blob: 6fcc910a50b9d7e3fbd3c15851aa0ee0cfc1b6af [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"
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include "xfs_types.h"
Nathan Scotta844f452005-11-02 14:38:42 +110021#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110023#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_trans.h"
Nathan Scotta844f452005-11-02 14:38:42 +110025#include "xfs_sb.h"
26#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_mount.h"
28#include "xfs_error.h"
29#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110030#include "xfs_alloc_btree.h"
31#include "xfs_ialloc_btree.h"
Christoph Hellwigee1a47a2013-04-21 14:53:46 -050032#include "xfs_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_dinode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include "xfs_inode.h"
Nathan Scotta844f452005-11-02 14:38:42 +110035#include "xfs_inode_item.h"
Nathan Scotta844f452005-11-02 14:38:42 +110036#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include "xfs_ialloc.h"
38#include "xfs_log_priv.h"
39#include "xfs_buf_item.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include "xfs_log_recover.h"
41#include "xfs_extfree_item.h"
42#include "xfs_trans_priv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include "xfs_quota.h"
Christoph Hellwig43355092008-03-27 18:01:08 +110044#include "xfs_utils.h"
Christoph Hellwig0e446be2012-11-12 22:54:24 +110045#include "xfs_cksum.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000046#include "xfs_trace.h"
Dave Chinner33479e02012-10-08 21:56:11 +110047#include "xfs_icache.h"
Dave Chinner28c8e412013-06-27 16:04:55 +100048#include "xfs_icreate_item.h"
Dave Chinnerd75afeb2013-04-03 16:11:29 +110049
50/* Need all the magic numbers and buffer ops structures from these headers */
Dave Chinnerf948dd72013-04-03 16:11:19 +110051#include "xfs_symlink.h"
Dave Chinnerd75afeb2013-04-03 16:11:29 +110052#include "xfs_da_btree.h"
53#include "xfs_dir2_format.h"
54#include "xfs_dir2_priv.h"
55#include "xfs_attr_leaf.h"
56#include "xfs_attr_remote.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -050058STATIC int
59xlog_find_zeroed(
60 struct xlog *,
61 xfs_daddr_t *);
62STATIC int
63xlog_clear_stale_blocks(
64 struct xlog *,
65 xfs_lsn_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -070066#if defined(DEBUG)
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -050067STATIC void
68xlog_recover_check_summary(
69 struct xlog *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070070#else
71#define xlog_recover_check_summary(log)
Linus Torvalds1da177e2005-04-16 15:20:36 -070072#endif
73
Linus Torvalds1da177e2005-04-16 15:20:36 -070074/*
Christoph Hellwigd5689ea2010-12-01 22:06:22 +000075 * This structure is used during recovery to record the buf log items which
76 * have been canceled and should not be replayed.
77 */
78struct xfs_buf_cancel {
79 xfs_daddr_t bc_blkno;
80 uint bc_len;
81 int bc_refcount;
82 struct list_head bc_list;
83};
84
85/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 * Sector aligned buffer routines for buffer create/read/write/access
87 */
88
Alex Elderff30a622010-04-13 15:22:58 +100089/*
90 * Verify the given count of basic blocks is valid number of blocks
91 * to specify for an operation involving the given XFS log buffer.
92 * Returns nonzero if the count is valid, 0 otherwise.
93 */
94
95static inline int
96xlog_buf_bbcount_valid(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -050097 struct xlog *log,
Alex Elderff30a622010-04-13 15:22:58 +100098 int bbcount)
99{
100 return bbcount > 0 && bbcount <= log->l_logBBsize;
101}
102
Alex Elder36adecf2010-04-13 15:21:13 +1000103/*
104 * Allocate a buffer to hold log data. The buffer needs to be able
105 * to map to a range of nbblks basic blocks at any valid (basic
106 * block) offset within the log.
107 */
Eric Sandeen5d77c0d2009-11-19 15:52:00 +0000108STATIC xfs_buf_t *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109xlog_get_bp(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500110 struct xlog *log,
Dave Chinner32281492009-01-22 15:37:47 +1100111 int nbblks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112{
Christoph Hellwigc8da0fa2011-07-08 14:36:25 +0200113 struct xfs_buf *bp;
114
Alex Elderff30a622010-04-13 15:22:58 +1000115 if (!xlog_buf_bbcount_valid(log, nbblks)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100116 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
Alex Elderff30a622010-04-13 15:22:58 +1000117 nbblks);
118 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
Dave Chinner32281492009-01-22 15:37:47 +1100119 return NULL;
120 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121
Alex Elder36adecf2010-04-13 15:21:13 +1000122 /*
123 * We do log I/O in units of log sectors (a power-of-2
124 * multiple of the basic block size), so we round up the
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300125 * requested size to accommodate the basic blocks required
Alex Elder36adecf2010-04-13 15:21:13 +1000126 * for complete log sectors.
127 *
128 * In addition, the buffer may be used for a non-sector-
129 * aligned block offset, in which case an I/O of the
130 * requested size could extend beyond the end of the
131 * buffer. If the requested size is only 1 basic block it
132 * will never straddle a sector boundary, so this won't be
133 * an issue. Nor will this be a problem if the log I/O is
134 * done in basic blocks (sector size 1). But otherwise we
135 * extend the buffer by one extra log sector to ensure
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300136 * there's space to accommodate this possibility.
Alex Elder36adecf2010-04-13 15:21:13 +1000137 */
Alex Elder69ce58f2010-04-20 17:09:59 +1000138 if (nbblks > 1 && log->l_sectBBsize > 1)
139 nbblks += log->l_sectBBsize;
140 nbblks = round_up(nbblks, log->l_sectBBsize);
Alex Elder36adecf2010-04-13 15:21:13 +1000141
Dave Chinnere70b73f2012-04-23 15:58:49 +1000142 bp = xfs_buf_get_uncached(log->l_mp->m_logdev_targp, nbblks, 0);
Christoph Hellwigc8da0fa2011-07-08 14:36:25 +0200143 if (bp)
144 xfs_buf_unlock(bp);
145 return bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146}
147
Eric Sandeen5d77c0d2009-11-19 15:52:00 +0000148STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149xlog_put_bp(
150 xfs_buf_t *bp)
151{
152 xfs_buf_free(bp);
153}
154
Alex Elder48389ef2010-04-20 17:10:21 +1000155/*
156 * Return the address of the start of the given block number's data
157 * in a log buffer. The buffer covers a log sector-aligned region.
158 */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100159STATIC xfs_caddr_t
160xlog_align(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500161 struct xlog *log,
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100162 xfs_daddr_t blk_no,
163 int nbblks,
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500164 struct xfs_buf *bp)
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100165{
Christoph Hellwigfdc07f42010-05-10 17:28:14 +0000166 xfs_daddr_t offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100167
Dave Chinner4e94b712012-04-23 15:58:51 +1000168 ASSERT(offset + nbblks <= bp->b_length);
Chandra Seetharaman62926042011-07-22 23:40:15 +0000169 return bp->b_addr + BBTOB(offset);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100170}
171
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172
173/*
174 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
175 */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100176STATIC int
177xlog_bread_noalign(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500178 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 xfs_daddr_t blk_no,
180 int nbblks,
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500181 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182{
183 int error;
184
Alex Elderff30a622010-04-13 15:22:58 +1000185 if (!xlog_buf_bbcount_valid(log, nbblks)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100186 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
Alex Elderff30a622010-04-13 15:22:58 +1000187 nbblks);
188 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
Dave Chinner32281492009-01-22 15:37:47 +1100189 return EFSCORRUPTED;
190 }
191
Alex Elder69ce58f2010-04-20 17:09:59 +1000192 blk_no = round_down(blk_no, log->l_sectBBsize);
193 nbblks = round_up(nbblks, log->l_sectBBsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194
195 ASSERT(nbblks > 0);
Dave Chinner4e94b712012-04-23 15:58:51 +1000196 ASSERT(nbblks <= bp->b_length);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197
198 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
199 XFS_BUF_READ(bp);
Dave Chinneraa0e8832012-04-23 15:58:52 +1000200 bp->b_io_length = nbblks;
Dave Chinner0e95f192012-04-23 15:58:46 +1000201 bp->b_error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202
203 xfsbdstrat(log->l_mp, bp);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +0000204 error = xfs_buf_iowait(bp);
David Chinnerd64e31a2008-04-10 12:22:17 +1000205 if (error)
Christoph Hellwig901796a2011-10-10 16:52:49 +0000206 xfs_buf_ioerror_alert(bp, __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 return error;
208}
209
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100210STATIC int
211xlog_bread(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500212 struct xlog *log,
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100213 xfs_daddr_t blk_no,
214 int nbblks,
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500215 struct xfs_buf *bp,
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100216 xfs_caddr_t *offset)
217{
218 int error;
219
220 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
221 if (error)
222 return error;
223
224 *offset = xlog_align(log, blk_no, nbblks, bp);
225 return 0;
226}
227
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228/*
Dave Chinner44396472011-04-21 09:34:27 +0000229 * Read at an offset into the buffer. Returns with the buffer in it's original
230 * state regardless of the result of the read.
231 */
232STATIC int
233xlog_bread_offset(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500234 struct xlog *log,
Dave Chinner44396472011-04-21 09:34:27 +0000235 xfs_daddr_t blk_no, /* block to read from */
236 int nbblks, /* blocks to read */
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500237 struct xfs_buf *bp,
Dave Chinner44396472011-04-21 09:34:27 +0000238 xfs_caddr_t offset)
239{
Chandra Seetharaman62926042011-07-22 23:40:15 +0000240 xfs_caddr_t orig_offset = bp->b_addr;
Dave Chinner4e94b712012-04-23 15:58:51 +1000241 int orig_len = BBTOB(bp->b_length);
Dave Chinner44396472011-04-21 09:34:27 +0000242 int error, error2;
243
Chandra Seetharaman02fe03d2011-07-22 23:40:22 +0000244 error = xfs_buf_associate_memory(bp, offset, BBTOB(nbblks));
Dave Chinner44396472011-04-21 09:34:27 +0000245 if (error)
246 return error;
247
248 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
249
250 /* must reset buffer pointer even on error */
Chandra Seetharaman02fe03d2011-07-22 23:40:22 +0000251 error2 = xfs_buf_associate_memory(bp, orig_offset, orig_len);
Dave Chinner44396472011-04-21 09:34:27 +0000252 if (error)
253 return error;
254 return error2;
255}
256
257/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 * Write out the buffer at the given block for the given number of blocks.
259 * The buffer is kept locked across the write and is returned locked.
260 * This can only be used for synchronous log writes.
261 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000262STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263xlog_bwrite(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500264 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 xfs_daddr_t blk_no,
266 int nbblks,
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500267 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268{
269 int error;
270
Alex Elderff30a622010-04-13 15:22:58 +1000271 if (!xlog_buf_bbcount_valid(log, nbblks)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100272 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
Alex Elderff30a622010-04-13 15:22:58 +1000273 nbblks);
274 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
Dave Chinner32281492009-01-22 15:37:47 +1100275 return EFSCORRUPTED;
276 }
277
Alex Elder69ce58f2010-04-20 17:09:59 +1000278 blk_no = round_down(blk_no, log->l_sectBBsize);
279 nbblks = round_up(nbblks, log->l_sectBBsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280
281 ASSERT(nbblks > 0);
Dave Chinner4e94b712012-04-23 15:58:51 +1000282 ASSERT(nbblks <= bp->b_length);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283
284 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
285 XFS_BUF_ZEROFLAGS(bp);
Chandra Seetharaman72790aa2011-07-22 23:40:04 +0000286 xfs_buf_hold(bp);
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200287 xfs_buf_lock(bp);
Dave Chinneraa0e8832012-04-23 15:58:52 +1000288 bp->b_io_length = nbblks;
Dave Chinner0e95f192012-04-23 15:58:46 +1000289 bp->b_error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290
Christoph Hellwigc2b006c2011-08-23 08:28:07 +0000291 error = xfs_bwrite(bp);
Christoph Hellwig901796a2011-10-10 16:52:49 +0000292 if (error)
293 xfs_buf_ioerror_alert(bp, __func__);
Christoph Hellwigc2b006c2011-08-23 08:28:07 +0000294 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 return error;
296}
297
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298#ifdef DEBUG
299/*
300 * dump debug superblock and log record information
301 */
302STATIC void
303xlog_header_check_dump(
304 xfs_mount_t *mp,
305 xlog_rec_header_t *head)
306{
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100307 xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d\n",
Joe Perches03daa572009-12-14 18:01:10 -0800308 __func__, &mp->m_sb.sb_uuid, XLOG_FMT);
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100309 xfs_debug(mp, " log : uuid = %pU, fmt = %d\n",
Joe Perches03daa572009-12-14 18:01:10 -0800310 &head->h_fs_uuid, be32_to_cpu(head->h_fmt));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311}
312#else
313#define xlog_header_check_dump(mp, head)
314#endif
315
316/*
317 * check log record header for recovery
318 */
319STATIC int
320xlog_header_check_recover(
321 xfs_mount_t *mp,
322 xlog_rec_header_t *head)
323{
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200324 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325
326 /*
327 * IRIX doesn't write the h_fmt field and leaves it zeroed
328 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
329 * a dirty log created in IRIX.
330 */
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200331 if (unlikely(head->h_fmt != cpu_to_be32(XLOG_FMT))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100332 xfs_warn(mp,
333 "dirty log written in incompatible format - can't recover");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 xlog_header_check_dump(mp, head);
335 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
336 XFS_ERRLEVEL_HIGH, mp);
337 return XFS_ERROR(EFSCORRUPTED);
338 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100339 xfs_warn(mp,
340 "dirty log entry has mismatched uuid - can't recover");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 xlog_header_check_dump(mp, head);
342 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
343 XFS_ERRLEVEL_HIGH, mp);
344 return XFS_ERROR(EFSCORRUPTED);
345 }
346 return 0;
347}
348
349/*
350 * read the head block of the log and check the header
351 */
352STATIC int
353xlog_header_check_mount(
354 xfs_mount_t *mp,
355 xlog_rec_header_t *head)
356{
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200357 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358
359 if (uuid_is_nil(&head->h_fs_uuid)) {
360 /*
361 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
362 * h_fs_uuid is nil, we assume this log was last mounted
363 * by IRIX and continue.
364 */
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100365 xfs_warn(mp, "nil uuid in log - IRIX style log");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100367 xfs_warn(mp, "log has mismatched uuid - can't recover");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 xlog_header_check_dump(mp, head);
369 XFS_ERROR_REPORT("xlog_header_check_mount",
370 XFS_ERRLEVEL_HIGH, mp);
371 return XFS_ERROR(EFSCORRUPTED);
372 }
373 return 0;
374}
375
376STATIC void
377xlog_recover_iodone(
378 struct xfs_buf *bp)
379{
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +0000380 if (bp->b_error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 /*
382 * We're not going to bother about retrying
383 * this during recovery. One strike!
384 */
Christoph Hellwig901796a2011-10-10 16:52:49 +0000385 xfs_buf_ioerror_alert(bp, __func__);
Dave Chinnerebad8612010-09-22 10:47:20 +1000386 xfs_force_shutdown(bp->b_target->bt_mount,
387 SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 }
Christoph Hellwigcb669ca2011-07-13 13:43:49 +0200389 bp->b_iodone = NULL;
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +0000390 xfs_buf_ioend(bp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391}
392
393/*
394 * This routine finds (to an approximation) the first block in the physical
395 * log which contains the given cycle. It uses a binary search algorithm.
396 * Note that the algorithm can not be perfect because the disk will not
397 * necessarily be perfect.
398 */
David Chinnera8272ce2007-11-23 16:28:09 +1100399STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400xlog_find_cycle_start(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500401 struct xlog *log,
402 struct xfs_buf *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 xfs_daddr_t first_blk,
404 xfs_daddr_t *last_blk,
405 uint cycle)
406{
407 xfs_caddr_t offset;
408 xfs_daddr_t mid_blk;
Alex Eldere3bb2e32010-04-15 18:17:30 +0000409 xfs_daddr_t end_blk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 uint mid_cycle;
411 int error;
412
Alex Eldere3bb2e32010-04-15 18:17:30 +0000413 end_blk = *last_blk;
414 mid_blk = BLK_AVG(first_blk, end_blk);
415 while (mid_blk != first_blk && mid_blk != end_blk) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100416 error = xlog_bread(log, mid_blk, 1, bp, &offset);
417 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 return error;
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000419 mid_cycle = xlog_get_cycle(offset);
Alex Eldere3bb2e32010-04-15 18:17:30 +0000420 if (mid_cycle == cycle)
421 end_blk = mid_blk; /* last_half_cycle == mid_cycle */
422 else
423 first_blk = mid_blk; /* first_half_cycle == mid_cycle */
424 mid_blk = BLK_AVG(first_blk, end_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 }
Alex Eldere3bb2e32010-04-15 18:17:30 +0000426 ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) ||
427 (mid_blk == end_blk && mid_blk-1 == first_blk));
428
429 *last_blk = end_blk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430
431 return 0;
432}
433
434/*
Alex Elder3f943d82010-04-15 18:17:34 +0000435 * Check that a range of blocks does not contain stop_on_cycle_no.
436 * Fill in *new_blk with the block offset where such a block is
437 * found, or with -1 (an invalid block number) if there is no such
438 * block in the range. The scan needs to occur from front to back
439 * and the pointer into the region must be updated since a later
440 * routine will need to perform another test.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 */
442STATIC int
443xlog_find_verify_cycle(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500444 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 xfs_daddr_t start_blk,
446 int nbblks,
447 uint stop_on_cycle_no,
448 xfs_daddr_t *new_blk)
449{
450 xfs_daddr_t i, j;
451 uint cycle;
452 xfs_buf_t *bp;
453 xfs_daddr_t bufblks;
454 xfs_caddr_t buf = NULL;
455 int error = 0;
456
Alex Elder6881a222010-04-13 15:22:29 +1000457 /*
458 * Greedily allocate a buffer big enough to handle the full
459 * range of basic blocks we'll be examining. If that fails,
460 * try a smaller size. We need to be able to read at least
461 * a log sector, or we're out of luck.
462 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 bufblks = 1 << ffs(nbblks);
Dave Chinner81158e02012-04-27 19:45:22 +1000464 while (bufblks > log->l_logBBsize)
465 bufblks >>= 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 while (!(bp = xlog_get_bp(log, bufblks))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 bufblks >>= 1;
Alex Elder69ce58f2010-04-20 17:09:59 +1000468 if (bufblks < log->l_sectBBsize)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 return ENOMEM;
470 }
471
472 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
473 int bcount;
474
475 bcount = min(bufblks, (start_blk + nbblks - i));
476
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100477 error = xlog_bread(log, i, bcount, bp, &buf);
478 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 goto out;
480
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 for (j = 0; j < bcount; j++) {
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000482 cycle = xlog_get_cycle(buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 if (cycle == stop_on_cycle_no) {
484 *new_blk = i+j;
485 goto out;
486 }
487
488 buf += BBSIZE;
489 }
490 }
491
492 *new_blk = -1;
493
494out:
495 xlog_put_bp(bp);
496 return error;
497}
498
499/*
500 * Potentially backup over partial log record write.
501 *
502 * In the typical case, last_blk is the number of the block directly after
503 * a good log record. Therefore, we subtract one to get the block number
504 * of the last block in the given buffer. extra_bblks contains the number
505 * of blocks we would have read on a previous read. This happens when the
506 * last log record is split over the end of the physical log.
507 *
508 * extra_bblks is the number of blocks potentially verified on a previous
509 * call to this routine.
510 */
511STATIC int
512xlog_find_verify_log_record(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500513 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 xfs_daddr_t start_blk,
515 xfs_daddr_t *last_blk,
516 int extra_bblks)
517{
518 xfs_daddr_t i;
519 xfs_buf_t *bp;
520 xfs_caddr_t offset = NULL;
521 xlog_rec_header_t *head = NULL;
522 int error = 0;
523 int smallmem = 0;
524 int num_blks = *last_blk - start_blk;
525 int xhdrs;
526
527 ASSERT(start_blk != 0 || *last_blk != start_blk);
528
529 if (!(bp = xlog_get_bp(log, num_blks))) {
530 if (!(bp = xlog_get_bp(log, 1)))
531 return ENOMEM;
532 smallmem = 1;
533 } else {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100534 error = xlog_bread(log, start_blk, num_blks, bp, &offset);
535 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 offset += ((num_blks - 1) << BBSHIFT);
538 }
539
540 for (i = (*last_blk) - 1; i >= 0; i--) {
541 if (i < start_blk) {
542 /* valid log record not found */
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100543 xfs_warn(log->l_mp,
544 "Log inconsistent (didn't find previous header)");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 ASSERT(0);
546 error = XFS_ERROR(EIO);
547 goto out;
548 }
549
550 if (smallmem) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100551 error = xlog_bread(log, i, 1, bp, &offset);
552 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 }
555
556 head = (xlog_rec_header_t *)offset;
557
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200558 if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 break;
560
561 if (!smallmem)
562 offset -= BBSIZE;
563 }
564
565 /*
566 * We hit the beginning of the physical log & still no header. Return
567 * to caller. If caller can handle a return of -1, then this routine
568 * will be called again for the end of the physical log.
569 */
570 if (i == -1) {
571 error = -1;
572 goto out;
573 }
574
575 /*
576 * We have the final block of the good log (the first block
577 * of the log record _before_ the head. So we check the uuid.
578 */
579 if ((error = xlog_header_check_mount(log->l_mp, head)))
580 goto out;
581
582 /*
583 * We may have found a log record header before we expected one.
584 * last_blk will be the 1st block # with a given cycle #. We may end
585 * up reading an entire log record. In this case, we don't want to
586 * reset last_blk. Only when last_blk points in the middle of a log
587 * record do we update last_blk.
588 */
Eric Sandeen62118702008-03-06 13:44:28 +1100589 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000590 uint h_size = be32_to_cpu(head->h_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591
592 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
593 if (h_size % XLOG_HEADER_CYCLE_SIZE)
594 xhdrs++;
595 } else {
596 xhdrs = 1;
597 }
598
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000599 if (*last_blk - i + extra_bblks !=
600 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 *last_blk = i;
602
603out:
604 xlog_put_bp(bp);
605 return error;
606}
607
608/*
609 * Head is defined to be the point of the log where the next log write
610 * write could go. This means that incomplete LR writes at the end are
611 * eliminated when calculating the head. We aren't guaranteed that previous
612 * LR have complete transactions. We only know that a cycle number of
613 * current cycle number -1 won't be present in the log if we start writing
614 * from our current block number.
615 *
616 * last_blk contains the block number of the first block with a given
617 * cycle number.
618 *
619 * Return: zero if normal, non-zero if error.
620 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000621STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622xlog_find_head(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500623 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 xfs_daddr_t *return_head_blk)
625{
626 xfs_buf_t *bp;
627 xfs_caddr_t offset;
628 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
629 int num_scan_bblks;
630 uint first_half_cycle, last_half_cycle;
631 uint stop_on_cycle;
632 int error, log_bbnum = log->l_logBBsize;
633
634 /* Is the end of the log device zeroed? */
635 if ((error = xlog_find_zeroed(log, &first_blk)) == -1) {
636 *return_head_blk = first_blk;
637
638 /* Is the whole lot zeroed? */
639 if (!first_blk) {
640 /* Linux XFS shouldn't generate totally zeroed logs -
641 * mkfs etc write a dummy unmount record to a fresh
642 * log so we can store the uuid in there
643 */
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100644 xfs_warn(log->l_mp, "totally zeroed log");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 }
646
647 return 0;
648 } else if (error) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100649 xfs_warn(log->l_mp, "empty log check failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 return error;
651 }
652
653 first_blk = 0; /* get cycle # of 1st block */
654 bp = xlog_get_bp(log, 1);
655 if (!bp)
656 return ENOMEM;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100657
658 error = xlog_bread(log, 0, 1, bp, &offset);
659 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 goto bp_err;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100661
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000662 first_half_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663
664 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100665 error = xlog_bread(log, last_blk, 1, bp, &offset);
666 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 goto bp_err;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100668
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000669 last_half_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 ASSERT(last_half_cycle != 0);
671
672 /*
673 * If the 1st half cycle number is equal to the last half cycle number,
674 * then the entire log is stamped with the same cycle number. In this
675 * case, head_blk can't be set to zero (which makes sense). The below
676 * math doesn't work out properly with head_blk equal to zero. Instead,
677 * we set it to log_bbnum which is an invalid block number, but this
678 * value makes the math correct. If head_blk doesn't changed through
679 * all the tests below, *head_blk is set to zero at the very end rather
680 * than log_bbnum. In a sense, log_bbnum and zero are the same block
681 * in a circular file.
682 */
683 if (first_half_cycle == last_half_cycle) {
684 /*
685 * In this case we believe that the entire log should have
686 * cycle number last_half_cycle. We need to scan backwards
687 * from the end verifying that there are no holes still
688 * containing last_half_cycle - 1. If we find such a hole,
689 * then the start of that hole will be the new head. The
690 * simple case looks like
691 * x | x ... | x - 1 | x
692 * Another case that fits this picture would be
693 * x | x + 1 | x ... | x
Nathan Scottc41564b2006-03-29 08:55:14 +1000694 * In this case the head really is somewhere at the end of the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 * log, as one of the latest writes at the beginning was
696 * incomplete.
697 * One more case is
698 * x | x + 1 | x ... | x - 1 | x
699 * This is really the combination of the above two cases, and
700 * the head has to end up at the start of the x-1 hole at the
701 * end of the log.
702 *
703 * In the 256k log case, we will read from the beginning to the
704 * end of the log and search for cycle numbers equal to x-1.
705 * We don't worry about the x+1 blocks that we encounter,
706 * because we know that they cannot be the head since the log
707 * started with x.
708 */
709 head_blk = log_bbnum;
710 stop_on_cycle = last_half_cycle - 1;
711 } else {
712 /*
713 * In this case we want to find the first block with cycle
714 * number matching last_half_cycle. We expect the log to be
715 * some variation on
Alex Elder3f943d82010-04-15 18:17:34 +0000716 * x + 1 ... | x ... | x
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 * The first block with cycle number x (last_half_cycle) will
718 * be where the new head belongs. First we do a binary search
719 * for the first occurrence of last_half_cycle. The binary
720 * search may not be totally accurate, so then we scan back
721 * from there looking for occurrences of last_half_cycle before
722 * us. If that backwards scan wraps around the beginning of
723 * the log, then we look for occurrences of last_half_cycle - 1
724 * at the end of the log. The cases we're looking for look
725 * like
Alex Elder3f943d82010-04-15 18:17:34 +0000726 * v binary search stopped here
727 * x + 1 ... | x | x + 1 | x ... | x
728 * ^ but we want to locate this spot
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 * or
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 * <---------> less than scan distance
Alex Elder3f943d82010-04-15 18:17:34 +0000731 * x + 1 ... | x ... | x - 1 | x
732 * ^ we want to locate this spot
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 */
734 stop_on_cycle = last_half_cycle;
735 if ((error = xlog_find_cycle_start(log, bp, first_blk,
736 &head_blk, last_half_cycle)))
737 goto bp_err;
738 }
739
740 /*
741 * Now validate the answer. Scan back some number of maximum possible
742 * blocks and make sure each one has the expected cycle number. The
743 * maximum is determined by the total possible amount of buffering
744 * in the in-core log. The following number can be made tighter if
745 * we actually look at the block size of the filesystem.
746 */
747 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
748 if (head_blk >= num_scan_bblks) {
749 /*
750 * We are guaranteed that the entire check can be performed
751 * in one buffer.
752 */
753 start_blk = head_blk - num_scan_bblks;
754 if ((error = xlog_find_verify_cycle(log,
755 start_blk, num_scan_bblks,
756 stop_on_cycle, &new_blk)))
757 goto bp_err;
758 if (new_blk != -1)
759 head_blk = new_blk;
760 } else { /* need to read 2 parts of log */
761 /*
762 * We are going to scan backwards in the log in two parts.
763 * First we scan the physical end of the log. In this part
764 * of the log, we are looking for blocks with cycle number
765 * last_half_cycle - 1.
766 * If we find one, then we know that the log starts there, as
767 * we've found a hole that didn't get written in going around
768 * the end of the physical log. The simple case for this is
769 * x + 1 ... | x ... | x - 1 | x
770 * <---------> less than scan distance
771 * If all of the blocks at the end of the log have cycle number
772 * last_half_cycle, then we check the blocks at the start of
773 * the log looking for occurrences of last_half_cycle. If we
774 * find one, then our current estimate for the location of the
775 * first occurrence of last_half_cycle is wrong and we move
776 * back to the hole we've found. This case looks like
777 * x + 1 ... | x | x + 1 | x ...
778 * ^ binary search stopped here
779 * Another case we need to handle that only occurs in 256k
780 * logs is
781 * x + 1 ... | x ... | x+1 | x ...
782 * ^ binary search stops here
783 * In a 256k log, the scan at the end of the log will see the
784 * x + 1 blocks. We need to skip past those since that is
785 * certainly not the head of the log. By searching for
786 * last_half_cycle-1 we accomplish that.
787 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 ASSERT(head_blk <= INT_MAX &&
Alex Elder3f943d82010-04-15 18:17:34 +0000789 (xfs_daddr_t) num_scan_bblks >= head_blk);
790 start_blk = log_bbnum - (num_scan_bblks - head_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 if ((error = xlog_find_verify_cycle(log, start_blk,
792 num_scan_bblks - (int)head_blk,
793 (stop_on_cycle - 1), &new_blk)))
794 goto bp_err;
795 if (new_blk != -1) {
796 head_blk = new_blk;
Alex Elder9db127e2010-04-15 18:17:26 +0000797 goto validate_head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 }
799
800 /*
801 * Scan beginning of log now. The last part of the physical
802 * log is good. This scan needs to verify that it doesn't find
803 * the last_half_cycle.
804 */
805 start_blk = 0;
806 ASSERT(head_blk <= INT_MAX);
807 if ((error = xlog_find_verify_cycle(log,
808 start_blk, (int)head_blk,
809 stop_on_cycle, &new_blk)))
810 goto bp_err;
811 if (new_blk != -1)
812 head_blk = new_blk;
813 }
814
Alex Elder9db127e2010-04-15 18:17:26 +0000815validate_head:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 /*
817 * Now we need to make sure head_blk is not pointing to a block in
818 * the middle of a log record.
819 */
820 num_scan_bblks = XLOG_REC_SHIFT(log);
821 if (head_blk >= num_scan_bblks) {
822 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
823
824 /* start ptr at last block ptr before head_blk */
825 if ((error = xlog_find_verify_log_record(log, start_blk,
826 &head_blk, 0)) == -1) {
827 error = XFS_ERROR(EIO);
828 goto bp_err;
829 } else if (error)
830 goto bp_err;
831 } else {
832 start_blk = 0;
833 ASSERT(head_blk <= INT_MAX);
834 if ((error = xlog_find_verify_log_record(log, start_blk,
835 &head_blk, 0)) == -1) {
836 /* We hit the beginning of the log during our search */
Alex Elder3f943d82010-04-15 18:17:34 +0000837 start_blk = log_bbnum - (num_scan_bblks - head_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 new_blk = log_bbnum;
839 ASSERT(start_blk <= INT_MAX &&
840 (xfs_daddr_t) log_bbnum-start_blk >= 0);
841 ASSERT(head_blk <= INT_MAX);
842 if ((error = xlog_find_verify_log_record(log,
843 start_blk, &new_blk,
844 (int)head_blk)) == -1) {
845 error = XFS_ERROR(EIO);
846 goto bp_err;
847 } else if (error)
848 goto bp_err;
849 if (new_blk != log_bbnum)
850 head_blk = new_blk;
851 } else if (error)
852 goto bp_err;
853 }
854
855 xlog_put_bp(bp);
856 if (head_blk == log_bbnum)
857 *return_head_blk = 0;
858 else
859 *return_head_blk = head_blk;
860 /*
861 * When returning here, we have a good block number. Bad block
862 * means that during a previous crash, we didn't have a clean break
863 * from cycle number N to cycle number N-1. In this case, we need
864 * to find the first block with cycle number N-1.
865 */
866 return 0;
867
868 bp_err:
869 xlog_put_bp(bp);
870
871 if (error)
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100872 xfs_warn(log->l_mp, "failed to find log head");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 return error;
874}
875
876/*
877 * Find the sync block number or the tail of the log.
878 *
879 * This will be the block number of the last record to have its
880 * associated buffers synced to disk. Every log record header has
881 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
882 * to get a sync block number. The only concern is to figure out which
883 * log record header to believe.
884 *
885 * The following algorithm uses the log record header with the largest
886 * lsn. The entire log record does not need to be valid. We only care
887 * that the header is valid.
888 *
889 * We could speed up search by using current head_blk buffer, but it is not
890 * available.
891 */
Eric Sandeen5d77c0d2009-11-19 15:52:00 +0000892STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893xlog_find_tail(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500894 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 xfs_daddr_t *head_blk,
Eric Sandeen65be6052006-01-11 15:34:19 +1100896 xfs_daddr_t *tail_blk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897{
898 xlog_rec_header_t *rhead;
899 xlog_op_header_t *op_head;
900 xfs_caddr_t offset = NULL;
901 xfs_buf_t *bp;
902 int error, i, found;
903 xfs_daddr_t umount_data_blk;
904 xfs_daddr_t after_umount_blk;
905 xfs_lsn_t tail_lsn;
906 int hblks;
907
908 found = 0;
909
910 /*
911 * Find previous log record
912 */
913 if ((error = xlog_find_head(log, head_blk)))
914 return error;
915
916 bp = xlog_get_bp(log, 1);
917 if (!bp)
918 return ENOMEM;
919 if (*head_blk == 0) { /* special case */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100920 error = xlog_bread(log, 0, 1, bp, &offset);
921 if (error)
Alex Elder9db127e2010-04-15 18:17:26 +0000922 goto done;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100923
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000924 if (xlog_get_cycle(offset) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 *tail_blk = 0;
926 /* leave all other log inited values alone */
Alex Elder9db127e2010-04-15 18:17:26 +0000927 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 }
929 }
930
931 /*
932 * Search backwards looking for log record header block
933 */
934 ASSERT(*head_blk < INT_MAX);
935 for (i = (int)(*head_blk) - 1; i >= 0; i--) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100936 error = xlog_bread(log, i, 1, bp, &offset);
937 if (error)
Alex Elder9db127e2010-04-15 18:17:26 +0000938 goto done;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100939
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200940 if (*(__be32 *)offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 found = 1;
942 break;
943 }
944 }
945 /*
946 * If we haven't found the log record header block, start looking
947 * again from the end of the physical log. XXXmiken: There should be
948 * a check here to make sure we didn't search more than N blocks in
949 * the previous code.
950 */
951 if (!found) {
952 for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100953 error = xlog_bread(log, i, 1, bp, &offset);
954 if (error)
Alex Elder9db127e2010-04-15 18:17:26 +0000955 goto done;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100956
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200957 if (*(__be32 *)offset ==
958 cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 found = 2;
960 break;
961 }
962 }
963 }
964 if (!found) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100965 xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 ASSERT(0);
967 return XFS_ERROR(EIO);
968 }
969
970 /* find blk_no of tail of log */
971 rhead = (xlog_rec_header_t *)offset;
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000972 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973
974 /*
975 * Reset log values according to the state of the log when we
976 * crashed. In the case where head_blk == 0, we bump curr_cycle
977 * one because the next write starts a new cycle rather than
978 * continuing the cycle of the last good log record. At this
979 * point we have guaranteed that all partial log records have been
980 * accounted for. Therefore, we know that the last good log record
981 * written was complete and ended exactly on the end boundary
982 * of the physical log.
983 */
984 log->l_prev_block = i;
985 log->l_curr_block = (int)*head_blk;
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000986 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 if (found == 2)
988 log->l_curr_cycle++;
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100989 atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn));
Dave Chinner84f3c682010-12-03 22:11:29 +1100990 atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn));
Christoph Hellwig28496962012-02-20 02:31:25 +0000991 xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle,
Dave Chinnera69ed032010-12-21 12:08:20 +1100992 BBTOB(log->l_curr_block));
Christoph Hellwig28496962012-02-20 02:31:25 +0000993 xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle,
Dave Chinnera69ed032010-12-21 12:08:20 +1100994 BBTOB(log->l_curr_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995
996 /*
997 * Look for unmount record. If we find it, then we know there
998 * was a clean unmount. Since 'i' could be the last block in
999 * the physical log, we convert to a log block before comparing
1000 * to the head_blk.
1001 *
1002 * Save the current tail lsn to use to pass to
1003 * xlog_clear_stale_blocks() below. We won't want to clear the
1004 * unmount record if there is one, so we pass the lsn of the
1005 * unmount record rather than the block after it.
1006 */
Eric Sandeen62118702008-03-06 13:44:28 +11001007 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001008 int h_size = be32_to_cpu(rhead->h_size);
1009 int h_version = be32_to_cpu(rhead->h_version);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010
1011 if ((h_version & XLOG_VERSION_2) &&
1012 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
1013 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
1014 if (h_size % XLOG_HEADER_CYCLE_SIZE)
1015 hblks++;
1016 } else {
1017 hblks = 1;
1018 }
1019 } else {
1020 hblks = 1;
1021 }
1022 after_umount_blk = (i + hblks + (int)
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001023 BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize;
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001024 tail_lsn = atomic64_read(&log->l_tail_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 if (*head_blk == after_umount_blk &&
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001026 be32_to_cpu(rhead->h_num_logops) == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 umount_data_blk = (i + hblks) % log->l_logBBsize;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001028 error = xlog_bread(log, umount_data_blk, 1, bp, &offset);
1029 if (error)
Alex Elder9db127e2010-04-15 18:17:26 +00001030 goto done;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001031
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 op_head = (xlog_op_header_t *)offset;
1033 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
1034 /*
1035 * Set tail and last sync so that newly written
1036 * log records will point recovery to after the
1037 * current unmount record.
1038 */
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001039 xlog_assign_atomic_lsn(&log->l_tail_lsn,
1040 log->l_curr_cycle, after_umount_blk);
1041 xlog_assign_atomic_lsn(&log->l_last_sync_lsn,
1042 log->l_curr_cycle, after_umount_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 *tail_blk = after_umount_blk;
David Chinner92821e22007-05-24 15:26:31 +10001044
1045 /*
1046 * Note that the unmount was clean. If the unmount
1047 * was not clean, we need to know this to rebuild the
1048 * superblock counters from the perag headers if we
1049 * have a filesystem using non-persistent counters.
1050 */
1051 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 }
1053 }
1054
1055 /*
1056 * Make sure that there are no blocks in front of the head
1057 * with the same cycle number as the head. This can happen
1058 * because we allow multiple outstanding log writes concurrently,
1059 * and the later writes might make it out before earlier ones.
1060 *
1061 * We use the lsn from before modifying it so that we'll never
1062 * overwrite the unmount record after a clean unmount.
1063 *
1064 * Do this only if we are going to recover the filesystem
1065 *
1066 * NOTE: This used to say "if (!readonly)"
1067 * However on Linux, we can & do recover a read-only filesystem.
1068 * We only skip recovery if NORECOVERY is specified on mount,
1069 * in which case we would not be here.
1070 *
1071 * But... if the -device- itself is readonly, just skip this.
1072 * We can't recover this device anyway, so it won't matter.
1073 */
Alex Elder9db127e2010-04-15 18:17:26 +00001074 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 error = xlog_clear_stale_blocks(log, tail_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076
Alex Elder9db127e2010-04-15 18:17:26 +00001077done:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 xlog_put_bp(bp);
1079
1080 if (error)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001081 xfs_warn(log->l_mp, "failed to locate log tail");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 return error;
1083}
1084
1085/*
1086 * Is the log zeroed at all?
1087 *
1088 * The last binary search should be changed to perform an X block read
1089 * once X becomes small enough. You can then search linearly through
1090 * the X blocks. This will cut down on the number of reads we need to do.
1091 *
1092 * If the log is partially zeroed, this routine will pass back the blkno
1093 * of the first block with cycle number 0. It won't have a complete LR
1094 * preceding it.
1095 *
1096 * Return:
1097 * 0 => the log is completely written to
1098 * -1 => use *blk_no as the first block of the log
1099 * >0 => error has occurred
1100 */
David Chinnera8272ce2007-11-23 16:28:09 +11001101STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102xlog_find_zeroed(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001103 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 xfs_daddr_t *blk_no)
1105{
1106 xfs_buf_t *bp;
1107 xfs_caddr_t offset;
1108 uint first_cycle, last_cycle;
1109 xfs_daddr_t new_blk, last_blk, start_blk;
1110 xfs_daddr_t num_scan_bblks;
1111 int error, log_bbnum = log->l_logBBsize;
1112
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10001113 *blk_no = 0;
1114
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 /* check totally zeroed log */
1116 bp = xlog_get_bp(log, 1);
1117 if (!bp)
1118 return ENOMEM;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001119 error = xlog_bread(log, 0, 1, bp, &offset);
1120 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 goto bp_err;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001122
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10001123 first_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 if (first_cycle == 0) { /* completely zeroed log */
1125 *blk_no = 0;
1126 xlog_put_bp(bp);
1127 return -1;
1128 }
1129
1130 /* check partially zeroed log */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001131 error = xlog_bread(log, log_bbnum-1, 1, bp, &offset);
1132 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 goto bp_err;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001134
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10001135 last_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 if (last_cycle != 0) { /* log completely written to */
1137 xlog_put_bp(bp);
1138 return 0;
1139 } else if (first_cycle != 1) {
1140 /*
1141 * If the cycle of the last block is zero, the cycle of
1142 * the first block must be 1. If it's not, maybe we're
1143 * not looking at a log... Bail out.
1144 */
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001145 xfs_warn(log->l_mp,
1146 "Log inconsistent or not a log (last==0, first!=1)");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 return XFS_ERROR(EINVAL);
1148 }
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 */
1184 if ((error = xlog_find_verify_log_record(log, start_blk,
1185 &last_blk, 0)) == -1) {
1186 error = XFS_ERROR(EIO);
1187 goto bp_err;
1188 } else if (error)
1189 goto bp_err;
1190
1191 *blk_no = last_blk;
1192bp_err:
1193 xlog_put_bp(bp);
1194 if (error)
1195 return error;
1196 return -1;
1197}
1198
1199/*
1200 * These are simple subroutines used by xlog_clear_stale_blocks() below
1201 * to initialize a buffer full of empty log record headers and write
1202 * them into the log.
1203 */
1204STATIC void
1205xlog_add_record(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001206 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 xfs_caddr_t buf,
1208 int cycle,
1209 int block,
1210 int tail_cycle,
1211 int tail_block)
1212{
1213 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1214
1215 memset(buf, 0, BBSIZE);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001216 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1217 recp->h_cycle = cpu_to_be32(cycle);
1218 recp->h_version = cpu_to_be32(
Eric Sandeen62118702008-03-06 13:44:28 +11001219 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001220 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1221 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1222 recp->h_fmt = cpu_to_be32(XLOG_FMT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1224}
1225
1226STATIC int
1227xlog_write_log_records(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001228 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 int cycle,
1230 int start_block,
1231 int blocks,
1232 int tail_cycle,
1233 int tail_block)
1234{
1235 xfs_caddr_t offset;
1236 xfs_buf_t *bp;
1237 int balign, ealign;
Alex Elder69ce58f2010-04-20 17:09:59 +10001238 int sectbb = log->l_sectBBsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 int end_block = start_block + blocks;
1240 int bufblks;
1241 int error = 0;
1242 int i, j = 0;
1243
Alex Elder6881a222010-04-13 15:22:29 +10001244 /*
1245 * Greedily allocate a buffer big enough to handle the full
1246 * range of basic blocks to be written. If that fails, try
1247 * a smaller size. We need to be able to write at least a
1248 * log sector, or we're out of luck.
1249 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 bufblks = 1 << ffs(blocks);
Dave Chinner81158e02012-04-27 19:45:22 +10001251 while (bufblks > log->l_logBBsize)
1252 bufblks >>= 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 while (!(bp = xlog_get_bp(log, bufblks))) {
1254 bufblks >>= 1;
Alex Elder69ce58f2010-04-20 17:09:59 +10001255 if (bufblks < sectbb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 return ENOMEM;
1257 }
1258
1259 /* We may need to do a read at the start to fill in part of
1260 * the buffer in the starting sector not covered by the first
1261 * write below.
1262 */
Alex Elder5c17f532010-04-13 15:22:48 +10001263 balign = round_down(start_block, sectbb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 if (balign != start_block) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001265 error = xlog_bread_noalign(log, start_block, 1, bp);
1266 if (error)
1267 goto out_put_bp;
1268
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 j = start_block - balign;
1270 }
1271
1272 for (i = start_block; i < end_block; i += bufblks) {
1273 int bcount, endcount;
1274
1275 bcount = min(bufblks, end_block - start_block);
1276 endcount = bcount - j;
1277
1278 /* We may need to do a read at the end to fill in part of
1279 * the buffer in the final sector not covered by the write.
1280 * If this is the same sector as the above read, skip it.
1281 */
Alex Elder5c17f532010-04-13 15:22:48 +10001282 ealign = round_down(end_block, sectbb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 if (j == 0 && (start_block + endcount > ealign)) {
Chandra Seetharaman62926042011-07-22 23:40:15 +00001284 offset = bp->b_addr + BBTOB(ealign - start_block);
Dave Chinner44396472011-04-21 09:34:27 +00001285 error = xlog_bread_offset(log, ealign, sectbb,
1286 bp, offset);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001287 if (error)
1288 break;
1289
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 }
1291
1292 offset = xlog_align(log, start_block, endcount, bp);
1293 for (; j < endcount; j++) {
1294 xlog_add_record(log, offset, cycle, i+j,
1295 tail_cycle, tail_block);
1296 offset += BBSIZE;
1297 }
1298 error = xlog_bwrite(log, start_block, endcount, bp);
1299 if (error)
1300 break;
1301 start_block += endcount;
1302 j = 0;
1303 }
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001304
1305 out_put_bp:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 xlog_put_bp(bp);
1307 return error;
1308}
1309
1310/*
1311 * This routine is called to blow away any incomplete log writes out
1312 * in front of the log head. We do this so that we won't become confused
1313 * if we come up, write only a little bit more, and then crash again.
1314 * If we leave the partial log records out there, this situation could
1315 * cause us to think those partial writes are valid blocks since they
1316 * have the current cycle number. We get rid of them by overwriting them
1317 * with empty log records with the old cycle number rather than the
1318 * current one.
1319 *
1320 * The tail lsn is passed in rather than taken from
1321 * the log so that we will not write over the unmount record after a
1322 * clean unmount in a 512 block log. Doing so would leave the log without
1323 * any valid log records in it until a new one was written. If we crashed
1324 * during that time we would not be able to recover.
1325 */
1326STATIC int
1327xlog_clear_stale_blocks(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001328 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 xfs_lsn_t tail_lsn)
1330{
1331 int tail_cycle, head_cycle;
1332 int tail_block, head_block;
1333 int tail_distance, max_distance;
1334 int distance;
1335 int error;
1336
1337 tail_cycle = CYCLE_LSN(tail_lsn);
1338 tail_block = BLOCK_LSN(tail_lsn);
1339 head_cycle = log->l_curr_cycle;
1340 head_block = log->l_curr_block;
1341
1342 /*
1343 * Figure out the distance between the new head of the log
1344 * and the tail. We want to write over any blocks beyond the
1345 * head that we may have written just before the crash, but
1346 * we don't want to overwrite the tail of the log.
1347 */
1348 if (head_cycle == tail_cycle) {
1349 /*
1350 * The tail is behind the head in the physical log,
1351 * so the distance from the head to the tail is the
1352 * distance from the head to the end of the log plus
1353 * the distance from the beginning of the log to the
1354 * tail.
1355 */
1356 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1357 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1358 XFS_ERRLEVEL_LOW, log->l_mp);
1359 return XFS_ERROR(EFSCORRUPTED);
1360 }
1361 tail_distance = tail_block + (log->l_logBBsize - head_block);
1362 } else {
1363 /*
1364 * The head is behind the tail in the physical log,
1365 * so the distance from the head to the tail is just
1366 * the tail block minus the head block.
1367 */
1368 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1369 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1370 XFS_ERRLEVEL_LOW, log->l_mp);
1371 return XFS_ERROR(EFSCORRUPTED);
1372 }
1373 tail_distance = tail_block - head_block;
1374 }
1375
1376 /*
1377 * If the head is right up against the tail, we can't clear
1378 * anything.
1379 */
1380 if (tail_distance <= 0) {
1381 ASSERT(tail_distance == 0);
1382 return 0;
1383 }
1384
1385 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1386 /*
1387 * Take the smaller of the maximum amount of outstanding I/O
1388 * we could have and the distance to the tail to clear out.
1389 * We take the smaller so that we don't overwrite the tail and
1390 * we don't waste all day writing from the head to the tail
1391 * for no reason.
1392 */
1393 max_distance = MIN(max_distance, tail_distance);
1394
1395 if ((head_block + max_distance) <= log->l_logBBsize) {
1396 /*
1397 * We can stomp all the blocks we need to without
1398 * wrapping around the end of the log. Just do it
1399 * in a single write. Use the cycle number of the
1400 * current cycle minus one so that the log will look like:
1401 * n ... | n - 1 ...
1402 */
1403 error = xlog_write_log_records(log, (head_cycle - 1),
1404 head_block, max_distance, tail_cycle,
1405 tail_block);
1406 if (error)
1407 return error;
1408 } else {
1409 /*
1410 * We need to wrap around the end of the physical log in
1411 * order to clear all the blocks. Do it in two separate
1412 * I/Os. The first write should be from the head to the
1413 * end of the physical log, and it should use the current
1414 * cycle number minus one just like above.
1415 */
1416 distance = log->l_logBBsize - head_block;
1417 error = xlog_write_log_records(log, (head_cycle - 1),
1418 head_block, distance, tail_cycle,
1419 tail_block);
1420
1421 if (error)
1422 return error;
1423
1424 /*
1425 * Now write the blocks at the start of the physical log.
1426 * This writes the remainder of the blocks we want to clear.
1427 * It uses the current cycle number since we're now on the
1428 * same cycle as the head so that we get:
1429 * n ... n ... | n - 1 ...
1430 * ^^^^^ blocks we're writing
1431 */
1432 distance = max_distance - (log->l_logBBsize - head_block);
1433 error = xlog_write_log_records(log, head_cycle, 0, distance,
1434 tail_cycle, tail_block);
1435 if (error)
1436 return error;
1437 }
1438
1439 return 0;
1440}
1441
1442/******************************************************************************
1443 *
1444 * Log recover routines
1445 *
1446 ******************************************************************************
1447 */
1448
1449STATIC xlog_recover_t *
1450xlog_recover_find_tid(
Dave Chinnerf0a76952010-01-11 11:49:57 +00001451 struct hlist_head *head,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 xlog_tid_t tid)
1453{
Dave Chinnerf0a76952010-01-11 11:49:57 +00001454 xlog_recover_t *trans;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455
Sasha Levinb67bfe02013-02-27 17:06:00 -08001456 hlist_for_each_entry(trans, head, r_list) {
Dave Chinnerf0a76952010-01-11 11:49:57 +00001457 if (trans->r_log_tid == tid)
1458 return trans;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 }
Dave Chinnerf0a76952010-01-11 11:49:57 +00001460 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461}
1462
1463STATIC void
Dave Chinnerf0a76952010-01-11 11:49:57 +00001464xlog_recover_new_tid(
1465 struct hlist_head *head,
1466 xlog_tid_t tid,
1467 xfs_lsn_t lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468{
Dave Chinnerf0a76952010-01-11 11:49:57 +00001469 xlog_recover_t *trans;
1470
1471 trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
1472 trans->r_log_tid = tid;
1473 trans->r_lsn = lsn;
1474 INIT_LIST_HEAD(&trans->r_itemq);
1475
1476 INIT_HLIST_NODE(&trans->r_list);
1477 hlist_add_head(&trans->r_list, head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478}
1479
1480STATIC void
1481xlog_recover_add_item(
Dave Chinnerf0a76952010-01-11 11:49:57 +00001482 struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483{
1484 xlog_recover_item_t *item;
1485
1486 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
Dave Chinnerf0a76952010-01-11 11:49:57 +00001487 INIT_LIST_HEAD(&item->ri_list);
1488 list_add_tail(&item->ri_list, head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489}
1490
1491STATIC int
1492xlog_recover_add_to_cont_trans(
Mark Tinguelyad223e62012-06-14 09:22:15 -05001493 struct xlog *log,
1494 struct xlog_recover *trans,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 xfs_caddr_t dp,
1496 int len)
1497{
1498 xlog_recover_item_t *item;
1499 xfs_caddr_t ptr, old_ptr;
1500 int old_len;
1501
Dave Chinnerf0a76952010-01-11 11:49:57 +00001502 if (list_empty(&trans->r_itemq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 /* finish copying rest of trans header */
1504 xlog_recover_add_item(&trans->r_itemq);
1505 ptr = (xfs_caddr_t) &trans->r_theader +
1506 sizeof(xfs_trans_header_t) - len;
1507 memcpy(ptr, dp, len); /* d, s, l */
1508 return 0;
1509 }
Dave Chinnerf0a76952010-01-11 11:49:57 +00001510 /* take the tail entry */
1511 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512
1513 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
1514 old_len = item->ri_buf[item->ri_cnt-1].i_len;
1515
Mitsuo Hayasaka45053602012-01-27 06:37:26 +00001516 ptr = kmem_realloc(old_ptr, len+old_len, old_len, KM_SLEEP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 memcpy(&ptr[old_len], dp, len); /* d, s, l */
1518 item->ri_buf[item->ri_cnt-1].i_len += len;
1519 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
Dave Chinner9abbc532010-04-13 15:06:46 +10001520 trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 return 0;
1522}
1523
1524/*
1525 * The next region to add is the start of a new region. It could be
1526 * a whole region or it could be the first part of a new region. Because
1527 * of this, the assumption here is that the type and size fields of all
1528 * format structures fit into the first 32 bits of the structure.
1529 *
1530 * This works because all regions must be 32 bit aligned. Therefore, we
1531 * either have both fields or we have neither field. In the case we have
1532 * neither field, the data part of the region is zero length. We only have
1533 * a log_op_header and can throw away the header since a new one will appear
1534 * later. If we have at least 4 bytes, then we can determine how many regions
1535 * will appear in the current log item.
1536 */
1537STATIC int
1538xlog_recover_add_to_trans(
Mark Tinguelyad223e62012-06-14 09:22:15 -05001539 struct xlog *log,
1540 struct xlog_recover *trans,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 xfs_caddr_t dp,
1542 int len)
1543{
1544 xfs_inode_log_format_t *in_f; /* any will do */
1545 xlog_recover_item_t *item;
1546 xfs_caddr_t ptr;
1547
1548 if (!len)
1549 return 0;
Dave Chinnerf0a76952010-01-11 11:49:57 +00001550 if (list_empty(&trans->r_itemq)) {
David Chinner5a792c42008-10-30 17:40:09 +11001551 /* we need to catch log corruptions here */
1552 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001553 xfs_warn(log->l_mp, "%s: bad header magic number",
1554 __func__);
David Chinner5a792c42008-10-30 17:40:09 +11001555 ASSERT(0);
1556 return XFS_ERROR(EIO);
1557 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 if (len == sizeof(xfs_trans_header_t))
1559 xlog_recover_add_item(&trans->r_itemq);
1560 memcpy(&trans->r_theader, dp, len); /* d, s, l */
1561 return 0;
1562 }
1563
1564 ptr = kmem_alloc(len, KM_SLEEP);
1565 memcpy(ptr, dp, len);
1566 in_f = (xfs_inode_log_format_t *)ptr;
1567
Dave Chinnerf0a76952010-01-11 11:49:57 +00001568 /* take the tail entry */
1569 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
1570 if (item->ri_total != 0 &&
1571 item->ri_total == item->ri_cnt) {
1572 /* tail item is in use, get a new one */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 xlog_recover_add_item(&trans->r_itemq);
Dave Chinnerf0a76952010-01-11 11:49:57 +00001574 item = list_entry(trans->r_itemq.prev,
1575 xlog_recover_item_t, ri_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577
1578 if (item->ri_total == 0) { /* first region to be added */
Christoph Hellwige8fa6b42009-03-03 14:48:36 -05001579 if (in_f->ilf_size == 0 ||
1580 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001581 xfs_warn(log->l_mp,
1582 "bad number of regions (%d) in inode log format",
Christoph Hellwige8fa6b42009-03-03 14:48:36 -05001583 in_f->ilf_size);
1584 ASSERT(0);
1585 return XFS_ERROR(EIO);
1586 }
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;
1658 LIST_HEAD(sort_list);
Dave Chinnera775ad72013-06-05 12:09:07 +10001659 LIST_HEAD(cancel_list);
1660 LIST_HEAD(buffer_list);
1661 LIST_HEAD(inode_buffer_list);
1662 LIST_HEAD(inode_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663
Dave Chinnerf0a76952010-01-11 11:49:57 +00001664 list_splice_init(&trans->r_itemq, &sort_list);
1665 list_for_each_entry_safe(item, n, &sort_list, ri_list) {
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10001666 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
Dave Chinnerf0a76952010-01-11 11:49:57 +00001667
1668 switch (ITEM_TYPE(item)) {
Dave Chinner28c8e412013-06-27 16:04:55 +10001669 case XFS_LI_ICREATE:
1670 list_move_tail(&item->ri_list, &buffer_list);
1671 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 case XFS_LI_BUF:
Dave Chinnera775ad72013-06-05 12:09:07 +10001673 if (buf_f->blf_flags & XFS_BLF_CANCEL) {
Dave Chinner9abbc532010-04-13 15:06:46 +10001674 trace_xfs_log_recover_item_reorder_head(log,
1675 trans, item, pass);
Dave Chinnera775ad72013-06-05 12:09:07 +10001676 list_move(&item->ri_list, &cancel_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 break;
1678 }
Dave Chinnera775ad72013-06-05 12:09:07 +10001679 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
1680 list_move(&item->ri_list, &inode_buffer_list);
1681 break;
1682 }
1683 list_move_tail(&item->ri_list, &buffer_list);
1684 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 case XFS_LI_INODE:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 case XFS_LI_DQUOT:
1687 case XFS_LI_QUOTAOFF:
1688 case XFS_LI_EFD:
1689 case XFS_LI_EFI:
Dave Chinner9abbc532010-04-13 15:06:46 +10001690 trace_xfs_log_recover_item_reorder_tail(log,
1691 trans, item, pass);
Dave Chinnera775ad72013-06-05 12:09:07 +10001692 list_move_tail(&item->ri_list, &inode_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 break;
1694 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001695 xfs_warn(log->l_mp,
1696 "%s: unrecognized type of log operation",
1697 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698 ASSERT(0);
1699 return XFS_ERROR(EIO);
1700 }
Dave Chinnerf0a76952010-01-11 11:49:57 +00001701 }
1702 ASSERT(list_empty(&sort_list));
Dave Chinnera775ad72013-06-05 12:09:07 +10001703 if (!list_empty(&buffer_list))
1704 list_splice(&buffer_list, &trans->r_itemq);
1705 if (!list_empty(&inode_list))
1706 list_splice_tail(&inode_list, &trans->r_itemq);
1707 if (!list_empty(&inode_buffer_list))
1708 list_splice_tail(&inode_buffer_list, &trans->r_itemq);
1709 if (!list_empty(&cancel_list))
1710 list_splice_tail(&cancel_list, &trans->r_itemq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 return 0;
1712}
1713
1714/*
1715 * Build up the table of buf cancel records so that we don't replay
1716 * cancelled data in the second pass. For buffer records that are
1717 * not cancel records, there is nothing to do here so we just return.
1718 *
1719 * If we get a cancel record which is already in the table, this indicates
1720 * that the buffer was cancelled multiple times. In order to ensure
1721 * that during pass 2 we keep the record in the table until we reach its
1722 * last occurrence in the log, we keep a reference count in the cancel
1723 * record in the table to tell us how many times we expect to see this
1724 * record during the second pass.
1725 */
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001726STATIC int
1727xlog_recover_buffer_pass1(
Mark Tinguelyad223e62012-06-14 09:22:15 -05001728 struct xlog *log,
1729 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001731 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001732 struct list_head *bucket;
1733 struct xfs_buf_cancel *bcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734
1735 /*
1736 * If this isn't a cancel buffer item, then just return.
1737 */
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001738 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
Dave Chinner9abbc532010-04-13 15:06:46 +10001739 trace_xfs_log_recover_buf_not_cancel(log, buf_f);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001740 return 0;
Dave Chinner9abbc532010-04-13 15:06:46 +10001741 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742
1743 /*
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001744 * Insert an xfs_buf_cancel record into the hash table of them.
1745 * If there is already an identical record, bump its reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 */
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001747 bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno);
1748 list_for_each_entry(bcp, bucket, bc_list) {
1749 if (bcp->bc_blkno == buf_f->blf_blkno &&
1750 bcp->bc_len == buf_f->blf_len) {
1751 bcp->bc_refcount++;
Dave Chinner9abbc532010-04-13 15:06:46 +10001752 trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001753 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 }
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001756
1757 bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP);
1758 bcp->bc_blkno = buf_f->blf_blkno;
1759 bcp->bc_len = buf_f->blf_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 bcp->bc_refcount = 1;
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001761 list_add_tail(&bcp->bc_list, bucket);
1762
Dave Chinner9abbc532010-04-13 15:06:46 +10001763 trace_xfs_log_recover_buf_cancel_add(log, buf_f);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001764 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765}
1766
1767/*
1768 * Check to see whether the buffer being recovered has a corresponding
1769 * entry in the buffer cancel record table. If it does then return 1
1770 * so that it will be cancelled, otherwise return 0. If the buffer is
Dave Chinnerc1155412010-05-07 11:05:19 +10001771 * actually a buffer cancel item (XFS_BLF_CANCEL is set), then decrement
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 * the refcount on the entry in the table and remove it from the table
1773 * if this is the last reference.
1774 *
1775 * We remove the cancel record from the table when we encounter its
1776 * last occurrence in the log so that if the same buffer is re-used
1777 * again after its last cancellation we actually replay the changes
1778 * made at that point.
1779 */
1780STATIC int
1781xlog_check_buffer_cancelled(
Mark Tinguelyad223e62012-06-14 09:22:15 -05001782 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 xfs_daddr_t blkno,
1784 uint len,
1785 ushort flags)
1786{
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001787 struct list_head *bucket;
1788 struct xfs_buf_cancel *bcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789
1790 if (log->l_buf_cancel_table == NULL) {
1791 /*
1792 * There is nothing in the table built in pass one,
1793 * so this buffer must not be cancelled.
1794 */
Dave Chinnerc1155412010-05-07 11:05:19 +10001795 ASSERT(!(flags & XFS_BLF_CANCEL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 return 0;
1797 }
1798
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001799 /*
1800 * Search for an entry in the cancel table that matches our buffer.
1801 */
1802 bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno);
1803 list_for_each_entry(bcp, bucket, bc_list) {
1804 if (bcp->bc_blkno == blkno && bcp->bc_len == len)
1805 goto found;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806 }
1807
1808 /*
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001809 * We didn't find a corresponding entry in the table, so return 0 so
1810 * that the buffer is NOT cancelled.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 */
Dave Chinnerc1155412010-05-07 11:05:19 +10001812 ASSERT(!(flags & XFS_BLF_CANCEL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813 return 0;
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001814
1815found:
1816 /*
1817 * We've go a match, so return 1 so that the recovery of this buffer
1818 * is cancelled. If this buffer is actually a buffer cancel log
1819 * item, then decrement the refcount on the one in the table and
1820 * remove it if this is the last reference.
1821 */
1822 if (flags & XFS_BLF_CANCEL) {
1823 if (--bcp->bc_refcount == 0) {
1824 list_del(&bcp->bc_list);
1825 kmem_free(bcp);
1826 }
1827 }
1828 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829}
1830
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831/*
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001832 * Perform recovery for a buffer full of inodes. In these buffers, the only
1833 * data which should be recovered is that which corresponds to the
1834 * di_next_unlinked pointers in the on disk inode structures. The rest of the
1835 * data for the inodes is always logged through the inodes themselves rather
1836 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 *
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001838 * The only time when buffers full of inodes are fully recovered is when the
1839 * buffer is full of newly allocated inodes. In this case the buffer will
1840 * not be marked as an inode buffer and so will be sent to
1841 * xlog_recover_do_reg_buffer() below during recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842 */
1843STATIC int
1844xlog_recover_do_inode_buffer(
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001845 struct xfs_mount *mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846 xlog_recover_item_t *item,
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001847 struct xfs_buf *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848 xfs_buf_log_format_t *buf_f)
1849{
1850 int i;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001851 int item_index = 0;
1852 int bit = 0;
1853 int nbits = 0;
1854 int reg_buf_offset = 0;
1855 int reg_buf_bytes = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 int next_unlinked_offset;
1857 int inodes_per_buf;
1858 xfs_agino_t *logged_nextp;
1859 xfs_agino_t *buffer_nextp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860
Dave Chinner9abbc532010-04-13 15:06:46 +10001861 trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f);
Dave Chinner9222a9c2013-06-12 12:19:06 +10001862
1863 /*
1864 * Post recovery validation only works properly on CRC enabled
1865 * filesystems.
1866 */
1867 if (xfs_sb_version_hascrc(&mp->m_sb))
1868 bp->b_ops = &xfs_inode_buf_ops;
Dave Chinner9abbc532010-04-13 15:06:46 +10001869
Dave Chinneraa0e8832012-04-23 15:58:52 +10001870 inodes_per_buf = BBTOB(bp->b_io_length) >> mp->m_sb.sb_inodelog;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 for (i = 0; i < inodes_per_buf; i++) {
1872 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
1873 offsetof(xfs_dinode_t, di_next_unlinked);
1874
1875 while (next_unlinked_offset >=
1876 (reg_buf_offset + reg_buf_bytes)) {
1877 /*
1878 * The next di_next_unlinked field is beyond
1879 * the current logged region. Find the next
1880 * logged region that contains or is beyond
1881 * the current di_next_unlinked field.
1882 */
1883 bit += nbits;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001884 bit = xfs_next_bit(buf_f->blf_data_map,
1885 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886
1887 /*
1888 * If there are no more logged regions in the
1889 * buffer, then we're done.
1890 */
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001891 if (bit == -1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001894 nbits = xfs_contig_bits(buf_f->blf_data_map,
1895 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896 ASSERT(nbits > 0);
Dave Chinnerc1155412010-05-07 11:05:19 +10001897 reg_buf_offset = bit << XFS_BLF_SHIFT;
1898 reg_buf_bytes = nbits << XFS_BLF_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 item_index++;
1900 }
1901
1902 /*
1903 * If the current logged region starts after the current
1904 * di_next_unlinked field, then move on to the next
1905 * di_next_unlinked field.
1906 */
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001907 if (next_unlinked_offset < reg_buf_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909
1910 ASSERT(item->ri_buf[item_index].i_addr != NULL);
Dave Chinnerc1155412010-05-07 11:05:19 +10001911 ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0);
Dave Chinneraa0e8832012-04-23 15:58:52 +10001912 ASSERT((reg_buf_offset + reg_buf_bytes) <=
1913 BBTOB(bp->b_io_length));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914
1915 /*
1916 * The current logged region contains a copy of the
1917 * current di_next_unlinked field. Extract its value
1918 * and copy it to the buffer copy.
1919 */
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10001920 logged_nextp = item->ri_buf[item_index].i_addr +
1921 next_unlinked_offset - reg_buf_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 if (unlikely(*logged_nextp == 0)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001923 xfs_alert(mp,
1924 "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
1925 "Trying to replay bad (0) inode di_next_unlinked field.",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 item, bp);
1927 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1928 XFS_ERRLEVEL_LOW, mp);
1929 return XFS_ERROR(EFSCORRUPTED);
1930 }
1931
1932 buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
1933 next_unlinked_offset);
Tim Shimmin87c199c2006-06-09 14:56:16 +10001934 *buffer_nextp = *logged_nextp;
Dave Chinner0a32c262013-06-05 12:09:08 +10001935
1936 /*
1937 * If necessary, recalculate the CRC in the on-disk inode. We
1938 * have to leave the inode in a consistent state for whoever
1939 * reads it next....
1940 */
1941 xfs_dinode_calc_crc(mp, (struct xfs_dinode *)
1942 xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize));
1943
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 }
1945
1946 return 0;
1947}
1948
1949/*
Dave Chinnerd75afeb2013-04-03 16:11:29 +11001950 * Validate the recovered buffer is of the correct type and attach the
1951 * appropriate buffer operations to them for writeback. Magic numbers are in a
1952 * few places:
1953 * the first 16 bits of the buffer (inode buffer, dquot buffer),
1954 * the first 32 bits of the buffer (most blocks),
1955 * inside a struct xfs_da_blkinfo at the start of the buffer.
1956 */
1957static void
1958xlog_recovery_validate_buf_type(
1959 struct xfs_mount *mp,
1960 struct xfs_buf *bp,
1961 xfs_buf_log_format_t *buf_f)
1962{
1963 struct xfs_da_blkinfo *info = bp->b_addr;
1964 __uint32_t magic32;
1965 __uint16_t magic16;
1966 __uint16_t magicda;
1967
1968 magic32 = be32_to_cpu(*(__be32 *)bp->b_addr);
1969 magic16 = be16_to_cpu(*(__be16*)bp->b_addr);
1970 magicda = be16_to_cpu(info->magic);
Dave Chinner61fe1352013-04-03 16:11:30 +11001971 switch (xfs_blft_from_flags(buf_f)) {
1972 case XFS_BLFT_BTREE_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11001973 switch (magic32) {
1974 case XFS_ABTB_CRC_MAGIC:
1975 case XFS_ABTC_CRC_MAGIC:
1976 case XFS_ABTB_MAGIC:
1977 case XFS_ABTC_MAGIC:
1978 bp->b_ops = &xfs_allocbt_buf_ops;
1979 break;
1980 case XFS_IBT_CRC_MAGIC:
1981 case XFS_IBT_MAGIC:
1982 bp->b_ops = &xfs_inobt_buf_ops;
1983 break;
1984 case XFS_BMAP_CRC_MAGIC:
1985 case XFS_BMAP_MAGIC:
1986 bp->b_ops = &xfs_bmbt_buf_ops;
1987 break;
1988 default:
1989 xfs_warn(mp, "Bad btree block magic!");
1990 ASSERT(0);
1991 break;
1992 }
1993 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11001994 case XFS_BLFT_AGF_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11001995 if (magic32 != XFS_AGF_MAGIC) {
1996 xfs_warn(mp, "Bad AGF block magic!");
1997 ASSERT(0);
1998 break;
1999 }
2000 bp->b_ops = &xfs_agf_buf_ops;
2001 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002002 case XFS_BLFT_AGFL_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002003 if (!xfs_sb_version_hascrc(&mp->m_sb))
2004 break;
2005 if (magic32 != XFS_AGFL_MAGIC) {
2006 xfs_warn(mp, "Bad AGFL block magic!");
2007 ASSERT(0);
2008 break;
2009 }
2010 bp->b_ops = &xfs_agfl_buf_ops;
2011 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002012 case XFS_BLFT_AGI_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002013 if (magic32 != XFS_AGI_MAGIC) {
2014 xfs_warn(mp, "Bad AGI block magic!");
2015 ASSERT(0);
2016 break;
2017 }
2018 bp->b_ops = &xfs_agi_buf_ops;
2019 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002020 case XFS_BLFT_UDQUOT_BUF:
2021 case XFS_BLFT_PDQUOT_BUF:
2022 case XFS_BLFT_GDQUOT_BUF:
Dave Chinner123887e2013-04-30 21:39:33 +10002023#ifdef CONFIG_XFS_QUOTA
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002024 if (magic16 != XFS_DQUOT_MAGIC) {
2025 xfs_warn(mp, "Bad DQUOT block magic!");
2026 ASSERT(0);
2027 break;
2028 }
2029 bp->b_ops = &xfs_dquot_buf_ops;
Dave Chinner123887e2013-04-30 21:39:33 +10002030#else
2031 xfs_alert(mp,
2032 "Trying to recover dquots without QUOTA support built in!");
2033 ASSERT(0);
2034#endif
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002035 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002036 case XFS_BLFT_DINO_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002037 /*
2038 * we get here with inode allocation buffers, not buffers that
2039 * track unlinked list changes.
2040 */
2041 if (magic16 != XFS_DINODE_MAGIC) {
2042 xfs_warn(mp, "Bad INODE block magic!");
2043 ASSERT(0);
2044 break;
2045 }
2046 bp->b_ops = &xfs_inode_buf_ops;
2047 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002048 case XFS_BLFT_SYMLINK_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002049 if (magic32 != XFS_SYMLINK_MAGIC) {
2050 xfs_warn(mp, "Bad symlink block magic!");
2051 ASSERT(0);
2052 break;
2053 }
2054 bp->b_ops = &xfs_symlink_buf_ops;
2055 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002056 case XFS_BLFT_DIR_BLOCK_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002057 if (magic32 != XFS_DIR2_BLOCK_MAGIC &&
2058 magic32 != XFS_DIR3_BLOCK_MAGIC) {
2059 xfs_warn(mp, "Bad dir block magic!");
2060 ASSERT(0);
2061 break;
2062 }
2063 bp->b_ops = &xfs_dir3_block_buf_ops;
2064 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002065 case XFS_BLFT_DIR_DATA_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002066 if (magic32 != XFS_DIR2_DATA_MAGIC &&
2067 magic32 != XFS_DIR3_DATA_MAGIC) {
2068 xfs_warn(mp, "Bad dir data magic!");
2069 ASSERT(0);
2070 break;
2071 }
2072 bp->b_ops = &xfs_dir3_data_buf_ops;
2073 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002074 case XFS_BLFT_DIR_FREE_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002075 if (magic32 != XFS_DIR2_FREE_MAGIC &&
2076 magic32 != XFS_DIR3_FREE_MAGIC) {
2077 xfs_warn(mp, "Bad dir3 free magic!");
2078 ASSERT(0);
2079 break;
2080 }
2081 bp->b_ops = &xfs_dir3_free_buf_ops;
2082 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002083 case XFS_BLFT_DIR_LEAF1_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002084 if (magicda != XFS_DIR2_LEAF1_MAGIC &&
2085 magicda != XFS_DIR3_LEAF1_MAGIC) {
2086 xfs_warn(mp, "Bad dir leaf1 magic!");
2087 ASSERT(0);
2088 break;
2089 }
2090 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
2091 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002092 case XFS_BLFT_DIR_LEAFN_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002093 if (magicda != XFS_DIR2_LEAFN_MAGIC &&
2094 magicda != XFS_DIR3_LEAFN_MAGIC) {
2095 xfs_warn(mp, "Bad dir leafn magic!");
2096 ASSERT(0);
2097 break;
2098 }
2099 bp->b_ops = &xfs_dir3_leafn_buf_ops;
2100 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002101 case XFS_BLFT_DA_NODE_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002102 if (magicda != XFS_DA_NODE_MAGIC &&
2103 magicda != XFS_DA3_NODE_MAGIC) {
2104 xfs_warn(mp, "Bad da node magic!");
2105 ASSERT(0);
2106 break;
2107 }
2108 bp->b_ops = &xfs_da3_node_buf_ops;
2109 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002110 case XFS_BLFT_ATTR_LEAF_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002111 if (magicda != XFS_ATTR_LEAF_MAGIC &&
2112 magicda != XFS_ATTR3_LEAF_MAGIC) {
2113 xfs_warn(mp, "Bad attr leaf magic!");
2114 ASSERT(0);
2115 break;
2116 }
2117 bp->b_ops = &xfs_attr3_leaf_buf_ops;
2118 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002119 case XFS_BLFT_ATTR_RMT_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002120 if (!xfs_sb_version_hascrc(&mp->m_sb))
2121 break;
Dave Chinnercab09a82013-04-30 21:39:36 +10002122 if (magic32 != XFS_ATTR3_RMT_MAGIC) {
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002123 xfs_warn(mp, "Bad attr remote magic!");
2124 ASSERT(0);
2125 break;
2126 }
2127 bp->b_ops = &xfs_attr3_rmt_buf_ops;
2128 break;
Dave Chinner04a1e6c2013-04-03 16:11:31 +11002129 case XFS_BLFT_SB_BUF:
2130 if (magic32 != XFS_SB_MAGIC) {
2131 xfs_warn(mp, "Bad SB block magic!");
2132 ASSERT(0);
2133 break;
2134 }
2135 bp->b_ops = &xfs_sb_buf_ops;
2136 break;
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002137 default:
Dave Chinner61fe1352013-04-03 16:11:30 +11002138 xfs_warn(mp, "Unknown buffer type %d!",
2139 xfs_blft_from_flags(buf_f));
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002140 break;
2141 }
2142}
2143
2144/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 * Perform a 'normal' buffer recovery. Each logged region of the
2146 * buffer should be copied over the corresponding region in the
2147 * given buffer. The bitmap in the buf log format structure indicates
2148 * where to place the logged data.
2149 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150STATIC void
2151xlog_recover_do_reg_buffer(
Dave Chinner9abbc532010-04-13 15:06:46 +10002152 struct xfs_mount *mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153 xlog_recover_item_t *item,
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002154 struct xfs_buf *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155 xfs_buf_log_format_t *buf_f)
2156{
2157 int i;
2158 int bit;
2159 int nbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160 int error;
2161
Dave Chinner9abbc532010-04-13 15:06:46 +10002162 trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f);
2163
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 bit = 0;
2165 i = 1; /* 0 is the buf format structure */
2166 while (1) {
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002167 bit = xfs_next_bit(buf_f->blf_data_map,
2168 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 if (bit == -1)
2170 break;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002171 nbits = xfs_contig_bits(buf_f->blf_data_map,
2172 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173 ASSERT(nbits > 0);
Christoph Hellwig4b809162007-08-16 15:37:36 +10002174 ASSERT(item->ri_buf[i].i_addr != NULL);
Dave Chinnerc1155412010-05-07 11:05:19 +10002175 ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0);
Dave Chinneraa0e8832012-04-23 15:58:52 +10002176 ASSERT(BBTOB(bp->b_io_length) >=
2177 ((uint)bit << XFS_BLF_SHIFT) + (nbits << XFS_BLF_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178
2179 /*
Dave Chinner709da6a2013-05-27 16:38:23 +10002180 * The dirty regions logged in the buffer, even though
2181 * contiguous, may span multiple chunks. This is because the
2182 * dirty region may span a physical page boundary in a buffer
2183 * and hence be split into two separate vectors for writing into
2184 * the log. Hence we need to trim nbits back to the length of
2185 * the current region being copied out of the log.
2186 */
2187 if (item->ri_buf[i].i_len < (nbits << XFS_BLF_SHIFT))
2188 nbits = item->ri_buf[i].i_len >> XFS_BLF_SHIFT;
2189
2190 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191 * Do a sanity check if this is a dquot buffer. Just checking
2192 * the first dquot in the buffer should do. XXXThis is
2193 * probably a good thing to do for other buf types also.
2194 */
2195 error = 0;
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002196 if (buf_f->blf_flags &
Dave Chinnerc1155412010-05-07 11:05:19 +10002197 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002198 if (item->ri_buf[i].i_addr == NULL) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002199 xfs_alert(mp,
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002200 "XFS: NULL dquot in %s.", __func__);
2201 goto next;
2202 }
Jan Rekorajski8ec6dba2009-11-16 11:57:02 +00002203 if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002204 xfs_alert(mp,
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002205 "XFS: dquot too small (%d) in %s.",
2206 item->ri_buf[i].i_len, __func__);
2207 goto next;
2208 }
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002209 error = xfs_qm_dqcheck(mp, item->ri_buf[i].i_addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210 -1, 0, XFS_QMOPT_DOWARN,
2211 "dquot_buf_recover");
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002212 if (error)
2213 goto next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 }
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002215
2216 memcpy(xfs_buf_offset(bp,
Dave Chinnerc1155412010-05-07 11:05:19 +10002217 (uint)bit << XFS_BLF_SHIFT), /* dest */
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002218 item->ri_buf[i].i_addr, /* source */
Dave Chinnerc1155412010-05-07 11:05:19 +10002219 nbits<<XFS_BLF_SHIFT); /* length */
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002220 next:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221 i++;
2222 bit += nbits;
2223 }
2224
2225 /* Shouldn't be any more regions */
2226 ASSERT(i == item->ri_total);
Christoph Hellwigee1a47a2013-04-21 14:53:46 -05002227
Dave Chinner9222a9c2013-06-12 12:19:06 +10002228 /*
2229 * We can only do post recovery validation on items on CRC enabled
2230 * fielsystems as we need to know when the buffer was written to be able
2231 * to determine if we should have replayed the item. If we replay old
2232 * metadata over a newer buffer, then it will enter a temporarily
2233 * inconsistent state resulting in verification failures. Hence for now
2234 * just avoid the verification stage for non-crc filesystems
2235 */
2236 if (xfs_sb_version_hascrc(&mp->m_sb))
2237 xlog_recovery_validate_buf_type(mp, bp, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238}
2239
2240/*
2241 * Do some primitive error checking on ondisk dquot data structures.
2242 */
2243int
2244xfs_qm_dqcheck(
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002245 struct xfs_mount *mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 xfs_disk_dquot_t *ddq,
2247 xfs_dqid_t id,
2248 uint type, /* used only when IO_dorepair is true */
2249 uint flags,
2250 char *str)
2251{
2252 xfs_dqblk_t *d = (xfs_dqblk_t *)ddq;
2253 int errs = 0;
2254
2255 /*
2256 * We can encounter an uninitialized dquot buffer for 2 reasons:
2257 * 1. If we crash while deleting the quotainode(s), and those blks got
2258 * used for user data. This is because we take the path of regular
2259 * file deletion; however, the size field of quotainodes is never
2260 * updated, so all the tricks that we play in itruncate_finish
2261 * don't quite matter.
2262 *
2263 * 2. We don't play the quota buffers when there's a quotaoff logitem.
2264 * But the allocation will be replayed so we'll end up with an
2265 * uninitialized quota block.
2266 *
2267 * This is all fine; things are still consistent, and we haven't lost
2268 * any quota information. Just don't complain about bad dquot blks.
2269 */
Christoph Hellwig69ef9212011-07-08 14:36:05 +02002270 if (ddq->d_magic != cpu_to_be16(XFS_DQUOT_MAGIC)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002272 xfs_alert(mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002274 str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 errs++;
2276 }
Christoph Hellwig1149d962005-11-02 15:01:12 +11002277 if (ddq->d_version != XFS_DQUOT_VERSION) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002279 xfs_alert(mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002281 str, id, ddq->d_version, XFS_DQUOT_VERSION);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282 errs++;
2283 }
2284
Christoph Hellwig1149d962005-11-02 15:01:12 +11002285 if (ddq->d_flags != XFS_DQ_USER &&
2286 ddq->d_flags != XFS_DQ_PROJ &&
2287 ddq->d_flags != XFS_DQ_GROUP) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002289 xfs_alert(mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002291 str, id, ddq->d_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 errs++;
2293 }
2294
Christoph Hellwig1149d962005-11-02 15:01:12 +11002295 if (id != -1 && id != be32_to_cpu(ddq->d_id)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002297 xfs_alert(mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298 "%s : ondisk-dquot 0x%p, ID mismatch: "
2299 "0x%x expected, found id 0x%x",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002300 str, ddq, id, be32_to_cpu(ddq->d_id));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301 errs++;
2302 }
2303
2304 if (!errs && ddq->d_id) {
Christoph Hellwig1149d962005-11-02 15:01:12 +11002305 if (ddq->d_blk_softlimit &&
Mitsuo Hayasakad0a3fe62012-02-06 12:50:07 +00002306 be64_to_cpu(ddq->d_bcount) >
Christoph Hellwig1149d962005-11-02 15:01:12 +11002307 be64_to_cpu(ddq->d_blk_softlimit)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 if (!ddq->d_btimer) {
2309 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002310 xfs_alert(mp,
2311 "%s : Dquot ID 0x%x (0x%p) BLK TIMER NOT STARTED",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002312 str, (int)be32_to_cpu(ddq->d_id), ddq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313 errs++;
2314 }
2315 }
Christoph Hellwig1149d962005-11-02 15:01:12 +11002316 if (ddq->d_ino_softlimit &&
Mitsuo Hayasakad0a3fe62012-02-06 12:50:07 +00002317 be64_to_cpu(ddq->d_icount) >
Christoph Hellwig1149d962005-11-02 15:01:12 +11002318 be64_to_cpu(ddq->d_ino_softlimit)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319 if (!ddq->d_itimer) {
2320 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002321 xfs_alert(mp,
2322 "%s : Dquot ID 0x%x (0x%p) INODE TIMER NOT STARTED",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002323 str, (int)be32_to_cpu(ddq->d_id), ddq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324 errs++;
2325 }
2326 }
Christoph Hellwig1149d962005-11-02 15:01:12 +11002327 if (ddq->d_rtb_softlimit &&
Mitsuo Hayasakad0a3fe62012-02-06 12:50:07 +00002328 be64_to_cpu(ddq->d_rtbcount) >
Christoph Hellwig1149d962005-11-02 15:01:12 +11002329 be64_to_cpu(ddq->d_rtb_softlimit)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 if (!ddq->d_rtbtimer) {
2331 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002332 xfs_alert(mp,
2333 "%s : Dquot ID 0x%x (0x%p) RTBLK TIMER NOT STARTED",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002334 str, (int)be32_to_cpu(ddq->d_id), ddq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 errs++;
2336 }
2337 }
2338 }
2339
2340 if (!errs || !(flags & XFS_QMOPT_DQREPAIR))
2341 return errs;
2342
2343 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002344 xfs_notice(mp, "Re-initializing dquot ID 0x%x", id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345
2346 /*
2347 * Typically, a repair is only requested by quotacheck.
2348 */
2349 ASSERT(id != -1);
2350 ASSERT(flags & XFS_QMOPT_DQREPAIR);
2351 memset(d, 0, sizeof(xfs_dqblk_t));
Christoph Hellwig1149d962005-11-02 15:01:12 +11002352
2353 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
2354 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
2355 d->dd_diskdq.d_flags = type;
2356 d->dd_diskdq.d_id = cpu_to_be32(id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357
Dave Chinner6fcdc592013-06-03 15:28:46 +10002358 if (xfs_sb_version_hascrc(&mp->m_sb)) {
2359 uuid_copy(&d->dd_uuid, &mp->m_sb.sb_uuid);
2360 xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
2361 XFS_DQUOT_CRC_OFF);
2362 }
2363
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364 return errs;
2365}
2366
2367/*
2368 * Perform a dquot buffer recovery.
2369 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2370 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2371 * Else, treat it as a regular buffer and do recovery.
2372 */
2373STATIC void
2374xlog_recover_do_dquot_buffer(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002375 struct xfs_mount *mp,
2376 struct xlog *log,
2377 struct xlog_recover_item *item,
2378 struct xfs_buf *bp,
2379 struct xfs_buf_log_format *buf_f)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380{
2381 uint type;
2382
Dave Chinner9abbc532010-04-13 15:06:46 +10002383 trace_xfs_log_recover_buf_dquot_buf(log, buf_f);
2384
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 /*
2386 * Filesystems are required to send in quota flags at mount time.
2387 */
2388 if (mp->m_qflags == 0) {
2389 return;
2390 }
2391
2392 type = 0;
Dave Chinnerc1155412010-05-07 11:05:19 +10002393 if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394 type |= XFS_DQ_USER;
Dave Chinnerc1155412010-05-07 11:05:19 +10002395 if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF)
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002396 type |= XFS_DQ_PROJ;
Dave Chinnerc1155412010-05-07 11:05:19 +10002397 if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398 type |= XFS_DQ_GROUP;
2399 /*
2400 * This type of quotas was turned off, so ignore this buffer
2401 */
2402 if (log->l_quotaoffs_flag & type)
2403 return;
2404
Dave Chinner9abbc532010-04-13 15:06:46 +10002405 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406}
2407
2408/*
2409 * This routine replays a modification made to a buffer at runtime.
2410 * There are actually two types of buffer, regular and inode, which
2411 * are handled differently. Inode buffers are handled differently
2412 * in that we only recover a specific set of data from them, namely
2413 * the inode di_next_unlinked fields. This is because all other inode
2414 * data is actually logged via inode records and any data we replay
2415 * here which overlaps that may be stale.
2416 *
2417 * When meta-data buffers are freed at run time we log a buffer item
Dave Chinnerc1155412010-05-07 11:05:19 +10002418 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419 * of the buffer in the log should not be replayed at recovery time.
2420 * This is so that if the blocks covered by the buffer are reused for
2421 * file data before we crash we don't end up replaying old, freed
2422 * meta-data into a user's file.
2423 *
2424 * To handle the cancellation of buffer log items, we make two passes
2425 * over the log during recovery. During the first we build a table of
2426 * those buffers which have been cancelled, and during the second we
2427 * only replay those buffers which do not have corresponding cancel
2428 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2429 * for more details on the implementation of the table of cancel records.
2430 */
2431STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002432xlog_recover_buffer_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002433 struct xlog *log,
2434 struct list_head *buffer_list,
2435 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436{
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002437 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002438 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439 xfs_buf_t *bp;
2440 int error;
Christoph Hellwig6ad112b2009-11-24 18:02:23 +00002441 uint buf_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002443 /*
2444 * In this pass we only want to recover all the buffers which have
2445 * not been cancelled and are not cancellation buffers themselves.
2446 */
2447 if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno,
2448 buf_f->blf_len, buf_f->blf_flags)) {
2449 trace_xfs_log_recover_buf_cancel(log, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451 }
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002452
Dave Chinner9abbc532010-04-13 15:06:46 +10002453 trace_xfs_log_recover_buf_recover(log, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454
Dave Chinnera8acad72012-04-23 15:58:54 +10002455 buf_flags = 0;
Dave Chinner611c9942012-04-23 15:59:07 +10002456 if (buf_f->blf_flags & XFS_BLF_INODE_BUF)
2457 buf_flags |= XBF_UNMAPPED;
Christoph Hellwig6ad112b2009-11-24 18:02:23 +00002458
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002459 bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len,
Dave Chinnerc3f8fc72012-11-12 22:54:01 +11002460 buf_flags, NULL);
Chandra Seetharamanac4d6882011-08-03 02:18:29 +00002461 if (!bp)
2462 return XFS_ERROR(ENOMEM);
Chandra Seetharamane5702802011-08-03 02:18:34 +00002463 error = bp->b_error;
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +00002464 if (error) {
Christoph Hellwig901796a2011-10-10 16:52:49 +00002465 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466 xfs_buf_relse(bp);
2467 return error;
2468 }
2469
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002470 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002472 } else if (buf_f->blf_flags &
Dave Chinnerc1155412010-05-07 11:05:19 +10002473 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474 xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2475 } else {
Dave Chinner9abbc532010-04-13 15:06:46 +10002476 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477 }
2478 if (error)
2479 return XFS_ERROR(error);
2480
2481 /*
2482 * Perform delayed write on the buffer. Asynchronous writes will be
2483 * slower when taking into account all the buffers to be flushed.
2484 *
2485 * Also make sure that only inode buffers with good sizes stay in
2486 * the buffer cache. The kernel moves inodes in buffers of 1 block
2487 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2488 * buffers in the log can be a different size if the log was generated
2489 * by an older kernel using unclustered inode buffers or a newer kernel
2490 * running with a different inode cluster size. Regardless, if the
2491 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2492 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2493 * the buffer out of the buffer cache so that the buffer won't
2494 * overlap with future reads of those inodes.
2495 */
2496 if (XFS_DINODE_MAGIC ==
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002497 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
Dave Chinneraa0e8832012-04-23 15:58:52 +10002498 (BBTOB(bp->b_io_length) != MAX(log->l_mp->m_sb.sb_blocksize,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499 (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) {
Christoph Hellwigc867cb62011-10-10 16:52:46 +00002500 xfs_buf_stale(bp);
Christoph Hellwigc2b006c2011-08-23 08:28:07 +00002501 error = xfs_bwrite(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502 } else {
Dave Chinnerebad8612010-09-22 10:47:20 +10002503 ASSERT(bp->b_target->bt_mount == mp);
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02002504 bp->b_iodone = xlog_recover_iodone;
Christoph Hellwig43ff2122012-04-23 15:58:39 +10002505 xfs_buf_delwri_queue(bp, buffer_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506 }
2507
Christoph Hellwigc2b006c2011-08-23 08:28:07 +00002508 xfs_buf_relse(bp);
2509 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510}
2511
2512STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002513xlog_recover_inode_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002514 struct xlog *log,
2515 struct list_head *buffer_list,
2516 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517{
2518 xfs_inode_log_format_t *in_f;
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002519 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520 xfs_buf_t *bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521 xfs_dinode_t *dip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522 int len;
2523 xfs_caddr_t src;
2524 xfs_caddr_t dest;
2525 int error;
2526 int attr_index;
2527 uint fields;
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002528 xfs_icdinode_t *dicp;
Christoph Hellwig93848a92013-04-03 16:11:17 +11002529 uint isize;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002530 int need_free = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531
Tim Shimmin6d192a92006-06-09 14:55:38 +10002532 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002533 in_f = item->ri_buf[0].i_addr;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002534 } else {
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002535 in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002536 need_free = 1;
2537 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2538 if (error)
2539 goto error;
2540 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002541
2542 /*
2543 * Inode buffers can be freed, look out for it,
2544 * and do not replay the inode.
2545 */
Christoph Hellwiga1941892008-11-28 14:23:40 +11002546 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
2547 in_f->ilf_len, 0)) {
Tim Shimmin6d192a92006-06-09 14:55:38 +10002548 error = 0;
Dave Chinner9abbc532010-04-13 15:06:46 +10002549 trace_xfs_log_recover_inode_cancel(log, in_f);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002550 goto error;
2551 }
Dave Chinner9abbc532010-04-13 15:06:46 +10002552 trace_xfs_log_recover_inode_recover(log, in_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553
Dave Chinnerc3f8fc72012-11-12 22:54:01 +11002554 bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len, 0,
Christoph Hellwig93848a92013-04-03 16:11:17 +11002555 &xfs_inode_buf_ops);
Chandra Seetharamanac4d6882011-08-03 02:18:29 +00002556 if (!bp) {
2557 error = ENOMEM;
2558 goto error;
2559 }
Chandra Seetharamane5702802011-08-03 02:18:34 +00002560 error = bp->b_error;
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +00002561 if (error) {
Christoph Hellwig901796a2011-10-10 16:52:49 +00002562 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563 xfs_buf_relse(bp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002564 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
Christoph Hellwiga1941892008-11-28 14:23:40 +11002567 dip = (xfs_dinode_t *)xfs_buf_offset(bp, in_f->ilf_boffset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568
2569 /*
2570 * Make sure the place we're flushing out to really looks
2571 * like an inode!
2572 */
Christoph Hellwig69ef9212011-07-08 14:36:05 +02002573 if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574 xfs_buf_relse(bp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002575 xfs_alert(mp,
2576 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
2577 __func__, dip, bp, in_f->ilf_ino);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002578 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 XFS_ERRLEVEL_LOW, mp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002580 error = EFSCORRUPTED;
2581 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582 }
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002583 dicp = item->ri_buf[1].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
2585 xfs_buf_relse(bp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002586 xfs_alert(mp,
2587 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
2588 __func__, item, in_f->ilf_ino);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002589 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590 XFS_ERRLEVEL_LOW, mp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002591 error = EFSCORRUPTED;
2592 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593 }
2594
2595 /* Skip replay when the on disk inode is newer than the log one */
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002596 if (dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597 /*
2598 * Deal with the wrap case, DI_MAX_FLUSH is less
2599 * than smaller numbers
2600 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002601 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002602 dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603 /* do nothing */
2604 } else {
2605 xfs_buf_relse(bp);
Dave Chinner9abbc532010-04-13 15:06:46 +10002606 trace_xfs_log_recover_inode_skip(log, in_f);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002607 error = 0;
2608 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609 }
2610 }
2611 /* Take the opportunity to reset the flush iteration count */
2612 dicp->di_flushiter = 0;
2613
Al Viroabbede12011-07-26 02:31:30 -04002614 if (unlikely(S_ISREG(dicp->di_mode))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2616 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002617 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618 XFS_ERRLEVEL_LOW, mp, dicp);
2619 xfs_buf_relse(bp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002620 xfs_alert(mp,
2621 "%s: Bad regular inode log record, rec ptr 0x%p, "
2622 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2623 __func__, item, dip, bp, in_f->ilf_ino);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002624 error = EFSCORRUPTED;
2625 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626 }
Al Viroabbede12011-07-26 02:31:30 -04002627 } else if (unlikely(S_ISDIR(dicp->di_mode))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2629 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2630 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002631 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632 XFS_ERRLEVEL_LOW, mp, dicp);
2633 xfs_buf_relse(bp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002634 xfs_alert(mp,
2635 "%s: Bad dir inode log record, rec ptr 0x%p, "
2636 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2637 __func__, item, dip, bp, in_f->ilf_ino);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002638 error = EFSCORRUPTED;
2639 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640 }
2641 }
2642 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002643 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644 XFS_ERRLEVEL_LOW, mp, dicp);
2645 xfs_buf_relse(bp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002646 xfs_alert(mp,
2647 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2648 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2649 __func__, item, dip, bp, in_f->ilf_ino,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650 dicp->di_nextents + dicp->di_anextents,
2651 dicp->di_nblocks);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002652 error = EFSCORRUPTED;
2653 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654 }
2655 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002656 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657 XFS_ERRLEVEL_LOW, mp, dicp);
2658 xfs_buf_relse(bp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002659 xfs_alert(mp,
2660 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2661 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__,
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002662 item, dip, bp, in_f->ilf_ino, dicp->di_forkoff);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002663 error = EFSCORRUPTED;
2664 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665 }
Christoph Hellwig93848a92013-04-03 16:11:17 +11002666 isize = xfs_icdinode_size(dicp->di_version);
2667 if (unlikely(item->ri_buf[1].i_len > isize)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002668 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669 XFS_ERRLEVEL_LOW, mp, dicp);
2670 xfs_buf_relse(bp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002671 xfs_alert(mp,
2672 "%s: Bad inode log record length %d, rec ptr 0x%p",
2673 __func__, item->ri_buf[1].i_len, item);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002674 error = EFSCORRUPTED;
2675 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676 }
2677
2678 /* The core is in in-core format */
Christoph Hellwig93848a92013-04-03 16:11:17 +11002679 xfs_dinode_to_disk(dip, dicp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680
2681 /* the rest is in on-disk format */
Christoph Hellwig93848a92013-04-03 16:11:17 +11002682 if (item->ri_buf[1].i_len > isize) {
2683 memcpy((char *)dip + isize,
2684 item->ri_buf[1].i_addr + isize,
2685 item->ri_buf[1].i_len - isize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686 }
2687
2688 fields = in_f->ilf_fields;
2689 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2690 case XFS_ILOG_DEV:
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002691 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692 break;
2693 case XFS_ILOG_UUID:
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002694 memcpy(XFS_DFORK_DPTR(dip),
2695 &in_f->ilf_u.ilfu_uuid,
2696 sizeof(uuid_t));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697 break;
2698 }
2699
2700 if (in_f->ilf_size == 2)
2701 goto write_inode_buffer;
2702 len = item->ri_buf[2].i_len;
2703 src = item->ri_buf[2].i_addr;
2704 ASSERT(in_f->ilf_size <= 4);
2705 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2706 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2707 (len == in_f->ilf_dsize));
2708
2709 switch (fields & XFS_ILOG_DFORK) {
2710 case XFS_ILOG_DDATA:
2711 case XFS_ILOG_DEXT:
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002712 memcpy(XFS_DFORK_DPTR(dip), src, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713 break;
2714
2715 case XFS_ILOG_DBROOT:
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002716 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002717 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718 XFS_DFORK_DSIZE(dip, mp));
2719 break;
2720
2721 default:
2722 /*
2723 * There are no data fork flags set.
2724 */
2725 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2726 break;
2727 }
2728
2729 /*
2730 * If we logged any attribute data, recover it. There may or
2731 * may not have been any other non-core data logged in this
2732 * transaction.
2733 */
2734 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2735 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2736 attr_index = 3;
2737 } else {
2738 attr_index = 2;
2739 }
2740 len = item->ri_buf[attr_index].i_len;
2741 src = item->ri_buf[attr_index].i_addr;
2742 ASSERT(len == in_f->ilf_asize);
2743
2744 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2745 case XFS_ILOG_ADATA:
2746 case XFS_ILOG_AEXT:
2747 dest = XFS_DFORK_APTR(dip);
2748 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2749 memcpy(dest, src, len);
2750 break;
2751
2752 case XFS_ILOG_ABROOT:
2753 dest = XFS_DFORK_APTR(dip);
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002754 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
2755 len, (xfs_bmdr_block_t*)dest,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002756 XFS_DFORK_ASIZE(dip, mp));
2757 break;
2758
2759 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002760 xfs_warn(log->l_mp, "%s: Invalid flag", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761 ASSERT(0);
2762 xfs_buf_relse(bp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002763 error = EIO;
2764 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765 }
2766 }
2767
2768write_inode_buffer:
Christoph Hellwig93848a92013-04-03 16:11:17 +11002769 /* re-generate the checksum. */
2770 xfs_dinode_calc_crc(log->l_mp, dip);
2771
Dave Chinnerebad8612010-09-22 10:47:20 +10002772 ASSERT(bp->b_target->bt_mount == mp);
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02002773 bp->b_iodone = xlog_recover_iodone;
Christoph Hellwig43ff2122012-04-23 15:58:39 +10002774 xfs_buf_delwri_queue(bp, buffer_list);
Christoph Hellwig61551f12011-08-23 08:28:06 +00002775 xfs_buf_relse(bp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002776error:
2777 if (need_free)
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002778 kmem_free(in_f);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002779 return XFS_ERROR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780}
2781
2782/*
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002783 * Recover QUOTAOFF records. We simply make a note of it in the xlog
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2785 * of that type.
2786 */
2787STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002788xlog_recover_quotaoff_pass1(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002789 struct xlog *log,
2790 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002792 xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793 ASSERT(qoff_f);
2794
2795 /*
2796 * The logitem format's flag tells us if this was user quotaoff,
Nathan Scott77a7cce2006-01-11 15:35:57 +11002797 * group/project quotaoff or both.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798 */
2799 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2800 log->l_quotaoffs_flag |= XFS_DQ_USER;
Nathan Scott77a7cce2006-01-11 15:35:57 +11002801 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2802 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2804 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2805
2806 return (0);
2807}
2808
2809/*
2810 * Recover a dquot record
2811 */
2812STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002813xlog_recover_dquot_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002814 struct xlog *log,
2815 struct list_head *buffer_list,
2816 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002818 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002819 xfs_buf_t *bp;
2820 struct xfs_disk_dquot *ddq, *recddq;
2821 int error;
2822 xfs_dq_logformat_t *dq_f;
2823 uint type;
2824
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825
2826 /*
2827 * Filesystems are required to send in quota flags at mount time.
2828 */
2829 if (mp->m_qflags == 0)
2830 return (0);
2831
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002832 recddq = item->ri_buf[1].i_addr;
2833 if (recddq == NULL) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002834 xfs_alert(log->l_mp, "NULL dquot in %s.", __func__);
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002835 return XFS_ERROR(EIO);
2836 }
Jan Rekorajski8ec6dba2009-11-16 11:57:02 +00002837 if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002838 xfs_alert(log->l_mp, "dquot too small (%d) in %s.",
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002839 item->ri_buf[1].i_len, __func__);
2840 return XFS_ERROR(EIO);
2841 }
2842
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843 /*
2844 * This type of quotas was turned off, so ignore this record.
2845 */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002846 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002847 ASSERT(type);
2848 if (log->l_quotaoffs_flag & type)
2849 return (0);
2850
2851 /*
2852 * At this point we know that quota was _not_ turned off.
2853 * Since the mount flags are not indicating to us otherwise, this
2854 * must mean that quota is on, and the dquot needs to be replayed.
2855 * Remember that we may not have fully recovered the superblock yet,
2856 * so we can't do the usual trick of looking at the SB quota bits.
2857 *
2858 * The other possibility, of course, is that the quota subsystem was
2859 * removed since the last mount - ENOSYS.
2860 */
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002861 dq_f = item->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862 ASSERT(dq_f);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002863 error = xfs_qm_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2864 "xlog_recover_dquot_pass2 (log copy)");
2865 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002866 return XFS_ERROR(EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867 ASSERT(dq_f->qlf_len == 1);
2868
Dave Chinner7ca790a2012-04-23 15:58:55 +10002869 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dq_f->qlf_blkno,
Dave Chinnerc3f8fc72012-11-12 22:54:01 +11002870 XFS_FSB_TO_BB(mp, dq_f->qlf_len), 0, &bp,
2871 NULL);
Dave Chinner7ca790a2012-04-23 15:58:55 +10002872 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002873 return error;
Dave Chinner7ca790a2012-04-23 15:58:55 +10002874
Linus Torvalds1da177e2005-04-16 15:20:36 -07002875 ASSERT(bp);
2876 ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
2877
2878 /*
2879 * At least the magic num portion should be on disk because this
2880 * was among a chunk of dquots created earlier, and we did some
2881 * minimal initialization then.
2882 */
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002883 error = xfs_qm_dqcheck(mp, ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2884 "xlog_recover_dquot_pass2");
2885 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002886 xfs_buf_relse(bp);
2887 return XFS_ERROR(EIO);
2888 }
2889
2890 memcpy(ddq, recddq, item->ri_buf[1].i_len);
Dave Chinner6fcdc592013-06-03 15:28:46 +10002891 if (xfs_sb_version_hascrc(&mp->m_sb)) {
2892 xfs_update_cksum((char *)ddq, sizeof(struct xfs_dqblk),
2893 XFS_DQUOT_CRC_OFF);
2894 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002895
2896 ASSERT(dq_f->qlf_size == 2);
Dave Chinnerebad8612010-09-22 10:47:20 +10002897 ASSERT(bp->b_target->bt_mount == mp);
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02002898 bp->b_iodone = xlog_recover_iodone;
Christoph Hellwig43ff2122012-04-23 15:58:39 +10002899 xfs_buf_delwri_queue(bp, buffer_list);
Christoph Hellwig61551f12011-08-23 08:28:06 +00002900 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002901
2902 return (0);
2903}
2904
2905/*
2906 * This routine is called to create an in-core extent free intent
2907 * item from the efi format structure which was logged on disk.
2908 * It allocates an in-core efi, copies the extents from the format
2909 * structure into it, and adds the efi to the AIL with the given
2910 * LSN.
2911 */
Tim Shimmin6d192a92006-06-09 14:55:38 +10002912STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002913xlog_recover_efi_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002914 struct xlog *log,
2915 struct xlog_recover_item *item,
2916 xfs_lsn_t lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002917{
Tim Shimmin6d192a92006-06-09 14:55:38 +10002918 int error;
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002919 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002920 xfs_efi_log_item_t *efip;
2921 xfs_efi_log_format_t *efi_formatp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002922
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002923 efi_formatp = item->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924
Linus Torvalds1da177e2005-04-16 15:20:36 -07002925 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002926 if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
2927 &(efip->efi_format)))) {
2928 xfs_efi_item_free(efip);
2929 return error;
2930 }
Dave Chinnerb199c8a2010-12-20 11:59:49 +11002931 atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002932
David Chinnera9c21c12008-10-30 17:39:35 +11002933 spin_lock(&log->l_ailp->xa_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934 /*
David Chinner783a2f62008-10-30 17:39:58 +11002935 * xfs_trans_ail_update() drops the AIL lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002936 */
Dave Chinnere6059942010-12-20 12:34:26 +11002937 xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002938 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939}
2940
2941
2942/*
2943 * This routine is called when an efd format structure is found in
2944 * a committed transaction in the log. It's purpose is to cancel
2945 * the corresponding efi if it was still in the log. To do this
2946 * it searches the AIL for the efi with an id equal to that in the
2947 * efd format structure. If we find it, we remove the efi from the
2948 * AIL and free it.
2949 */
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002950STATIC int
2951xlog_recover_efd_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002952 struct xlog *log,
2953 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002954{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002955 xfs_efd_log_format_t *efd_formatp;
2956 xfs_efi_log_item_t *efip = NULL;
2957 xfs_log_item_t *lip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002958 __uint64_t efi_id;
David Chinner27d8d5f2008-10-30 17:38:39 +11002959 struct xfs_ail_cursor cur;
David Chinner783a2f62008-10-30 17:39:58 +11002960 struct xfs_ail *ailp = log->l_ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002961
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002962 efd_formatp = item->ri_buf[0].i_addr;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002963 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
2964 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
2965 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
2966 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002967 efi_id = efd_formatp->efd_efi_id;
2968
2969 /*
2970 * Search for the efi with the id in the efd format structure
2971 * in the AIL.
2972 */
David Chinnera9c21c12008-10-30 17:39:35 +11002973 spin_lock(&ailp->xa_lock);
2974 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975 while (lip != NULL) {
2976 if (lip->li_type == XFS_LI_EFI) {
2977 efip = (xfs_efi_log_item_t *)lip;
2978 if (efip->efi_format.efi_id == efi_id) {
2979 /*
David Chinner783a2f62008-10-30 17:39:58 +11002980 * xfs_trans_ail_delete() drops the
Linus Torvalds1da177e2005-04-16 15:20:36 -07002981 * AIL lock.
2982 */
Dave Chinner04913fd2012-04-23 15:58:41 +10002983 xfs_trans_ail_delete(ailp, lip,
2984 SHUTDOWN_CORRUPT_INCORE);
David Chinner8ae2c0f2007-11-23 16:28:17 +11002985 xfs_efi_item_free(efip);
David Chinnera9c21c12008-10-30 17:39:35 +11002986 spin_lock(&ailp->xa_lock);
David Chinner27d8d5f2008-10-30 17:38:39 +11002987 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002988 }
2989 }
David Chinnera9c21c12008-10-30 17:39:35 +11002990 lip = xfs_trans_ail_cursor_next(ailp, &cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002991 }
David Chinnera9c21c12008-10-30 17:39:35 +11002992 xfs_trans_ail_cursor_done(ailp, &cur);
2993 spin_unlock(&ailp->xa_lock);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002994
2995 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002996}
2997
2998/*
Dave Chinner28c8e412013-06-27 16:04:55 +10002999 * This routine is called when an inode create format structure is found in a
3000 * committed transaction in the log. It's purpose is to initialise the inodes
3001 * being allocated on disk. This requires us to get inode cluster buffers that
3002 * match the range to be intialised, stamped with inode templates and written
3003 * by delayed write so that subsequent modifications will hit the cached buffer
3004 * and only need writing out at the end of recovery.
3005 */
3006STATIC int
3007xlog_recover_do_icreate_pass2(
3008 struct xlog *log,
3009 struct list_head *buffer_list,
3010 xlog_recover_item_t *item)
3011{
3012 struct xfs_mount *mp = log->l_mp;
3013 struct xfs_icreate_log *icl;
3014 xfs_agnumber_t agno;
3015 xfs_agblock_t agbno;
3016 unsigned int count;
3017 unsigned int isize;
3018 xfs_agblock_t length;
3019
3020 icl = (struct xfs_icreate_log *)item->ri_buf[0].i_addr;
3021 if (icl->icl_type != XFS_LI_ICREATE) {
3022 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad type");
3023 return EINVAL;
3024 }
3025
3026 if (icl->icl_size != 1) {
3027 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad icl size");
3028 return EINVAL;
3029 }
3030
3031 agno = be32_to_cpu(icl->icl_ag);
3032 if (agno >= mp->m_sb.sb_agcount) {
3033 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agno");
3034 return EINVAL;
3035 }
3036 agbno = be32_to_cpu(icl->icl_agbno);
3037 if (!agbno || agbno == NULLAGBLOCK || agbno >= mp->m_sb.sb_agblocks) {
3038 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agbno");
3039 return EINVAL;
3040 }
3041 isize = be32_to_cpu(icl->icl_isize);
3042 if (isize != mp->m_sb.sb_inodesize) {
3043 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad isize");
3044 return EINVAL;
3045 }
3046 count = be32_to_cpu(icl->icl_count);
3047 if (!count) {
3048 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count");
3049 return EINVAL;
3050 }
3051 length = be32_to_cpu(icl->icl_length);
3052 if (!length || length >= mp->m_sb.sb_agblocks) {
3053 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad length");
3054 return EINVAL;
3055 }
3056
3057 /* existing allocation is fixed value */
3058 ASSERT(count == XFS_IALLOC_INODES(mp));
3059 ASSERT(length == XFS_IALLOC_BLOCKS(mp));
3060 if (count != XFS_IALLOC_INODES(mp) ||
3061 length != XFS_IALLOC_BLOCKS(mp)) {
3062 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count 2");
3063 return EINVAL;
3064 }
3065
3066 /*
3067 * Inode buffers can be freed. Do not replay the inode initialisation as
3068 * we could be overwriting something written after this inode buffer was
3069 * cancelled.
3070 *
3071 * XXX: we need to iterate all buffers and only init those that are not
3072 * cancelled. I think that a more fine grained factoring of
3073 * xfs_ialloc_inode_init may be appropriate here to enable this to be
3074 * done easily.
3075 */
3076 if (xlog_check_buffer_cancelled(log,
3077 XFS_AGB_TO_DADDR(mp, agno, agbno), length, 0))
3078 return 0;
3079
3080 xfs_ialloc_inode_init(mp, NULL, buffer_list, agno, agbno, length,
3081 be32_to_cpu(icl->icl_gen));
3082 return 0;
3083}
3084
3085/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003086 * Free up any resources allocated by the transaction
3087 *
3088 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
3089 */
3090STATIC void
3091xlog_recover_free_trans(
Christoph Hellwigd0450942010-12-01 22:06:23 +00003092 struct xlog_recover *trans)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003093{
Dave Chinnerf0a76952010-01-11 11:49:57 +00003094 xlog_recover_item_t *item, *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003095 int i;
3096
Dave Chinnerf0a76952010-01-11 11:49:57 +00003097 list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
3098 /* Free the regions in the item. */
3099 list_del(&item->ri_list);
3100 for (i = 0; i < item->ri_cnt; i++)
3101 kmem_free(item->ri_buf[i].i_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003102 /* Free the item itself */
Dave Chinnerf0a76952010-01-11 11:49:57 +00003103 kmem_free(item->ri_buf);
3104 kmem_free(item);
3105 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003106 /* Free the transaction recover structure */
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003107 kmem_free(trans);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003108}
3109
3110STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003111xlog_recover_commit_pass1(
Mark Tinguelyad223e62012-06-14 09:22:15 -05003112 struct xlog *log,
3113 struct xlog_recover *trans,
3114 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003115{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003116 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1);
Christoph Hellwigd0450942010-12-01 22:06:23 +00003117
3118 switch (ITEM_TYPE(item)) {
3119 case XFS_LI_BUF:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003120 return xlog_recover_buffer_pass1(log, item);
Christoph Hellwigd0450942010-12-01 22:06:23 +00003121 case XFS_LI_QUOTAOFF:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003122 return xlog_recover_quotaoff_pass1(log, item);
3123 case XFS_LI_INODE:
3124 case XFS_LI_EFI:
3125 case XFS_LI_EFD:
3126 case XFS_LI_DQUOT:
Dave Chinner28c8e412013-06-27 16:04:55 +10003127 case XFS_LI_ICREATE:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003128 /* nothing to do in pass 1 */
3129 return 0;
Christoph Hellwigd0450942010-12-01 22:06:23 +00003130 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003131 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
3132 __func__, ITEM_TYPE(item));
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003133 ASSERT(0);
3134 return XFS_ERROR(EIO);
3135 }
3136}
3137
3138STATIC int
3139xlog_recover_commit_pass2(
Mark Tinguelyad223e62012-06-14 09:22:15 -05003140 struct xlog *log,
3141 struct xlog_recover *trans,
3142 struct list_head *buffer_list,
3143 struct xlog_recover_item *item)
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003144{
3145 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2);
3146
3147 switch (ITEM_TYPE(item)) {
3148 case XFS_LI_BUF:
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003149 return xlog_recover_buffer_pass2(log, buffer_list, item);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003150 case XFS_LI_INODE:
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003151 return xlog_recover_inode_pass2(log, buffer_list, item);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003152 case XFS_LI_EFI:
3153 return xlog_recover_efi_pass2(log, item, trans->r_lsn);
3154 case XFS_LI_EFD:
3155 return xlog_recover_efd_pass2(log, item);
3156 case XFS_LI_DQUOT:
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003157 return xlog_recover_dquot_pass2(log, buffer_list, item);
Dave Chinner28c8e412013-06-27 16:04:55 +10003158 case XFS_LI_ICREATE:
3159 return xlog_recover_do_icreate_pass2(log, buffer_list, item);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003160 case XFS_LI_QUOTAOFF:
3161 /* nothing to do in pass2 */
3162 return 0;
3163 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003164 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
3165 __func__, ITEM_TYPE(item));
Christoph Hellwigd0450942010-12-01 22:06:23 +00003166 ASSERT(0);
3167 return XFS_ERROR(EIO);
3168 }
3169}
3170
3171/*
3172 * Perform the transaction.
3173 *
3174 * If the transaction modifies a buffer or inode, do it now. Otherwise,
3175 * EFIs and EFDs get queued up by adding entries into the AIL for them.
3176 */
3177STATIC int
3178xlog_recover_commit_trans(
Mark Tinguelyad223e62012-06-14 09:22:15 -05003179 struct xlog *log,
Christoph Hellwigd0450942010-12-01 22:06:23 +00003180 struct xlog_recover *trans,
3181 int pass)
3182{
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003183 int error = 0, error2;
Christoph Hellwigd0450942010-12-01 22:06:23 +00003184 xlog_recover_item_t *item;
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003185 LIST_HEAD (buffer_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003186
Dave Chinnerf0a76952010-01-11 11:49:57 +00003187 hlist_del(&trans->r_list);
Christoph Hellwigd0450942010-12-01 22:06:23 +00003188
3189 error = xlog_recover_reorder_trans(log, trans, pass);
3190 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003191 return error;
Christoph Hellwigd0450942010-12-01 22:06:23 +00003192
3193 list_for_each_entry(item, &trans->r_itemq, ri_list) {
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003194 switch (pass) {
3195 case XLOG_RECOVER_PASS1:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003196 error = xlog_recover_commit_pass1(log, trans, item);
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003197 break;
3198 case XLOG_RECOVER_PASS2:
3199 error = xlog_recover_commit_pass2(log, trans,
3200 &buffer_list, item);
3201 break;
3202 default:
3203 ASSERT(0);
3204 }
3205
Christoph Hellwigd0450942010-12-01 22:06:23 +00003206 if (error)
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003207 goto out;
Christoph Hellwigd0450942010-12-01 22:06:23 +00003208 }
3209
3210 xlog_recover_free_trans(trans);
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003211
3212out:
3213 error2 = xfs_buf_delwri_submit(&buffer_list);
3214 return error ? error : error2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003215}
3216
3217STATIC int
3218xlog_recover_unmount_trans(
Mark Tinguelyad223e62012-06-14 09:22:15 -05003219 struct xlog *log,
3220 struct xlog_recover *trans)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003221{
3222 /* Do nothing now */
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003223 xfs_warn(log->l_mp, "%s: Unmount LR", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003224 return 0;
3225}
3226
3227/*
3228 * There are two valid states of the r_state field. 0 indicates that the
3229 * transaction structure is in a normal state. We have either seen the
3230 * start of the transaction or the last operation we added was not a partial
3231 * operation. If the last operation we added to the transaction was a
3232 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
3233 *
3234 * NOTE: skip LRs with 0 data length.
3235 */
3236STATIC int
3237xlog_recover_process_data(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003238 struct xlog *log,
Dave Chinnerf0a76952010-01-11 11:49:57 +00003239 struct hlist_head rhash[],
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003240 struct xlog_rec_header *rhead,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003241 xfs_caddr_t dp,
3242 int pass)
3243{
3244 xfs_caddr_t lp;
3245 int num_logops;
3246 xlog_op_header_t *ohead;
3247 xlog_recover_t *trans;
3248 xlog_tid_t tid;
3249 int error;
3250 unsigned long hash;
3251 uint flags;
3252
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003253 lp = dp + be32_to_cpu(rhead->h_len);
3254 num_logops = be32_to_cpu(rhead->h_num_logops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003255
3256 /* check the log format matches our own - else we can't recover */
3257 if (xlog_header_check_recover(log->l_mp, rhead))
3258 return (XFS_ERROR(EIO));
3259
3260 while ((dp < lp) && num_logops) {
3261 ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
3262 ohead = (xlog_op_header_t *)dp;
3263 dp += sizeof(xlog_op_header_t);
3264 if (ohead->oh_clientid != XFS_TRANSACTION &&
3265 ohead->oh_clientid != XFS_LOG) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003266 xfs_warn(log->l_mp, "%s: bad clientid 0x%x",
3267 __func__, ohead->oh_clientid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003268 ASSERT(0);
3269 return (XFS_ERROR(EIO));
3270 }
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10003271 tid = be32_to_cpu(ohead->oh_tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003272 hash = XLOG_RHASH(tid);
Dave Chinnerf0a76952010-01-11 11:49:57 +00003273 trans = xlog_recover_find_tid(&rhash[hash], tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003274 if (trans == NULL) { /* not found; add new tid */
3275 if (ohead->oh_flags & XLOG_START_TRANS)
3276 xlog_recover_new_tid(&rhash[hash], tid,
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003277 be64_to_cpu(rhead->h_lsn));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003278 } else {
Lachlan McIlroy9742bb92008-01-10 16:43:36 +11003279 if (dp + be32_to_cpu(ohead->oh_len) > lp) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003280 xfs_warn(log->l_mp, "%s: bad length 0x%x",
3281 __func__, be32_to_cpu(ohead->oh_len));
Lachlan McIlroy9742bb92008-01-10 16:43:36 +11003282 WARN_ON(1);
3283 return (XFS_ERROR(EIO));
3284 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003285 flags = ohead->oh_flags & ~XLOG_END_TRANS;
3286 if (flags & XLOG_WAS_CONT_TRANS)
3287 flags &= ~XLOG_CONTINUE_TRANS;
3288 switch (flags) {
3289 case XLOG_COMMIT_TRANS:
3290 error = xlog_recover_commit_trans(log,
Dave Chinnerf0a76952010-01-11 11:49:57 +00003291 trans, pass);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003292 break;
3293 case XLOG_UNMOUNT_TRANS:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003294 error = xlog_recover_unmount_trans(log, trans);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003295 break;
3296 case XLOG_WAS_CONT_TRANS:
Dave Chinner9abbc532010-04-13 15:06:46 +10003297 error = xlog_recover_add_to_cont_trans(log,
3298 trans, dp,
3299 be32_to_cpu(ohead->oh_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003300 break;
3301 case XLOG_START_TRANS:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003302 xfs_warn(log->l_mp, "%s: bad transaction",
3303 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003304 ASSERT(0);
3305 error = XFS_ERROR(EIO);
3306 break;
3307 case 0:
3308 case XLOG_CONTINUE_TRANS:
Dave Chinner9abbc532010-04-13 15:06:46 +10003309 error = xlog_recover_add_to_trans(log, trans,
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10003310 dp, be32_to_cpu(ohead->oh_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003311 break;
3312 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003313 xfs_warn(log->l_mp, "%s: bad flag 0x%x",
3314 __func__, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003315 ASSERT(0);
3316 error = XFS_ERROR(EIO);
3317 break;
3318 }
3319 if (error)
3320 return error;
3321 }
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10003322 dp += be32_to_cpu(ohead->oh_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003323 num_logops--;
3324 }
3325 return 0;
3326}
3327
3328/*
3329 * Process an extent free intent item that was recovered from
3330 * the log. We need to free the extents that it describes.
3331 */
David Chinner3c1e2bb2008-04-10 12:21:11 +10003332STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003333xlog_recover_process_efi(
3334 xfs_mount_t *mp,
3335 xfs_efi_log_item_t *efip)
3336{
3337 xfs_efd_log_item_t *efdp;
3338 xfs_trans_t *tp;
3339 int i;
David Chinner3c1e2bb2008-04-10 12:21:11 +10003340 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003341 xfs_extent_t *extp;
3342 xfs_fsblock_t startblock_fsb;
3343
Dave Chinnerb199c8a2010-12-20 11:59:49 +11003344 ASSERT(!test_bit(XFS_EFI_RECOVERED, &efip->efi_flags));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003345
3346 /*
3347 * First check the validity of the extents described by the
3348 * EFI. If any are bad, then assume that all are bad and
3349 * just toss the EFI.
3350 */
3351 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3352 extp = &(efip->efi_format.efi_extents[i]);
3353 startblock_fsb = XFS_BB_TO_FSB(mp,
3354 XFS_FSB_TO_DADDR(mp, extp->ext_start));
3355 if ((startblock_fsb == 0) ||
3356 (extp->ext_len == 0) ||
3357 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
3358 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
3359 /*
3360 * This will pull the EFI from the AIL and
3361 * free the memory associated with it.
3362 */
Dave Chinner666d6442013-04-03 14:09:21 +11003363 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003364 xfs_efi_release(efip, efip->efi_format.efi_nextents);
David Chinner3c1e2bb2008-04-10 12:21:11 +10003365 return XFS_ERROR(EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003366 }
3367 }
3368
3369 tp = xfs_trans_alloc(mp, 0);
David Chinner3c1e2bb2008-04-10 12:21:11 +10003370 error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 0, 0);
David Chinnerfc6149d2008-04-10 12:21:53 +10003371 if (error)
3372 goto abort_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003373 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
3374
3375 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3376 extp = &(efip->efi_format.efi_extents[i]);
David Chinnerfc6149d2008-04-10 12:21:53 +10003377 error = xfs_free_extent(tp, extp->ext_start, extp->ext_len);
3378 if (error)
3379 goto abort_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003380 xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
3381 extp->ext_len);
3382 }
3383
Dave Chinnerb199c8a2010-12-20 11:59:49 +11003384 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
David Chinnere5720ee2008-04-10 12:21:18 +10003385 error = xfs_trans_commit(tp, 0);
David Chinner3c1e2bb2008-04-10 12:21:11 +10003386 return error;
David Chinnerfc6149d2008-04-10 12:21:53 +10003387
3388abort_error:
3389 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3390 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003391}
3392
3393/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003394 * When this is called, all of the EFIs which did not have
3395 * corresponding EFDs should be in the AIL. What we do now
3396 * is free the extents associated with each one.
3397 *
3398 * Since we process the EFIs in normal transactions, they
3399 * will be removed at some point after the commit. This prevents
3400 * us from just walking down the list processing each one.
3401 * We'll use a flag in the EFI to skip those that we've already
3402 * processed and use the AIL iteration mechanism's generation
3403 * count to try to speed this up at least a bit.
3404 *
3405 * When we start, we know that the EFIs are the only things in
3406 * the AIL. As we process them, however, other items are added
3407 * to the AIL. Since everything added to the AIL must come after
3408 * everything already in the AIL, we stop processing as soon as
3409 * we see something other than an EFI in the AIL.
3410 */
David Chinner3c1e2bb2008-04-10 12:21:11 +10003411STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003412xlog_recover_process_efis(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003413 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003414{
3415 xfs_log_item_t *lip;
3416 xfs_efi_log_item_t *efip;
David Chinner3c1e2bb2008-04-10 12:21:11 +10003417 int error = 0;
David Chinner27d8d5f2008-10-30 17:38:39 +11003418 struct xfs_ail_cursor cur;
David Chinnera9c21c12008-10-30 17:39:35 +11003419 struct xfs_ail *ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003420
David Chinnera9c21c12008-10-30 17:39:35 +11003421 ailp = log->l_ailp;
3422 spin_lock(&ailp->xa_lock);
3423 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003424 while (lip != NULL) {
3425 /*
3426 * We're done when we see something other than an EFI.
David Chinner27d8d5f2008-10-30 17:38:39 +11003427 * There should be no EFIs left in the AIL now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003428 */
3429 if (lip->li_type != XFS_LI_EFI) {
David Chinner27d8d5f2008-10-30 17:38:39 +11003430#ifdef DEBUG
David Chinnera9c21c12008-10-30 17:39:35 +11003431 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
David Chinner27d8d5f2008-10-30 17:38:39 +11003432 ASSERT(lip->li_type != XFS_LI_EFI);
3433#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003434 break;
3435 }
3436
3437 /*
3438 * Skip EFIs that we've already processed.
3439 */
3440 efip = (xfs_efi_log_item_t *)lip;
Dave Chinnerb199c8a2010-12-20 11:59:49 +11003441 if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) {
David Chinnera9c21c12008-10-30 17:39:35 +11003442 lip = xfs_trans_ail_cursor_next(ailp, &cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003443 continue;
3444 }
3445
David Chinnera9c21c12008-10-30 17:39:35 +11003446 spin_unlock(&ailp->xa_lock);
3447 error = xlog_recover_process_efi(log->l_mp, efip);
3448 spin_lock(&ailp->xa_lock);
David Chinner27d8d5f2008-10-30 17:38:39 +11003449 if (error)
3450 goto out;
David Chinnera9c21c12008-10-30 17:39:35 +11003451 lip = xfs_trans_ail_cursor_next(ailp, &cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003452 }
David Chinner27d8d5f2008-10-30 17:38:39 +11003453out:
David Chinnera9c21c12008-10-30 17:39:35 +11003454 xfs_trans_ail_cursor_done(ailp, &cur);
3455 spin_unlock(&ailp->xa_lock);
David Chinner3c1e2bb2008-04-10 12:21:11 +10003456 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003457}
3458
3459/*
3460 * This routine performs a transaction to null out a bad inode pointer
3461 * in an agi unlinked inode hash bucket.
3462 */
3463STATIC void
3464xlog_recover_clear_agi_bucket(
3465 xfs_mount_t *mp,
3466 xfs_agnumber_t agno,
3467 int bucket)
3468{
3469 xfs_trans_t *tp;
3470 xfs_agi_t *agi;
3471 xfs_buf_t *agibp;
3472 int offset;
3473 int error;
3474
3475 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003476 error = xfs_trans_reserve(tp, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp),
3477 0, 0, 0);
David Chinnere5720ee2008-04-10 12:21:18 +10003478 if (error)
3479 goto out_abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003480
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003481 error = xfs_read_agi(mp, tp, agno, &agibp);
3482 if (error)
David Chinnere5720ee2008-04-10 12:21:18 +10003483 goto out_abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003484
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003485 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11003486 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003487 offset = offsetof(xfs_agi_t, agi_unlinked) +
3488 (sizeof(xfs_agino_t) * bucket);
3489 xfs_trans_log_buf(tp, agibp, offset,
3490 (offset + sizeof(xfs_agino_t) - 1));
3491
David Chinnere5720ee2008-04-10 12:21:18 +10003492 error = xfs_trans_commit(tp, 0);
3493 if (error)
3494 goto out_error;
3495 return;
3496
3497out_abort:
3498 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3499out_error:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003500 xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno);
David Chinnere5720ee2008-04-10 12:21:18 +10003501 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003502}
3503
Christoph Hellwig23fac502008-11-28 14:23:40 +11003504STATIC xfs_agino_t
3505xlog_recover_process_one_iunlink(
3506 struct xfs_mount *mp,
3507 xfs_agnumber_t agno,
3508 xfs_agino_t agino,
3509 int bucket)
3510{
3511 struct xfs_buf *ibp;
3512 struct xfs_dinode *dip;
3513 struct xfs_inode *ip;
3514 xfs_ino_t ino;
3515 int error;
3516
3517 ino = XFS_AGINO_TO_INO(mp, agno, agino);
Dave Chinner7b6259e2010-06-24 11:35:17 +10003518 error = xfs_iget(mp, NULL, ino, 0, 0, &ip);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003519 if (error)
3520 goto fail;
3521
3522 /*
3523 * Get the on disk inode to find the next inode in the bucket.
3524 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04003525 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &ibp, 0, 0);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003526 if (error)
Christoph Hellwig0e446672008-11-28 14:23:42 +11003527 goto fail_iput;
Christoph Hellwig23fac502008-11-28 14:23:40 +11003528
Christoph Hellwig23fac502008-11-28 14:23:40 +11003529 ASSERT(ip->i_d.di_nlink == 0);
Christoph Hellwig0e446672008-11-28 14:23:42 +11003530 ASSERT(ip->i_d.di_mode != 0);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003531
3532 /* setup for the next pass */
3533 agino = be32_to_cpu(dip->di_next_unlinked);
3534 xfs_buf_relse(ibp);
3535
3536 /*
3537 * Prevent any DMAPI event from being sent when the reference on
3538 * the inode is dropped.
3539 */
3540 ip->i_d.di_dmevmask = 0;
3541
Christoph Hellwig0e446672008-11-28 14:23:42 +11003542 IRELE(ip);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003543 return agino;
3544
Christoph Hellwig0e446672008-11-28 14:23:42 +11003545 fail_iput:
3546 IRELE(ip);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003547 fail:
3548 /*
3549 * We can't read in the inode this bucket points to, or this inode
3550 * is messed up. Just ditch this bucket of inodes. We will lose
3551 * some inodes and space, but at least we won't hang.
3552 *
3553 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3554 * clear the inode pointer in the bucket.
3555 */
3556 xlog_recover_clear_agi_bucket(mp, agno, bucket);
3557 return NULLAGINO;
3558}
3559
Linus Torvalds1da177e2005-04-16 15:20:36 -07003560/*
3561 * xlog_iunlink_recover
3562 *
3563 * This is called during recovery to process any inodes which
3564 * we unlinked but not freed when the system crashed. These
3565 * inodes will be on the lists in the AGI blocks. What we do
3566 * here is scan all the AGIs and fully truncate and free any
3567 * inodes found on the lists. Each inode is removed from the
3568 * lists when it has been fully truncated and is freed. The
3569 * freeing of the inode and its removal from the list must be
3570 * atomic.
3571 */
Eric Sandeend96f8f82009-07-02 00:09:33 -05003572STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07003573xlog_recover_process_iunlinks(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003574 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003575{
3576 xfs_mount_t *mp;
3577 xfs_agnumber_t agno;
3578 xfs_agi_t *agi;
3579 xfs_buf_t *agibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003580 xfs_agino_t agino;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003581 int bucket;
3582 int error;
3583 uint mp_dmevmask;
3584
3585 mp = log->l_mp;
3586
3587 /*
3588 * Prevent any DMAPI event from being sent while in this function.
3589 */
3590 mp_dmevmask = mp->m_dmevmask;
3591 mp->m_dmevmask = 0;
3592
3593 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3594 /*
3595 * Find the agi for this ag.
3596 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003597 error = xfs_read_agi(mp, NULL, agno, &agibp);
3598 if (error) {
3599 /*
3600 * AGI is b0rked. Don't process it.
3601 *
3602 * We should probably mark the filesystem as corrupt
3603 * after we've recovered all the ag's we can....
3604 */
3605 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003606 }
Jan Karad97d32e2012-03-15 09:34:02 +00003607 /*
3608 * Unlock the buffer so that it can be acquired in the normal
3609 * course of the transaction to truncate and free each inode.
3610 * Because we are not racing with anyone else here for the AGI
3611 * buffer, we don't even need to hold it locked to read the
3612 * initial unlinked bucket entries out of the buffer. We keep
3613 * buffer reference though, so that it stays pinned in memory
3614 * while we need the buffer.
3615 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003616 agi = XFS_BUF_TO_AGI(agibp);
Jan Karad97d32e2012-03-15 09:34:02 +00003617 xfs_buf_unlock(agibp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003618
3619 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
Christoph Hellwig16259e72005-11-02 15:11:25 +11003620 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003621 while (agino != NULLAGINO) {
Christoph Hellwig23fac502008-11-28 14:23:40 +11003622 agino = xlog_recover_process_one_iunlink(mp,
3623 agno, agino, bucket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003624 }
3625 }
Jan Karad97d32e2012-03-15 09:34:02 +00003626 xfs_buf_rele(agibp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003627 }
3628
3629 mp->m_dmevmask = mp_dmevmask;
3630}
3631
Linus Torvalds1da177e2005-04-16 15:20:36 -07003632/*
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003633 * Upack the log buffer data and crc check it. If the check fails, issue a
3634 * warning if and only if the CRC in the header is non-zero. This makes the
3635 * check an advisory warning, and the zero CRC check will prevent failure
3636 * warnings from being emitted when upgrading the kernel from one that does not
3637 * add CRCs by default.
3638 *
3639 * When filesystems are CRC enabled, this CRC mismatch becomes a fatal log
3640 * corruption failure
Linus Torvalds1da177e2005-04-16 15:20:36 -07003641 */
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003642STATIC int
3643xlog_unpack_data_crc(
3644 struct xlog_rec_header *rhead,
3645 xfs_caddr_t dp,
3646 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003647{
Dave Chinnerf9668a02012-11-28 13:01:03 +11003648 __le32 crc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003649
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003650 crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len));
3651 if (crc != rhead->h_crc) {
3652 if (rhead->h_crc || xfs_sb_version_hascrc(&log->l_mp->m_sb)) {
3653 xfs_alert(log->l_mp,
3654 "log record CRC mismatch: found 0x%x, expected 0x%x.\n",
Dave Chinnerf9668a02012-11-28 13:01:03 +11003655 le32_to_cpu(rhead->h_crc),
3656 le32_to_cpu(crc));
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003657 xfs_hex_dump(dp, 32);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003658 }
3659
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003660 /*
3661 * If we've detected a log record corruption, then we can't
3662 * recover past this point. Abort recovery if we are enforcing
3663 * CRC protection by punting an error back up the stack.
3664 */
3665 if (xfs_sb_version_hascrc(&log->l_mp->m_sb))
3666 return EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003667 }
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003668
3669 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003670}
3671
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003672STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003673xlog_unpack_data(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003674 struct xlog_rec_header *rhead,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003675 xfs_caddr_t dp,
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003676 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003677{
3678 int i, j, k;
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003679 int error;
3680
3681 error = xlog_unpack_data_crc(rhead, dp, log);
3682 if (error)
3683 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003684
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003685 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07003686 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003687 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003688 dp += BBSIZE;
3689 }
3690
Eric Sandeen62118702008-03-06 13:44:28 +11003691 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Christoph Hellwigb28708d2008-11-28 14:23:38 +11003692 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003693 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003694 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3695 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003696 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003697 dp += BBSIZE;
3698 }
3699 }
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003700
3701 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003702}
3703
3704STATIC int
3705xlog_valid_rec_header(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003706 struct xlog *log,
3707 struct xlog_rec_header *rhead,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003708 xfs_daddr_t blkno)
3709{
3710 int hlen;
3711
Christoph Hellwig69ef9212011-07-08 14:36:05 +02003712 if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003713 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3714 XFS_ERRLEVEL_LOW, log->l_mp);
3715 return XFS_ERROR(EFSCORRUPTED);
3716 }
3717 if (unlikely(
3718 (!rhead->h_version ||
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003719 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003720 xfs_warn(log->l_mp, "%s: unrecognised log version (%d).",
Harvey Harrison34a622b2008-04-10 12:19:21 +10003721 __func__, be32_to_cpu(rhead->h_version));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003722 return XFS_ERROR(EIO);
3723 }
3724
3725 /* LR body must have data or it wouldn't have been written */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003726 hlen = be32_to_cpu(rhead->h_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003727 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
3728 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3729 XFS_ERRLEVEL_LOW, log->l_mp);
3730 return XFS_ERROR(EFSCORRUPTED);
3731 }
3732 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
3733 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3734 XFS_ERRLEVEL_LOW, log->l_mp);
3735 return XFS_ERROR(EFSCORRUPTED);
3736 }
3737 return 0;
3738}
3739
3740/*
3741 * Read the log from tail to head and process the log records found.
3742 * Handle the two cases where the tail and head are in the same cycle
3743 * and where the active portion of the log wraps around the end of
3744 * the physical log separately. The pass parameter is passed through
3745 * to the routines called to process the data and is not looked at
3746 * here.
3747 */
3748STATIC int
3749xlog_do_recovery_pass(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003750 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003751 xfs_daddr_t head_blk,
3752 xfs_daddr_t tail_blk,
3753 int pass)
3754{
3755 xlog_rec_header_t *rhead;
3756 xfs_daddr_t blk_no;
Andy Polingfc5bc4c2009-11-03 17:26:47 +00003757 xfs_caddr_t offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003758 xfs_buf_t *hbp, *dbp;
3759 int error = 0, h_size;
3760 int bblks, split_bblks;
3761 int hblks, split_hblks, wrapped_hblks;
Dave Chinnerf0a76952010-01-11 11:49:57 +00003762 struct hlist_head rhash[XLOG_RHASH_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003763
3764 ASSERT(head_blk != tail_blk);
3765
3766 /*
3767 * Read the header of the tail block and get the iclog buffer size from
3768 * h_size. Use this to tell how many sectors make up the log header.
3769 */
Eric Sandeen62118702008-03-06 13:44:28 +11003770 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003771 /*
3772 * When using variable length iclogs, read first sector of
3773 * iclog header and extract the header size from it. Get a
3774 * new hbp that is the correct size.
3775 */
3776 hbp = xlog_get_bp(log, 1);
3777 if (!hbp)
3778 return ENOMEM;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003779
3780 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
3781 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003782 goto bread_err1;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003783
Linus Torvalds1da177e2005-04-16 15:20:36 -07003784 rhead = (xlog_rec_header_t *)offset;
3785 error = xlog_valid_rec_header(log, rhead, tail_blk);
3786 if (error)
3787 goto bread_err1;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003788 h_size = be32_to_cpu(rhead->h_size);
3789 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07003790 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
3791 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
3792 if (h_size % XLOG_HEADER_CYCLE_SIZE)
3793 hblks++;
3794 xlog_put_bp(hbp);
3795 hbp = xlog_get_bp(log, hblks);
3796 } else {
3797 hblks = 1;
3798 }
3799 } else {
Alex Elder69ce58f2010-04-20 17:09:59 +10003800 ASSERT(log->l_sectBBsize == 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003801 hblks = 1;
3802 hbp = xlog_get_bp(log, 1);
3803 h_size = XLOG_BIG_RECORD_BSIZE;
3804 }
3805
3806 if (!hbp)
3807 return ENOMEM;
3808 dbp = xlog_get_bp(log, BTOBB(h_size));
3809 if (!dbp) {
3810 xlog_put_bp(hbp);
3811 return ENOMEM;
3812 }
3813
3814 memset(rhash, 0, sizeof(rhash));
3815 if (tail_blk <= head_blk) {
3816 for (blk_no = tail_blk; blk_no < head_blk; ) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003817 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
3818 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003819 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003820
Linus Torvalds1da177e2005-04-16 15:20:36 -07003821 rhead = (xlog_rec_header_t *)offset;
3822 error = xlog_valid_rec_header(log, rhead, blk_no);
3823 if (error)
3824 goto bread_err2;
3825
3826 /* blocks in data section */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003827 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003828 error = xlog_bread(log, blk_no + hblks, bblks, dbp,
3829 &offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003830 if (error)
3831 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003832
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003833 error = xlog_unpack_data(rhead, offset, log);
3834 if (error)
3835 goto bread_err2;
3836
3837 error = xlog_recover_process_data(log,
3838 rhash, rhead, offset, pass);
3839 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003840 goto bread_err2;
3841 blk_no += bblks + hblks;
3842 }
3843 } else {
3844 /*
3845 * Perform recovery around the end of the physical log.
3846 * When the head is not on the same cycle number as the tail,
3847 * we can't do a sequential recovery as above.
3848 */
3849 blk_no = tail_blk;
3850 while (blk_no < log->l_logBBsize) {
3851 /*
3852 * Check for header wrapping around physical end-of-log
3853 */
Chandra Seetharaman62926042011-07-22 23:40:15 +00003854 offset = hbp->b_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003855 split_hblks = 0;
3856 wrapped_hblks = 0;
3857 if (blk_no + hblks <= log->l_logBBsize) {
3858 /* Read header in one read */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003859 error = xlog_bread(log, blk_no, hblks, hbp,
3860 &offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003861 if (error)
3862 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003863 } else {
3864 /* This LR is split across physical log end */
3865 if (blk_no != log->l_logBBsize) {
3866 /* some data before physical log end */
3867 ASSERT(blk_no <= INT_MAX);
3868 split_hblks = log->l_logBBsize - (int)blk_no;
3869 ASSERT(split_hblks > 0);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003870 error = xlog_bread(log, blk_no,
3871 split_hblks, hbp,
3872 &offset);
3873 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003874 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003875 }
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003876
Linus Torvalds1da177e2005-04-16 15:20:36 -07003877 /*
3878 * Note: this black magic still works with
3879 * large sector sizes (non-512) only because:
3880 * - we increased the buffer size originally
3881 * by 1 sector giving us enough extra space
3882 * for the second read;
3883 * - the log start is guaranteed to be sector
3884 * aligned;
3885 * - we read the log end (LR header start)
3886 * _first_, then the log start (LR header end)
3887 * - order is important.
3888 */
David Chinner234f56a2008-04-10 12:24:24 +10003889 wrapped_hblks = hblks - split_hblks;
Dave Chinner44396472011-04-21 09:34:27 +00003890 error = xlog_bread_offset(log, 0,
3891 wrapped_hblks, hbp,
3892 offset + BBTOB(split_hblks));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003893 if (error)
3894 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003895 }
3896 rhead = (xlog_rec_header_t *)offset;
3897 error = xlog_valid_rec_header(log, rhead,
3898 split_hblks ? blk_no : 0);
3899 if (error)
3900 goto bread_err2;
3901
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003902 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003903 blk_no += hblks;
3904
3905 /* Read in data for log record */
3906 if (blk_no + bblks <= log->l_logBBsize) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003907 error = xlog_bread(log, blk_no, bblks, dbp,
3908 &offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003909 if (error)
3910 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003911 } else {
3912 /* This log record is split across the
3913 * physical end of log */
Chandra Seetharaman62926042011-07-22 23:40:15 +00003914 offset = dbp->b_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003915 split_bblks = 0;
3916 if (blk_no != log->l_logBBsize) {
3917 /* some data is before the physical
3918 * end of log */
3919 ASSERT(!wrapped_hblks);
3920 ASSERT(blk_no <= INT_MAX);
3921 split_bblks =
3922 log->l_logBBsize - (int)blk_no;
3923 ASSERT(split_bblks > 0);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003924 error = xlog_bread(log, blk_no,
3925 split_bblks, dbp,
3926 &offset);
3927 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003928 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003929 }
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003930
Linus Torvalds1da177e2005-04-16 15:20:36 -07003931 /*
3932 * Note: this black magic still works with
3933 * large sector sizes (non-512) only because:
3934 * - we increased the buffer size originally
3935 * by 1 sector giving us enough extra space
3936 * for the second read;
3937 * - the log start is guaranteed to be sector
3938 * aligned;
3939 * - we read the log end (LR header start)
3940 * _first_, then the log start (LR header end)
3941 * - order is important.
3942 */
Dave Chinner44396472011-04-21 09:34:27 +00003943 error = xlog_bread_offset(log, 0,
Dave Chinner009507b2012-11-02 11:38:44 +11003944 bblks - split_bblks, dbp,
Dave Chinner44396472011-04-21 09:34:27 +00003945 offset + BBTOB(split_bblks));
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003946 if (error)
3947 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003948 }
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003949
3950 error = xlog_unpack_data(rhead, offset, log);
3951 if (error)
3952 goto bread_err2;
3953
3954 error = xlog_recover_process_data(log, rhash,
3955 rhead, offset, pass);
3956 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003957 goto bread_err2;
3958 blk_no += bblks;
3959 }
3960
3961 ASSERT(blk_no >= log->l_logBBsize);
3962 blk_no -= log->l_logBBsize;
3963
3964 /* read first part of physical log */
3965 while (blk_no < head_blk) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003966 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
3967 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003968 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003969
Linus Torvalds1da177e2005-04-16 15:20:36 -07003970 rhead = (xlog_rec_header_t *)offset;
3971 error = xlog_valid_rec_header(log, rhead, blk_no);
3972 if (error)
3973 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003974
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003975 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003976 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
3977 &offset);
3978 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003979 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003980
Christoph Hellwig0e446be2012-11-12 22:54:24 +11003981 error = xlog_unpack_data(rhead, offset, log);
3982 if (error)
3983 goto bread_err2;
3984
3985 error = xlog_recover_process_data(log, rhash,
3986 rhead, offset, pass);
3987 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003988 goto bread_err2;
3989 blk_no += bblks + hblks;
3990 }
3991 }
3992
3993 bread_err2:
3994 xlog_put_bp(dbp);
3995 bread_err1:
3996 xlog_put_bp(hbp);
3997 return error;
3998}
3999
4000/*
4001 * Do the recovery of the log. We actually do this in two phases.
4002 * The two passes are necessary in order to implement the function
4003 * of cancelling a record written into the log. The first pass
4004 * determines those things which have been cancelled, and the
4005 * second pass replays log items normally except for those which
4006 * have been cancelled. The handling of the replay and cancellations
4007 * takes place in the log item type specific routines.
4008 *
4009 * The table of items which have cancel records in the log is allocated
4010 * and freed at this level, since only here do we know when all of
4011 * the log recovery has been completed.
4012 */
4013STATIC int
4014xlog_do_log_recovery(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004015 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004016 xfs_daddr_t head_blk,
4017 xfs_daddr_t tail_blk)
4018{
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00004019 int error, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004020
4021 ASSERT(head_blk != tail_blk);
4022
4023 /*
4024 * First do a pass to find all of the cancelled buf log items.
4025 * Store them in the buf_cancel_table for use in the second pass.
4026 */
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00004027 log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE *
4028 sizeof(struct list_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07004029 KM_SLEEP);
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00004030 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
4031 INIT_LIST_HEAD(&log->l_buf_cancel_table[i]);
4032
Linus Torvalds1da177e2005-04-16 15:20:36 -07004033 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
4034 XLOG_RECOVER_PASS1);
4035 if (error != 0) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10004036 kmem_free(log->l_buf_cancel_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004037 log->l_buf_cancel_table = NULL;
4038 return error;
4039 }
4040 /*
4041 * Then do a second pass to actually recover the items in the log.
4042 * When it is complete free the table of buf cancel items.
4043 */
4044 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
4045 XLOG_RECOVER_PASS2);
4046#ifdef DEBUG
Tim Shimmin6d192a92006-06-09 14:55:38 +10004047 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004048 int i;
4049
4050 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00004051 ASSERT(list_empty(&log->l_buf_cancel_table[i]));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004052 }
4053#endif /* DEBUG */
4054
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10004055 kmem_free(log->l_buf_cancel_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004056 log->l_buf_cancel_table = NULL;
4057
4058 return error;
4059}
4060
4061/*
4062 * Do the actual recovery
4063 */
4064STATIC int
4065xlog_do_recover(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004066 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004067 xfs_daddr_t head_blk,
4068 xfs_daddr_t tail_blk)
4069{
4070 int error;
4071 xfs_buf_t *bp;
4072 xfs_sb_t *sbp;
4073
4074 /*
4075 * First replay the images in the log.
4076 */
4077 error = xlog_do_log_recovery(log, head_blk, tail_blk);
Christoph Hellwig43ff2122012-04-23 15:58:39 +10004078 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004079 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004080
4081 /*
4082 * If IO errors happened during recovery, bail out.
4083 */
4084 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
4085 return (EIO);
4086 }
4087
4088 /*
4089 * We now update the tail_lsn since much of the recovery has completed
4090 * and there may be space available to use. If there were no extent
4091 * or iunlinks, we can free up the entire log and set the tail_lsn to
4092 * be the last_sync_lsn. This was set in xlog_find_tail to be the
4093 * lsn of the last known good LR on disk. If there are extent frees
4094 * or iunlinks they will have some entries in the AIL; so we look at
4095 * the AIL to determine how to set the tail_lsn.
4096 */
4097 xlog_assign_tail_lsn(log->l_mp);
4098
4099 /*
4100 * Now that we've finished replaying all buffer and inode
Dave Chinner98021822012-11-12 22:54:03 +11004101 * updates, re-read in the superblock and reverify it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004102 */
4103 bp = xfs_getsb(log->l_mp, 0);
4104 XFS_BUF_UNDONE(bp);
Lachlan McIlroybebf9632007-10-15 13:18:02 +10004105 ASSERT(!(XFS_BUF_ISWRITE(bp)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004106 XFS_BUF_READ(bp);
Lachlan McIlroybebf9632007-10-15 13:18:02 +10004107 XFS_BUF_UNASYNC(bp);
Dave Chinner1813dd62012-11-14 17:54:40 +11004108 bp->b_ops = &xfs_sb_buf_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004109 xfsbdstrat(log->l_mp, bp);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +00004110 error = xfs_buf_iowait(bp);
David Chinnerd64e31a2008-04-10 12:22:17 +10004111 if (error) {
Christoph Hellwig901796a2011-10-10 16:52:49 +00004112 xfs_buf_ioerror_alert(bp, __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004113 ASSERT(0);
4114 xfs_buf_relse(bp);
4115 return error;
4116 }
4117
4118 /* Convert superblock from on-disk format */
4119 sbp = &log->l_mp->m_sb;
Dave Chinner98021822012-11-12 22:54:03 +11004120 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004121 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
Eric Sandeen62118702008-03-06 13:44:28 +11004122 ASSERT(xfs_sb_good_version(sbp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004123 xfs_buf_relse(bp);
4124
Lachlan McIlroy5478eea2007-02-10 18:36:29 +11004125 /* We've re-read the superblock so re-initialize per-cpu counters */
4126 xfs_icsb_reinit_counters(log->l_mp);
4127
Linus Torvalds1da177e2005-04-16 15:20:36 -07004128 xlog_recover_check_summary(log);
4129
4130 /* Normal transactions can now occur */
4131 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
4132 return 0;
4133}
4134
4135/*
4136 * Perform recovery and re-initialize some log variables in xlog_find_tail.
4137 *
4138 * Return error or zero.
4139 */
4140int
4141xlog_recover(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004142 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004143{
4144 xfs_daddr_t head_blk, tail_blk;
4145 int error;
4146
4147 /* find the tail of the log */
Eric Sandeen65be6052006-01-11 15:34:19 +11004148 if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004149 return error;
4150
4151 if (tail_blk != head_blk) {
4152 /* There used to be a comment here:
4153 *
4154 * disallow recovery on read-only mounts. note -- mount
4155 * checks for ENOSPC and turns it into an intelligent
4156 * error message.
4157 * ...but this is no longer true. Now, unless you specify
4158 * NORECOVERY (in which case this function would never be
4159 * called), we just go ahead and recover. We do this all
4160 * under the vfs layer, so we can get away with it unless
4161 * the device itself is read-only, in which case we fail.
4162 */
Utako Kusaka3a02ee12007-05-08 13:50:06 +10004163 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004164 return error;
4165 }
4166
Dave Chinnere721f502013-04-03 16:11:32 +11004167 /*
4168 * Version 5 superblock log feature mask validation. We know the
4169 * log is dirty so check if there are any unknown log features
4170 * in what we need to recover. If there are unknown features
4171 * (e.g. unsupported transactions, then simply reject the
4172 * attempt at recovery before touching anything.
4173 */
4174 if (XFS_SB_VERSION_NUM(&log->l_mp->m_sb) == XFS_SB_VERSION_5 &&
4175 xfs_sb_has_incompat_log_feature(&log->l_mp->m_sb,
4176 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
4177 xfs_warn(log->l_mp,
4178"Superblock has unknown incompatible log features (0x%x) enabled.\n"
4179"The log can not be fully and/or safely recovered by this kernel.\n"
4180"Please recover the log on a kernel that supports the unknown features.",
4181 (log->l_mp->m_sb.sb_features_log_incompat &
4182 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
4183 return EINVAL;
4184 }
4185
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004186 xfs_notice(log->l_mp, "Starting recovery (logdev: %s)",
4187 log->l_mp->m_logname ? log->l_mp->m_logname
4188 : "internal");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004189
4190 error = xlog_do_recover(log, head_blk, tail_blk);
4191 log->l_flags |= XLOG_RECOVERY_NEEDED;
4192 }
4193 return error;
4194}
4195
4196/*
4197 * In the first part of recovery we replay inodes and buffers and build
4198 * up the list of extent free items which need to be processed. Here
4199 * we process the extent free items and clean up the on disk unlinked
4200 * inode lists. This is separated from the first part of recovery so
4201 * that the root and real-time bitmap inodes can be read in from disk in
4202 * between the two stages. This is necessary so that we can free space
4203 * in the real-time portion of the file system.
4204 */
4205int
4206xlog_recover_finish(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004207 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004208{
4209 /*
4210 * Now we're ready to do the transactions needed for the
4211 * rest of recovery. Start with completing all the extent
4212 * free intent records and then process the unlinked inode
4213 * lists. At this point, we essentially run in normal mode
4214 * except that we're still performing recovery actions
4215 * rather than accepting new requests.
4216 */
4217 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
David Chinner3c1e2bb2008-04-10 12:21:11 +10004218 int error;
4219 error = xlog_recover_process_efis(log);
4220 if (error) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004221 xfs_alert(log->l_mp, "Failed to recover EFIs");
David Chinner3c1e2bb2008-04-10 12:21:11 +10004222 return error;
4223 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004224 /*
4225 * Sync the log to get all the EFIs out of the AIL.
4226 * This isn't absolutely necessary, but it helps in
4227 * case the unlink transactions would have problems
4228 * pushing the EFIs out of the way.
4229 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +00004230 xfs_log_force(log->l_mp, XFS_LOG_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004231
Christoph Hellwig42490232008-08-13 16:49:32 +10004232 xlog_recover_process_iunlinks(log);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004233
4234 xlog_recover_check_summary(log);
4235
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004236 xfs_notice(log->l_mp, "Ending recovery (logdev: %s)",
4237 log->l_mp->m_logname ? log->l_mp->m_logname
4238 : "internal");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004239 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
4240 } else {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004241 xfs_info(log->l_mp, "Ending clean mount");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004242 }
4243 return 0;
4244}
4245
4246
4247#if defined(DEBUG)
4248/*
4249 * Read all of the agf and agi counters and check that they
4250 * are consistent with the superblock counters.
4251 */
4252void
4253xlog_recover_check_summary(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004254 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004255{
4256 xfs_mount_t *mp;
4257 xfs_agf_t *agfp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004258 xfs_buf_t *agfbp;
4259 xfs_buf_t *agibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004260 xfs_agnumber_t agno;
4261 __uint64_t freeblks;
4262 __uint64_t itotal;
4263 __uint64_t ifree;
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11004264 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004265
4266 mp = log->l_mp;
4267
4268 freeblks = 0LL;
4269 itotal = 0LL;
4270 ifree = 0LL;
4271 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
From: Christoph Hellwig48056212008-11-28 14:23:38 +11004272 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
4273 if (error) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004274 xfs_alert(mp, "%s agf read failed agno %d error %d",
4275 __func__, agno, error);
From: Christoph Hellwig48056212008-11-28 14:23:38 +11004276 } else {
4277 agfp = XFS_BUF_TO_AGF(agfbp);
4278 freeblks += be32_to_cpu(agfp->agf_freeblks) +
4279 be32_to_cpu(agfp->agf_flcount);
4280 xfs_buf_relse(agfbp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004281 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004282
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11004283 error = xfs_read_agi(mp, NULL, agno, &agibp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004284 if (error) {
4285 xfs_alert(mp, "%s agi read failed agno %d error %d",
4286 __func__, agno, error);
4287 } else {
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11004288 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004289
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11004290 itotal += be32_to_cpu(agi->agi_count);
4291 ifree += be32_to_cpu(agi->agi_freecount);
4292 xfs_buf_relse(agibp);
4293 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004294 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004295}
4296#endif /* DEBUG */