efd2013-12-13 15:51:48 +11002562 (__uint32_t)log->l_mp->m_inode_cluster_size))) {
Christoph Hellwigc867cb62011-10-10 16:52:46 +00002563 xfs_buf_stale(bp);
Christoph Hellwigc2b006c2011-08-23 08:28:07 +00002564 error = xfs_bwrite(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 } else {
Dave Chinnerebad8612010-09-22 10:47:20 +10002566 ASSERT(bp->b_target->bt_mount == mp);
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02002567 bp->b_iodone = xlog_recover_iodone;
Christoph Hellwig43ff2122012-04-23 15:58:39 +10002568 xfs_buf_delwri_queue(bp, buffer_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569 }
2570
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002571out_release:
Christoph Hellwigc2b006c2011-08-23 08:28:07 +00002572 xfs_buf_relse(bp);
2573 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574}
2575
Dave Chinner638f44162013-08-30 10:23:45 +10002576/*
2577 * Inode fork owner changes
2578 *
2579 * If we have been told that we have to reparent the inode fork, it's because an
2580 * extent swap operation on a CRC enabled filesystem has been done and we are
2581 * replaying it. We need to walk the BMBT of the appropriate fork and change the
2582 * owners of it.
2583 *
2584 * The complexity here is that we don't have an inode context to work with, so
2585 * after we've replayed the inode we need to instantiate one. This is where the
2586 * fun begins.
2587 *
2588 * We are in the middle of log recovery, so we can't run transactions. That
2589 * means we cannot use cache coherent inode instantiation via xfs_iget(), as
2590 * that will result in the corresponding iput() running the inode through
2591 * xfs_inactive(). If we've just replayed an inode core that changes the link
2592 * count to zero (i.e. it's been unlinked), then xfs_inactive() will run
2593 * transactions (bad!).
2594 *
2595 * So, to avoid this, we instantiate an inode directly from the inode core we've
2596 * just recovered. We have the buffer still locked, and all we really need to
2597 * instantiate is the inode core and the forks being modified. We can do this
2598 * manually, then run the inode btree owner change, and then tear down the
2599 * xfs_inode without having to run any transactions at all.
2600 *
2601 * Also, because we don't have a transaction context available here but need to
2602 * gather all the buffers we modify for writeback so we pass the buffer_list
2603 * instead for the operation to use.
2604 */
2605
2606STATIC int
2607xfs_recover_inode_owner_change(
2608 struct xfs_mount *mp,
2609 struct xfs_dinode *dip,
2610 struct xfs_inode_log_format *in_f,
2611 struct list_head *buffer_list)
2612{
2613 struct xfs_inode *ip;
2614 int error;
2615
2616 ASSERT(in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER));
2617
2618 ip = xfs_inode_alloc(mp, in_f->ilf_ino);
2619 if (!ip)
Dave Chinner24513372014-06-25 14:58:08 +10002620 return -ENOMEM;
Dave Chinner638f44162013-08-30 10:23:45 +10002621
2622 /* instantiate the inode */
2623 xfs_dinode_from_disk(&ip->i_d, dip);
2624 ASSERT(ip->i_d.di_version >= 3);
2625
2626 error = xfs_iformat_fork(ip, dip);
2627 if (error)
2628 goto out_free_ip;
2629
2630
2631 if (in_f->ilf_fields & XFS_ILOG_DOWNER) {
2632 ASSERT(in_f->ilf_fields & XFS_ILOG_DBROOT);
2633 error = xfs_bmbt_change_owner(NULL, ip, XFS_DATA_FORK,
2634 ip->i_ino, buffer_list);
2635 if (error)
2636 goto out_free_ip;
2637 }
2638
2639 if (in_f->ilf_fields & XFS_ILOG_AOWNER) {
2640 ASSERT(in_f->ilf_fields & XFS_ILOG_ABROOT);
2641 error = xfs_bmbt_change_owner(NULL, ip, XFS_ATTR_FORK,
2642 ip->i_ino, buffer_list);
2643 if (error)
2644 goto out_free_ip;
2645 }
2646
2647out_free_ip:
2648 xfs_inode_free(ip);
2649 return error;
2650}
2651
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002653xlog_recover_inode_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002654 struct xlog *log,
2655 struct list_head *buffer_list,
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002656 struct xlog_recover_item *item,
2657 xfs_lsn_t current_lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658{
2659 xfs_inode_log_format_t *in_f;
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002660 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661 xfs_buf_t *bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 xfs_dinode_t *dip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663 int len;
2664 xfs_caddr_t src;
2665 xfs_caddr_t dest;
2666 int error;
2667 int attr_index;
2668 uint fields;
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002669 xfs_icdinode_t *dicp;
Christoph Hellwig93848a92013-04-03 16:11:17 +11002670 uint isize;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002671 int need_free = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672
Tim Shimmin6d192a92006-06-09 14:55:38 +10002673 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002674 in_f = item->ri_buf[0].i_addr;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002675 } else {
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002676 in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002677 need_free = 1;
2678 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2679 if (error)
2680 goto error;
2681 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682
2683 /*
2684 * Inode buffers can be freed, look out for it,
2685 * and do not replay the inode.
2686 */
Christoph Hellwiga1941892008-11-28 14:23:40 +11002687 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
2688 in_f->ilf_len, 0)) {
Tim Shimmin6d192a92006-06-09 14:55:38 +10002689 error = 0;
Dave Chinner9abbc532010-04-13 15:06:46 +10002690 trace_xfs_log_recover_inode_cancel(log, in_f);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002691 goto error;
2692 }
Dave Chinner9abbc532010-04-13 15:06:46 +10002693 trace_xfs_log_recover_inode_recover(log, in_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002694
Dave Chinnerc3f8fc72012-11-12 22:54:01 +11002695 bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len, 0,
Christoph Hellwig93848a92013-04-03 16:11:17 +11002696 &xfs_inode_buf_ops);
Chandra Seetharamanac4d6882011-08-03 02:18:29 +00002697 if (!bp) {
Dave Chinner24513372014-06-25 14:58:08 +10002698 error = -ENOMEM;
Chandra Seetharamanac4d6882011-08-03 02:18:29 +00002699 goto error;
2700 }
Chandra Seetharamane5702802011-08-03 02:18:34 +00002701 error = bp->b_error;
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +00002702 if (error) {
Christoph Hellwig901796a2011-10-10 16:52:49 +00002703 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)");
Dave Chinner638f44162013-08-30 10:23:45 +10002704 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
Christoph Hellwiga1941892008-11-28 14:23:40 +11002707 dip = (xfs_dinode_t *)xfs_buf_offset(bp, in_f->ilf_boffset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708
2709 /*
2710 * Make sure the place we're flushing out to really looks
2711 * like an inode!
2712 */
Christoph Hellwig69ef9212011-07-08 14:36:05 +02002713 if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002714 xfs_alert(mp,
2715 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
2716 __func__, dip, bp, in_f->ilf_ino);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002717 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718 XFS_ERRLEVEL_LOW, mp);
Dave Chinner24513372014-06-25 14:58:08 +10002719 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002720 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002721 }
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002722 dicp = item->ri_buf[1].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002723 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002724 xfs_alert(mp,
2725 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
2726 __func__, item, in_f->ilf_ino);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002727 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728 XFS_ERRLEVEL_LOW, mp);
Dave Chinner24513372014-06-25 14:58:08 +10002729 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002730 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002731 }
2732
Dave Chinnere60896d2013-07-24 15:47:30 +10002733 /*
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002734 * If the inode has an LSN in it, recover the inode only if it's less
Dave Chinner638f44162013-08-30 10:23:45 +10002735 * than the lsn of the transaction we are replaying. Note: we still
2736 * need to replay an owner change even though the inode is more recent
2737 * than the transaction as there is no guarantee that all the btree
2738 * blocks are more recent than this transaction, too.
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002739 */
2740 if (dip->di_version >= 3) {
2741 xfs_lsn_t lsn = be64_to_cpu(dip->di_lsn);
2742
2743 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
2744 trace_xfs_log_recover_inode_skip(log, in_f);
2745 error = 0;
Dave Chinner638f44162013-08-30 10:23:45 +10002746 goto out_owner_change;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002747 }
2748 }
2749
2750 /*
Dave Chinnere60896d2013-07-24 15:47:30 +10002751 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
2752 * are transactional and if ordering is necessary we can determine that
2753 * more accurately by the LSN field in the V3 inode core. Don't trust
2754 * the inode versions we might be changing them here - use the
2755 * superblock flag to determine whether we need to look at di_flushiter
2756 * to skip replay when the on disk inode is newer than the log one
2757 */
2758 if (!xfs_sb_version_hascrc(&mp->m_sb) &&
2759 dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760 /*
2761 * Deal with the wrap case, DI_MAX_FLUSH is less
2762 * than smaller numbers
2763 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002764 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002765 dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002766 /* do nothing */
2767 } else {
Dave Chinner9abbc532010-04-13 15:06:46 +10002768 trace_xfs_log_recover_inode_skip(log, in_f);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002769 error = 0;
Dave Chinner638f44162013-08-30 10:23:45 +10002770 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771 }
2772 }
Dave Chinnere60896d2013-07-24 15:47:30 +10002773
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774 /* Take the opportunity to reset the flush iteration count */
2775 dicp->di_flushiter = 0;
2776
Al Viroabbede12011-07-26 02:31:30 -04002777 if (unlikely(S_ISREG(dicp->di_mode))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2779 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002780 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781 XFS_ERRLEVEL_LOW, mp, dicp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002782 xfs_alert(mp,
2783 "%s: Bad regular inode log record, rec ptr 0x%p, "
2784 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2785 __func__, item, dip, bp, in_f->ilf_ino);
Dave Chinner24513372014-06-25 14:58:08 +10002786 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002787 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788 }
Al Viroabbede12011-07-26 02:31:30 -04002789 } else if (unlikely(S_ISDIR(dicp->di_mode))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002790 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2791 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2792 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002793 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794 XFS_ERRLEVEL_LOW, mp, dicp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002795 xfs_alert(mp,
2796 "%s: Bad dir inode log record, rec ptr 0x%p, "
2797 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2798 __func__, item, dip, bp, in_f->ilf_ino);
Dave Chinner24513372014-06-25 14:58:08 +10002799 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002800 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002801 }
2802 }
2803 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002804 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002805 XFS_ERRLEVEL_LOW, mp, dicp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002806 xfs_alert(mp,
2807 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2808 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2809 __func__, item, dip, bp, in_f->ilf_ino,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810 dicp->di_nextents + dicp->di_anextents,
2811 dicp->di_nblocks);
Dave Chinner24513372014-06-25 14:58:08 +10002812 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002813 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814 }
2815 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002816 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817 XFS_ERRLEVEL_LOW, mp, dicp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002818 xfs_alert(mp,
2819 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2820 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__,
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002821 item, dip, bp, in_f->ilf_ino, dicp->di_forkoff);
Dave Chinner24513372014-06-25 14:58:08 +10002822 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002823 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002824 }
Christoph Hellwig93848a92013-04-03 16:11:17 +11002825 isize = xfs_icdinode_size(dicp->di_version);
2826 if (unlikely(item->ri_buf[1].i_len > isize)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002827 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828 XFS_ERRLEVEL_LOW, mp, dicp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002829 xfs_alert(mp,
2830 "%s: Bad inode log record length %d, rec ptr 0x%p",
2831 __func__, item->ri_buf[1].i_len, item);
Dave Chinner24513372014-06-25 14:58:08 +10002832 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002833 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834 }
2835
2836 /* The core is in in-core format */
Christoph Hellwig93848a92013-04-03 16:11:17 +11002837 xfs_dinode_to_disk(dip, dicp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838
2839 /* the rest is in on-disk format */
Christoph Hellwig93848a92013-04-03 16:11:17 +11002840 if (item->ri_buf[1].i_len > isize) {
2841 memcpy((char *)dip + isize,
2842 item->ri_buf[1].i_addr + isize,
2843 item->ri_buf[1].i_len - isize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844 }
2845
2846 fields = in_f->ilf_fields;
2847 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2848 case XFS_ILOG_DEV:
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002849 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002850 break;
2851 case XFS_ILOG_UUID:
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002852 memcpy(XFS_DFORK_DPTR(dip),
2853 &in_f->ilf_u.ilfu_uuid,
2854 sizeof(uuid_t));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855 break;
2856 }
2857
2858 if (in_f->ilf_size == 2)
Dave Chinner638f44162013-08-30 10:23:45 +10002859 goto out_owner_change;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002860 len = item->ri_buf[2].i_len;
2861 src = item->ri_buf[2].i_addr;
2862 ASSERT(in_f->ilf_size <= 4);
2863 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2864 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2865 (len == in_f->ilf_dsize));
2866
2867 switch (fields & XFS_ILOG_DFORK) {
2868 case XFS_ILOG_DDATA:
2869 case XFS_ILOG_DEXT:
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002870 memcpy(XFS_DFORK_DPTR(dip), src, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002871 break;
2872
2873 case XFS_ILOG_DBROOT:
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002874 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002875 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002876 XFS_DFORK_DSIZE(dip, mp));
2877 break;
2878
2879 default:
2880 /*
2881 * There are no data fork flags set.
2882 */
2883 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2884 break;
2885 }
2886
2887 /*
2888 * If we logged any attribute data, recover it. There may or
2889 * may not have been any other non-core data logged in this
2890 * transaction.
2891 */
2892 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2893 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2894 attr_index = 3;
2895 } else {
2896 attr_index = 2;
2897 }
2898 len = item->ri_buf[attr_index].i_len;
2899 src = item->ri_buf[attr_index].i_addr;
2900 ASSERT(len == in_f->ilf_asize);
2901
2902 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2903 case XFS_ILOG_ADATA:
2904 case XFS_ILOG_AEXT:
2905 dest = XFS_DFORK_APTR(dip);
2906 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2907 memcpy(dest, src, len);
2908 break;
2909
2910 case XFS_ILOG_ABROOT:
2911 dest = XFS_DFORK_APTR(dip);
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002912 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
2913 len, (xfs_bmdr_block_t*)dest,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002914 XFS_DFORK_ASIZE(dip, mp));
2915 break;
2916
2917 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002918 xfs_warn(log->l_mp, "%s: Invalid flag", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002919 ASSERT(0);
Dave Chinner24513372014-06-25 14:58:08 +10002920 error = -EIO;
Dave Chinner638f44162013-08-30 10:23:45 +10002921 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002922 }
2923 }
2924
Dave Chinner638f44162013-08-30 10:23:45 +10002925out_owner_change:
2926 if (in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER))
2927 error = xfs_recover_inode_owner_change(mp, dip, in_f,
2928 buffer_list);
Christoph Hellwig93848a92013-04-03 16:11:17 +11002929 /* re-generate the checksum. */
2930 xfs_dinode_calc_crc(log->l_mp, dip);
2931
Dave Chinnerebad8612010-09-22 10:47:20 +10002932 ASSERT(bp->b_target->bt_mount == mp);
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02002933 bp->b_iodone = xlog_recover_iodone;
Christoph Hellwig43ff2122012-04-23 15:58:39 +10002934 xfs_buf_delwri_queue(bp, buffer_list);
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002935
2936out_release:
Christoph Hellwig61551f12011-08-23 08:28:06 +00002937 xfs_buf_relse(bp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002938error:
2939 if (need_free)
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002940 kmem_free(in_f);
Eric Sandeenb474c7a2014-06-22 15:04:54 +10002941 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002942}
2943
2944/*
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002945 * Recover QUOTAOFF records. We simply make a note of it in the xlog
Linus Torvalds1da177e2005-04-16 15:20:36 -07002946 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2947 * of that type.
2948 */
2949STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002950xlog_recover_quotaoff_pass1(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002951 struct xlog *log,
2952 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002953{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002954 xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002955 ASSERT(qoff_f);
2956
2957 /*
2958 * The logitem format's flag tells us if this was user quotaoff,
Nathan Scott77a7cce2006-01-11 15:35:57 +11002959 * group/project quotaoff or both.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002960 */
2961 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2962 log->l_quotaoffs_flag |= XFS_DQ_USER;
Nathan Scott77a7cce2006-01-11 15:35:57 +11002963 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2964 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002965 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2966 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2967
Eric Sandeend99831f2014-06-22 15:03:54 +10002968 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002969}
2970
2971/*
2972 * Recover a dquot record
2973 */
2974STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002975xlog_recover_dquot_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002976 struct xlog *log,
2977 struct list_head *buffer_list,
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002978 struct xlog_recover_item *item,
2979 xfs_lsn_t current_lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002980{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002981 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002982 xfs_buf_t *bp;
2983 struct xfs_disk_dquot *ddq, *recddq;
2984 int error;
2985 xfs_dq_logformat_t *dq_f;
2986 uint type;
2987
Linus Torvalds1da177e2005-04-16 15:20:36 -07002988
2989 /*
2990 * Filesystems are required to send in quota flags at mount time.
2991 */
2992 if (mp->m_qflags == 0)
Eric Sandeend99831f2014-06-22 15:03:54 +10002993 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002994
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002995 recddq = item->ri_buf[1].i_addr;
2996 if (recddq == NULL) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002997 xfs_alert(log->l_mp, "NULL dquot in %s.", __func__);
Dave Chinner24513372014-06-25 14:58:08 +10002998 return -EIO;
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002999 }
Jan Rekorajski8ec6dba2009-11-16 11:57:02 +00003000 if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003001 xfs_alert(log->l_mp, "dquot too small (%d) in %s.",
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02003002 item->ri_buf[1].i_len, __func__);
Dave Chinner24513372014-06-25 14:58:08 +10003003 return -EIO;
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02003004 }
3005
Linus Torvalds1da177e2005-04-16 15:20:36 -07003006 /*
3007 * This type of quotas was turned off, so ignore this record.
3008 */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003009 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003010 ASSERT(type);
3011 if (log->l_quotaoffs_flag & type)
Eric Sandeend99831f2014-06-22 15:03:54 +10003012 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003013
3014 /*
3015 * At this point we know that quota was _not_ turned off.
3016 * Since the mount flags are not indicating to us otherwise, this
3017 * must mean that quota is on, and the dquot needs to be replayed.
3018 * Remember that we may not have fully recovered the superblock yet,
3019 * so we can't do the usual trick of looking at the SB quota bits.
3020 *
3021 * The other possibility, of course, is that the quota subsystem was
3022 * removed since the last mount - ENOSYS.
3023 */
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10003024 dq_f = item->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003025 ASSERT(dq_f);
Dave Chinner9aede1d2013-10-15 09:17:52 +11003026 error = xfs_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003027 "xlog_recover_dquot_pass2 (log copy)");
3028 if (error)
Dave Chinner24513372014-06-25 14:58:08 +10003029 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003030 ASSERT(dq_f->qlf_len == 1);
3031
Dave Chinnerad3714b2014-08-04 12:59:31 +10003032 /*
3033 * At this point we are assuming that the dquots have been allocated
3034 * and hence the buffer has valid dquots stamped in it. It should,
3035 * therefore, pass verifier validation. If the dquot is bad, then the
3036 * we'll return an error here, so we don't need to specifically check
3037 * the dquot in the buffer after the verifier has run.
3038 */
Dave Chinner7ca790a2012-04-23 15:58:55 +10003039 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dq_f->qlf_blkno,
Dave Chinnerc3f8fc72012-11-12 22:54:01 +11003040 XFS_FSB_TO_BB(mp, dq_f->qlf_len), 0, &bp,
Dave Chinnerad3714b2014-08-04 12:59:31 +10003041 &xfs_dquot_buf_ops);
Dave Chinner7ca790a2012-04-23 15:58:55 +10003042 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003043 return error;
Dave Chinner7ca790a2012-04-23 15:58:55 +10003044
Linus Torvalds1da177e2005-04-16 15:20:36 -07003045 ASSERT(bp);
3046 ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
3047
3048 /*
Dave Chinner50d5c8d2013-08-28 21:22:47 +10003049 * If the dquot has an LSN in it, recover the dquot only if it's less
3050 * than the lsn of the transaction we are replaying.
3051 */
3052 if (xfs_sb_version_hascrc(&mp->m_sb)) {
3053 struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddq;
3054 xfs_lsn_t lsn = be64_to_cpu(dqb->dd_lsn);
3055
3056 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
3057 goto out_release;
3058 }
3059 }
3060
Linus Torvalds1da177e2005-04-16 15:20:36 -07003061 memcpy(ddq, recddq, item->ri_buf[1].i_len);
Dave Chinner6fcdc592013-06-03 15:28:46 +10003062 if (xfs_sb_version_hascrc(&mp->m_sb)) {
3063 xfs_update_cksum((char *)ddq, sizeof(struct xfs_dqblk),
3064 XFS_DQUOT_CRC_OFF);
3065 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003066
3067 ASSERT(dq_f->qlf_size == 2);
Dave Chinnerebad8612010-09-22 10:47:20 +10003068 ASSERT(bp->b_target->bt_mount == mp);
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02003069 bp->b_iodone = xlog_recover_iodone;
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003070 xfs_buf_delwri_queue(bp, buffer_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003071
Dave Chinner50d5c8d2013-08-28 21:22:47 +10003072out_release:
3073 xfs_buf_relse(bp);
3074 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003075}
3076
3077/*
3078 * This routine is called to create an in-core extent free intent
3079 * item from the efi format structure which was logged on disk.
3080 * It allocates an in-core efi, copies the extents from the format
3081 * structure into it, and adds the efi to the AIL with the given
3082 * LSN.
3083 */
Tim Shimmin6d192a92006-06-09 14:55:38 +10003084STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003085xlog_recover_efi_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003086 struct xlog *log,
3087 struct xlog_recover_item *item,
3088 xfs_lsn_t lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003089{
Tim Shimmin6d192a92006-06-09 14:55:38 +10003090 int error;
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003091 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003092 xfs_efi_log_item_t *efip;
3093 xfs_efi_log_format_t *efi_formatp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003094
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10003095 efi_formatp = item->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003096
Linus Torvalds1da177e2005-04-16 15:20:36 -07003097 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
Tim Shimmin6d192a92006-06-09 14:55:38 +10003098 if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
3099 &(efip->efi_format)))) {
3100 xfs_efi_item_free(efip);
3101 return error;
3102 }
Dave Chinnerb199c8a2010-12-20 11:59:49 +11003103 atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003104
David Chinnera9c21c12008-10-30 17:39:35 +11003105 spin_lock(&log->l_ailp->xa_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003106 /*
David Chinner783a2f62008-10-30 17:39:58 +11003107 * xfs_trans_ail_update() drops the AIL lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003108 */
Dave Chinnere6059942010-12-20 12:34:26 +11003109 xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn);
Tim Shimmin6d192a92006-06-09 14:55:38 +10003110 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003111}
3112
3113
3114/*
3115 * This routine is called when an efd format structure is found in
3116 * a committed transaction in the log. It's purpose is to cancel
3117 * the corresponding efi if it was still in the log. To do this
3118 * it searches the AIL for the efi with an id equal to that in the
3119 * efd format structure. If we find it, we remove the efi from the
3120 * AIL and free it.
3121 */
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003122STATIC int
3123xlog_recover_efd_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003124 struct xlog *log,
3125 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003126{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003127 xfs_efd_log_format_t *efd_formatp;
3128 xfs_efi_log_item_t *efip = NULL;
3129 xfs_log_item_t *lip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003130 __uint64_t efi_id;
David Chinner27d8d5f2008-10-30 17:38:39 +11003131 struct xfs_ail_cursor cur;
David Chinner783a2f62008-10-30 17:39:58 +11003132 struct xfs_ail *ailp = log->l_ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003133
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10003134 efd_formatp = item->ri_buf[0].i_addr;
Tim Shimmin6d192a92006-06-09 14:55:38 +10003135 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
3136 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
3137 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
3138 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003139 efi_id = efd_formatp->efd_efi_id;
3140
3141 /*
3142 * Search for the efi with the id in the efd format structure
3143 * in the AIL.
3144 */
David Chinnera9c21c12008-10-30 17:39:35 +11003145 spin_lock(&ailp->xa_lock);
3146 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003147 while (lip != NULL) {
3148 if (lip->li_type == XFS_LI_EFI) {
3149 efip = (xfs_efi_log_item_t *)lip;
3150 if (efip->efi_format.efi_id == efi_id) {
3151 /*
David Chinner783a2f62008-10-30 17:39:58 +11003152 * xfs_trans_ail_delete() drops the
Linus Torvalds1da177e2005-04-16 15:20:36 -07003153 * AIL lock.
3154 */
Dave Chinner04913fd2012-04-23 15:58:41 +10003155 xfs_trans_ail_delete(ailp, lip,
3156 SHUTDOWN_CORRUPT_INCORE);
David Chinner8ae2c0f2007-11-23 16:28:17 +11003157 xfs_efi_item_free(efip);
David Chinnera9c21c12008-10-30 17:39:35 +11003158 spin_lock(&ailp->xa_lock);
David Chinner27d8d5f2008-10-30 17:38:39 +11003159 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003160 }
3161 }
David Chinnera9c21c12008-10-30 17:39:35 +11003162 lip = xfs_trans_ail_cursor_next(ailp, &cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003163 }
Eric Sandeene4a1e292014-04-14 19:06:05 +10003164 xfs_trans_ail_cursor_done(&cur);
David Chinnera9c21c12008-10-30 17:39:35 +11003165 spin_unlock(&ailp->xa_lock);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003166
3167 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003168}
3169
3170/*
Dave Chinner28c8e412013-06-27 16:04:55 +10003171 * This routine is called when an inode create format structure is found in a
3172 * committed transaction in the log. It's purpose is to initialise the inodes
3173 * being allocated on disk. This requires us to get inode cluster buffers that
3174 * match the range to be intialised, stamped with inode templates and written
3175 * by delayed write so that subsequent modifications will hit the cached buffer
3176 * and only need writing out at the end of recovery.
3177 */
3178STATIC int
3179xlog_recover_do_icreate_pass2(
3180 struct xlog *log,
3181 struct list_head *buffer_list,
3182 xlog_recover_item_t *item)
3183{
3184 struct xfs_mount *mp = log->l_mp;
3185 struct xfs_icreate_log *icl;
3186 xfs_agnumber_t agno;
3187 xfs_agblock_t agbno;
3188 unsigned int count;
3189 unsigned int isize;
3190 xfs_agblock_t length;
3191
3192 icl = (struct xfs_icreate_log *)item->ri_buf[0].i_addr;
3193 if (icl->icl_type != XFS_LI_ICREATE) {
3194 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad type");
Dave Chinner24513372014-06-25 14:58:08 +10003195 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003196 }
3197
3198 if (icl->icl_size != 1) {
3199 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad icl size");
Dave Chinner24513372014-06-25 14:58:08 +10003200 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003201 }
3202
3203 agno = be32_to_cpu(icl->icl_ag);
3204 if (agno >= mp->m_sb.sb_agcount) {
3205 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agno");
Dave Chinner24513372014-06-25 14:58:08 +10003206 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003207 }
3208 agbno = be32_to_cpu(icl->icl_agbno);
3209 if (!agbno || agbno == NULLAGBLOCK || agbno >= mp->m_sb.sb_agblocks) {
3210 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agbno");
Dave Chinner24513372014-06-25 14:58:08 +10003211 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003212 }
3213 isize = be32_to_cpu(icl->icl_isize);
3214 if (isize != mp->m_sb.sb_inodesize) {
3215 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad isize");
Dave Chinner24513372014-06-25 14:58:08 +10003216 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003217 }
3218 count = be32_to_cpu(icl->icl_count);
3219 if (!count) {
3220 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count");
Dave Chinner24513372014-06-25 14:58:08 +10003221 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003222 }
3223 length = be32_to_cpu(icl->icl_length);
3224 if (!length || length >= mp->m_sb.sb_agblocks) {
3225 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad length");
Dave Chinner24513372014-06-25 14:58:08 +10003226 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003227 }
3228
3229 /* existing allocation is fixed value */
Jie Liu71783432013-12-13 15:51:46 +11003230 ASSERT(count == mp->m_ialloc_inos);
Jie Liu126cd102013-12-13 15:51:48 +11003231 ASSERT(length == mp->m_ialloc_blks);
Jie Liu71783432013-12-13 15:51:46 +11003232 if (count != mp->m_ialloc_inos ||
Jie Liu126cd102013-12-13 15:51:48 +11003233 length != mp->m_ialloc_blks) {
Dave Chinner28c8e412013-06-27 16:04:55 +10003234 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count 2");
Dave Chinner24513372014-06-25 14:58:08 +10003235 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003236 }
3237
3238 /*
3239 * Inode buffers can be freed. Do not replay the inode initialisation as
3240 * we could be overwriting something written after this inode buffer was
3241 * cancelled.
3242 *
3243 * XXX: we need to iterate all buffers and only init those that are not
3244 * cancelled. I think that a more fine grained factoring of
3245 * xfs_ialloc_inode_init may be appropriate here to enable this to be
3246 * done easily.
3247 */
3248 if (xlog_check_buffer_cancelled(log,
3249 XFS_AGB_TO_DADDR(mp, agno, agbno), length, 0))
3250 return 0;
3251
3252 xfs_ialloc_inode_init(mp, NULL, buffer_list, agno, agbno, length,
3253 be32_to_cpu(icl->icl_gen));
3254 return 0;
3255}
3256
3257/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003258 * Free up any resources allocated by the transaction
3259 *
3260 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
3261 */
3262STATIC void
3263xlog_recover_free_trans(
Christoph Hellwigd0450942010-12-01 22:06:23 +00003264 struct xlog_recover *trans)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003265{
Dave Chinnerf0a76952010-01-11 11:49:57 +00003266 xlog_recover_item_t *item, *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003267 int i;
3268
Dave Chinnerf0a76952010-01-11 11:49:57 +00003269 list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
3270 /* Free the regions in the item. */
3271 list_del(&item->ri_list);
3272 for (i = 0; i < item->ri_cnt; i++)
3273 kmem_free(item->ri_buf[i].i_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003274 /* Free the item itself */
Dave Chinnerf0a76952010-01-11 11:49:57 +00003275 kmem_free(item->ri_buf);
3276 kmem_free(item);
3277 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003278 /* Free the transaction recover structure */
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003279 kmem_free(trans);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003280}
3281
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003282STATIC void
3283xlog_recover_buffer_ra_pass2(
3284 struct xlog *log,
3285 struct xlog_recover_item *item)
3286{
3287 struct xfs_buf_log_format *buf_f = item->ri_buf[0].i_addr;
3288 struct xfs_mount *mp = log->l_mp;
3289
Dave Chinner84a5b732013-08-27 08:10:53 +10003290 if (xlog_peek_buffer_cancelled(log, buf_f->blf_blkno,
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003291 buf_f->blf_len, buf_f->blf_flags)) {
3292 return;
3293 }
3294
3295 xfs_buf_readahead(mp->m_ddev_targp, buf_f->blf_blkno,
3296 buf_f->blf_len, NULL);
3297}
3298
3299STATIC void
3300xlog_recover_inode_ra_pass2(
3301 struct xlog *log,
3302 struct xlog_recover_item *item)
3303{
3304 struct xfs_inode_log_format ilf_buf;
3305 struct xfs_inode_log_format *ilfp;
3306 struct xfs_mount *mp = log->l_mp;
3307 int error;
3308
3309 if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) {
3310 ilfp = item->ri_buf[0].i_addr;
3311 } else {
3312 ilfp = &ilf_buf;
3313 memset(ilfp, 0, sizeof(*ilfp));
3314 error = xfs_inode_item_format_convert(&item->ri_buf[0], ilfp);
3315 if (error)
3316 return;
3317 }
3318
Dave Chinner84a5b732013-08-27 08:10:53 +10003319 if (xlog_peek_buffer_cancelled(log, ilfp->ilf_blkno, ilfp->ilf_len, 0))
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003320 return;
3321
3322 xfs_buf_readahead(mp->m_ddev_targp, ilfp->ilf_blkno,
Dave Chinnerd8914002013-08-27 11:39:37 +10003323 ilfp->ilf_len, &xfs_inode_buf_ra_ops);
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003324}
3325
3326STATIC void
3327xlog_recover_dquot_ra_pass2(
3328 struct xlog *log,
3329 struct xlog_recover_item *item)
3330{
3331 struct xfs_mount *mp = log->l_mp;
3332 struct xfs_disk_dquot *recddq;
3333 struct xfs_dq_logformat *dq_f;
3334 uint type;
3335
3336
3337 if (mp->m_qflags == 0)
3338 return;
3339
3340 recddq = item->ri_buf[1].i_addr;
3341 if (recddq == NULL)
3342 return;
3343 if (item->ri_buf[1].i_len < sizeof(struct xfs_disk_dquot))
3344 return;
3345
3346 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
3347 ASSERT(type);
3348 if (log->l_quotaoffs_flag & type)
3349 return;
3350
3351 dq_f = item->ri_buf[0].i_addr;
3352 ASSERT(dq_f);
3353 ASSERT(dq_f->qlf_len == 1);
3354
3355 xfs_buf_readahead(mp->m_ddev_targp, dq_f->qlf_blkno,
Dave Chinner0f0d3342013-08-27 13:25:43 +10003356 XFS_FSB_TO_BB(mp, dq_f->qlf_len), NULL);
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003357}
3358
3359STATIC void
3360xlog_recover_ra_pass2(
3361 struct xlog *log,
3362 struct xlog_recover_item *item)
3363{
3364 switch (ITEM_TYPE(item)) {
3365 case XFS_LI_BUF:
3366 xlog_recover_buffer_ra_pass2(log, item);
3367 break;
3368 case XFS_LI_INODE:
3369 xlog_recover_inode_ra_pass2(log, item);
3370 break;
3371 case XFS_LI_DQUOT:
3372 xlog_recover_dquot_ra_pass2(log, item);
3373 break;
3374 case XFS_LI_EFI:
3375 case XFS_LI_EFD:
3376 case XFS_LI_QUOTAOFF:
3377 default:
3378 break;
3379 }
3380}
3381
Linus Torvalds1da177e2005-04-16 15:20:36 -07003382STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003383xlog_recover_commit_pass1(
Mark Tinguelyad223e62012-06-14 09:22:15 -05003384 struct xlog *log,
3385 struct xlog_recover *trans,
3386 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003387{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003388 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1);
Christoph Hellwigd0450942010-12-01 22:06:23 +00003389
3390 switch (ITEM_TYPE(item)) {
3391 case XFS_LI_BUF:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003392 return xlog_recover_buffer_pass1(log, item);
Christoph Hellwigd0450942010-12-01 22:06:23 +00003393 case XFS_LI_QUOTAOFF:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003394 return xlog_recover_quotaoff_pass1(log, item);
3395 case XFS_LI_INODE:
3396 case XFS_LI_EFI:
3397 case XFS_LI_EFD:
3398 case XFS_LI_DQUOT:
Dave Chinner28c8e412013-06-27 16:04:55 +10003399 case XFS_LI_ICREATE:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003400 /* nothing to do in pass 1 */
3401 return 0;
Christoph Hellwigd0450942010-12-01 22:06:23 +00003402 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003403 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
3404 __func__, ITEM_TYPE(item));
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003405 ASSERT(0);
Dave Chinner24513372014-06-25 14:58:08 +10003406 return -EIO;
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003407 }
3408}
3409
3410STATIC int
3411xlog_recover_commit_pass2(
Mark Tinguelyad223e62012-06-14 09:22:15 -05003412 struct xlog *log,
3413 struct xlog_recover *trans,
3414 struct list_head *buffer_list,
3415 struct xlog_recover_item *item)
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003416{
3417 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2);
3418
3419 switch (ITEM_TYPE(item)) {
3420 case XFS_LI_BUF:
Dave Chinner50d5c8d2013-08-28 21:22:47 +10003421 return xlog_recover_buffer_pass2(log, buffer_list, item,
3422 trans->r_lsn);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003423 case XFS_LI_INODE:
Dave Chinner50d5c8d2013-08-28 21:22:47 +10003424 return xlog_recover_inode_pass2(log, buffer_list, item,
3425 trans->r_lsn);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003426 case XFS_LI_EFI:
3427 return xlog_recover_efi_pass2(log, item, trans->r_lsn);
3428 case XFS_LI_EFD:
3429 return xlog_recover_efd_pass2(log, item);
3430 case XFS_LI_DQUOT:
Dave Chinner50d5c8d2013-08-28 21:22:47 +10003431 return xlog_recover_dquot_pass2(log, buffer_list, item,
3432 trans->r_lsn);
Dave Chinner28c8e412013-06-27 16:04:55 +10003433 case XFS_LI_ICREATE:
3434 return xlog_recover_do_icreate_pass2(log, buffer_list, item);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003435 case XFS_LI_QUOTAOFF:
3436 /* nothing to do in pass2 */
3437 return 0;
3438 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003439 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
3440 __func__, ITEM_TYPE(item));
Christoph Hellwigd0450942010-12-01 22:06:23 +00003441 ASSERT(0);
Dave Chinner24513372014-06-25 14:58:08 +10003442 return -EIO;
Christoph Hellwigd0450942010-12-01 22:06:23 +00003443 }
3444}
3445
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003446STATIC int
3447xlog_recover_items_pass2(
3448 struct xlog *log,
3449 struct xlog_recover *trans,
3450 struct list_head *buffer_list,
3451 struct list_head *item_list)
3452{
3453 struct xlog_recover_item *item;
3454 int error = 0;
3455
3456 list_for_each_entry(item, item_list, ri_list) {
3457 error = xlog_recover_commit_pass2(log, trans,
3458 buffer_list, item);
3459 if (error)
3460 return error;
3461 }
3462
3463 return error;
3464}
3465
Christoph Hellwigd0450942010-12-01 22:06:23 +00003466/*
3467 * Perform the transaction.
3468 *
3469 * If the transaction modifies a buffer or inode, do it now. Otherwise,
3470 * EFIs and EFDs get queued up by adding entries into the AIL for them.
3471 */
3472STATIC int
3473xlog_recover_commit_trans(
Mark Tinguelyad223e62012-06-14 09:22:15 -05003474 struct xlog *log,
Christoph Hellwigd0450942010-12-01 22:06:23 +00003475 struct xlog_recover *trans,
3476 int pass)
3477{
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003478 int error = 0;
3479 int error2;
3480 int items_queued = 0;
3481 struct xlog_recover_item *item;
3482 struct xlog_recover_item *next;
3483 LIST_HEAD (buffer_list);
3484 LIST_HEAD (ra_list);
3485 LIST_HEAD (done_list);
3486
3487 #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100
Linus Torvalds1da177e2005-04-16 15:20:36 -07003488
Dave Chinnerf0a76952010-01-11 11:49:57 +00003489 hlist_del(&trans->r_list);
Christoph Hellwigd0450942010-12-01 22:06:23 +00003490
3491 error = xlog_recover_reorder_trans(log, trans, pass);
3492 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003493 return error;
Christoph Hellwigd0450942010-12-01 22:06:23 +00003494
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003495 list_for_each_entry_safe(item, next, &trans->r_itemq, ri_list) {
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003496 switch (pass) {
3497 case XLOG_RECOVER_PASS1:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003498 error = xlog_recover_commit_pass1(log, trans, item);
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003499 break;
3500 case XLOG_RECOVER_PASS2:
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003501 xlog_recover_ra_pass2(log, item);
3502 list_move_tail(&item->ri_list, &ra_list);
3503 items_queued++;
3504 if (items_queued >= XLOG_RECOVER_COMMIT_QUEUE_MAX) {
3505 error = xlog_recover_items_pass2(log, trans,
3506 &buffer_list, &ra_list);
3507 list_splice_tail_init(&ra_list, &done_list);
3508 items_queued = 0;
3509 }
3510
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003511 break;
3512 default:
3513 ASSERT(0);
3514 }
3515
Christoph Hellwigd0450942010-12-01 22:06:23 +00003516 if (error)
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003517 goto out;
Christoph Hellwigd0450942010-12-01 22:06:23 +00003518 }
3519
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003520out:
3521 if (!list_empty(&ra_list)) {
3522 if (!error)
3523 error = xlog_recover_items_pass2(log, trans,
3524 &buffer_list, &ra_list);
3525 list_splice_tail_init(&ra_list, &done_list);
3526 }
3527
3528 if (!list_empty(&done_list))
3529 list_splice_init(&done_list, &trans->r_itemq);
3530
Christoph Hellwigd0450942010-12-01 22:06:23 +00003531 xlog_recover_free_trans(trans);
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003532
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003533 error2 = xfs_buf_delwri_submit(&buffer_list);
3534 return error ? error : error2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003535}
3536
3537STATIC int
3538xlog_recover_unmount_trans(
Eric Sandeenbbe4c662014-04-14 19:06:25 +10003539 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003540{
3541 /* Do nothing now */
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003542 xfs_warn(log->l_mp, "%s: Unmount LR", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003543 return 0;
3544}
3545
3546/*
3547 * There are two valid states of the r_state field. 0 indicates that the
3548 * transaction structure is in a normal state. We have either seen the
3549 * start of the transaction or the last operation we added was not a partial
3550 * operation. If the last operation we added to the transaction was a
3551 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
3552 *
3553 * NOTE: skip LRs with 0 data length.
3554 */
3555STATIC int
3556xlog_recover_process_data(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003557 struct xlog *log,
Dave Chinnerf0a76952010-01-11 11:49:57 +00003558 struct hlist_head rhash[],
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003559 struct xlog_rec_header *rhead,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003560 xfs_caddr_t dp,
3561 int pass)
3562{
3563 xfs_caddr_t lp;
3564 int num_logops;
3565 xlog_op_header_t *ohead;
3566 xlog_recover_t *trans;
3567 xlog_tid_t tid;
3568 int error;
3569 unsigned long hash;
3570 uint flags;
3571
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003572 lp = dp + be32_to_cpu(rhead->h_len);
3573 num_logops = be32_to_cpu(rhead->h_num_logops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003574
3575 /* check the log format matches our own - else we can't recover */
3576 if (xlog_header_check_recover(log->l_mp, rhead))
Dave Chinner24513372014-06-25 14:58:08 +10003577 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003578
3579 while ((dp < lp) && num_logops) {
3580 ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
3581 ohead = (xlog_op_header_t *)dp;
3582 dp += sizeof(xlog_op_header_t);
3583 if (ohead->oh_clientid != XFS_TRANSACTION &&
3584 ohead->oh_clientid != XFS_LOG) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003585 xfs_warn(log->l_mp, "%s: bad clientid 0x%x",
3586 __func__, ohead->oh_clientid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003587 ASSERT(0);
Dave Chinner24513372014-06-25 14:58:08 +10003588 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003589 }
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10003590 tid = be32_to_cpu(ohead->oh_tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003591 hash = XLOG_RHASH(tid);
Dave Chinnerf0a76952010-01-11 11:49:57 +00003592 trans = xlog_recover_find_tid(&rhash[hash], tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003593 if (trans == NULL) { /* not found; add new tid */
3594 if (ohead->oh_flags & XLOG_START_TRANS)
3595 xlog_recover_new_tid(&rhash[hash], tid,
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003596 be64_to_cpu(rhead->h_lsn));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003597 } else {
Lachlan McIlroy9742bb92008-01-10 16:43:36 +11003598 if (dp + be32_to_cpu(ohead->oh_len) > lp) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003599 xfs_warn(log->l_mp, "%s: bad length 0x%x",
3600 __func__, be32_to_cpu(ohead->oh_len));
Lachlan McIlroy9742bb92008-01-10 16:43:36 +11003601 WARN_ON(1);
Dave Chinner24513372014-06-25 14:58:08 +10003602 return -EIO;
Lachlan McIlroy9742bb92008-01-10 16:43:36 +11003603 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003604 flags = ohead->oh_flags & ~XLOG_END_TRANS;
3605 if (flags & XLOG_WAS_CONT_TRANS)
3606 flags &= ~XLOG_CONTINUE_TRANS;
3607 switch (flags) {
3608 case XLOG_COMMIT_TRANS:
3609 error = xlog_recover_commit_trans(log,
Dave Chinnerf0a76952010-01-11 11:49:57 +00003610 trans, pass);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003611 break;
3612 case XLOG_UNMOUNT_TRANS:
Eric Sandeenbbe4c662014-04-14 19:06:25 +10003613 error = xlog_recover_unmount_trans(log);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003614 break;
3615 case XLOG_WAS_CONT_TRANS:
Dave Chinner9abbc532010-04-13 15:06:46 +10003616 error = xlog_recover_add_to_cont_trans(log,
3617 trans, dp,
3618 be32_to_cpu(ohead->oh_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003619 break;
3620 case XLOG_START_TRANS:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003621 xfs_warn(log->l_mp, "%s: bad transaction",
3622 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003623 ASSERT(0);
Dave Chinner24513372014-06-25 14:58:08 +10003624 error = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003625 break;
3626 case 0:
3627 case XLOG_CONTINUE_TRANS:
Dave Chinner9abbc532010-04-13 15:06:46 +10003628 error = xlog_recover_add_to_trans(log, trans,
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10003629 dp, be32_to_cpu(ohead->oh_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003630 break;
3631 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003632 xfs_warn(log->l_mp, "%s: bad flag 0x%x",
3633 __func__, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003634 ASSERT(0);
Dave Chinner24513372014-06-25 14:58:08 +10003635 error = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003636 break;
3637 }
Mark Tinguely2a841082013-10-02 07:51:12 -05003638 if (error) {
3639 xlog_recover_free_trans(trans);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003640 return error;
Mark Tinguely2a841082013-10-02 07:51:12 -05003641 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003642 }
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10003643 dp += be32_to_cpu(ohead->oh_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003644 num_logops--;
3645 }
3646 return 0;
3647}
3648
3649/*
3650 * Process an extent free intent item that was recovered from
3651 * the log. We need to free the extents that it describes.
3652 */
David Chinner3c1e2bb2008-04-10 12:21:11 +10003653STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003654xlog_recover_process_efi(
3655 xfs_mount_t *mp,
3656 xfs_efi_log_item_t *efip)
3657{
3658 xfs_efd_log_item_t *efdp;
3659 xfs_trans_t *tp;
3660 int i;
David Chinner3c1e2bb2008-04-10 12:21:11 +10003661 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003662 xfs_extent_t *extp;
3663 xfs_fsblock_t startblock_fsb;
3664
Dave Chinnerb199c8a2010-12-20 11:59:49 +11003665 ASSERT(!test_bit(XFS_EFI_RECOVERED, &efip->efi_flags));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003666
3667 /*
3668 * First check the validity of the extents described by the
3669 * EFI. If any are bad, then assume that all are bad and
3670 * just toss the EFI.
3671 */
3672 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3673 extp = &(efip->efi_format.efi_extents[i]);
3674 startblock_fsb = XFS_BB_TO_FSB(mp,
3675 XFS_FSB_TO_DADDR(mp, extp->ext_start));
3676 if ((startblock_fsb == 0) ||
3677 (extp->ext_len == 0) ||
3678 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
3679 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
3680 /*
3681 * This will pull the EFI from the AIL and
3682 * free the memory associated with it.
3683 */
Dave Chinner666d6442013-04-03 14:09:21 +11003684 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003685 xfs_efi_release(efip, efip->efi_format.efi_nextents);
Dave Chinner24513372014-06-25 14:58:08 +10003686 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003687 }
3688 }
3689
3690 tp = xfs_trans_alloc(mp, 0);
Jie Liu3d3c8b52013-08-12 20:49:59 +10003691 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
David Chinnerfc6149d2008-04-10 12:21:53 +10003692 if (error)
3693 goto abort_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003694 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
3695
3696 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3697 extp = &(efip->efi_format.efi_extents[i]);
David Chinnerfc6149d2008-04-10 12:21:53 +10003698 error = xfs_free_extent(tp, extp->ext_start, extp->ext_len);
3699 if (error)
3700 goto abort_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003701 xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
3702 extp->ext_len);
3703 }
3704
Dave Chinnerb199c8a2010-12-20 11:59:49 +11003705 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
David Chinnere5720ee2008-04-10 12:21:18 +10003706 error = xfs_trans_commit(tp, 0);
David Chinner3c1e2bb2008-04-10 12:21:11 +10003707 return error;
David Chinnerfc6149d2008-04-10 12:21:53 +10003708
3709abort_error:
3710 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3711 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003712}
3713
3714/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003715 * When this is called, all of the EFIs which did not have
3716 * corresponding EFDs should be in the AIL. What we do now
3717 * is free the extents associated with each one.
3718 *
3719 * Since we process the EFIs in normal transactions, they
3720 * will be removed at some point after the commit. This prevents
3721 * us from just walking down the list processing each one.
3722 * We'll use a flag in the EFI to skip those that we've already
3723 * processed and use the AIL iteration mechanism's generation
3724 * count to try to speed this up at least a bit.
3725 *
3726 * When we start, we know that the EFIs are the only things in
3727 * the AIL. As we process them, however, other items are added
3728 * to the AIL. Since everything added to the AIL must come after
3729 * everything already in the AIL, we stop processing as soon as
3730 * we see something other than an EFI in the AIL.
3731 */
David Chinner3c1e2bb2008-04-10 12:21:11 +10003732STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003733xlog_recover_process_efis(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003734 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003735{
3736 xfs_log_item_t *lip;
3737 xfs_efi_log_item_t *efip;
David Chinner3c1e2bb2008-04-10 12:21:11 +10003738 int error = 0;
David Chinner27d8d5f2008-10-30 17:38:39 +11003739 struct xfs_ail_cursor cur;
David Chinnera9c21c12008-10-30 17:39:35 +11003740 struct xfs_ail *ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003741
David Chinnera9c21c12008-10-30 17:39:35 +11003742 ailp = log->l_ailp;
3743 spin_lock(&ailp->xa_lock);
3744 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003745 while (lip != NULL) {
3746 /*
3747 * We're done when we see something other than an EFI.
David Chinner27d8d5f2008-10-30 17:38:39 +11003748 * There should be no EFIs left in the AIL now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003749 */
3750 if (lip->li_type != XFS_LI_EFI) {
David Chinner27d8d5f2008-10-30 17:38:39 +11003751#ifdef DEBUG
David Chinnera9c21c12008-10-30 17:39:35 +11003752 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
David Chinner27d8d5f2008-10-30 17:38:39 +11003753 ASSERT(lip->li_type != XFS_LI_EFI);
3754#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003755 break;
3756 }
3757
3758 /*
3759 * Skip EFIs that we've already processed.
3760 */
3761 efip = (xfs_efi_log_item_t *)lip;
Dave Chinnerb199c8a2010-12-20 11:59:49 +11003762 if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) {
David Chinnera9c21c12008-10-30 17:39:35 +11003763 lip = xfs_trans_ail_cursor_next(ailp, &cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003764 continue;
3765 }
3766
David Chinnera9c21c12008-10-30 17:39:35 +11003767 spin_unlock(&ailp->xa_lock);
3768 error = xlog_recover_process_efi(log->l_mp, efip);
3769 spin_lock(&ailp->xa_lock);
David Chinner27d8d5f2008-10-30 17:38:39 +11003770 if (error)
3771 goto out;
David Chinnera9c21c12008-10-30 17:39:35 +11003772 lip = xfs_trans_ail_cursor_next(ailp, &cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003773 }
David Chinner27d8d5f2008-10-30 17:38:39 +11003774out:
Eric Sandeene4a1e292014-04-14 19:06:05 +10003775 xfs_trans_ail_cursor_done(&cur);
David Chinnera9c21c12008-10-30 17:39:35 +11003776 spin_unlock(&ailp->xa_lock);
David Chinner3c1e2bb2008-04-10 12:21:11 +10003777 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003778}
3779
3780/*
3781 * This routine performs a transaction to null out a bad inode pointer
3782 * in an agi unlinked inode hash bucket.
3783 */
3784STATIC void
3785xlog_recover_clear_agi_bucket(
3786 xfs_mount_t *mp,
3787 xfs_agnumber_t agno,
3788 int bucket)
3789{
3790 xfs_trans_t *tp;
3791 xfs_agi_t *agi;
3792 xfs_buf_t *agibp;
3793 int offset;
3794 int error;
3795
3796 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
Jie Liu3d3c8b52013-08-12 20:49:59 +10003797 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_clearagi, 0, 0);
David Chinnere5720ee2008-04-10 12:21:18 +10003798 if (error)
3799 goto out_abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003800
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003801 error = xfs_read_agi(mp, tp, agno, &agibp);
3802 if (error)
David Chinnere5720ee2008-04-10 12:21:18 +10003803 goto out_abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003804
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003805 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11003806 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003807 offset = offsetof(xfs_agi_t, agi_unlinked) +
3808 (sizeof(xfs_agino_t) * bucket);
3809 xfs_trans_log_buf(tp, agibp, offset,
3810 (offset + sizeof(xfs_agino_t) - 1));
3811
David Chinnere5720ee2008-04-10 12:21:18 +10003812 error = xfs_trans_commit(tp, 0);
3813 if (error)
3814 goto out_error;
3815 return;
3816
3817out_abort:
3818 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3819out_error:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003820 xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno);
David Chinnere5720ee2008-04-10 12:21:18 +10003821 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003822}
3823
Christoph Hellwig23fac502008-11-28 14:23:40 +11003824STATIC xfs_agino_t
3825xlog_recover_process_one_iunlink(
3826 struct xfs_mount *mp,
3827 xfs_agnumber_t agno,
3828 xfs_agino_t agino,
3829 int bucket)
3830{
3831 struct xfs_buf *ibp;
3832 struct xfs_dinode *dip;
3833 struct xfs_inode *ip;
3834 xfs_ino_t ino;
3835 int error;
3836
3837 ino = XFS_AGINO_TO_INO(mp, agno, agino);
Dave Chinner7b6259e2010-06-24 11:35:17 +10003838 error = xfs_iget(mp, NULL, ino, 0, 0, &ip);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003839 if (error)
3840 goto fail;
3841
3842 /*
3843 * Get the on disk inode to find the next inode in the bucket.
3844 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04003845 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &ibp, 0, 0);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003846 if (error)
Christoph Hellwig0e446672008-11-28 14:23:42 +11003847 goto fail_iput;
Christoph Hellwig23fac502008-11-28 14:23:40 +11003848
Christoph Hellwig23fac502008-11-28 14:23:40 +11003849 ASSERT(ip->i_d.di_nlink == 0);
Christoph Hellwig0e446672008-11-28 14:23:42 +11003850 ASSERT(ip->i_d.di_mode != 0);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003851
3852 /* setup for the next pass */
3853 agino = be32_to_cpu(dip->di_next_unlinked);
3854 xfs_buf_relse(ibp);
3855
3856 /*
3857 * Prevent any DMAPI event from being sent when the reference on
3858 * the inode is dropped.
3859 */
3860 ip->i_d.di_dmevmask = 0;
3861
Christoph Hellwig0e446672008-11-28 14:23:42 +11003862 IRELE(ip);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003863 return agino;
3864
Christoph Hellwig0e446672008-11-28 14:23:42 +11003865 fail_iput:
3866 IRELE(ip);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003867 fail:
3868 /*
3869 * We can't read in the inode this bucket points to, or this inode
3870 * is messed up. Just ditch this bucket of inodes. We will lose
3871 * some inodes and space, but at least we won't hang.
3872 *
3873 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3874 * clear the inode pointer in the bucket.
3875 */
3876 xlog_recover_clear_agi_bucket(mp, agno, bucket);
3877 return NULLAGINO;
3878}
3879
Linus Torvalds1da177e2005-04-16 15:20:36 -07003880/*
3881 * xlog_iunlink_recover
3882 *
3883 * This is called during recovery to process any inodes which
3884 * we unlinked but not freed when the system crashed. These
3885 * inodes will be on the lists in the AGI blocks. What we do
3886 * here is scan all the AGIs and fully truncate and free any
3887 * inodes found on the lists. Each inode is removed from the
3888 * lists when it has been fully truncated and is freed. The
3889 * freeing of the inode and its removal from the list must be
3890 * atomic.
3891 */
Eric Sandeend96f8f82009-07-02 00:09:33 -05003892STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07003893xlog_recover_process_iunlinks(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003894 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003895{
3896 xfs_mount_t *mp;
3897 xfs_agnumber_t agno;
3898 xfs_agi_t *agi;
3899 xfs_buf_t *agibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003900 xfs_agino_t agino;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003901 int bucket;
3902 int error;
3903 uint mp_dmevmask;
3904
3905 mp = log->l_mp;
3906
3907 /*
3908 * Prevent any DMAPI event from being sent while in this function.
3909 */
3910 mp_dmevmask = mp->m_dmevmask;
3911 mp->m_dmevmask = 0;
3912
3913 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3914 /*
3915 * Find the agi for this ag.
3916 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003917 error = xfs_read_agi(mp, NULL, agno, &agibp);
3918 if (error) {
3919 /*
3920 * AGI is b0rked. Don't process it.
3921 *
3922 * We should probably mark the filesystem as corrupt
3923 * after we've recovered all the ag's we can....
3924 */
3925 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003926 }
Jan Karad97d32e2012-03-15 09:34:02 +00003927 /*
3928 * Unlock the buffer so that it can be acquired in the normal
3929 * course of the transaction to truncate and free each inode.
3930 * Because we are not racing with anyone else here for the AGI
3931 * buffer, we don't even need to hold it locked to read the
3932 * initial unlinked bucket entries out of the buffer. We keep
3933 * buffer reference though, so that it stays pinned in memory
3934 * while we need the buffer.
3935 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003936 agi = XFS_BUF_TO_AGI(agibp);
Jan Karad97d32e2012-03-15 09:34:02 +00003937 xfs_buf_unlock(agibp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003938
3939 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
Christoph Hellwig16259e72005-11-02 15:11:25 +11003940 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003941 while (agino != NULLAGINO) {
Christoph Hellwig23fac502008-11-28 14:23:40 +11003942 agino = xlog_recover_process_one_iunlink(mp,
3943 agno, agino, bucket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003944 }
3945 }
Jan Karad97d32e2012-03-15 09:34:02 +00003946 xfs_buf_rele(agibp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003947 }
3948
3949 mp->m_dmevmask = mp_dmevmask;
3950}
3951
Linus Torvalds1da177e2005-04-16 15:20:36 -07003952/*
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003953 * Upack the log buffer data and crc check it. If the check fails, issue a
3954 * warning if and only if the CRC in the header is non-zero. This makes the
3955 * check an advisory warning, and the zero CRC check will prevent failure
3956 * warnings from being emitted when upgrading the kernel from one that does not
3957 * add CRCs by default.
3958 *
3959 * When filesystems are CRC enabled, this CRC mismatch becomes a fatal log
3960 * corruption failure
Linus Torvalds1da177e2005-04-16 15:20:36 -07003961 */
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003962STATIC int
3963xlog_unpack_data_crc(
3964 struct xlog_rec_header *rhead,
3965 xfs_caddr_t dp,
3966 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003967{
Dave Chinnerf9668a02012-11-28 13:01:03 +11003968 __le32 crc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003969
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003970 crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len));
3971 if (crc != rhead->h_crc) {
3972 if (rhead->h_crc || xfs_sb_version_hascrc(&log->l_mp->m_sb)) {
3973 xfs_alert(log->l_mp,
Eric Sandeen08e96e12013-10-11 20:59:05 -05003974 "log record CRC mismatch: found 0x%x, expected 0x%x.",
Dave Chinnerf9668a02012-11-28 13:01:03 +11003975 le32_to_cpu(rhead->h_crc),
3976 le32_to_cpu(crc));
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003977 xfs_hex_dump(dp, 32);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003978 }
3979
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003980 /*
3981 * If we've detected a log record corruption, then we can't
3982 * recover past this point. Abort recovery if we are enforcing
3983 * CRC protection by punting an error back up the stack.
3984 */
3985 if (xfs_sb_version_hascrc(&log->l_mp->m_sb))
Dave Chinner24513372014-06-25 14:58:08 +10003986 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003987 }
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003988
3989 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003990}
3991
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003992STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003993xlog_unpack_data(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003994 struct xlog_rec_header *rhead,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003995 xfs_caddr_t dp,
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003996 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003997{
3998 int i, j, k;
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003999 int error;
4000
4001 error = xlog_unpack_data_crc(rhead, dp, log);
4002 if (error)
4003 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004004
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004005 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07004006 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004007 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07004008 dp += BBSIZE;
4009 }
4010
Eric Sandeen62118702008-03-06 13:44:28 +11004011 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Christoph Hellwigb28708d2008-11-28 14:23:38 +11004012 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004013 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004014 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
4015 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004016 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
Linus Torvalds1da177e2005-04-16 15:20:36 -07004017 dp += BBSIZE;
4018 }
4019 }
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004020
4021 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004022}
4023
4024STATIC int
4025xlog_valid_rec_header(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004026 struct xlog *log,
4027 struct xlog_rec_header *rhead,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004028 xfs_daddr_t blkno)
4029{
4030 int hlen;
4031
Christoph Hellwig69ef9212011-07-08 14:36:05 +02004032 if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004033 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
4034 XFS_ERRLEVEL_LOW, log->l_mp);
Dave Chinner24513372014-06-25 14:58:08 +10004035 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004036 }
4037 if (unlikely(
4038 (!rhead->h_version ||
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004039 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004040 xfs_warn(log->l_mp, "%s: unrecognised log version (%d).",
Harvey Harrison34a622b2008-04-10 12:19:21 +10004041 __func__, be32_to_cpu(rhead->h_version));
Dave Chinner24513372014-06-25 14:58:08 +10004042 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004043 }
4044
4045 /* LR body must have data or it wouldn't have been written */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004046 hlen = be32_to_cpu(rhead->h_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004047 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
4048 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
4049 XFS_ERRLEVEL_LOW, log->l_mp);
Dave Chinner24513372014-06-25 14:58:08 +10004050 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004051 }
4052 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
4053 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
4054 XFS_ERRLEVEL_LOW, log->l_mp);
Dave Chinner24513372014-06-25 14:58:08 +10004055 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004056 }
4057 return 0;
4058}
4059
4060/*
4061 * Read the log from tail to head and process the log records found.
4062 * Handle the two cases where the tail and head are in the same cycle
4063 * and where the active portion of the log wraps around the end of
4064 * the physical log separately. The pass parameter is passed through
4065 * to the routines called to process the data and is not looked at
4066 * here.
4067 */
4068STATIC int
4069xlog_do_recovery_pass(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004070 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004071 xfs_daddr_t head_blk,
4072 xfs_daddr_t tail_blk,
4073 int pass)
4074{
4075 xlog_rec_header_t *rhead;
4076 xfs_daddr_t blk_no;
Andy Polingfc5bc4c2009-11-03 17:26:47 +00004077 xfs_caddr_t offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004078 xfs_buf_t *hbp, *dbp;
4079 int error = 0, h_size;
4080 int bblks, split_bblks;
4081 int hblks, split_hblks, wrapped_hblks;
Dave Chinnerf0a76952010-01-11 11:49:57 +00004082 struct hlist_head rhash[XLOG_RHASH_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -07004083
4084 ASSERT(head_blk != tail_blk);
4085
4086 /*
4087 * Read the header of the tail block and get the iclog buffer size from
4088 * h_size. Use this to tell how many sectors make up the log header.
4089 */
Eric Sandeen62118702008-03-06 13:44:28 +11004090 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004091 /*
4092 * When using variable length iclogs, read first sector of
4093 * iclog header and extract the header size from it. Get a
4094 * new hbp that is the correct size.
4095 */
4096 hbp = xlog_get_bp(log, 1);
4097 if (!hbp)
Dave Chinner24513372014-06-25 14:58:08 +10004098 return -ENOMEM;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004099
4100 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
4101 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004102 goto bread_err1;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004103
Linus Torvalds1da177e2005-04-16 15:20:36 -07004104 rhead = (xlog_rec_header_t *)offset;
4105 error = xlog_valid_rec_header(log, rhead, tail_blk);
4106 if (error)
4107 goto bread_err1;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004108 h_size = be32_to_cpu(rhead->h_size);
4109 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07004110 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
4111 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
4112 if (h_size % XLOG_HEADER_CYCLE_SIZE)
4113 hblks++;
4114 xlog_put_bp(hbp);
4115 hbp = xlog_get_bp(log, hblks);
4116 } else {
4117 hblks = 1;
4118 }
4119 } else {
Alex Elder69ce58f2010-04-20 17:09:59 +10004120 ASSERT(log->l_sectBBsize == 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004121 hblks = 1;
4122 hbp = xlog_get_bp(log, 1);
4123 h_size = XLOG_BIG_RECORD_BSIZE;
4124 }
4125
4126 if (!hbp)
Dave Chinner24513372014-06-25 14:58:08 +10004127 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004128 dbp = xlog_get_bp(log, BTOBB(h_size));
4129 if (!dbp) {
4130 xlog_put_bp(hbp);
Dave Chinner24513372014-06-25 14:58:08 +10004131 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004132 }
4133
4134 memset(rhash, 0, sizeof(rhash));
4135 if (tail_blk <= head_blk) {
4136 for (blk_no = tail_blk; blk_no < head_blk; ) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004137 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
4138 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004139 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004140
Linus Torvalds1da177e2005-04-16 15:20:36 -07004141 rhead = (xlog_rec_header_t *)offset;
4142 error = xlog_valid_rec_header(log, rhead, blk_no);
4143 if (error)
4144 goto bread_err2;
4145
4146 /* blocks in data section */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004147 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004148 error = xlog_bread(log, blk_no + hblks, bblks, dbp,
4149 &offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004150 if (error)
4151 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004152
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004153 error = xlog_unpack_data(rhead, offset, log);
4154 if (error)
4155 goto bread_err2;
4156
4157 error = xlog_recover_process_data(log,
4158 rhash, rhead, offset, pass);
4159 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004160 goto bread_err2;
4161 blk_no += bblks + hblks;
4162 }
4163 } else {
4164 /*
4165 * Perform recovery around the end of the physical log.
4166 * When the head is not on the same cycle number as the tail,
4167 * we can't do a sequential recovery as above.
4168 */
4169 blk_no = tail_blk;
4170 while (blk_no < log->l_logBBsize) {
4171 /*
4172 * Check for header wrapping around physical end-of-log
4173 */
Chandra Seetharaman62926042011-07-22 23:40:15 +00004174 offset = hbp->b_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004175 split_hblks = 0;
4176 wrapped_hblks = 0;
4177 if (blk_no + hblks <= log->l_logBBsize) {
4178 /* Read header in one read */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004179 error = xlog_bread(log, blk_no, hblks, hbp,
4180 &offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004181 if (error)
4182 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004183 } else {
4184 /* This LR is split across physical log end */
4185 if (blk_no != log->l_logBBsize) {
4186 /* some data before physical log end */
4187 ASSERT(blk_no <= INT_MAX);
4188 split_hblks = log->l_logBBsize - (int)blk_no;
4189 ASSERT(split_hblks > 0);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004190 error = xlog_bread(log, blk_no,
4191 split_hblks, hbp,
4192 &offset);
4193 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004194 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004195 }
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004196
Linus Torvalds1da177e2005-04-16 15:20:36 -07004197 /*
4198 * Note: this black magic still works with
4199 * large sector sizes (non-512) only because:
4200 * - we increased the buffer size originally
4201 * by 1 sector giving us enough extra space
4202 * for the second read;
4203 * - the log start is guaranteed to be sector
4204 * aligned;
4205 * - we read the log end (LR header start)
4206 * _first_, then the log start (LR header end)
4207 * - order is important.
4208 */
David Chinner234f56a2008-04-10 12:24:24 +10004209 wrapped_hblks = hblks - split_hblks;
Dave Chinner44396472011-04-21 09:34:27 +00004210 error = xlog_bread_offset(log, 0,
4211 wrapped_hblks, hbp,
4212 offset + BBTOB(split_hblks));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004213 if (error)
4214 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004215 }
4216 rhead = (xlog_rec_header_t *)offset;
4217 error = xlog_valid_rec_header(log, rhead,
4218 split_hblks ? blk_no : 0);
4219 if (error)
4220 goto bread_err2;
4221
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004222 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004223 blk_no += hblks;
4224
4225 /* Read in data for log record */
4226 if (blk_no + bblks <= log->l_logBBsize) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004227 error = xlog_bread(log, blk_no, bblks, dbp,
4228 &offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004229 if (error)
4230 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004231 } else {
4232 /* This log record is split across the
4233 * physical end of log */
Chandra Seetharaman62926042011-07-22 23:40:15 +00004234 offset = dbp->b_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004235 split_bblks = 0;
4236 if (blk_no != log->l_logBBsize) {
4237 /* some data is before the physical
4238 * end of log */
4239 ASSERT(!wrapped_hblks);
4240 ASSERT(blk_no <= INT_MAX);
4241 split_bblks =
4242 log->l_logBBsize - (int)blk_no;
4243 ASSERT(split_bblks > 0);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004244 error = xlog_bread(log, blk_no,
4245 split_bblks, dbp,
4246 &offset);
4247 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004248 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004249 }
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004250
Linus Torvalds1da177e2005-04-16 15:20:36 -07004251 /*
4252 * Note: this black magic still works with
4253 * large sector sizes (non-512) only because:
4254 * - we increased the buffer size originally
4255 * by 1 sector giving us enough extra space
4256 * for the second read;
4257 * - the log start is guaranteed to be sector
4258 * aligned;
4259 * - we read the log end (LR header start)
4260 * _first_, then the log start (LR header end)
4261 * - order is important.
4262 */
Dave Chinner44396472011-04-21 09:34:27 +00004263 error = xlog_bread_offset(log, 0,
Dave Chinner009507b2012-11-02 11:38:44 +11004264 bblks - split_bblks, dbp,
Dave Chinner44396472011-04-21 09:34:27 +00004265 offset + BBTOB(split_bblks));
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004266 if (error)
4267 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004268 }
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004269
4270 error = xlog_unpack_data(rhead, offset, log);
4271 if (error)
4272 goto bread_err2;
4273
4274 error = xlog_recover_process_data(log, rhash,
4275 rhead, offset, pass);
4276 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004277 goto bread_err2;
4278 blk_no += bblks;
4279 }
4280
4281 ASSERT(blk_no >= log->l_logBBsize);
4282 blk_no -= log->l_logBBsize;
4283
4284 /* read first part of physical log */
4285 while (blk_no < head_blk) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004286 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
4287 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004288 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004289
Linus Torvalds1da177e2005-04-16 15:20:36 -07004290 rhead = (xlog_rec_header_t *)offset;
4291 error = xlog_valid_rec_header(log, rhead, blk_no);
4292 if (error)
4293 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004294
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004295 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004296 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
4297 &offset);
4298 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004299 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004300
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004301 error = xlog_unpack_data(rhead, offset, log);
4302 if (error)
4303 goto bread_err2;
4304
4305 error = xlog_recover_process_data(log, rhash,
4306 rhead, offset, pass);
4307 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004308 goto bread_err2;
4309 blk_no += bblks + hblks;
4310 }
4311 }
4312
4313 bread_err2:
4314 xlog_put_bp(dbp);
4315 bread_err1:
4316 xlog_put_bp(hbp);
4317 return error;
4318}
4319
4320/*
4321 * Do the recovery of the log. We actually do this in two phases.
4322 * The two passes are necessary in order to implement the function
4323 * of cancelling a record written into the log. The first pass
4324 * determines those things which have been cancelled, and the
4325 * second pass replays log items normally except for those which
4326 * have been cancelled. The handling of the replay and cancellations
4327 * takes place in the log item type specific routines.
4328 *
4329 * The table of items which have cancel records in the log is allocated
4330 * and freed at this level, since only here do we know when all of
4331 * the log recovery has been completed.
4332 */
4333STATIC int
4334xlog_do_log_recovery(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004335 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004336 xfs_daddr_t head_blk,
4337 xfs_daddr_t tail_blk)
4338{
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00004339 int error, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004340
4341 ASSERT(head_blk != tail_blk);
4342
4343 /*
4344 * First do a pass to find all of the cancelled buf log items.
4345 * Store them in the buf_cancel_table for use in the second pass.
4346 */
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00004347 log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE *
4348 sizeof(struct list_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07004349 KM_SLEEP);
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00004350 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
4351 INIT_LIST_HEAD(&log->l_buf_cancel_table[i]);
4352
Linus Torvalds1da177e2005-04-16 15:20:36 -07004353 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
4354 XLOG_RECOVER_PASS1);
4355 if (error != 0) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10004356 kmem_free(log->l_buf_cancel_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004357 log->l_buf_cancel_table = NULL;
4358 return error;
4359 }
4360 /*
4361 * Then do a second pass to actually recover the items in the log.
4362 * When it is complete free the table of buf cancel items.
4363 */
4364 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
4365 XLOG_RECOVER_PASS2);
4366#ifdef DEBUG
Tim Shimmin6d192a92006-06-09 14:55:38 +10004367 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004368 int i;
4369
4370 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00004371 ASSERT(list_empty(&log->l_buf_cancel_table[i]));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004372 }
4373#endif /* DEBUG */
4374
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10004375 kmem_free(log->l_buf_cancel_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004376 log->l_buf_cancel_table = NULL;
4377
4378 return error;
4379}
4380
4381/*
4382 * Do the actual recovery
4383 */
4384STATIC int
4385xlog_do_recover(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004386 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004387 xfs_daddr_t head_blk,
4388 xfs_daddr_t tail_blk)
4389{
4390 int error;
4391 xfs_buf_t *bp;
4392 xfs_sb_t *sbp;
4393
4394 /*
4395 * First replay the images in the log.
4396 */
4397 error = xlog_do_log_recovery(log, head_blk, tail_blk);
Christoph Hellwig43ff2122012-04-23 15:58:39 +10004398 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004399 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004400
4401 /*
4402 * If IO errors happened during recovery, bail out.
4403 */
4404 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
Dave Chinner24513372014-06-25 14:58:08 +10004405 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004406 }
4407
4408 /*
4409 * We now update the tail_lsn since much of the recovery has completed
4410 * and there may be space available to use. If there were no extent
4411 * or iunlinks, we can free up the entire log and set the tail_lsn to
4412 * be the last_sync_lsn. This was set in xlog_find_tail to be the
4413 * lsn of the last known good LR on disk. If there are extent frees
4414 * or iunlinks they will have some entries in the AIL; so we look at
4415 * the AIL to determine how to set the tail_lsn.
4416 */
4417 xlog_assign_tail_lsn(log->l_mp);
4418
4419 /*
4420 * Now that we've finished replaying all buffer and inode
Dave Chinner98021822012-11-12 22:54:03 +11004421 * updates, re-read in the superblock and reverify it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004422 */
4423 bp = xfs_getsb(log->l_mp, 0);
4424 XFS_BUF_UNDONE(bp);
Lachlan McIlroybebf9632007-10-15 13:18:02 +10004425 ASSERT(!(XFS_BUF_ISWRITE(bp)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004426 XFS_BUF_READ(bp);
Lachlan McIlroybebf9632007-10-15 13:18:02 +10004427 XFS_BUF_UNASYNC(bp);
Dave Chinner1813dd62012-11-14 17:54:40 +11004428 bp->b_ops = &xfs_sb_buf_ops;
Christoph Hellwig83a0adc2013-12-17 00:03:52 -08004429
4430 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
4431 xfs_buf_relse(bp);
Dave Chinner24513372014-06-25 14:58:08 +10004432 return -EIO;
Christoph Hellwig83a0adc2013-12-17 00:03:52 -08004433 }
4434
4435 xfs_buf_iorequest(bp);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +00004436 error = xfs_buf_iowait(bp);
David Chinnerd64e31a2008-04-10 12:22:17 +10004437 if (error) {
Christoph Hellwig901796a2011-10-10 16:52:49 +00004438 xfs_buf_ioerror_alert(bp, __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004439 ASSERT(0);
4440 xfs_buf_relse(bp);
4441 return error;
4442 }
4443
4444 /* Convert superblock from on-disk format */
4445 sbp = &log->l_mp->m_sb;
Dave Chinner98021822012-11-12 22:54:03 +11004446 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004447 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
Eric Sandeen62118702008-03-06 13:44:28 +11004448 ASSERT(xfs_sb_good_version(sbp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004449 xfs_buf_relse(bp);
4450
Lachlan McIlroy5478eea2007-02-10 18:36:29 +11004451 /* We've re-read the superblock so re-initialize per-cpu counters */
4452 xfs_icsb_reinit_counters(log->l_mp);
4453
Linus Torvalds1da177e2005-04-16 15:20:36 -07004454 xlog_recover_check_summary(log);
4455
4456 /* Normal transactions can now occur */
4457 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
4458 return 0;
4459}
4460
4461/*
4462 * Perform recovery and re-initialize some log variables in xlog_find_tail.
4463 *
4464 * Return error or zero.
4465 */
4466int
4467xlog_recover(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004468 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004469{
4470 xfs_daddr_t head_blk, tail_blk;
4471 int error;
4472
4473 /* find the tail of the log */
Eric Sandeen65be6052006-01-11 15:34:19 +11004474 if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004475 return error;
4476
4477 if (tail_blk != head_blk) {
4478 /* There used to be a comment here:
4479 *
4480 * disallow recovery on read-only mounts. note -- mount
4481 * checks for ENOSPC and turns it into an intelligent
4482 * error message.
4483 * ...but this is no longer true. Now, unless you specify
4484 * NORECOVERY (in which case this function would never be
4485 * called), we just go ahead and recover. We do this all
4486 * under the vfs layer, so we can get away with it unless
4487 * the device itself is read-only, in which case we fail.
4488 */
Utako Kusaka3a02ee12007-05-08 13:50:06 +10004489 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004490 return error;
4491 }
4492
Dave Chinnere721f502013-04-03 16:11:32 +11004493 /*
4494 * Version 5 superblock log feature mask validation. We know the
4495 * log is dirty so check if there are any unknown log features
4496 * in what we need to recover. If there are unknown features
4497 * (e.g. unsupported transactions, then simply reject the
4498 * attempt at recovery before touching anything.
4499 */
4500 if (XFS_SB_VERSION_NUM(&log->l_mp->m_sb) == XFS_SB_VERSION_5 &&
4501 xfs_sb_has_incompat_log_feature(&log->l_mp->m_sb,
4502 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
4503 xfs_warn(log->l_mp,
4504"Superblock has unknown incompatible log features (0x%x) enabled.\n"
4505"The log can not be fully and/or safely recovered by this kernel.\n"
4506"Please recover the log on a kernel that supports the unknown features.",
4507 (log->l_mp->m_sb.sb_features_log_incompat &
4508 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
Dave Chinner24513372014-06-25 14:58:08 +10004509 return -EINVAL;
Dave Chinnere721f502013-04-03 16:11:32 +11004510 }
4511
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004512 xfs_notice(log->l_mp, "Starting recovery (logdev: %s)",
4513 log->l_mp->m_logname ? log->l_mp->m_logname
4514 : "internal");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004515
4516 error = xlog_do_recover(log, head_blk, tail_blk);
4517 log->l_flags |= XLOG_RECOVERY_NEEDED;
4518 }
4519 return error;
4520}
4521
4522/*
4523 * In the first part of recovery we replay inodes and buffers and build
4524 * up the list of extent free items which need to be processed. Here
4525 * we process the extent free items and clean up the on disk unlinked
4526 * inode lists. This is separated from the first part of recovery so
4527 * that the root and real-time bitmap inodes can be read in from disk in
4528 * between the two stages. This is necessary so that we can free space
4529 * in the real-time portion of the file system.
4530 */
4531int
4532xlog_recover_finish(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004533 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004534{
4535 /*
4536 * Now we're ready to do the transactions needed for the
4537 * rest of recovery. Start with completing all the extent
4538 * free intent records and then process the unlinked inode
4539 * lists. At this point, we essentially run in normal mode
4540 * except that we're still performing recovery actions
4541 * rather than accepting new requests.
4542 */
4543 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
David Chinner3c1e2bb2008-04-10 12:21:11 +10004544 int error;
4545 error = xlog_recover_process_efis(log);
4546 if (error) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004547 xfs_alert(log->l_mp, "Failed to recover EFIs");
David Chinner3c1e2bb2008-04-10 12:21:11 +10004548 return error;
4549 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004550 /*
4551 * Sync the log to get all the EFIs out of the AIL.
4552 * This isn't absolutely necessary, but it helps in
4553 * case the unlink transactions would have problems
4554 * pushing the EFIs out of the way.
4555 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +00004556 xfs_log_force(log->l_mp, XFS_LOG_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004557
Christoph Hellwig42490232008-08-13 16:49:32 +10004558 xlog_recover_process_iunlinks(log);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004559
4560 xlog_recover_check_summary(log);
4561
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004562 xfs_notice(log->l_mp, "Ending recovery (logdev: %s)",
4563 log->l_mp->m_logname ? log->l_mp->m_logname
4564 : "internal");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004565 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
4566 } else {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004567 xfs_info(log->l_mp, "Ending clean mount");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004568 }
4569 return 0;
4570}
4571
4572
4573#if defined(DEBUG)
4574/*
4575 * Read all of the agf and agi counters and check that they
4576 * are consistent with the superblock counters.
4577 */
4578void
4579xlog_recover_check_summary(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004580 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004581{
4582 xfs_mount_t *mp;
4583 xfs_agf_t *agfp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004584 xfs_buf_t *agfbp;
4585 xfs_buf_t *agibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004586 xfs_agnumber_t agno;
4587 __uint64_t freeblks;
4588 __uint64_t itotal;
4589 __uint64_t ifree;
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11004590 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004591
4592 mp = log->l_mp;
4593
4594 freeblks = 0LL;
4595 itotal = 0LL;
4596 ifree = 0LL;
4597 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
From: Christoph Hellwig48056212008-11-28 14:23:38 +11004598 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
4599 if (error) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004600 xfs_alert(mp, "%s agf read failed agno %d error %d",
4601 __func__, agno, error);
From: Christoph Hellwig48056212008-11-28 14:23:38 +11004602 } else {
4603 agfp = XFS_BUF_TO_AGF(agfbp);
4604 freeblks += be32_to_cpu(agfp->agf_freeblks) +
4605 be32_to_cpu(agfp->agf_flcount);
4606 xfs_buf_relse(agfbp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004607 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004608
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11004609 error = xfs_read_agi(mp, NULL, agno, &agibp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004610 if (error) {
4611 xfs_alert(mp, "%s agi read failed agno %d error %d",
4612 __func__, agno, error);
4613 } else {
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11004614 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004615
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11004616 itotal += be32_to_cpu(agi->agi_count);
4617 ifree += be32_to_cpu(agi->agi_freecount);
4618 xfs_buf_relse(agibp);
4619 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004620 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004621}
4622#endif /* DEBUG */