blob: 512a0945d52ac4e023e2181510e70e65f3da04e8 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Tim Shimmin87c199c2006-06-09 14:56:16 +10002 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
Nathan Scott7b718762005-11-02 14:58:39 +11003 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110019#include "xfs_fs.h"
Dave Chinner70a9883c2013-10-23 10:36:05 +110020#include "xfs_shared.h"
Dave Chinner239880e2013-10-23 10:50:10 +110021#include "xfs_format.h"
22#include "xfs_log_format.h"
23#include "xfs_trans_resv.h"
Nathan Scotta844f452005-11-02 14:38:42 +110024#include "xfs_bit.h"
Nathan Scotta844f452005-11-02 14:38:42 +110025#include "xfs_sb.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include "xfs_mount.h"
Dave Chinner57062782013-10-15 09:17:51 +110027#include "xfs_da_format.h"
Dave Chinner9a2cc412014-12-04 09:43:17 +110028#include "xfs_da_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include "xfs_inode.h"
Dave Chinner239880e2013-10-23 10:50:10 +110030#include "xfs_trans.h"
Dave Chinner239880e2013-10-23 10:50:10 +110031#include "xfs_log.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include "xfs_log_priv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_log_recover.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110034#include "xfs_inode_item.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include "xfs_extfree_item.h"
36#include "xfs_trans_priv.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110037#include "xfs_alloc.h"
38#include "xfs_ialloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include "xfs_quota.h"
Christoph Hellwig0e446be2012-11-12 22:54:24 +110040#include "xfs_cksum.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000041#include "xfs_trace.h"
Dave Chinner33479e02012-10-08 21:56:11 +110042#include "xfs_icache.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110043#include "xfs_bmap_btree.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110044#include "xfs_error.h"
Dave Chinner2b9ab5a2013-08-12 20:49:37 +100045#include "xfs_dir2.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
Dave Chinnerfc06c6d2013-08-12 20:49:22 +100047#define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
48
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -050049STATIC int
50xlog_find_zeroed(
51 struct xlog *,
52 xfs_daddr_t *);
53STATIC int
54xlog_clear_stale_blocks(
55 struct xlog *,
56 xfs_lsn_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -070057#if defined(DEBUG)
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -050058STATIC void
59xlog_recover_check_summary(
60 struct xlog *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070061#else
62#define xlog_recover_check_summary(log)
Linus Torvalds1da177e2005-04-16 15:20:36 -070063#endif
64
Linus Torvalds1da177e2005-04-16 15:20:36 -070065/*
Christoph Hellwigd5689ea2010-12-01 22:06:22 +000066 * This structure is used during recovery to record the buf log items which
67 * have been canceled and should not be replayed.
68 */
69struct xfs_buf_cancel {
70 xfs_daddr_t bc_blkno;
71 uint bc_len;
72 int bc_refcount;
73 struct list_head bc_list;
74};
75
76/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 * Sector aligned buffer routines for buffer create/read/write/access
78 */
79
Alex Elderff30a622010-04-13 15:22:58 +100080/*
81 * Verify the given count of basic blocks is valid number of blocks
82 * to specify for an operation involving the given XFS log buffer.
83 * Returns nonzero if the count is valid, 0 otherwise.
84 */
85
86static inline int
87xlog_buf_bbcount_valid(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -050088 struct xlog *log,
Alex Elderff30a622010-04-13 15:22:58 +100089 int bbcount)
90{
91 return bbcount > 0 && bbcount <= log->l_logBBsize;
92}
93
Alex Elder36adecf2010-04-13 15:21:13 +100094/*
95 * Allocate a buffer to hold log data. The buffer needs to be able
96 * to map to a range of nbblks basic blocks at any valid (basic
97 * block) offset within the log.
98 */
Eric Sandeen5d77c0d2009-11-19 15:52:00 +000099STATIC xfs_buf_t *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100xlog_get_bp(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500101 struct xlog *log,
Dave Chinner32281492009-01-22 15:37:47 +1100102 int nbblks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103{
Christoph Hellwigc8da0fa2011-07-08 14:36:25 +0200104 struct xfs_buf *bp;
105
Alex Elderff30a622010-04-13 15:22:58 +1000106 if (!xlog_buf_bbcount_valid(log, nbblks)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100107 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
Alex Elderff30a622010-04-13 15:22:58 +1000108 nbblks);
109 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
Dave Chinner32281492009-01-22 15:37:47 +1100110 return NULL;
111 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112
Alex Elder36adecf2010-04-13 15:21:13 +1000113 /*
114 * We do log I/O in units of log sectors (a power-of-2
115 * multiple of the basic block size), so we round up the
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300116 * requested size to accommodate the basic blocks required
Alex Elder36adecf2010-04-13 15:21:13 +1000117 * for complete log sectors.
118 *
119 * In addition, the buffer may be used for a non-sector-
120 * aligned block offset, in which case an I/O of the
121 * requested size could extend beyond the end of the
122 * buffer. If the requested size is only 1 basic block it
123 * will never straddle a sector boundary, so this won't be
124 * an issue. Nor will this be a problem if the log I/O is
125 * done in basic blocks (sector size 1). But otherwise we
126 * extend the buffer by one extra log sector to ensure
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300127 * there's space to accommodate this possibility.
Alex Elder36adecf2010-04-13 15:21:13 +1000128 */
Alex Elder69ce58f2010-04-20 17:09:59 +1000129 if (nbblks > 1 && log->l_sectBBsize > 1)
130 nbblks += log->l_sectBBsize;
131 nbblks = round_up(nbblks, log->l_sectBBsize);
Alex Elder36adecf2010-04-13 15:21:13 +1000132
Dave Chinnere70b73f2012-04-23 15:58:49 +1000133 bp = xfs_buf_get_uncached(log->l_mp->m_logdev_targp, nbblks, 0);
Christoph Hellwigc8da0fa2011-07-08 14:36:25 +0200134 if (bp)
135 xfs_buf_unlock(bp);
136 return bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137}
138
Eric Sandeen5d77c0d2009-11-19 15:52:00 +0000139STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140xlog_put_bp(
141 xfs_buf_t *bp)
142{
143 xfs_buf_free(bp);
144}
145
Alex Elder48389ef2010-04-20 17:10:21 +1000146/*
147 * Return the address of the start of the given block number's data
148 * in a log buffer. The buffer covers a log sector-aligned region.
149 */
Christoph Hellwigb2a922c2015-06-22 09:45:10 +1000150STATIC char *
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100151xlog_align(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500152 struct xlog *log,
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100153 xfs_daddr_t blk_no,
154 int nbblks,
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500155 struct xfs_buf *bp)
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100156{
Christoph Hellwigfdc07f42010-05-10 17:28:14 +0000157 xfs_daddr_t offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100158
Dave Chinner4e94b712012-04-23 15:58:51 +1000159 ASSERT(offset + nbblks <= bp->b_length);
Chandra Seetharaman62926042011-07-22 23:40:15 +0000160 return bp->b_addr + BBTOB(offset);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100161}
162
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163
164/*
165 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
166 */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100167STATIC int
168xlog_bread_noalign(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500169 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 xfs_daddr_t blk_no,
171 int nbblks,
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500172 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173{
174 int error;
175
Alex Elderff30a622010-04-13 15:22:58 +1000176 if (!xlog_buf_bbcount_valid(log, nbblks)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100177 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
Alex Elderff30a622010-04-13 15:22:58 +1000178 nbblks);
179 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
Dave Chinner24513372014-06-25 14:58:08 +1000180 return -EFSCORRUPTED;
Dave Chinner32281492009-01-22 15:37:47 +1100181 }
182
Alex Elder69ce58f2010-04-20 17:09:59 +1000183 blk_no = round_down(blk_no, log->l_sectBBsize);
184 nbblks = round_up(nbblks, log->l_sectBBsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185
186 ASSERT(nbblks > 0);
Dave Chinner4e94b712012-04-23 15:58:51 +1000187 ASSERT(nbblks <= bp->b_length);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
189 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
190 XFS_BUF_READ(bp);
Dave Chinneraa0e8832012-04-23 15:58:52 +1000191 bp->b_io_length = nbblks;
Dave Chinner0e95f192012-04-23 15:58:46 +1000192 bp->b_error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193
Dave Chinner595bff72014-10-02 09:05:14 +1000194 error = xfs_buf_submit_wait(bp);
195 if (error && !XFS_FORCED_SHUTDOWN(log->l_mp))
Christoph Hellwig901796a2011-10-10 16:52:49 +0000196 xfs_buf_ioerror_alert(bp, __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 return error;
198}
199
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100200STATIC int
201xlog_bread(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500202 struct xlog *log,
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100203 xfs_daddr_t blk_no,
204 int nbblks,
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500205 struct xfs_buf *bp,
Christoph Hellwigb2a922c2015-06-22 09:45:10 +1000206 char **offset)
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100207{
208 int error;
209
210 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
211 if (error)
212 return error;
213
214 *offset = xlog_align(log, blk_no, nbblks, bp);
215 return 0;
216}
217
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218/*
Dave Chinner44396472011-04-21 09:34:27 +0000219 * Read at an offset into the buffer. Returns with the buffer in it's original
220 * state regardless of the result of the read.
221 */
222STATIC int
223xlog_bread_offset(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500224 struct xlog *log,
Dave Chinner44396472011-04-21 09:34:27 +0000225 xfs_daddr_t blk_no, /* block to read from */
226 int nbblks, /* blocks to read */
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500227 struct xfs_buf *bp,
Christoph Hellwigb2a922c2015-06-22 09:45:10 +1000228 char *offset)
Dave Chinner44396472011-04-21 09:34:27 +0000229{
Christoph Hellwigb2a922c2015-06-22 09:45:10 +1000230 char *orig_offset = bp->b_addr;
Dave Chinner4e94b712012-04-23 15:58:51 +1000231 int orig_len = BBTOB(bp->b_length);
Dave Chinner44396472011-04-21 09:34:27 +0000232 int error, error2;
233
Chandra Seetharaman02fe03d2011-07-22 23:40:22 +0000234 error = xfs_buf_associate_memory(bp, offset, BBTOB(nbblks));
Dave Chinner44396472011-04-21 09:34:27 +0000235 if (error)
236 return error;
237
238 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
239
240 /* must reset buffer pointer even on error */
Chandra Seetharaman02fe03d2011-07-22 23:40:22 +0000241 error2 = xfs_buf_associate_memory(bp, orig_offset, orig_len);
Dave Chinner44396472011-04-21 09:34:27 +0000242 if (error)
243 return error;
244 return error2;
245}
246
247/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 * Write out the buffer at the given block for the given number of blocks.
249 * The buffer is kept locked across the write and is returned locked.
250 * This can only be used for synchronous log writes.
251 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000252STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253xlog_bwrite(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500254 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 xfs_daddr_t blk_no,
256 int nbblks,
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500257 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258{
259 int error;
260
Alex Elderff30a622010-04-13 15:22:58 +1000261 if (!xlog_buf_bbcount_valid(log, nbblks)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100262 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
Alex Elderff30a622010-04-13 15:22:58 +1000263 nbblks);
264 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
Dave Chinner24513372014-06-25 14:58:08 +1000265 return -EFSCORRUPTED;
Dave Chinner32281492009-01-22 15:37:47 +1100266 }
267
Alex Elder69ce58f2010-04-20 17:09:59 +1000268 blk_no = round_down(blk_no, log->l_sectBBsize);
269 nbblks = round_up(nbblks, log->l_sectBBsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270
271 ASSERT(nbblks > 0);
Dave Chinner4e94b712012-04-23 15:58:51 +1000272 ASSERT(nbblks <= bp->b_length);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273
274 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
275 XFS_BUF_ZEROFLAGS(bp);
Chandra Seetharaman72790aa2011-07-22 23:40:04 +0000276 xfs_buf_hold(bp);
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200277 xfs_buf_lock(bp);
Dave Chinneraa0e8832012-04-23 15:58:52 +1000278 bp->b_io_length = nbblks;
Dave Chinner0e95f192012-04-23 15:58:46 +1000279 bp->b_error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280
Christoph Hellwigc2b006c2011-08-23 08:28:07 +0000281 error = xfs_bwrite(bp);
Christoph Hellwig901796a2011-10-10 16:52:49 +0000282 if (error)
283 xfs_buf_ioerror_alert(bp, __func__);
Christoph Hellwigc2b006c2011-08-23 08:28:07 +0000284 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 return error;
286}
287
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288#ifdef DEBUG
289/*
290 * dump debug superblock and log record information
291 */
292STATIC void
293xlog_header_check_dump(
294 xfs_mount_t *mp,
295 xlog_rec_header_t *head)
296{
Eric Sandeen08e96e12013-10-11 20:59:05 -0500297 xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d",
Joe Perches03daa572009-12-14 18:01:10 -0800298 __func__, &mp->m_sb.sb_uuid, XLOG_FMT);
Eric Sandeen08e96e12013-10-11 20:59:05 -0500299 xfs_debug(mp, " log : uuid = %pU, fmt = %d",
Joe Perches03daa572009-12-14 18:01:10 -0800300 &head->h_fs_uuid, be32_to_cpu(head->h_fmt));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301}
302#else
303#define xlog_header_check_dump(mp, head)
304#endif
305
306/*
307 * check log record header for recovery
308 */
309STATIC int
310xlog_header_check_recover(
311 xfs_mount_t *mp,
312 xlog_rec_header_t *head)
313{
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200314 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315
316 /*
317 * IRIX doesn't write the h_fmt field and leaves it zeroed
318 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
319 * a dirty log created in IRIX.
320 */
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200321 if (unlikely(head->h_fmt != cpu_to_be32(XLOG_FMT))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100322 xfs_warn(mp,
323 "dirty log written in incompatible format - can't recover");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 xlog_header_check_dump(mp, head);
325 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
326 XFS_ERRLEVEL_HIGH, mp);
Dave Chinner24513372014-06-25 14:58:08 +1000327 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100329 xfs_warn(mp,
330 "dirty log entry has mismatched uuid - can't recover");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 xlog_header_check_dump(mp, head);
332 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
333 XFS_ERRLEVEL_HIGH, mp);
Dave Chinner24513372014-06-25 14:58:08 +1000334 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 }
336 return 0;
337}
338
339/*
340 * read the head block of the log and check the header
341 */
342STATIC int
343xlog_header_check_mount(
344 xfs_mount_t *mp,
345 xlog_rec_header_t *head)
346{
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200347 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348
349 if (uuid_is_nil(&head->h_fs_uuid)) {
350 /*
351 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
352 * h_fs_uuid is nil, we assume this log was last mounted
353 * by IRIX and continue.
354 */
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100355 xfs_warn(mp, "nil uuid in log - IRIX style log");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100357 xfs_warn(mp, "log has mismatched uuid - can't recover");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 xlog_header_check_dump(mp, head);
359 XFS_ERROR_REPORT("xlog_header_check_mount",
360 XFS_ERRLEVEL_HIGH, mp);
Dave Chinner24513372014-06-25 14:58:08 +1000361 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 }
363 return 0;
364}
365
366STATIC void
367xlog_recover_iodone(
368 struct xfs_buf *bp)
369{
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +0000370 if (bp->b_error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 /*
372 * We're not going to bother about retrying
373 * this during recovery. One strike!
374 */
Dave Chinner595bff72014-10-02 09:05:14 +1000375 if (!XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) {
376 xfs_buf_ioerror_alert(bp, __func__);
377 xfs_force_shutdown(bp->b_target->bt_mount,
378 SHUTDOWN_META_IO_ERROR);
379 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 }
Christoph Hellwigcb669ca2011-07-13 13:43:49 +0200381 bp->b_iodone = NULL;
Dave Chinnere8aaba92014-10-02 09:04:22 +1000382 xfs_buf_ioend(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383}
384
385/*
386 * This routine finds (to an approximation) the first block in the physical
387 * log which contains the given cycle. It uses a binary search algorithm.
388 * Note that the algorithm can not be perfect because the disk will not
389 * necessarily be perfect.
390 */
David Chinnera8272ce2007-11-23 16:28:09 +1100391STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392xlog_find_cycle_start(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500393 struct xlog *log,
394 struct xfs_buf *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 xfs_daddr_t first_blk,
396 xfs_daddr_t *last_blk,
397 uint cycle)
398{
Christoph Hellwigb2a922c2015-06-22 09:45:10 +1000399 char *offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 xfs_daddr_t mid_blk;
Alex Eldere3bb2e32010-04-15 18:17:30 +0000401 xfs_daddr_t end_blk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 uint mid_cycle;
403 int error;
404
Alex Eldere3bb2e32010-04-15 18:17:30 +0000405 end_blk = *last_blk;
406 mid_blk = BLK_AVG(first_blk, end_blk);
407 while (mid_blk != first_blk && mid_blk != end_blk) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100408 error = xlog_bread(log, mid_blk, 1, bp, &offset);
409 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 return error;
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000411 mid_cycle = xlog_get_cycle(offset);
Alex Eldere3bb2e32010-04-15 18:17:30 +0000412 if (mid_cycle == cycle)
413 end_blk = mid_blk; /* last_half_cycle == mid_cycle */
414 else
415 first_blk = mid_blk; /* first_half_cycle == mid_cycle */
416 mid_blk = BLK_AVG(first_blk, end_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 }
Alex Eldere3bb2e32010-04-15 18:17:30 +0000418 ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) ||
419 (mid_blk == end_blk && mid_blk-1 == first_blk));
420
421 *last_blk = end_blk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422
423 return 0;
424}
425
426/*
Alex Elder3f943d82010-04-15 18:17:34 +0000427 * Check that a range of blocks does not contain stop_on_cycle_no.
428 * Fill in *new_blk with the block offset where such a block is
429 * found, or with -1 (an invalid block number) if there is no such
430 * block in the range. The scan needs to occur from front to back
431 * and the pointer into the region must be updated since a later
432 * routine will need to perform another test.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 */
434STATIC int
435xlog_find_verify_cycle(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500436 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 xfs_daddr_t start_blk,
438 int nbblks,
439 uint stop_on_cycle_no,
440 xfs_daddr_t *new_blk)
441{
442 xfs_daddr_t i, j;
443 uint cycle;
444 xfs_buf_t *bp;
445 xfs_daddr_t bufblks;
Christoph Hellwigb2a922c2015-06-22 09:45:10 +1000446 char *buf = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 int error = 0;
448
Alex Elder6881a222010-04-13 15:22:29 +1000449 /*
450 * Greedily allocate a buffer big enough to handle the full
451 * range of basic blocks we'll be examining. If that fails,
452 * try a smaller size. We need to be able to read at least
453 * a log sector, or we're out of luck.
454 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 bufblks = 1 << ffs(nbblks);
Dave Chinner81158e02012-04-27 19:45:22 +1000456 while (bufblks > log->l_logBBsize)
457 bufblks >>= 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 while (!(bp = xlog_get_bp(log, bufblks))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 bufblks >>= 1;
Alex Elder69ce58f2010-04-20 17:09:59 +1000460 if (bufblks < log->l_sectBBsize)
Dave Chinner24513372014-06-25 14:58:08 +1000461 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 }
463
464 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
465 int bcount;
466
467 bcount = min(bufblks, (start_blk + nbblks - i));
468
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100469 error = xlog_bread(log, i, bcount, bp, &buf);
470 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 goto out;
472
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 for (j = 0; j < bcount; j++) {
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000474 cycle = xlog_get_cycle(buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 if (cycle == stop_on_cycle_no) {
476 *new_blk = i+j;
477 goto out;
478 }
479
480 buf += BBSIZE;
481 }
482 }
483
484 *new_blk = -1;
485
486out:
487 xlog_put_bp(bp);
488 return error;
489}
490
491/*
492 * Potentially backup over partial log record write.
493 *
494 * In the typical case, last_blk is the number of the block directly after
495 * a good log record. Therefore, we subtract one to get the block number
496 * of the last block in the given buffer. extra_bblks contains the number
497 * of blocks we would have read on a previous read. This happens when the
498 * last log record is split over the end of the physical log.
499 *
500 * extra_bblks is the number of blocks potentially verified on a previous
501 * call to this routine.
502 */
503STATIC int
504xlog_find_verify_log_record(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500505 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 xfs_daddr_t start_blk,
507 xfs_daddr_t *last_blk,
508 int extra_bblks)
509{
510 xfs_daddr_t i;
511 xfs_buf_t *bp;
Christoph Hellwigb2a922c2015-06-22 09:45:10 +1000512 char *offset = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 xlog_rec_header_t *head = NULL;
514 int error = 0;
515 int smallmem = 0;
516 int num_blks = *last_blk - start_blk;
517 int xhdrs;
518
519 ASSERT(start_blk != 0 || *last_blk != start_blk);
520
521 if (!(bp = xlog_get_bp(log, num_blks))) {
522 if (!(bp = xlog_get_bp(log, 1)))
Dave Chinner24513372014-06-25 14:58:08 +1000523 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 smallmem = 1;
525 } else {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100526 error = xlog_bread(log, start_blk, num_blks, bp, &offset);
527 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 offset += ((num_blks - 1) << BBSHIFT);
530 }
531
532 for (i = (*last_blk) - 1; i >= 0; i--) {
533 if (i < start_blk) {
534 /* valid log record not found */
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100535 xfs_warn(log->l_mp,
536 "Log inconsistent (didn't find previous header)");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 ASSERT(0);
Dave Chinner24513372014-06-25 14:58:08 +1000538 error = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 goto out;
540 }
541
542 if (smallmem) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100543 error = xlog_bread(log, i, 1, bp, &offset);
544 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 }
547
548 head = (xlog_rec_header_t *)offset;
549
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200550 if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 break;
552
553 if (!smallmem)
554 offset -= BBSIZE;
555 }
556
557 /*
558 * We hit the beginning of the physical log & still no header. Return
559 * to caller. If caller can handle a return of -1, then this routine
560 * will be called again for the end of the physical log.
561 */
562 if (i == -1) {
Dave Chinner24513372014-06-25 14:58:08 +1000563 error = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 goto out;
565 }
566
567 /*
568 * We have the final block of the good log (the first block
569 * of the log record _before_ the head. So we check the uuid.
570 */
571 if ((error = xlog_header_check_mount(log->l_mp, head)))
572 goto out;
573
574 /*
575 * We may have found a log record header before we expected one.
576 * last_blk will be the 1st block # with a given cycle #. We may end
577 * up reading an entire log record. In this case, we don't want to
578 * reset last_blk. Only when last_blk points in the middle of a log
579 * record do we update last_blk.
580 */
Eric Sandeen62118702008-03-06 13:44:28 +1100581 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000582 uint h_size = be32_to_cpu(head->h_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583
584 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
585 if (h_size % XLOG_HEADER_CYCLE_SIZE)
586 xhdrs++;
587 } else {
588 xhdrs = 1;
589 }
590
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000591 if (*last_blk - i + extra_bblks !=
592 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 *last_blk = i;
594
595out:
596 xlog_put_bp(bp);
597 return error;
598}
599
600/*
601 * Head is defined to be the point of the log where the next log write
Zhi Yong Wu0a94da22013-08-07 10:11:08 +0000602 * could go. This means that incomplete LR writes at the end are
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 * eliminated when calculating the head. We aren't guaranteed that previous
604 * LR have complete transactions. We only know that a cycle number of
605 * current cycle number -1 won't be present in the log if we start writing
606 * from our current block number.
607 *
608 * last_blk contains the block number of the first block with a given
609 * cycle number.
610 *
611 * Return: zero if normal, non-zero if error.
612 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000613STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614xlog_find_head(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500615 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 xfs_daddr_t *return_head_blk)
617{
618 xfs_buf_t *bp;
Christoph Hellwigb2a922c2015-06-22 09:45:10 +1000619 char *offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
621 int num_scan_bblks;
622 uint first_half_cycle, last_half_cycle;
623 uint stop_on_cycle;
624 int error, log_bbnum = log->l_logBBsize;
625
626 /* Is the end of the log device zeroed? */
Dave Chinner24513372014-06-25 14:58:08 +1000627 error = xlog_find_zeroed(log, &first_blk);
628 if (error < 0) {
629 xfs_warn(log->l_mp, "empty log check failed");
630 return error;
631 }
632 if (error == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 *return_head_blk = first_blk;
634
635 /* Is the whole lot zeroed? */
636 if (!first_blk) {
637 /* Linux XFS shouldn't generate totally zeroed logs -
638 * mkfs etc write a dummy unmount record to a fresh
639 * log so we can store the uuid in there
640 */
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100641 xfs_warn(log->l_mp, "totally zeroed log");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 }
643
644 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 }
646
647 first_blk = 0; /* get cycle # of 1st block */
648 bp = xlog_get_bp(log, 1);
649 if (!bp)
Dave Chinner24513372014-06-25 14:58:08 +1000650 return -ENOMEM;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100651
652 error = xlog_bread(log, 0, 1, bp, &offset);
653 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 goto bp_err;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100655
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000656 first_half_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657
658 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100659 error = xlog_bread(log, last_blk, 1, bp, &offset);
660 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 goto bp_err;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100662
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000663 last_half_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 ASSERT(last_half_cycle != 0);
665
666 /*
667 * If the 1st half cycle number is equal to the last half cycle number,
668 * then the entire log is stamped with the same cycle number. In this
669 * case, head_blk can't be set to zero (which makes sense). The below
670 * math doesn't work out properly with head_blk equal to zero. Instead,
671 * we set it to log_bbnum which is an invalid block number, but this
672 * value makes the math correct. If head_blk doesn't changed through
673 * all the tests below, *head_blk is set to zero at the very end rather
674 * than log_bbnum. In a sense, log_bbnum and zero are the same block
675 * in a circular file.
676 */
677 if (first_half_cycle == last_half_cycle) {
678 /*
679 * In this case we believe that the entire log should have
680 * cycle number last_half_cycle. We need to scan backwards
681 * from the end verifying that there are no holes still
682 * containing last_half_cycle - 1. If we find such a hole,
683 * then the start of that hole will be the new head. The
684 * simple case looks like
685 * x | x ... | x - 1 | x
686 * Another case that fits this picture would be
687 * x | x + 1 | x ... | x
Nathan Scottc41564b2006-03-29 08:55:14 +1000688 * In this case the head really is somewhere at the end of the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 * log, as one of the latest writes at the beginning was
690 * incomplete.
691 * One more case is
692 * x | x + 1 | x ... | x - 1 | x
693 * This is really the combination of the above two cases, and
694 * the head has to end up at the start of the x-1 hole at the
695 * end of the log.
696 *
697 * In the 256k log case, we will read from the beginning to the
698 * end of the log and search for cycle numbers equal to x-1.
699 * We don't worry about the x+1 blocks that we encounter,
700 * because we know that they cannot be the head since the log
701 * started with x.
702 */
703 head_blk = log_bbnum;
704 stop_on_cycle = last_half_cycle - 1;
705 } else {
706 /*
707 * In this case we want to find the first block with cycle
708 * number matching last_half_cycle. We expect the log to be
709 * some variation on
Alex Elder3f943d82010-04-15 18:17:34 +0000710 * x + 1 ... | x ... | x
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 * The first block with cycle number x (last_half_cycle) will
712 * be where the new head belongs. First we do a binary search
713 * for the first occurrence of last_half_cycle. The binary
714 * search may not be totally accurate, so then we scan back
715 * from there looking for occurrences of last_half_cycle before
716 * us. If that backwards scan wraps around the beginning of
717 * the log, then we look for occurrences of last_half_cycle - 1
718 * at the end of the log. The cases we're looking for look
719 * like
Alex Elder3f943d82010-04-15 18:17:34 +0000720 * v binary search stopped here
721 * x + 1 ... | x | x + 1 | x ... | x
722 * ^ but we want to locate this spot
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 * or
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 * <---------> less than scan distance
Alex Elder3f943d82010-04-15 18:17:34 +0000725 * x + 1 ... | x ... | x - 1 | x
726 * ^ we want to locate this spot
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 */
728 stop_on_cycle = last_half_cycle;
729 if ((error = xlog_find_cycle_start(log, bp, first_blk,
730 &head_blk, last_half_cycle)))
731 goto bp_err;
732 }
733
734 /*
735 * Now validate the answer. Scan back some number of maximum possible
736 * blocks and make sure each one has the expected cycle number. The
737 * maximum is determined by the total possible amount of buffering
738 * in the in-core log. The following number can be made tighter if
739 * we actually look at the block size of the filesystem.
740 */
741 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
742 if (head_blk >= num_scan_bblks) {
743 /*
744 * We are guaranteed that the entire check can be performed
745 * in one buffer.
746 */
747 start_blk = head_blk - num_scan_bblks;
748 if ((error = xlog_find_verify_cycle(log,
749 start_blk, num_scan_bblks,
750 stop_on_cycle, &new_blk)))
751 goto bp_err;
752 if (new_blk != -1)
753 head_blk = new_blk;
754 } else { /* need to read 2 parts of log */
755 /*
756 * We are going to scan backwards in the log in two parts.
757 * First we scan the physical end of the log. In this part
758 * of the log, we are looking for blocks with cycle number
759 * last_half_cycle - 1.
760 * If we find one, then we know that the log starts there, as
761 * we've found a hole that didn't get written in going around
762 * the end of the physical log. The simple case for this is
763 * x + 1 ... | x ... | x - 1 | x
764 * <---------> less than scan distance
765 * If all of the blocks at the end of the log have cycle number
766 * last_half_cycle, then we check the blocks at the start of
767 * the log looking for occurrences of last_half_cycle. If we
768 * find one, then our current estimate for the location of the
769 * first occurrence of last_half_cycle is wrong and we move
770 * back to the hole we've found. This case looks like
771 * x + 1 ... | x | x + 1 | x ...
772 * ^ binary search stopped here
773 * Another case we need to handle that only occurs in 256k
774 * logs is
775 * x + 1 ... | x ... | x+1 | x ...
776 * ^ binary search stops here
777 * In a 256k log, the scan at the end of the log will see the
778 * x + 1 blocks. We need to skip past those since that is
779 * certainly not the head of the log. By searching for
780 * last_half_cycle-1 we accomplish that.
781 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 ASSERT(head_blk <= INT_MAX &&
Alex Elder3f943d82010-04-15 18:17:34 +0000783 (xfs_daddr_t) num_scan_bblks >= head_blk);
784 start_blk = log_bbnum - (num_scan_bblks - head_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 if ((error = xlog_find_verify_cycle(log, start_blk,
786 num_scan_bblks - (int)head_blk,
787 (stop_on_cycle - 1), &new_blk)))
788 goto bp_err;
789 if (new_blk != -1) {
790 head_blk = new_blk;
Alex Elder9db127e2010-04-15 18:17:26 +0000791 goto validate_head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 }
793
794 /*
795 * Scan beginning of log now. The last part of the physical
796 * log is good. This scan needs to verify that it doesn't find
797 * the last_half_cycle.
798 */
799 start_blk = 0;
800 ASSERT(head_blk <= INT_MAX);
801 if ((error = xlog_find_verify_cycle(log,
802 start_blk, (int)head_blk,
803 stop_on_cycle, &new_blk)))
804 goto bp_err;
805 if (new_blk != -1)
806 head_blk = new_blk;
807 }
808
Alex Elder9db127e2010-04-15 18:17:26 +0000809validate_head:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 /*
811 * Now we need to make sure head_blk is not pointing to a block in
812 * the middle of a log record.
813 */
814 num_scan_bblks = XLOG_REC_SHIFT(log);
815 if (head_blk >= num_scan_bblks) {
816 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
817
818 /* start ptr at last block ptr before head_blk */
Dave Chinner24513372014-06-25 14:58:08 +1000819 error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0);
820 if (error == 1)
821 error = -EIO;
822 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 goto bp_err;
824 } else {
825 start_blk = 0;
826 ASSERT(head_blk <= INT_MAX);
Dave Chinner24513372014-06-25 14:58:08 +1000827 error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0);
828 if (error < 0)
829 goto bp_err;
830 if (error == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 /* We hit the beginning of the log during our search */
Alex Elder3f943d82010-04-15 18:17:34 +0000832 start_blk = log_bbnum - (num_scan_bblks - head_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 new_blk = log_bbnum;
834 ASSERT(start_blk <= INT_MAX &&
835 (xfs_daddr_t) log_bbnum-start_blk >= 0);
836 ASSERT(head_blk <= INT_MAX);
Dave Chinner24513372014-06-25 14:58:08 +1000837 error = xlog_find_verify_log_record(log, start_blk,
838 &new_blk, (int)head_blk);
839 if (error == 1)
840 error = -EIO;
841 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 goto bp_err;
843 if (new_blk != log_bbnum)
844 head_blk = new_blk;
845 } else if (error)
846 goto bp_err;
847 }
848
849 xlog_put_bp(bp);
850 if (head_blk == log_bbnum)
851 *return_head_blk = 0;
852 else
853 *return_head_blk = head_blk;
854 /*
855 * When returning here, we have a good block number. Bad block
856 * means that during a previous crash, we didn't have a clean break
857 * from cycle number N to cycle number N-1. In this case, we need
858 * to find the first block with cycle number N-1.
859 */
860 return 0;
861
862 bp_err:
863 xlog_put_bp(bp);
864
865 if (error)
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100866 xfs_warn(log->l_mp, "failed to find log head");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 return error;
868}
869
870/*
871 * Find the sync block number or the tail of the log.
872 *
873 * This will be the block number of the last record to have its
874 * associated buffers synced to disk. Every log record header has
875 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
876 * to get a sync block number. The only concern is to figure out which
877 * log record header to believe.
878 *
879 * The following algorithm uses the log record header with the largest
880 * lsn. The entire log record does not need to be valid. We only care
881 * that the header is valid.
882 *
883 * We could speed up search by using current head_blk buffer, but it is not
884 * available.
885 */
Eric Sandeen5d77c0d2009-11-19 15:52:00 +0000886STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887xlog_find_tail(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500888 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 xfs_daddr_t *head_blk,
Eric Sandeen65be6052006-01-11 15:34:19 +1100890 xfs_daddr_t *tail_blk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891{
892 xlog_rec_header_t *rhead;
893 xlog_op_header_t *op_head;
Christoph Hellwigb2a922c2015-06-22 09:45:10 +1000894 char *offset = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 xfs_buf_t *bp;
896 int error, i, found;
897 xfs_daddr_t umount_data_blk;
898 xfs_daddr_t after_umount_blk;
899 xfs_lsn_t tail_lsn;
900 int hblks;
901
902 found = 0;
903
904 /*
905 * Find previous log record
906 */
907 if ((error = xlog_find_head(log, head_blk)))
908 return error;
909
910 bp = xlog_get_bp(log, 1);
911 if (!bp)
Dave Chinner24513372014-06-25 14:58:08 +1000912 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 if (*head_blk == 0) { /* special case */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100914 error = xlog_bread(log, 0, 1, bp, &offset);
915 if (error)
Alex Elder9db127e2010-04-15 18:17:26 +0000916 goto done;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100917
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000918 if (xlog_get_cycle(offset) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 *tail_blk = 0;
920 /* leave all other log inited values alone */
Alex Elder9db127e2010-04-15 18:17:26 +0000921 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 }
923 }
924
925 /*
926 * Search backwards looking for log record header block
927 */
928 ASSERT(*head_blk < INT_MAX);
929 for (i = (int)(*head_blk) - 1; i >= 0; i--) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100930 error = xlog_bread(log, i, 1, bp, &offset);
931 if (error)
Alex Elder9db127e2010-04-15 18:17:26 +0000932 goto done;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100933
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200934 if (*(__be32 *)offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 found = 1;
936 break;
937 }
938 }
939 /*
940 * If we haven't found the log record header block, start looking
941 * again from the end of the physical log. XXXmiken: There should be
942 * a check here to make sure we didn't search more than N blocks in
943 * the previous code.
944 */
945 if (!found) {
946 for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100947 error = xlog_bread(log, i, 1, bp, &offset);
948 if (error)
Alex Elder9db127e2010-04-15 18:17:26 +0000949 goto done;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100950
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200951 if (*(__be32 *)offset ==
952 cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 found = 2;
954 break;
955 }
956 }
957 }
958 if (!found) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100959 xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__);
Eric Sandeen050a1952013-07-31 20:33:47 -0500960 xlog_put_bp(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 ASSERT(0);
Dave Chinner24513372014-06-25 14:58:08 +1000962 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 }
964
965 /* find blk_no of tail of log */
966 rhead = (xlog_rec_header_t *)offset;
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000967 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968
969 /*
970 * Reset log values according to the state of the log when we
971 * crashed. In the case where head_blk == 0, we bump curr_cycle
972 * one because the next write starts a new cycle rather than
973 * continuing the cycle of the last good log record. At this
974 * point we have guaranteed that all partial log records have been
975 * accounted for. Therefore, we know that the last good log record
976 * written was complete and ended exactly on the end boundary
977 * of the physical log.
978 */
979 log->l_prev_block = i;
980 log->l_curr_block = (int)*head_blk;
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000981 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 if (found == 2)
983 log->l_curr_cycle++;
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100984 atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn));
Dave Chinner84f3c682010-12-03 22:11:29 +1100985 atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn));
Christoph Hellwig28496962012-02-20 02:31:25 +0000986 xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle,
Dave Chinnera69ed032010-12-21 12:08:20 +1100987 BBTOB(log->l_curr_block));
Christoph Hellwig28496962012-02-20 02:31:25 +0000988 xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle,
Dave Chinnera69ed032010-12-21 12:08:20 +1100989 BBTOB(log->l_curr_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990
991 /*
992 * Look for unmount record. If we find it, then we know there
993 * was a clean unmount. Since 'i' could be the last block in
994 * the physical log, we convert to a log block before comparing
995 * to the head_blk.
996 *
997 * Save the current tail lsn to use to pass to
998 * xlog_clear_stale_blocks() below. We won't want to clear the
999 * unmount record if there is one, so we pass the lsn of the
1000 * unmount record rather than the block after it.
1001 */
Eric Sandeen62118702008-03-06 13:44:28 +11001002 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001003 int h_size = be32_to_cpu(rhead->h_size);
1004 int h_version = be32_to_cpu(rhead->h_version);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005
1006 if ((h_version & XLOG_VERSION_2) &&
1007 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
1008 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
1009 if (h_size % XLOG_HEADER_CYCLE_SIZE)
1010 hblks++;
1011 } else {
1012 hblks = 1;
1013 }
1014 } else {
1015 hblks = 1;
1016 }
1017 after_umount_blk = (i + hblks + (int)
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001018 BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize;
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001019 tail_lsn = atomic64_read(&log->l_tail_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 if (*head_blk == after_umount_blk &&
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001021 be32_to_cpu(rhead->h_num_logops) == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 umount_data_blk = (i + hblks) % log->l_logBBsize;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001023 error = xlog_bread(log, umount_data_blk, 1, bp, &offset);
1024 if (error)
Alex Elder9db127e2010-04-15 18:17:26 +00001025 goto done;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001026
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 op_head = (xlog_op_header_t *)offset;
1028 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
1029 /*
1030 * Set tail and last sync so that newly written
1031 * log records will point recovery to after the
1032 * current unmount record.
1033 */
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001034 xlog_assign_atomic_lsn(&log->l_tail_lsn,
1035 log->l_curr_cycle, after_umount_blk);
1036 xlog_assign_atomic_lsn(&log->l_last_sync_lsn,
1037 log->l_curr_cycle, after_umount_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 *tail_blk = after_umount_blk;
David Chinner92821e22007-05-24 15:26:31 +10001039
1040 /*
1041 * Note that the unmount was clean. If the unmount
1042 * was not clean, we need to know this to rebuild the
1043 * superblock counters from the perag headers if we
1044 * have a filesystem using non-persistent counters.
1045 */
1046 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 }
1048 }
1049
1050 /*
1051 * Make sure that there are no blocks in front of the head
1052 * with the same cycle number as the head. This can happen
1053 * because we allow multiple outstanding log writes concurrently,
1054 * and the later writes might make it out before earlier ones.
1055 *
1056 * We use the lsn from before modifying it so that we'll never
1057 * overwrite the unmount record after a clean unmount.
1058 *
1059 * Do this only if we are going to recover the filesystem
1060 *
1061 * NOTE: This used to say "if (!readonly)"
1062 * However on Linux, we can & do recover a read-only filesystem.
1063 * We only skip recovery if NORECOVERY is specified on mount,
1064 * in which case we would not be here.
1065 *
1066 * But... if the -device- itself is readonly, just skip this.
1067 * We can't recover this device anyway, so it won't matter.
1068 */
Alex Elder9db127e2010-04-15 18:17:26 +00001069 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 error = xlog_clear_stale_blocks(log, tail_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071
Alex Elder9db127e2010-04-15 18:17:26 +00001072done:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 xlog_put_bp(bp);
1074
1075 if (error)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001076 xfs_warn(log->l_mp, "failed to locate log tail");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 return error;
1078}
1079
1080/*
1081 * Is the log zeroed at all?
1082 *
1083 * The last binary search should be changed to perform an X block read
1084 * once X becomes small enough. You can then search linearly through
1085 * the X blocks. This will cut down on the number of reads we need to do.
1086 *
1087 * If the log is partially zeroed, this routine will pass back the blkno
1088 * of the first block with cycle number 0. It won't have a complete LR
1089 * preceding it.
1090 *
1091 * Return:
1092 * 0 => the log is completely written to
Dave Chinner24513372014-06-25 14:58:08 +10001093 * 1 => use *blk_no as the first block of the log
1094 * <0 => error has occurred
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 */
David Chinnera8272ce2007-11-23 16:28:09 +11001096STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097xlog_find_zeroed(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001098 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 xfs_daddr_t *blk_no)
1100{
1101 xfs_buf_t *bp;
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10001102 char *offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 uint first_cycle, last_cycle;
1104 xfs_daddr_t new_blk, last_blk, start_blk;
1105 xfs_daddr_t num_scan_bblks;
1106 int error, log_bbnum = log->l_logBBsize;
1107
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10001108 *blk_no = 0;
1109
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 /* check totally zeroed log */
1111 bp = xlog_get_bp(log, 1);
1112 if (!bp)
Dave Chinner24513372014-06-25 14:58:08 +10001113 return -ENOMEM;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001114 error = xlog_bread(log, 0, 1, bp, &offset);
1115 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 goto bp_err;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001117
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10001118 first_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 if (first_cycle == 0) { /* completely zeroed log */
1120 *blk_no = 0;
1121 xlog_put_bp(bp);
Dave Chinner24513372014-06-25 14:58:08 +10001122 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 }
1124
1125 /* check partially zeroed log */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001126 error = xlog_bread(log, log_bbnum-1, 1, bp, &offset);
1127 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 goto bp_err;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001129
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10001130 last_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 if (last_cycle != 0) { /* log completely written to */
1132 xlog_put_bp(bp);
1133 return 0;
1134 } else if (first_cycle != 1) {
1135 /*
1136 * If the cycle of the last block is zero, the cycle of
1137 * the first block must be 1. If it's not, maybe we're
1138 * not looking at a log... Bail out.
1139 */
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001140 xfs_warn(log->l_mp,
1141 "Log inconsistent or not a log (last==0, first!=1)");
Dave Chinner24513372014-06-25 14:58:08 +10001142 error = -EINVAL;
Eric Sandeen5d0a6542013-07-31 20:32:30 -05001143 goto bp_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 }
1145
1146 /* we have a partially zeroed log */
1147 last_blk = log_bbnum-1;
1148 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1149 goto bp_err;
1150
1151 /*
1152 * Validate the answer. Because there is no way to guarantee that
1153 * the entire log is made up of log records which are the same size,
1154 * we scan over the defined maximum blocks. At this point, the maximum
1155 * is not chosen to mean anything special. XXXmiken
1156 */
1157 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1158 ASSERT(num_scan_bblks <= INT_MAX);
1159
1160 if (last_blk < num_scan_bblks)
1161 num_scan_bblks = last_blk;
1162 start_blk = last_blk - num_scan_bblks;
1163
1164 /*
1165 * We search for any instances of cycle number 0 that occur before
1166 * our current estimate of the head. What we're trying to detect is
1167 * 1 ... | 0 | 1 | 0...
1168 * ^ binary search ends here
1169 */
1170 if ((error = xlog_find_verify_cycle(log, start_blk,
1171 (int)num_scan_bblks, 0, &new_blk)))
1172 goto bp_err;
1173 if (new_blk != -1)
1174 last_blk = new_blk;
1175
1176 /*
1177 * Potentially backup over partial log record write. We don't need
1178 * to search the end of the log because we know it is zero.
1179 */
Dave Chinner24513372014-06-25 14:58:08 +10001180 error = xlog_find_verify_log_record(log, start_blk, &last_blk, 0);
1181 if (error == 1)
1182 error = -EIO;
1183 if (error)
1184 goto bp_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185
1186 *blk_no = last_blk;
1187bp_err:
1188 xlog_put_bp(bp);
1189 if (error)
1190 return error;
Dave Chinner24513372014-06-25 14:58:08 +10001191 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192}
1193
1194/*
1195 * These are simple subroutines used by xlog_clear_stale_blocks() below
1196 * to initialize a buffer full of empty log record headers and write
1197 * them into the log.
1198 */
1199STATIC void
1200xlog_add_record(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001201 struct xlog *log,
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10001202 char *buf,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 int cycle,
1204 int block,
1205 int tail_cycle,
1206 int tail_block)
1207{
1208 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1209
1210 memset(buf, 0, BBSIZE);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001211 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1212 recp->h_cycle = cpu_to_be32(cycle);
1213 recp->h_version = cpu_to_be32(
Eric Sandeen62118702008-03-06 13:44:28 +11001214 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001215 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1216 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1217 recp->h_fmt = cpu_to_be32(XLOG_FMT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1219}
1220
1221STATIC int
1222xlog_write_log_records(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001223 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 int cycle,
1225 int start_block,
1226 int blocks,
1227 int tail_cycle,
1228 int tail_block)
1229{
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10001230 char *offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 xfs_buf_t *bp;
1232 int balign, ealign;
Alex Elder69ce58f2010-04-20 17:09:59 +10001233 int sectbb = log->l_sectBBsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 int end_block = start_block + blocks;
1235 int bufblks;
1236 int error = 0;
1237 int i, j = 0;
1238
Alex Elder6881a222010-04-13 15:22:29 +10001239 /*
1240 * Greedily allocate a buffer big enough to handle the full
1241 * range of basic blocks to be written. If that fails, try
1242 * a smaller size. We need to be able to write at least a
1243 * log sector, or we're out of luck.
1244 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 bufblks = 1 << ffs(blocks);
Dave Chinner81158e02012-04-27 19:45:22 +10001246 while (bufblks > log->l_logBBsize)
1247 bufblks >>= 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 while (!(bp = xlog_get_bp(log, bufblks))) {
1249 bufblks >>= 1;
Alex Elder69ce58f2010-04-20 17:09:59 +10001250 if (bufblks < sectbb)
Dave Chinner24513372014-06-25 14:58:08 +10001251 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 }
1253
1254 /* We may need to do a read at the start to fill in part of
1255 * the buffer in the starting sector not covered by the first
1256 * write below.
1257 */
Alex Elder5c17f532010-04-13 15:22:48 +10001258 balign = round_down(start_block, sectbb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 if (balign != start_block) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001260 error = xlog_bread_noalign(log, start_block, 1, bp);
1261 if (error)
1262 goto out_put_bp;
1263
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 j = start_block - balign;
1265 }
1266
1267 for (i = start_block; i < end_block; i += bufblks) {
1268 int bcount, endcount;
1269
1270 bcount = min(bufblks, end_block - start_block);
1271 endcount = bcount - j;
1272
1273 /* We may need to do a read at the end to fill in part of
1274 * the buffer in the final sector not covered by the write.
1275 * If this is the same sector as the above read, skip it.
1276 */
Alex Elder5c17f532010-04-13 15:22:48 +10001277 ealign = round_down(end_block, sectbb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 if (j == 0 && (start_block + endcount > ealign)) {
Chandra Seetharaman62926042011-07-22 23:40:15 +00001279 offset = bp->b_addr + BBTOB(ealign - start_block);
Dave Chinner44396472011-04-21 09:34:27 +00001280 error = xlog_bread_offset(log, ealign, sectbb,
1281 bp, offset);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001282 if (error)
1283 break;
1284
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 }
1286
1287 offset = xlog_align(log, start_block, endcount, bp);
1288 for (; j < endcount; j++) {
1289 xlog_add_record(log, offset, cycle, i+j,
1290 tail_cycle, tail_block);
1291 offset += BBSIZE;
1292 }
1293 error = xlog_bwrite(log, start_block, endcount, bp);
1294 if (error)
1295 break;
1296 start_block += endcount;
1297 j = 0;
1298 }
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001299
1300 out_put_bp:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 xlog_put_bp(bp);
1302 return error;
1303}
1304
1305/*
1306 * This routine is called to blow away any incomplete log writes out
1307 * in front of the log head. We do this so that we won't become confused
1308 * if we come up, write only a little bit more, and then crash again.
1309 * If we leave the partial log records out there, this situation could
1310 * cause us to think those partial writes are valid blocks since they
1311 * have the current cycle number. We get rid of them by overwriting them
1312 * with empty log records with the old cycle number rather than the
1313 * current one.
1314 *
1315 * The tail lsn is passed in rather than taken from
1316 * the log so that we will not write over the unmount record after a
1317 * clean unmount in a 512 block log. Doing so would leave the log without
1318 * any valid log records in it until a new one was written. If we crashed
1319 * during that time we would not be able to recover.
1320 */
1321STATIC int
1322xlog_clear_stale_blocks(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001323 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 xfs_lsn_t tail_lsn)
1325{
1326 int tail_cycle, head_cycle;
1327 int tail_block, head_block;
1328 int tail_distance, max_distance;
1329 int distance;
1330 int error;
1331
1332 tail_cycle = CYCLE_LSN(tail_lsn);
1333 tail_block = BLOCK_LSN(tail_lsn);
1334 head_cycle = log->l_curr_cycle;
1335 head_block = log->l_curr_block;
1336
1337 /*
1338 * Figure out the distance between the new head of the log
1339 * and the tail. We want to write over any blocks beyond the
1340 * head that we may have written just before the crash, but
1341 * we don't want to overwrite the tail of the log.
1342 */
1343 if (head_cycle == tail_cycle) {
1344 /*
1345 * The tail is behind the head in the physical log,
1346 * so the distance from the head to the tail is the
1347 * distance from the head to the end of the log plus
1348 * the distance from the beginning of the log to the
1349 * tail.
1350 */
1351 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1352 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1353 XFS_ERRLEVEL_LOW, log->l_mp);
Dave Chinner24513372014-06-25 14:58:08 +10001354 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 }
1356 tail_distance = tail_block + (log->l_logBBsize - head_block);
1357 } else {
1358 /*
1359 * The head is behind the tail in the physical log,
1360 * so the distance from the head to the tail is just
1361 * the tail block minus the head block.
1362 */
1363 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1364 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1365 XFS_ERRLEVEL_LOW, log->l_mp);
Dave Chinner24513372014-06-25 14:58:08 +10001366 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 }
1368 tail_distance = tail_block - head_block;
1369 }
1370
1371 /*
1372 * If the head is right up against the tail, we can't clear
1373 * anything.
1374 */
1375 if (tail_distance <= 0) {
1376 ASSERT(tail_distance == 0);
1377 return 0;
1378 }
1379
1380 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1381 /*
1382 * Take the smaller of the maximum amount of outstanding I/O
1383 * we could have and the distance to the tail to clear out.
1384 * We take the smaller so that we don't overwrite the tail and
1385 * we don't waste all day writing from the head to the tail
1386 * for no reason.
1387 */
1388 max_distance = MIN(max_distance, tail_distance);
1389
1390 if ((head_block + max_distance) <= log->l_logBBsize) {
1391 /*
1392 * We can stomp all the blocks we need to without
1393 * wrapping around the end of the log. Just do it
1394 * in a single write. Use the cycle number of the
1395 * current cycle minus one so that the log will look like:
1396 * n ... | n - 1 ...
1397 */
1398 error = xlog_write_log_records(log, (head_cycle - 1),
1399 head_block, max_distance, tail_cycle,
1400 tail_block);
1401 if (error)
1402 return error;
1403 } else {
1404 /*
1405 * We need to wrap around the end of the physical log in
1406 * order to clear all the blocks. Do it in two separate
1407 * I/Os. The first write should be from the head to the
1408 * end of the physical log, and it should use the current
1409 * cycle number minus one just like above.
1410 */
1411 distance = log->l_logBBsize - head_block;
1412 error = xlog_write_log_records(log, (head_cycle - 1),
1413 head_block, distance, tail_cycle,
1414 tail_block);
1415
1416 if (error)
1417 return error;
1418
1419 /*
1420 * Now write the blocks at the start of the physical log.
1421 * This writes the remainder of the blocks we want to clear.
1422 * It uses the current cycle number since we're now on the
1423 * same cycle as the head so that we get:
1424 * n ... n ... | n - 1 ...
1425 * ^^^^^ blocks we're writing
1426 */
1427 distance = max_distance - (log->l_logBBsize - head_block);
1428 error = xlog_write_log_records(log, head_cycle, 0, distance,
1429 tail_cycle, tail_block);
1430 if (error)
1431 return error;
1432 }
1433
1434 return 0;
1435}
1436
1437/******************************************************************************
1438 *
1439 * Log recover routines
1440 *
1441 ******************************************************************************
1442 */
1443
Dave Chinnerf0a76952010-01-11 11:49:57 +00001444/*
Dave Chinnera775ad72013-06-05 12:09:07 +10001445 * Sort the log items in the transaction.
1446 *
1447 * The ordering constraints are defined by the inode allocation and unlink
1448 * behaviour. The rules are:
1449 *
1450 * 1. Every item is only logged once in a given transaction. Hence it
1451 * represents the last logged state of the item. Hence ordering is
1452 * dependent on the order in which operations need to be performed so
1453 * required initial conditions are always met.
1454 *
1455 * 2. Cancelled buffers are recorded in pass 1 in a separate table and
1456 * there's nothing to replay from them so we can simply cull them
1457 * from the transaction. However, we can't do that until after we've
1458 * replayed all the other items because they may be dependent on the
1459 * cancelled buffer and replaying the cancelled buffer can remove it
1460 * form the cancelled buffer table. Hence they have tobe done last.
1461 *
1462 * 3. Inode allocation buffers must be replayed before inode items that
Dave Chinner28c8e412013-06-27 16:04:55 +10001463 * read the buffer and replay changes into it. For filesystems using the
1464 * ICREATE transactions, this means XFS_LI_ICREATE objects need to get
1465 * treated the same as inode allocation buffers as they create and
1466 * initialise the buffers directly.
Dave Chinnera775ad72013-06-05 12:09:07 +10001467 *
1468 * 4. Inode unlink buffers must be replayed after inode items are replayed.
1469 * This ensures that inodes are completely flushed to the inode buffer
1470 * in a "free" state before we remove the unlinked inode list pointer.
1471 *
1472 * Hence the ordering needs to be inode allocation buffers first, inode items
1473 * second, inode unlink buffers third and cancelled buffers last.
1474 *
1475 * But there's a problem with that - we can't tell an inode allocation buffer
1476 * apart from a regular buffer, so we can't separate them. We can, however,
1477 * tell an inode unlink buffer from the others, and so we can separate them out
1478 * from all the other buffers and move them to last.
1479 *
1480 * Hence, 4 lists, in order from head to tail:
Dave Chinner28c8e412013-06-27 16:04:55 +10001481 * - buffer_list for all buffers except cancelled/inode unlink buffers
1482 * - item_list for all non-buffer items
1483 * - inode_buffer_list for inode unlink buffers
1484 * - cancel_list for the cancelled buffers
1485 *
1486 * Note that we add objects to the tail of the lists so that first-to-last
1487 * ordering is preserved within the lists. Adding objects to the head of the
1488 * list means when we traverse from the head we walk them in last-to-first
1489 * order. For cancelled buffers and inode unlink buffers this doesn't matter,
1490 * but for all other items there may be specific ordering that we need to
1491 * preserve.
Dave Chinnerf0a76952010-01-11 11:49:57 +00001492 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493STATIC int
1494xlog_recover_reorder_trans(
Mark Tinguelyad223e62012-06-14 09:22:15 -05001495 struct xlog *log,
1496 struct xlog_recover *trans,
Dave Chinner9abbc532010-04-13 15:06:46 +10001497 int pass)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498{
Dave Chinnerf0a76952010-01-11 11:49:57 +00001499 xlog_recover_item_t *item, *n;
Mark Tinguely2a841082013-10-02 07:51:12 -05001500 int error = 0;
Dave Chinnerf0a76952010-01-11 11:49:57 +00001501 LIST_HEAD(sort_list);
Dave Chinnera775ad72013-06-05 12:09:07 +10001502 LIST_HEAD(cancel_list);
1503 LIST_HEAD(buffer_list);
1504 LIST_HEAD(inode_buffer_list);
1505 LIST_HEAD(inode_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506
Dave Chinnerf0a76952010-01-11 11:49:57 +00001507 list_splice_init(&trans->r_itemq, &sort_list);
1508 list_for_each_entry_safe(item, n, &sort_list, ri_list) {
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10001509 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
Dave Chinnerf0a76952010-01-11 11:49:57 +00001510
1511 switch (ITEM_TYPE(item)) {
Dave Chinner28c8e412013-06-27 16:04:55 +10001512 case XFS_LI_ICREATE:
1513 list_move_tail(&item->ri_list, &buffer_list);
1514 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 case XFS_LI_BUF:
Dave Chinnera775ad72013-06-05 12:09:07 +10001516 if (buf_f->blf_flags & XFS_BLF_CANCEL) {
Dave Chinner9abbc532010-04-13 15:06:46 +10001517 trace_xfs_log_recover_item_reorder_head(log,
1518 trans, item, pass);
Dave Chinnera775ad72013-06-05 12:09:07 +10001519 list_move(&item->ri_list, &cancel_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 break;
1521 }
Dave Chinnera775ad72013-06-05 12:09:07 +10001522 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
1523 list_move(&item->ri_list, &inode_buffer_list);
1524 break;
1525 }
1526 list_move_tail(&item->ri_list, &buffer_list);
1527 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 case XFS_LI_INODE:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 case XFS_LI_DQUOT:
1530 case XFS_LI_QUOTAOFF:
1531 case XFS_LI_EFD:
1532 case XFS_LI_EFI:
Dave Chinner9abbc532010-04-13 15:06:46 +10001533 trace_xfs_log_recover_item_reorder_tail(log,
1534 trans, item, pass);
Dave Chinnera775ad72013-06-05 12:09:07 +10001535 list_move_tail(&item->ri_list, &inode_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 break;
1537 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001538 xfs_warn(log->l_mp,
1539 "%s: unrecognized type of log operation",
1540 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 ASSERT(0);
Mark Tinguely2a841082013-10-02 07:51:12 -05001542 /*
1543 * return the remaining items back to the transaction
1544 * item list so they can be freed in caller.
1545 */
1546 if (!list_empty(&sort_list))
1547 list_splice_init(&sort_list, &trans->r_itemq);
Dave Chinner24513372014-06-25 14:58:08 +10001548 error = -EIO;
Mark Tinguely2a841082013-10-02 07:51:12 -05001549 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 }
Dave Chinnerf0a76952010-01-11 11:49:57 +00001551 }
Mark Tinguely2a841082013-10-02 07:51:12 -05001552out:
Dave Chinnerf0a76952010-01-11 11:49:57 +00001553 ASSERT(list_empty(&sort_list));
Dave Chinnera775ad72013-06-05 12:09:07 +10001554 if (!list_empty(&buffer_list))
1555 list_splice(&buffer_list, &trans->r_itemq);
1556 if (!list_empty(&inode_list))
1557 list_splice_tail(&inode_list, &trans->r_itemq);
1558 if (!list_empty(&inode_buffer_list))
1559 list_splice_tail(&inode_buffer_list, &trans->r_itemq);
1560 if (!list_empty(&cancel_list))
1561 list_splice_tail(&cancel_list, &trans->r_itemq);
Mark Tinguely2a841082013-10-02 07:51:12 -05001562 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563}
1564
1565/*
1566 * Build up the table of buf cancel records so that we don't replay
1567 * cancelled data in the second pass. For buffer records that are
1568 * not cancel records, there is nothing to do here so we just return.
1569 *
1570 * If we get a cancel record which is already in the table, this indicates
1571 * that the buffer was cancelled multiple times. In order to ensure
1572 * that during pass 2 we keep the record in the table until we reach its
1573 * last occurrence in the log, we keep a reference count in the cancel
1574 * record in the table to tell us how many times we expect to see this
1575 * record during the second pass.
1576 */
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001577STATIC int
1578xlog_recover_buffer_pass1(
Mark Tinguelyad223e62012-06-14 09:22:15 -05001579 struct xlog *log,
1580 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001582 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001583 struct list_head *bucket;
1584 struct xfs_buf_cancel *bcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585
1586 /*
1587 * If this isn't a cancel buffer item, then just return.
1588 */
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001589 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
Dave Chinner9abbc532010-04-13 15:06:46 +10001590 trace_xfs_log_recover_buf_not_cancel(log, buf_f);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001591 return 0;
Dave Chinner9abbc532010-04-13 15:06:46 +10001592 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593
1594 /*
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001595 * Insert an xfs_buf_cancel record into the hash table of them.
1596 * If there is already an identical record, bump its reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 */
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001598 bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno);
1599 list_for_each_entry(bcp, bucket, bc_list) {
1600 if (bcp->bc_blkno == buf_f->blf_blkno &&
1601 bcp->bc_len == buf_f->blf_len) {
1602 bcp->bc_refcount++;
Dave Chinner9abbc532010-04-13 15:06:46 +10001603 trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001604 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 }
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001607
1608 bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP);
1609 bcp->bc_blkno = buf_f->blf_blkno;
1610 bcp->bc_len = buf_f->blf_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 bcp->bc_refcount = 1;
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001612 list_add_tail(&bcp->bc_list, bucket);
1613
Dave Chinner9abbc532010-04-13 15:06:46 +10001614 trace_xfs_log_recover_buf_cancel_add(log, buf_f);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001615 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616}
1617
1618/*
1619 * Check to see whether the buffer being recovered has a corresponding
Dave Chinner84a5b732013-08-27 08:10:53 +10001620 * entry in the buffer cancel record table. If it is, return the cancel
1621 * buffer structure to the caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 */
Dave Chinner84a5b732013-08-27 08:10:53 +10001623STATIC struct xfs_buf_cancel *
1624xlog_peek_buffer_cancelled(
Mark Tinguelyad223e62012-06-14 09:22:15 -05001625 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 xfs_daddr_t blkno,
1627 uint len,
1628 ushort flags)
1629{
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001630 struct list_head *bucket;
1631 struct xfs_buf_cancel *bcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632
Dave Chinner84a5b732013-08-27 08:10:53 +10001633 if (!log->l_buf_cancel_table) {
1634 /* empty table means no cancelled buffers in the log */
Dave Chinnerc1155412010-05-07 11:05:19 +10001635 ASSERT(!(flags & XFS_BLF_CANCEL));
Dave Chinner84a5b732013-08-27 08:10:53 +10001636 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 }
1638
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001639 bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno);
1640 list_for_each_entry(bcp, bucket, bc_list) {
1641 if (bcp->bc_blkno == blkno && bcp->bc_len == len)
Dave Chinner84a5b732013-08-27 08:10:53 +10001642 return bcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 }
1644
1645 /*
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001646 * We didn't find a corresponding entry in the table, so return 0 so
1647 * that the buffer is NOT cancelled.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 */
Dave Chinnerc1155412010-05-07 11:05:19 +10001649 ASSERT(!(flags & XFS_BLF_CANCEL));
Dave Chinner84a5b732013-08-27 08:10:53 +10001650 return NULL;
1651}
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001652
Dave Chinner84a5b732013-08-27 08:10:53 +10001653/*
1654 * If the buffer is being cancelled then return 1 so that it will be cancelled,
1655 * otherwise return 0. If the buffer is actually a buffer cancel item
1656 * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the
1657 * table and remove it from the table if this is the last reference.
1658 *
1659 * We remove the cancel record from the table when we encounter its last
1660 * occurrence in the log so that if the same buffer is re-used again after its
1661 * last cancellation we actually replay the changes made at that point.
1662 */
1663STATIC int
1664xlog_check_buffer_cancelled(
1665 struct xlog *log,
1666 xfs_daddr_t blkno,
1667 uint len,
1668 ushort flags)
1669{
1670 struct xfs_buf_cancel *bcp;
1671
1672 bcp = xlog_peek_buffer_cancelled(log, blkno, len, flags);
1673 if (!bcp)
1674 return 0;
1675
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001676 /*
1677 * We've go a match, so return 1 so that the recovery of this buffer
1678 * is cancelled. If this buffer is actually a buffer cancel log
1679 * item, then decrement the refcount on the one in the table and
1680 * remove it if this is the last reference.
1681 */
1682 if (flags & XFS_BLF_CANCEL) {
1683 if (--bcp->bc_refcount == 0) {
1684 list_del(&bcp->bc_list);
1685 kmem_free(bcp);
1686 }
1687 }
1688 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689}
1690
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691/*
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001692 * Perform recovery for a buffer full of inodes. In these buffers, the only
1693 * data which should be recovered is that which corresponds to the
1694 * di_next_unlinked pointers in the on disk inode structures. The rest of the
1695 * data for the inodes is always logged through the inodes themselves rather
1696 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 *
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001698 * The only time when buffers full of inodes are fully recovered is when the
1699 * buffer is full of newly allocated inodes. In this case the buffer will
1700 * not be marked as an inode buffer and so will be sent to
1701 * xlog_recover_do_reg_buffer() below during recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 */
1703STATIC int
1704xlog_recover_do_inode_buffer(
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001705 struct xfs_mount *mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 xlog_recover_item_t *item,
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001707 struct xfs_buf *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 xfs_buf_log_format_t *buf_f)
1709{
1710 int i;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001711 int item_index = 0;
1712 int bit = 0;
1713 int nbits = 0;
1714 int reg_buf_offset = 0;
1715 int reg_buf_bytes = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 int next_unlinked_offset;
1717 int inodes_per_buf;
1718 xfs_agino_t *logged_nextp;
1719 xfs_agino_t *buffer_nextp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720
Dave Chinner9abbc532010-04-13 15:06:46 +10001721 trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f);
Dave Chinner9222a9c2013-06-12 12:19:06 +10001722
1723 /*
1724 * Post recovery validation only works properly on CRC enabled
1725 * filesystems.
1726 */
1727 if (xfs_sb_version_hascrc(&mp->m_sb))
1728 bp->b_ops = &xfs_inode_buf_ops;
Dave Chinner9abbc532010-04-13 15:06:46 +10001729
Dave Chinneraa0e8832012-04-23 15:58:52 +10001730 inodes_per_buf = BBTOB(bp->b_io_length) >> mp->m_sb.sb_inodelog;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 for (i = 0; i < inodes_per_buf; i++) {
1732 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
1733 offsetof(xfs_dinode_t, di_next_unlinked);
1734
1735 while (next_unlinked_offset >=
1736 (reg_buf_offset + reg_buf_bytes)) {
1737 /*
1738 * The next di_next_unlinked field is beyond
1739 * the current logged region. Find the next
1740 * logged region that contains or is beyond
1741 * the current di_next_unlinked field.
1742 */
1743 bit += nbits;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001744 bit = xfs_next_bit(buf_f->blf_data_map,
1745 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746
1747 /*
1748 * If there are no more logged regions in the
1749 * buffer, then we're done.
1750 */
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001751 if (bit == -1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001754 nbits = xfs_contig_bits(buf_f->blf_data_map,
1755 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 ASSERT(nbits > 0);
Dave Chinnerc1155412010-05-07 11:05:19 +10001757 reg_buf_offset = bit << XFS_BLF_SHIFT;
1758 reg_buf_bytes = nbits << XFS_BLF_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 item_index++;
1760 }
1761
1762 /*
1763 * If the current logged region starts after the current
1764 * di_next_unlinked field, then move on to the next
1765 * di_next_unlinked field.
1766 */
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001767 if (next_unlinked_offset < reg_buf_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769
1770 ASSERT(item->ri_buf[item_index].i_addr != NULL);
Dave Chinnerc1155412010-05-07 11:05:19 +10001771 ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0);
Dave Chinneraa0e8832012-04-23 15:58:52 +10001772 ASSERT((reg_buf_offset + reg_buf_bytes) <=
1773 BBTOB(bp->b_io_length));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774
1775 /*
1776 * The current logged region contains a copy of the
1777 * current di_next_unlinked field. Extract its value
1778 * and copy it to the buffer copy.
1779 */
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10001780 logged_nextp = item->ri_buf[item_index].i_addr +
1781 next_unlinked_offset - reg_buf_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 if (unlikely(*logged_nextp == 0)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001783 xfs_alert(mp,
1784 "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
1785 "Trying to replay bad (0) inode di_next_unlinked field.",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 item, bp);
1787 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1788 XFS_ERRLEVEL_LOW, mp);
Dave Chinner24513372014-06-25 14:58:08 +10001789 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 }
1791
Christoph Hellwig88ee2df2015-06-22 09:44:29 +10001792 buffer_nextp = xfs_buf_offset(bp, next_unlinked_offset);
Tim Shimmin87c199c2006-06-09 14:56:16 +10001793 *buffer_nextp = *logged_nextp;
Dave Chinner0a32c262013-06-05 12:09:08 +10001794
1795 /*
1796 * If necessary, recalculate the CRC in the on-disk inode. We
1797 * have to leave the inode in a consistent state for whoever
1798 * reads it next....
1799 */
Christoph Hellwig88ee2df2015-06-22 09:44:29 +10001800 xfs_dinode_calc_crc(mp,
Dave Chinner0a32c262013-06-05 12:09:08 +10001801 xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize));
1802
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 }
1804
1805 return 0;
1806}
1807
1808/*
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001809 * V5 filesystems know the age of the buffer on disk being recovered. We can
1810 * have newer objects on disk than we are replaying, and so for these cases we
1811 * don't want to replay the current change as that will make the buffer contents
1812 * temporarily invalid on disk.
1813 *
1814 * The magic number might not match the buffer type we are going to recover
1815 * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags. Hence
1816 * extract the LSN of the existing object in the buffer based on it's current
1817 * magic number. If we don't recognise the magic number in the buffer, then
1818 * return a LSN of -1 so that the caller knows it was an unrecognised block and
1819 * so can recover the buffer.
Dave Chinner566055d2013-09-24 16:01:16 +10001820 *
1821 * Note: we cannot rely solely on magic number matches to determine that the
1822 * buffer has a valid LSN - we also need to verify that it belongs to this
1823 * filesystem, so we need to extract the object's LSN and compare it to that
1824 * which we read from the superblock. If the UUIDs don't match, then we've got a
1825 * stale metadata block from an old filesystem instance that we need to recover
1826 * over the top of.
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001827 */
1828static xfs_lsn_t
1829xlog_recover_get_buf_lsn(
1830 struct xfs_mount *mp,
1831 struct xfs_buf *bp)
1832{
1833 __uint32_t magic32;
1834 __uint16_t magic16;
1835 __uint16_t magicda;
1836 void *blk = bp->b_addr;
Dave Chinner566055d2013-09-24 16:01:16 +10001837 uuid_t *uuid;
1838 xfs_lsn_t lsn = -1;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001839
1840 /* v4 filesystems always recover immediately */
1841 if (!xfs_sb_version_hascrc(&mp->m_sb))
1842 goto recover_immediately;
1843
1844 magic32 = be32_to_cpu(*(__be32 *)blk);
1845 switch (magic32) {
1846 case XFS_ABTB_CRC_MAGIC:
1847 case XFS_ABTC_CRC_MAGIC:
1848 case XFS_ABTB_MAGIC:
1849 case XFS_ABTC_MAGIC:
1850 case XFS_IBT_CRC_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10001851 case XFS_IBT_MAGIC: {
1852 struct xfs_btree_block *btb = blk;
1853
1854 lsn = be64_to_cpu(btb->bb_u.s.bb_lsn);
1855 uuid = &btb->bb_u.s.bb_uuid;
1856 break;
1857 }
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001858 case XFS_BMAP_CRC_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10001859 case XFS_BMAP_MAGIC: {
1860 struct xfs_btree_block *btb = blk;
1861
1862 lsn = be64_to_cpu(btb->bb_u.l.bb_lsn);
1863 uuid = &btb->bb_u.l.bb_uuid;
1864 break;
1865 }
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001866 case XFS_AGF_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10001867 lsn = be64_to_cpu(((struct xfs_agf *)blk)->agf_lsn);
1868 uuid = &((struct xfs_agf *)blk)->agf_uuid;
1869 break;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001870 case XFS_AGFL_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10001871 lsn = be64_to_cpu(((struct xfs_agfl *)blk)->agfl_lsn);
1872 uuid = &((struct xfs_agfl *)blk)->agfl_uuid;
1873 break;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001874 case XFS_AGI_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10001875 lsn = be64_to_cpu(((struct xfs_agi *)blk)->agi_lsn);
1876 uuid = &((struct xfs_agi *)blk)->agi_uuid;
1877 break;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001878 case XFS_SYMLINK_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10001879 lsn = be64_to_cpu(((struct xfs_dsymlink_hdr *)blk)->sl_lsn);
1880 uuid = &((struct xfs_dsymlink_hdr *)blk)->sl_uuid;
1881 break;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001882 case XFS_DIR3_BLOCK_MAGIC:
1883 case XFS_DIR3_DATA_MAGIC:
1884 case XFS_DIR3_FREE_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10001885 lsn = be64_to_cpu(((struct xfs_dir3_blk_hdr *)blk)->lsn);
1886 uuid = &((struct xfs_dir3_blk_hdr *)blk)->uuid;
1887 break;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001888 case XFS_ATTR3_RMT_MAGIC:
Dave Chinnere3c32ee2015-07-29 11:48:01 +10001889 /*
1890 * Remote attr blocks are written synchronously, rather than
1891 * being logged. That means they do not contain a valid LSN
1892 * (i.e. transactionally ordered) in them, and hence any time we
1893 * see a buffer to replay over the top of a remote attribute
1894 * block we should simply do so.
1895 */
1896 goto recover_immediately;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001897 case XFS_SB_MAGIC:
Dave Chinnerfcfbe2c2015-08-19 10:31:54 +10001898 /*
1899 * superblock uuids are magic. We may or may not have a
1900 * sb_meta_uuid on disk, but it will be set in the in-core
1901 * superblock. We set the uuid pointer for verification
1902 * according to the superblock feature mask to ensure we check
1903 * the relevant UUID in the superblock.
1904 */
Dave Chinner566055d2013-09-24 16:01:16 +10001905 lsn = be64_to_cpu(((struct xfs_dsb *)blk)->sb_lsn);
Dave Chinnerfcfbe2c2015-08-19 10:31:54 +10001906 if (xfs_sb_version_hasmetauuid(&mp->m_sb))
1907 uuid = &((struct xfs_dsb *)blk)->sb_meta_uuid;
1908 else
1909 uuid = &((struct xfs_dsb *)blk)->sb_uuid;
Dave Chinner566055d2013-09-24 16:01:16 +10001910 break;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001911 default:
1912 break;
1913 }
1914
Dave Chinner566055d2013-09-24 16:01:16 +10001915 if (lsn != (xfs_lsn_t)-1) {
Dave Chinnerfcfbe2c2015-08-19 10:31:54 +10001916 if (!uuid_equal(&mp->m_sb.sb_meta_uuid, uuid))
Dave Chinner566055d2013-09-24 16:01:16 +10001917 goto recover_immediately;
1918 return lsn;
1919 }
1920
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001921 magicda = be16_to_cpu(((struct xfs_da_blkinfo *)blk)->magic);
1922 switch (magicda) {
1923 case XFS_DIR3_LEAF1_MAGIC:
1924 case XFS_DIR3_LEAFN_MAGIC:
1925 case XFS_DA3_NODE_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10001926 lsn = be64_to_cpu(((struct xfs_da3_blkinfo *)blk)->lsn);
1927 uuid = &((struct xfs_da3_blkinfo *)blk)->uuid;
1928 break;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001929 default:
1930 break;
1931 }
1932
Dave Chinner566055d2013-09-24 16:01:16 +10001933 if (lsn != (xfs_lsn_t)-1) {
1934 if (!uuid_equal(&mp->m_sb.sb_uuid, uuid))
1935 goto recover_immediately;
1936 return lsn;
1937 }
1938
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001939 /*
1940 * We do individual object checks on dquot and inode buffers as they
1941 * have their own individual LSN records. Also, we could have a stale
1942 * buffer here, so we have to at least recognise these buffer types.
1943 *
1944 * A notd complexity here is inode unlinked list processing - it logs
1945 * the inode directly in the buffer, but we don't know which inodes have
1946 * been modified, and there is no global buffer LSN. Hence we need to
1947 * recover all inode buffer types immediately. This problem will be
1948 * fixed by logical logging of the unlinked list modifications.
1949 */
1950 magic16 = be16_to_cpu(*(__be16 *)blk);
1951 switch (magic16) {
1952 case XFS_DQUOT_MAGIC:
1953 case XFS_DINODE_MAGIC:
1954 goto recover_immediately;
1955 default:
1956 break;
1957 }
1958
1959 /* unknown buffer contents, recover immediately */
1960
1961recover_immediately:
1962 return (xfs_lsn_t)-1;
1963
1964}
1965
1966/*
Dave Chinnerd75afeb2013-04-03 16:11:29 +11001967 * Validate the recovered buffer is of the correct type and attach the
1968 * appropriate buffer operations to them for writeback. Magic numbers are in a
1969 * few places:
1970 * the first 16 bits of the buffer (inode buffer, dquot buffer),
1971 * the first 32 bits of the buffer (most blocks),
1972 * inside a struct xfs_da_blkinfo at the start of the buffer.
1973 */
1974static void
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001975xlog_recover_validate_buf_type(
Dave Chinnerd75afeb2013-04-03 16:11:29 +11001976 struct xfs_mount *mp,
1977 struct xfs_buf *bp,
1978 xfs_buf_log_format_t *buf_f)
1979{
1980 struct xfs_da_blkinfo *info = bp->b_addr;
1981 __uint32_t magic32;
1982 __uint16_t magic16;
1983 __uint16_t magicda;
1984
Dave Chinner67dc2882014-08-04 12:43:06 +10001985 /*
1986 * We can only do post recovery validation on items on CRC enabled
1987 * fielsystems as we need to know when the buffer was written to be able
1988 * to determine if we should have replayed the item. If we replay old
1989 * metadata over a newer buffer, then it will enter a temporarily
1990 * inconsistent state resulting in verification failures. Hence for now
1991 * just avoid the verification stage for non-crc filesystems
1992 */
1993 if (!xfs_sb_version_hascrc(&mp->m_sb))
1994 return;
1995
Dave Chinnerd75afeb2013-04-03 16:11:29 +11001996 magic32 = be32_to_cpu(*(__be32 *)bp->b_addr);
1997 magic16 = be16_to_cpu(*(__be16*)bp->b_addr);
1998 magicda = be16_to_cpu(info->magic);
Dave Chinner61fe1352013-04-03 16:11:30 +11001999 switch (xfs_blft_from_flags(buf_f)) {
2000 case XFS_BLFT_BTREE_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002001 switch (magic32) {
2002 case XFS_ABTB_CRC_MAGIC:
2003 case XFS_ABTC_CRC_MAGIC:
2004 case XFS_ABTB_MAGIC:
2005 case XFS_ABTC_MAGIC:
2006 bp->b_ops = &xfs_allocbt_buf_ops;
2007 break;
2008 case XFS_IBT_CRC_MAGIC:
Brian Fosteraafc3c22014-04-24 16:00:52 +10002009 case XFS_FIBT_CRC_MAGIC:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002010 case XFS_IBT_MAGIC:
Brian Fosteraafc3c22014-04-24 16:00:52 +10002011 case XFS_FIBT_MAGIC:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002012 bp->b_ops = &xfs_inobt_buf_ops;
2013 break;
2014 case XFS_BMAP_CRC_MAGIC:
2015 case XFS_BMAP_MAGIC:
2016 bp->b_ops = &xfs_bmbt_buf_ops;
2017 break;
2018 default:
2019 xfs_warn(mp, "Bad btree block magic!");
2020 ASSERT(0);
2021 break;
2022 }
2023 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002024 case XFS_BLFT_AGF_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002025 if (magic32 != XFS_AGF_MAGIC) {
2026 xfs_warn(mp, "Bad AGF block magic!");
2027 ASSERT(0);
2028 break;
2029 }
2030 bp->b_ops = &xfs_agf_buf_ops;
2031 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002032 case XFS_BLFT_AGFL_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002033 if (magic32 != XFS_AGFL_MAGIC) {
2034 xfs_warn(mp, "Bad AGFL block magic!");
2035 ASSERT(0);
2036 break;
2037 }
2038 bp->b_ops = &xfs_agfl_buf_ops;
2039 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002040 case XFS_BLFT_AGI_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002041 if (magic32 != XFS_AGI_MAGIC) {
2042 xfs_warn(mp, "Bad AGI block magic!");
2043 ASSERT(0);
2044 break;
2045 }
2046 bp->b_ops = &xfs_agi_buf_ops;
2047 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002048 case XFS_BLFT_UDQUOT_BUF:
2049 case XFS_BLFT_PDQUOT_BUF:
2050 case XFS_BLFT_GDQUOT_BUF:
Dave Chinner123887e2013-04-30 21:39:33 +10002051#ifdef CONFIG_XFS_QUOTA
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002052 if (magic16 != XFS_DQUOT_MAGIC) {
2053 xfs_warn(mp, "Bad DQUOT block magic!");
2054 ASSERT(0);
2055 break;
2056 }
2057 bp->b_ops = &xfs_dquot_buf_ops;
Dave Chinner123887e2013-04-30 21:39:33 +10002058#else
2059 xfs_alert(mp,
2060 "Trying to recover dquots without QUOTA support built in!");
2061 ASSERT(0);
2062#endif
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002063 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002064 case XFS_BLFT_DINO_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002065 if (magic16 != XFS_DINODE_MAGIC) {
2066 xfs_warn(mp, "Bad INODE block magic!");
2067 ASSERT(0);
2068 break;
2069 }
2070 bp->b_ops = &xfs_inode_buf_ops;
2071 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002072 case XFS_BLFT_SYMLINK_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002073 if (magic32 != XFS_SYMLINK_MAGIC) {
2074 xfs_warn(mp, "Bad symlink block magic!");
2075 ASSERT(0);
2076 break;
2077 }
2078 bp->b_ops = &xfs_symlink_buf_ops;
2079 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002080 case XFS_BLFT_DIR_BLOCK_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002081 if (magic32 != XFS_DIR2_BLOCK_MAGIC &&
2082 magic32 != XFS_DIR3_BLOCK_MAGIC) {
2083 xfs_warn(mp, "Bad dir block magic!");
2084 ASSERT(0);
2085 break;
2086 }
2087 bp->b_ops = &xfs_dir3_block_buf_ops;
2088 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002089 case XFS_BLFT_DIR_DATA_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002090 if (magic32 != XFS_DIR2_DATA_MAGIC &&
2091 magic32 != XFS_DIR3_DATA_MAGIC) {
2092 xfs_warn(mp, "Bad dir data magic!");
2093 ASSERT(0);
2094 break;
2095 }
2096 bp->b_ops = &xfs_dir3_data_buf_ops;
2097 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002098 case XFS_BLFT_DIR_FREE_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002099 if (magic32 != XFS_DIR2_FREE_MAGIC &&
2100 magic32 != XFS_DIR3_FREE_MAGIC) {
2101 xfs_warn(mp, "Bad dir3 free magic!");
2102 ASSERT(0);
2103 break;
2104 }
2105 bp->b_ops = &xfs_dir3_free_buf_ops;
2106 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002107 case XFS_BLFT_DIR_LEAF1_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002108 if (magicda != XFS_DIR2_LEAF1_MAGIC &&
2109 magicda != XFS_DIR3_LEAF1_MAGIC) {
2110 xfs_warn(mp, "Bad dir leaf1 magic!");
2111 ASSERT(0);
2112 break;
2113 }
2114 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
2115 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002116 case XFS_BLFT_DIR_LEAFN_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002117 if (magicda != XFS_DIR2_LEAFN_MAGIC &&
2118 magicda != XFS_DIR3_LEAFN_MAGIC) {
2119 xfs_warn(mp, "Bad dir leafn magic!");
2120 ASSERT(0);
2121 break;
2122 }
2123 bp->b_ops = &xfs_dir3_leafn_buf_ops;
2124 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002125 case XFS_BLFT_DA_NODE_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002126 if (magicda != XFS_DA_NODE_MAGIC &&
2127 magicda != XFS_DA3_NODE_MAGIC) {
2128 xfs_warn(mp, "Bad da node magic!");
2129 ASSERT(0);
2130 break;
2131 }
2132 bp->b_ops = &xfs_da3_node_buf_ops;
2133 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002134 case XFS_BLFT_ATTR_LEAF_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002135 if (magicda != XFS_ATTR_LEAF_MAGIC &&
2136 magicda != XFS_ATTR3_LEAF_MAGIC) {
2137 xfs_warn(mp, "Bad attr leaf magic!");
2138 ASSERT(0);
2139 break;
2140 }
2141 bp->b_ops = &xfs_attr3_leaf_buf_ops;
2142 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002143 case XFS_BLFT_ATTR_RMT_BUF:
Dave Chinnercab09a82013-04-30 21:39:36 +10002144 if (magic32 != XFS_ATTR3_RMT_MAGIC) {
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002145 xfs_warn(mp, "Bad attr remote magic!");
2146 ASSERT(0);
2147 break;
2148 }
2149 bp->b_ops = &xfs_attr3_rmt_buf_ops;
2150 break;
Dave Chinner04a1e6c2013-04-03 16:11:31 +11002151 case XFS_BLFT_SB_BUF:
2152 if (magic32 != XFS_SB_MAGIC) {
2153 xfs_warn(mp, "Bad SB block magic!");
2154 ASSERT(0);
2155 break;
2156 }
2157 bp->b_ops = &xfs_sb_buf_ops;
2158 break;
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002159 default:
Dave Chinner61fe1352013-04-03 16:11:30 +11002160 xfs_warn(mp, "Unknown buffer type %d!",
2161 xfs_blft_from_flags(buf_f));
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002162 break;
2163 }
2164}
2165
2166/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 * Perform a 'normal' buffer recovery. Each logged region of the
2168 * buffer should be copied over the corresponding region in the
2169 * given buffer. The bitmap in the buf log format structure indicates
2170 * where to place the logged data.
2171 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172STATIC void
2173xlog_recover_do_reg_buffer(
Dave Chinner9abbc532010-04-13 15:06:46 +10002174 struct xfs_mount *mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 xlog_recover_item_t *item,
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002176 struct xfs_buf *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177 xfs_buf_log_format_t *buf_f)
2178{
2179 int i;
2180 int bit;
2181 int nbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 int error;
2183
Dave Chinner9abbc532010-04-13 15:06:46 +10002184 trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f);
2185
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186 bit = 0;
2187 i = 1; /* 0 is the buf format structure */
2188 while (1) {
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002189 bit = xfs_next_bit(buf_f->blf_data_map,
2190 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191 if (bit == -1)
2192 break;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002193 nbits = xfs_contig_bits(buf_f->blf_data_map,
2194 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195 ASSERT(nbits > 0);
Christoph Hellwig4b809162007-08-16 15:37:36 +10002196 ASSERT(item->ri_buf[i].i_addr != NULL);
Dave Chinnerc1155412010-05-07 11:05:19 +10002197 ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0);
Dave Chinneraa0e8832012-04-23 15:58:52 +10002198 ASSERT(BBTOB(bp->b_io_length) >=
2199 ((uint)bit << XFS_BLF_SHIFT) + (nbits << XFS_BLF_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200
2201 /*
Dave Chinner709da6a2013-05-27 16:38:23 +10002202 * The dirty regions logged in the buffer, even though
2203 * contiguous, may span multiple chunks. This is because the
2204 * dirty region may span a physical page boundary in a buffer
2205 * and hence be split into two separate vectors for writing into
2206 * the log. Hence we need to trim nbits back to the length of
2207 * the current region being copied out of the log.
2208 */
2209 if (item->ri_buf[i].i_len < (nbits << XFS_BLF_SHIFT))
2210 nbits = item->ri_buf[i].i_len >> XFS_BLF_SHIFT;
2211
2212 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213 * Do a sanity check if this is a dquot buffer. Just checking
2214 * the first dquot in the buffer should do. XXXThis is
2215 * probably a good thing to do for other buf types also.
2216 */
2217 error = 0;
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002218 if (buf_f->blf_flags &
Dave Chinnerc1155412010-05-07 11:05:19 +10002219 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002220 if (item->ri_buf[i].i_addr == NULL) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002221 xfs_alert(mp,
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002222 "XFS: NULL dquot in %s.", __func__);
2223 goto next;
2224 }
Jan Rekorajski8ec6dba2009-11-16 11:57:02 +00002225 if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002226 xfs_alert(mp,
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002227 "XFS: dquot too small (%d) in %s.",
2228 item->ri_buf[i].i_len, __func__);
2229 goto next;
2230 }
Dave Chinner9aede1d2013-10-15 09:17:52 +11002231 error = xfs_dqcheck(mp, item->ri_buf[i].i_addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232 -1, 0, XFS_QMOPT_DOWARN,
2233 "dquot_buf_recover");
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002234 if (error)
2235 goto next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 }
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002237
2238 memcpy(xfs_buf_offset(bp,
Dave Chinnerc1155412010-05-07 11:05:19 +10002239 (uint)bit << XFS_BLF_SHIFT), /* dest */
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002240 item->ri_buf[i].i_addr, /* source */
Dave Chinnerc1155412010-05-07 11:05:19 +10002241 nbits<<XFS_BLF_SHIFT); /* length */
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002242 next:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243 i++;
2244 bit += nbits;
2245 }
2246
2247 /* Shouldn't be any more regions */
2248 ASSERT(i == item->ri_total);
Christoph Hellwigee1a47a2013-04-21 14:53:46 -05002249
Dave Chinner67dc2882014-08-04 12:43:06 +10002250 xlog_recover_validate_buf_type(mp, bp, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251}
2252
2253/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254 * Perform a dquot buffer recovery.
Zhi Yong Wu8ba701e2013-08-12 03:15:01 +00002255 * Simple algorithm: if we have found a QUOTAOFF log item of the same type
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2257 * Else, treat it as a regular buffer and do recovery.
Dave Chinnerad3714b2014-08-04 12:59:31 +10002258 *
2259 * Return false if the buffer was tossed and true if we recovered the buffer to
2260 * indicate to the caller if the buffer needs writing.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261 */
Dave Chinnerad3714b2014-08-04 12:59:31 +10002262STATIC bool
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263xlog_recover_do_dquot_buffer(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002264 struct xfs_mount *mp,
2265 struct xlog *log,
2266 struct xlog_recover_item *item,
2267 struct xfs_buf *bp,
2268 struct xfs_buf_log_format *buf_f)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269{
2270 uint type;
2271
Dave Chinner9abbc532010-04-13 15:06:46 +10002272 trace_xfs_log_recover_buf_dquot_buf(log, buf_f);
2273
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274 /*
2275 * Filesystems are required to send in quota flags at mount time.
2276 */
Dave Chinnerad3714b2014-08-04 12:59:31 +10002277 if (!mp->m_qflags)
2278 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279
2280 type = 0;
Dave Chinnerc1155412010-05-07 11:05:19 +10002281 if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282 type |= XFS_DQ_USER;
Dave Chinnerc1155412010-05-07 11:05:19 +10002283 if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF)
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002284 type |= XFS_DQ_PROJ;
Dave Chinnerc1155412010-05-07 11:05:19 +10002285 if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 type |= XFS_DQ_GROUP;
2287 /*
2288 * This type of quotas was turned off, so ignore this buffer
2289 */
2290 if (log->l_quotaoffs_flag & type)
Dave Chinnerad3714b2014-08-04 12:59:31 +10002291 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292
Dave Chinner9abbc532010-04-13 15:06:46 +10002293 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
Dave Chinnerad3714b2014-08-04 12:59:31 +10002294 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295}
2296
2297/*
2298 * This routine replays a modification made to a buffer at runtime.
2299 * There are actually two types of buffer, regular and inode, which
2300 * are handled differently. Inode buffers are handled differently
2301 * in that we only recover a specific set of data from them, namely
2302 * the inode di_next_unlinked fields. This is because all other inode
2303 * data is actually logged via inode records and any data we replay
2304 * here which overlaps that may be stale.
2305 *
2306 * When meta-data buffers are freed at run time we log a buffer item
Dave Chinnerc1155412010-05-07 11:05:19 +10002307 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 * of the buffer in the log should not be replayed at recovery time.
2309 * This is so that if the blocks covered by the buffer are reused for
2310 * file data before we crash we don't end up replaying old, freed
2311 * meta-data into a user's file.
2312 *
2313 * To handle the cancellation of buffer log items, we make two passes
2314 * over the log during recovery. During the first we build a table of
2315 * those buffers which have been cancelled, and during the second we
2316 * only replay those buffers which do not have corresponding cancel
Zhi Yong Wu34be5ff2013-08-07 10:11:07 +00002317 * records in the table. See xlog_recover_buffer_pass[1,2] above
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318 * for more details on the implementation of the table of cancel records.
2319 */
2320STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002321xlog_recover_buffer_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002322 struct xlog *log,
2323 struct list_head *buffer_list,
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002324 struct xlog_recover_item *item,
2325 xfs_lsn_t current_lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326{
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002327 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002328 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329 xfs_buf_t *bp;
2330 int error;
Christoph Hellwig6ad112b2009-11-24 18:02:23 +00002331 uint buf_flags;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002332 xfs_lsn_t lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002334 /*
2335 * In this pass we only want to recover all the buffers which have
2336 * not been cancelled and are not cancellation buffers themselves.
2337 */
2338 if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno,
2339 buf_f->blf_len, buf_f->blf_flags)) {
2340 trace_xfs_log_recover_buf_cancel(log, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342 }
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002343
Dave Chinner9abbc532010-04-13 15:06:46 +10002344 trace_xfs_log_recover_buf_recover(log, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345
Dave Chinnera8acad72012-04-23 15:58:54 +10002346 buf_flags = 0;
Dave Chinner611c9942012-04-23 15:59:07 +10002347 if (buf_f->blf_flags & XFS_BLF_INODE_BUF)
2348 buf_flags |= XBF_UNMAPPED;
Christoph Hellwig6ad112b2009-11-24 18:02:23 +00002349
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002350 bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len,
Dave Chinnerc3f8fc72012-11-12 22:54:01 +11002351 buf_flags, NULL);
Chandra Seetharamanac4d6882011-08-03 02:18:29 +00002352 if (!bp)
Dave Chinner24513372014-06-25 14:58:08 +10002353 return -ENOMEM;
Chandra Seetharamane5702802011-08-03 02:18:34 +00002354 error = bp->b_error;
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +00002355 if (error) {
Christoph Hellwig901796a2011-10-10 16:52:49 +00002356 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)");
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002357 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358 }
2359
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002360 /*
Dave Chinner67dc2882014-08-04 12:43:06 +10002361 * Recover the buffer only if we get an LSN from it and it's less than
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002362 * the lsn of the transaction we are replaying.
Dave Chinner67dc2882014-08-04 12:43:06 +10002363 *
2364 * Note that we have to be extremely careful of readahead here.
2365 * Readahead does not attach verfiers to the buffers so if we don't
2366 * actually do any replay after readahead because of the LSN we found
2367 * in the buffer if more recent than that current transaction then we
2368 * need to attach the verifier directly. Failure to do so can lead to
2369 * future recovery actions (e.g. EFI and unlinked list recovery) can
2370 * operate on the buffers and they won't get the verifier attached. This
2371 * can lead to blocks on disk having the correct content but a stale
2372 * CRC.
2373 *
2374 * It is safe to assume these clean buffers are currently up to date.
2375 * If the buffer is dirtied by a later transaction being replayed, then
2376 * the verifier will be reset to match whatever recover turns that
2377 * buffer into.
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002378 */
2379 lsn = xlog_recover_get_buf_lsn(mp, bp);
Dave Chinner67dc2882014-08-04 12:43:06 +10002380 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
2381 xlog_recover_validate_buf_type(mp, bp, buf_f);
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002382 goto out_release;
Dave Chinner67dc2882014-08-04 12:43:06 +10002383 }
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002384
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002385 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
Dave Chinnerad3714b2014-08-04 12:59:31 +10002387 if (error)
2388 goto out_release;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002389 } else if (buf_f->blf_flags &
Dave Chinnerc1155412010-05-07 11:05:19 +10002390 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
Dave Chinnerad3714b2014-08-04 12:59:31 +10002391 bool dirty;
2392
2393 dirty = xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2394 if (!dirty)
2395 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396 } else {
Dave Chinner9abbc532010-04-13 15:06:46 +10002397 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399
2400 /*
2401 * Perform delayed write on the buffer. Asynchronous writes will be
2402 * slower when taking into account all the buffers to be flushed.
2403 *
2404 * Also make sure that only inode buffers with good sizes stay in
2405 * the buffer cache. The kernel moves inodes in buffers of 1 block
Jie Liu0f49efd2013-12-13 15:51:48 +11002406 * or mp->m_inode_cluster_size bytes, whichever is bigger. The inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407 * buffers in the log can be a different size if the log was generated
2408 * by an older kernel using unclustered inode buffers or a newer kernel
2409 * running with a different inode cluster size. Regardless, if the
Jie Liu0f49efd2013-12-13 15:51:48 +11002410 * the inode buffer size isn't MAX(blocksize, mp->m_inode_cluster_size)
2411 * for *our* value of mp->m_inode_cluster_size, then we need to keep
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412 * the buffer out of the buffer cache so that the buffer won't
2413 * overlap with future reads of those inodes.
2414 */
2415 if (XFS_DINODE_MAGIC ==
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002416 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
Dave Chinneraa0e8832012-04-23 15:58:52 +10002417 (BBTOB(bp->b_io_length) != MAX(log->l_mp->m_sb.sb_blocksize,
Jie Liu0f49efd2013-12-13 15:51:48 +11002418 (__uint32_t)log->l_mp->m_inode_cluster_size))) {
Christoph Hellwigc867cb62011-10-10 16:52:46 +00002419 xfs_buf_stale(bp);
Christoph Hellwigc2b006c2011-08-23 08:28:07 +00002420 error = xfs_bwrite(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 } else {
Dave Chinnerebad8612010-09-22 10:47:20 +10002422 ASSERT(bp->b_target->bt_mount == mp);
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02002423 bp->b_iodone = xlog_recover_iodone;
Christoph Hellwig43ff2122012-04-23 15:58:39 +10002424 xfs_buf_delwri_queue(bp, buffer_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425 }
2426
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002427out_release:
Christoph Hellwigc2b006c2011-08-23 08:28:07 +00002428 xfs_buf_relse(bp);
2429 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430}
2431
Dave Chinner638f44162013-08-30 10:23:45 +10002432/*
2433 * Inode fork owner changes
2434 *
2435 * If we have been told that we have to reparent the inode fork, it's because an
2436 * extent swap operation on a CRC enabled filesystem has been done and we are
2437 * replaying it. We need to walk the BMBT of the appropriate fork and change the
2438 * owners of it.
2439 *
2440 * The complexity here is that we don't have an inode context to work with, so
2441 * after we've replayed the inode we need to instantiate one. This is where the
2442 * fun begins.
2443 *
2444 * We are in the middle of log recovery, so we can't run transactions. That
2445 * means we cannot use cache coherent inode instantiation via xfs_iget(), as
2446 * that will result in the corresponding iput() running the inode through
2447 * xfs_inactive(). If we've just replayed an inode core that changes the link
2448 * count to zero (i.e. it's been unlinked), then xfs_inactive() will run
2449 * transactions (bad!).
2450 *
2451 * So, to avoid this, we instantiate an inode directly from the inode core we've
2452 * just recovered. We have the buffer still locked, and all we really need to
2453 * instantiate is the inode core and the forks being modified. We can do this
2454 * manually, then run the inode btree owner change, and then tear down the
2455 * xfs_inode without having to run any transactions at all.
2456 *
2457 * Also, because we don't have a transaction context available here but need to
2458 * gather all the buffers we modify for writeback so we pass the buffer_list
2459 * instead for the operation to use.
2460 */
2461
2462STATIC int
2463xfs_recover_inode_owner_change(
2464 struct xfs_mount *mp,
2465 struct xfs_dinode *dip,
2466 struct xfs_inode_log_format *in_f,
2467 struct list_head *buffer_list)
2468{
2469 struct xfs_inode *ip;
2470 int error;
2471
2472 ASSERT(in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER));
2473
2474 ip = xfs_inode_alloc(mp, in_f->ilf_ino);
2475 if (!ip)
Dave Chinner24513372014-06-25 14:58:08 +10002476 return -ENOMEM;
Dave Chinner638f44162013-08-30 10:23:45 +10002477
2478 /* instantiate the inode */
2479 xfs_dinode_from_disk(&ip->i_d, dip);
2480 ASSERT(ip->i_d.di_version >= 3);
2481
2482 error = xfs_iformat_fork(ip, dip);
2483 if (error)
2484 goto out_free_ip;
2485
2486
2487 if (in_f->ilf_fields & XFS_ILOG_DOWNER) {
2488 ASSERT(in_f->ilf_fields & XFS_ILOG_DBROOT);
2489 error = xfs_bmbt_change_owner(NULL, ip, XFS_DATA_FORK,
2490 ip->i_ino, buffer_list);
2491 if (error)
2492 goto out_free_ip;
2493 }
2494
2495 if (in_f->ilf_fields & XFS_ILOG_AOWNER) {
2496 ASSERT(in_f->ilf_fields & XFS_ILOG_ABROOT);
2497 error = xfs_bmbt_change_owner(NULL, ip, XFS_ATTR_FORK,
2498 ip->i_ino, buffer_list);
2499 if (error)
2500 goto out_free_ip;
2501 }
2502
2503out_free_ip:
2504 xfs_inode_free(ip);
2505 return error;
2506}
2507
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002509xlog_recover_inode_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002510 struct xlog *log,
2511 struct list_head *buffer_list,
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002512 struct xlog_recover_item *item,
2513 xfs_lsn_t current_lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514{
2515 xfs_inode_log_format_t *in_f;
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002516 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517 xfs_buf_t *bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518 xfs_dinode_t *dip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519 int len;
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10002520 char *src;
2521 char *dest;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522 int error;
2523 int attr_index;
2524 uint fields;
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002525 xfs_icdinode_t *dicp;
Christoph Hellwig93848a92013-04-03 16:11:17 +11002526 uint isize;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002527 int need_free = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528
Tim Shimmin6d192a92006-06-09 14:55:38 +10002529 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002530 in_f = item->ri_buf[0].i_addr;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002531 } else {
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002532 in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002533 need_free = 1;
2534 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2535 if (error)
2536 goto error;
2537 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538
2539 /*
2540 * Inode buffers can be freed, look out for it,
2541 * and do not replay the inode.
2542 */
Christoph Hellwiga1941892008-11-28 14:23:40 +11002543 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
2544 in_f->ilf_len, 0)) {
Tim Shimmin6d192a92006-06-09 14:55:38 +10002545 error = 0;
Dave Chinner9abbc532010-04-13 15:06:46 +10002546 trace_xfs_log_recover_inode_cancel(log, in_f);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002547 goto error;
2548 }
Dave Chinner9abbc532010-04-13 15:06:46 +10002549 trace_xfs_log_recover_inode_recover(log, in_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550
Dave Chinnerc3f8fc72012-11-12 22:54:01 +11002551 bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len, 0,
Christoph Hellwig93848a92013-04-03 16:11:17 +11002552 &xfs_inode_buf_ops);
Chandra Seetharamanac4d6882011-08-03 02:18:29 +00002553 if (!bp) {
Dave Chinner24513372014-06-25 14:58:08 +10002554 error = -ENOMEM;
Chandra Seetharamanac4d6882011-08-03 02:18:29 +00002555 goto error;
2556 }
Chandra Seetharamane5702802011-08-03 02:18:34 +00002557 error = bp->b_error;
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +00002558 if (error) {
Christoph Hellwig901796a2011-10-10 16:52:49 +00002559 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)");
Dave Chinner638f44162013-08-30 10:23:45 +10002560 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
Christoph Hellwig88ee2df2015-06-22 09:44:29 +10002563 dip = xfs_buf_offset(bp, in_f->ilf_boffset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564
2565 /*
2566 * Make sure the place we're flushing out to really looks
2567 * like an inode!
2568 */
Christoph Hellwig69ef9212011-07-08 14:36:05 +02002569 if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002570 xfs_alert(mp,
2571 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
2572 __func__, dip, bp, in_f->ilf_ino);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002573 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574 XFS_ERRLEVEL_LOW, mp);
Dave Chinner24513372014-06-25 14:58:08 +10002575 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002576 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577 }
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002578 dicp = item->ri_buf[1].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002580 xfs_alert(mp,
2581 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
2582 __func__, item, in_f->ilf_ino);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002583 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 XFS_ERRLEVEL_LOW, mp);
Dave Chinner24513372014-06-25 14:58:08 +10002585 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002586 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587 }
2588
Dave Chinnere60896d2013-07-24 15:47:30 +10002589 /*
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002590 * If the inode has an LSN in it, recover the inode only if it's less
Dave Chinner638f44162013-08-30 10:23:45 +10002591 * than the lsn of the transaction we are replaying. Note: we still
2592 * need to replay an owner change even though the inode is more recent
2593 * than the transaction as there is no guarantee that all the btree
2594 * blocks are more recent than this transaction, too.
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002595 */
2596 if (dip->di_version >= 3) {
2597 xfs_lsn_t lsn = be64_to_cpu(dip->di_lsn);
2598
2599 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
2600 trace_xfs_log_recover_inode_skip(log, in_f);
2601 error = 0;
Dave Chinner638f44162013-08-30 10:23:45 +10002602 goto out_owner_change;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002603 }
2604 }
2605
2606 /*
Dave Chinnere60896d2013-07-24 15:47:30 +10002607 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
2608 * are transactional and if ordering is necessary we can determine that
2609 * more accurately by the LSN field in the V3 inode core. Don't trust
2610 * the inode versions we might be changing them here - use the
2611 * superblock flag to determine whether we need to look at di_flushiter
2612 * to skip replay when the on disk inode is newer than the log one
2613 */
2614 if (!xfs_sb_version_hascrc(&mp->m_sb) &&
2615 dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616 /*
2617 * Deal with the wrap case, DI_MAX_FLUSH is less
2618 * than smaller numbers
2619 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002620 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002621 dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622 /* do nothing */
2623 } else {
Dave Chinner9abbc532010-04-13 15:06:46 +10002624 trace_xfs_log_recover_inode_skip(log, in_f);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002625 error = 0;
Dave Chinner638f44162013-08-30 10:23:45 +10002626 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 }
2628 }
Dave Chinnere60896d2013-07-24 15:47:30 +10002629
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630 /* Take the opportunity to reset the flush iteration count */
2631 dicp->di_flushiter = 0;
2632
Al Viroabbede12011-07-26 02:31:30 -04002633 if (unlikely(S_ISREG(dicp->di_mode))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2635 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002636 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637 XFS_ERRLEVEL_LOW, mp, dicp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002638 xfs_alert(mp,
2639 "%s: Bad regular inode log record, rec ptr 0x%p, "
2640 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2641 __func__, item, dip, bp, in_f->ilf_ino);
Dave Chinner24513372014-06-25 14:58:08 +10002642 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002643 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644 }
Al Viroabbede12011-07-26 02:31:30 -04002645 } else if (unlikely(S_ISDIR(dicp->di_mode))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2647 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2648 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002649 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650 XFS_ERRLEVEL_LOW, mp, dicp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002651 xfs_alert(mp,
2652 "%s: Bad dir inode log record, rec ptr 0x%p, "
2653 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2654 __func__, item, dip, bp, in_f->ilf_ino);
Dave Chinner24513372014-06-25 14:58:08 +10002655 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002656 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657 }
2658 }
2659 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002660 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661 XFS_ERRLEVEL_LOW, mp, dicp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002662 xfs_alert(mp,
2663 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2664 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2665 __func__, item, dip, bp, in_f->ilf_ino,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 dicp->di_nextents + dicp->di_anextents,
2667 dicp->di_nblocks);
Dave Chinner24513372014-06-25 14:58:08 +10002668 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002669 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670 }
2671 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002672 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673 XFS_ERRLEVEL_LOW, mp, dicp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002674 xfs_alert(mp,
2675 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2676 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__,
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002677 item, dip, bp, in_f->ilf_ino, dicp->di_forkoff);
Dave Chinner24513372014-06-25 14:58:08 +10002678 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002679 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680 }
Christoph Hellwig93848a92013-04-03 16:11:17 +11002681 isize = xfs_icdinode_size(dicp->di_version);
2682 if (unlikely(item->ri_buf[1].i_len > isize)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002683 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002684 XFS_ERRLEVEL_LOW, mp, dicp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002685 xfs_alert(mp,
2686 "%s: Bad inode log record length %d, rec ptr 0x%p",
2687 __func__, item->ri_buf[1].i_len, item);
Dave Chinner24513372014-06-25 14:58:08 +10002688 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002689 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690 }
2691
2692 /* The core is in in-core format */
Christoph Hellwig93848a92013-04-03 16:11:17 +11002693 xfs_dinode_to_disk(dip, dicp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002694
2695 /* the rest is in on-disk format */
Christoph Hellwig93848a92013-04-03 16:11:17 +11002696 if (item->ri_buf[1].i_len > isize) {
2697 memcpy((char *)dip + isize,
2698 item->ri_buf[1].i_addr + isize,
2699 item->ri_buf[1].i_len - isize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700 }
2701
2702 fields = in_f->ilf_fields;
2703 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2704 case XFS_ILOG_DEV:
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002705 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706 break;
2707 case XFS_ILOG_UUID:
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002708 memcpy(XFS_DFORK_DPTR(dip),
2709 &in_f->ilf_u.ilfu_uuid,
2710 sizeof(uuid_t));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711 break;
2712 }
2713
2714 if (in_f->ilf_size == 2)
Dave Chinner638f44162013-08-30 10:23:45 +10002715 goto out_owner_change;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716 len = item->ri_buf[2].i_len;
2717 src = item->ri_buf[2].i_addr;
2718 ASSERT(in_f->ilf_size <= 4);
2719 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2720 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2721 (len == in_f->ilf_dsize));
2722
2723 switch (fields & XFS_ILOG_DFORK) {
2724 case XFS_ILOG_DDATA:
2725 case XFS_ILOG_DEXT:
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002726 memcpy(XFS_DFORK_DPTR(dip), src, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727 break;
2728
2729 case XFS_ILOG_DBROOT:
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002730 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002731 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732 XFS_DFORK_DSIZE(dip, mp));
2733 break;
2734
2735 default:
2736 /*
2737 * There are no data fork flags set.
2738 */
2739 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2740 break;
2741 }
2742
2743 /*
2744 * If we logged any attribute data, recover it. There may or
2745 * may not have been any other non-core data logged in this
2746 * transaction.
2747 */
2748 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2749 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2750 attr_index = 3;
2751 } else {
2752 attr_index = 2;
2753 }
2754 len = item->ri_buf[attr_index].i_len;
2755 src = item->ri_buf[attr_index].i_addr;
2756 ASSERT(len == in_f->ilf_asize);
2757
2758 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2759 case XFS_ILOG_ADATA:
2760 case XFS_ILOG_AEXT:
2761 dest = XFS_DFORK_APTR(dip);
2762 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2763 memcpy(dest, src, len);
2764 break;
2765
2766 case XFS_ILOG_ABROOT:
2767 dest = XFS_DFORK_APTR(dip);
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002768 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
2769 len, (xfs_bmdr_block_t*)dest,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770 XFS_DFORK_ASIZE(dip, mp));
2771 break;
2772
2773 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002774 xfs_warn(log->l_mp, "%s: Invalid flag", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002775 ASSERT(0);
Dave Chinner24513372014-06-25 14:58:08 +10002776 error = -EIO;
Dave Chinner638f44162013-08-30 10:23:45 +10002777 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778 }
2779 }
2780
Dave Chinner638f44162013-08-30 10:23:45 +10002781out_owner_change:
2782 if (in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER))
2783 error = xfs_recover_inode_owner_change(mp, dip, in_f,
2784 buffer_list);
Christoph Hellwig93848a92013-04-03 16:11:17 +11002785 /* re-generate the checksum. */
2786 xfs_dinode_calc_crc(log->l_mp, dip);
2787
Dave Chinnerebad8612010-09-22 10:47:20 +10002788 ASSERT(bp->b_target->bt_mount == mp);
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02002789 bp->b_iodone = xlog_recover_iodone;
Christoph Hellwig43ff2122012-04-23 15:58:39 +10002790 xfs_buf_delwri_queue(bp, buffer_list);
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002791
2792out_release:
Christoph Hellwig61551f12011-08-23 08:28:06 +00002793 xfs_buf_relse(bp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002794error:
2795 if (need_free)
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002796 kmem_free(in_f);
Eric Sandeenb474c7a2014-06-22 15:04:54 +10002797 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798}
2799
2800/*
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002801 * Recover QUOTAOFF records. We simply make a note of it in the xlog
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2803 * of that type.
2804 */
2805STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002806xlog_recover_quotaoff_pass1(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002807 struct xlog *log,
2808 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002809{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002810 xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811 ASSERT(qoff_f);
2812
2813 /*
2814 * The logitem format's flag tells us if this was user quotaoff,
Nathan Scott77a7cce2006-01-11 15:35:57 +11002815 * group/project quotaoff or both.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816 */
2817 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2818 log->l_quotaoffs_flag |= XFS_DQ_USER;
Nathan Scott77a7cce2006-01-11 15:35:57 +11002819 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2820 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2822 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2823
Eric Sandeend99831f2014-06-22 15:03:54 +10002824 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825}
2826
2827/*
2828 * Recover a dquot record
2829 */
2830STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002831xlog_recover_dquot_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002832 struct xlog *log,
2833 struct list_head *buffer_list,
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002834 struct xlog_recover_item *item,
2835 xfs_lsn_t current_lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002837 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838 xfs_buf_t *bp;
2839 struct xfs_disk_dquot *ddq, *recddq;
2840 int error;
2841 xfs_dq_logformat_t *dq_f;
2842 uint type;
2843
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844
2845 /*
2846 * Filesystems are required to send in quota flags at mount time.
2847 */
2848 if (mp->m_qflags == 0)
Eric Sandeend99831f2014-06-22 15:03:54 +10002849 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002850
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002851 recddq = item->ri_buf[1].i_addr;
2852 if (recddq == NULL) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002853 xfs_alert(log->l_mp, "NULL dquot in %s.", __func__);
Dave Chinner24513372014-06-25 14:58:08 +10002854 return -EIO;
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002855 }
Jan Rekorajski8ec6dba2009-11-16 11:57:02 +00002856 if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002857 xfs_alert(log->l_mp, "dquot too small (%d) in %s.",
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002858 item->ri_buf[1].i_len, __func__);
Dave Chinner24513372014-06-25 14:58:08 +10002859 return -EIO;
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002860 }
2861
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862 /*
2863 * This type of quotas was turned off, so ignore this record.
2864 */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002865 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002866 ASSERT(type);
2867 if (log->l_quotaoffs_flag & type)
Eric Sandeend99831f2014-06-22 15:03:54 +10002868 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002869
2870 /*
2871 * At this point we know that quota was _not_ turned off.
2872 * Since the mount flags are not indicating to us otherwise, this
2873 * must mean that quota is on, and the dquot needs to be replayed.
2874 * Remember that we may not have fully recovered the superblock yet,
2875 * so we can't do the usual trick of looking at the SB quota bits.
2876 *
2877 * The other possibility, of course, is that the quota subsystem was
2878 * removed since the last mount - ENOSYS.
2879 */
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002880 dq_f = item->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881 ASSERT(dq_f);
Dave Chinner9aede1d2013-10-15 09:17:52 +11002882 error = xfs_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002883 "xlog_recover_dquot_pass2 (log copy)");
2884 if (error)
Dave Chinner24513372014-06-25 14:58:08 +10002885 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002886 ASSERT(dq_f->qlf_len == 1);
2887
Dave Chinnerad3714b2014-08-04 12:59:31 +10002888 /*
2889 * At this point we are assuming that the dquots have been allocated
2890 * and hence the buffer has valid dquots stamped in it. It should,
2891 * therefore, pass verifier validation. If the dquot is bad, then the
2892 * we'll return an error here, so we don't need to specifically check
2893 * the dquot in the buffer after the verifier has run.
2894 */
Dave Chinner7ca790a2012-04-23 15:58:55 +10002895 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dq_f->qlf_blkno,
Dave Chinnerc3f8fc72012-11-12 22:54:01 +11002896 XFS_FSB_TO_BB(mp, dq_f->qlf_len), 0, &bp,
Dave Chinnerad3714b2014-08-04 12:59:31 +10002897 &xfs_dquot_buf_ops);
Dave Chinner7ca790a2012-04-23 15:58:55 +10002898 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002899 return error;
Dave Chinner7ca790a2012-04-23 15:58:55 +10002900
Linus Torvalds1da177e2005-04-16 15:20:36 -07002901 ASSERT(bp);
Christoph Hellwig88ee2df2015-06-22 09:44:29 +10002902 ddq = xfs_buf_offset(bp, dq_f->qlf_boffset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002903
2904 /*
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002905 * If the dquot has an LSN in it, recover the dquot only if it's less
2906 * than the lsn of the transaction we are replaying.
2907 */
2908 if (xfs_sb_version_hascrc(&mp->m_sb)) {
2909 struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddq;
2910 xfs_lsn_t lsn = be64_to_cpu(dqb->dd_lsn);
2911
2912 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
2913 goto out_release;
2914 }
2915 }
2916
Linus Torvalds1da177e2005-04-16 15:20:36 -07002917 memcpy(ddq, recddq, item->ri_buf[1].i_len);
Dave Chinner6fcdc592013-06-03 15:28:46 +10002918 if (xfs_sb_version_hascrc(&mp->m_sb)) {
2919 xfs_update_cksum((char *)ddq, sizeof(struct xfs_dqblk),
2920 XFS_DQUOT_CRC_OFF);
2921 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002922
2923 ASSERT(dq_f->qlf_size == 2);
Dave Chinnerebad8612010-09-22 10:47:20 +10002924 ASSERT(bp->b_target->bt_mount == mp);
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02002925 bp->b_iodone = xlog_recover_iodone;
Christoph Hellwig43ff2122012-04-23 15:58:39 +10002926 xfs_buf_delwri_queue(bp, buffer_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002927
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002928out_release:
2929 xfs_buf_relse(bp);
2930 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931}
2932
2933/*
2934 * This routine is called to create an in-core extent free intent
2935 * item from the efi format structure which was logged on disk.
2936 * It allocates an in-core efi, copies the extents from the format
2937 * structure into it, and adds the efi to the AIL with the given
2938 * LSN.
2939 */
Tim Shimmin6d192a92006-06-09 14:55:38 +10002940STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002941xlog_recover_efi_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002942 struct xlog *log,
2943 struct xlog_recover_item *item,
2944 xfs_lsn_t lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002945{
Brian Fostere32a1d12015-08-19 09:52:21 +10002946 int error;
2947 struct xfs_mount *mp = log->l_mp;
2948 struct xfs_efi_log_item *efip;
2949 struct xfs_efi_log_format *efi_formatp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002950
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002951 efi_formatp = item->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002952
Linus Torvalds1da177e2005-04-16 15:20:36 -07002953 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
Brian Fostere32a1d12015-08-19 09:52:21 +10002954 error = xfs_efi_copy_format(&item->ri_buf[0], &efip->efi_format);
2955 if (error) {
Tim Shimmin6d192a92006-06-09 14:55:38 +10002956 xfs_efi_item_free(efip);
2957 return error;
2958 }
Dave Chinnerb199c8a2010-12-20 11:59:49 +11002959 atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002960
David Chinnera9c21c12008-10-30 17:39:35 +11002961 spin_lock(&log->l_ailp->xa_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002962 /*
Brian Fostere32a1d12015-08-19 09:52:21 +10002963 * The EFI has two references. One for the EFD and one for EFI to ensure
2964 * it makes it into the AIL. Insert the EFI into the AIL directly and
2965 * drop the EFI reference. Note that xfs_trans_ail_update() drops the
2966 * AIL lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002967 */
Dave Chinnere6059942010-12-20 12:34:26 +11002968 xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn);
Brian Fostere32a1d12015-08-19 09:52:21 +10002969 xfs_efi_release(efip);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002970 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002971}
2972
2973
2974/*
Brian Fostere32a1d12015-08-19 09:52:21 +10002975 * This routine is called when an EFD format structure is found in a committed
2976 * transaction in the log. Its purpose is to cancel the corresponding EFI if it
2977 * was still in the log. To do this it searches the AIL for the EFI with an id
2978 * equal to that in the EFD format structure. If we find it we drop the EFD
2979 * reference, which removes the EFI from the AIL and frees it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002980 */
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002981STATIC int
2982xlog_recover_efd_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002983 struct xlog *log,
2984 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002985{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002986 xfs_efd_log_format_t *efd_formatp;
2987 xfs_efi_log_item_t *efip = NULL;
2988 xfs_log_item_t *lip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989 __uint64_t efi_id;
David Chinner27d8d5f2008-10-30 17:38:39 +11002990 struct xfs_ail_cursor cur;
David Chinner783a2f62008-10-30 17:39:58 +11002991 struct xfs_ail *ailp = log->l_ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002992
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002993 efd_formatp = item->ri_buf[0].i_addr;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002994 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
2995 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
2996 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
2997 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998 efi_id = efd_formatp->efd_efi_id;
2999
3000 /*
Brian Fostere32a1d12015-08-19 09:52:21 +10003001 * Search for the EFI with the id in the EFD format structure in the
3002 * AIL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003003 */
David Chinnera9c21c12008-10-30 17:39:35 +11003004 spin_lock(&ailp->xa_lock);
3005 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003006 while (lip != NULL) {
3007 if (lip->li_type == XFS_LI_EFI) {
3008 efip = (xfs_efi_log_item_t *)lip;
3009 if (efip->efi_format.efi_id == efi_id) {
3010 /*
Brian Fostere32a1d12015-08-19 09:52:21 +10003011 * Drop the EFD reference to the EFI. This
3012 * removes the EFI from the AIL and frees it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003013 */
Brian Fostere32a1d12015-08-19 09:52:21 +10003014 spin_unlock(&ailp->xa_lock);
3015 xfs_efi_release(efip);
David Chinnera9c21c12008-10-30 17:39:35 +11003016 spin_lock(&ailp->xa_lock);
David Chinner27d8d5f2008-10-30 17:38:39 +11003017 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003018 }
3019 }
David Chinnera9c21c12008-10-30 17:39:35 +11003020 lip = xfs_trans_ail_cursor_next(ailp, &cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021 }
Brian Fostere32a1d12015-08-19 09:52:21 +10003022
Eric Sandeene4a1e292014-04-14 19:06:05 +10003023 xfs_trans_ail_cursor_done(&cur);
David Chinnera9c21c12008-10-30 17:39:35 +11003024 spin_unlock(&ailp->xa_lock);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003025
3026 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003027}
3028
3029/*
Dave Chinner28c8e412013-06-27 16:04:55 +10003030 * This routine is called when an inode create format structure is found in a
3031 * committed transaction in the log. It's purpose is to initialise the inodes
3032 * being allocated on disk. This requires us to get inode cluster buffers that
3033 * match the range to be intialised, stamped with inode templates and written
3034 * by delayed write so that subsequent modifications will hit the cached buffer
3035 * and only need writing out at the end of recovery.
3036 */
3037STATIC int
3038xlog_recover_do_icreate_pass2(
3039 struct xlog *log,
3040 struct list_head *buffer_list,
3041 xlog_recover_item_t *item)
3042{
3043 struct xfs_mount *mp = log->l_mp;
3044 struct xfs_icreate_log *icl;
3045 xfs_agnumber_t agno;
3046 xfs_agblock_t agbno;
3047 unsigned int count;
3048 unsigned int isize;
3049 xfs_agblock_t length;
Brian Fosterfc0d1652015-08-19 09:59:38 +10003050 int blks_per_cluster;
3051 int bb_per_cluster;
3052 int cancel_count;
3053 int nbufs;
3054 int i;
Dave Chinner28c8e412013-06-27 16:04:55 +10003055
3056 icl = (struct xfs_icreate_log *)item->ri_buf[0].i_addr;
3057 if (icl->icl_type != XFS_LI_ICREATE) {
3058 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad type");
Dave Chinner24513372014-06-25 14:58:08 +10003059 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003060 }
3061
3062 if (icl->icl_size != 1) {
3063 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad icl size");
Dave Chinner24513372014-06-25 14:58:08 +10003064 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003065 }
3066
3067 agno = be32_to_cpu(icl->icl_ag);
3068 if (agno >= mp->m_sb.sb_agcount) {
3069 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agno");
Dave Chinner24513372014-06-25 14:58:08 +10003070 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003071 }
3072 agbno = be32_to_cpu(icl->icl_agbno);
3073 if (!agbno || agbno == NULLAGBLOCK || agbno >= mp->m_sb.sb_agblocks) {
3074 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agbno");
Dave Chinner24513372014-06-25 14:58:08 +10003075 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003076 }
3077 isize = be32_to_cpu(icl->icl_isize);
3078 if (isize != mp->m_sb.sb_inodesize) {
3079 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad isize");
Dave Chinner24513372014-06-25 14:58:08 +10003080 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003081 }
3082 count = be32_to_cpu(icl->icl_count);
3083 if (!count) {
3084 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count");
Dave Chinner24513372014-06-25 14:58:08 +10003085 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003086 }
3087 length = be32_to_cpu(icl->icl_length);
3088 if (!length || length >= mp->m_sb.sb_agblocks) {
3089 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad length");
Dave Chinner24513372014-06-25 14:58:08 +10003090 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003091 }
3092
Brian Foster7f43c902015-05-29 09:06:30 +10003093 /*
3094 * The inode chunk is either full or sparse and we only support
3095 * m_ialloc_min_blks sized sparse allocations at this time.
3096 */
3097 if (length != mp->m_ialloc_blks &&
3098 length != mp->m_ialloc_min_blks) {
3099 xfs_warn(log->l_mp,
3100 "%s: unsupported chunk length", __FUNCTION__);
3101 return -EINVAL;
3102 }
3103
3104 /* verify inode count is consistent with extent length */
3105 if ((count >> mp->m_sb.sb_inopblog) != length) {
3106 xfs_warn(log->l_mp,
3107 "%s: inconsistent inode count and chunk length",
3108 __FUNCTION__);
Dave Chinner24513372014-06-25 14:58:08 +10003109 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003110 }
3111
3112 /*
Brian Fosterfc0d1652015-08-19 09:59:38 +10003113 * The icreate transaction can cover multiple cluster buffers and these
3114 * buffers could have been freed and reused. Check the individual
3115 * buffers for cancellation so we don't overwrite anything written after
3116 * a cancellation.
Dave Chinner28c8e412013-06-27 16:04:55 +10003117 */
Brian Fosterfc0d1652015-08-19 09:59:38 +10003118 blks_per_cluster = xfs_icluster_size_fsb(mp);
3119 bb_per_cluster = XFS_FSB_TO_BB(mp, blks_per_cluster);
3120 nbufs = length / blks_per_cluster;
3121 for (i = 0, cancel_count = 0; i < nbufs; i++) {
3122 xfs_daddr_t daddr;
Dave Chinner28c8e412013-06-27 16:04:55 +10003123
Brian Fosterfc0d1652015-08-19 09:59:38 +10003124 daddr = XFS_AGB_TO_DADDR(mp, agno,
3125 agbno + i * blks_per_cluster);
3126 if (xlog_check_buffer_cancelled(log, daddr, bb_per_cluster, 0))
3127 cancel_count++;
3128 }
3129
3130 /*
3131 * We currently only use icreate for a single allocation at a time. This
3132 * means we should expect either all or none of the buffers to be
3133 * cancelled. Be conservative and skip replay if at least one buffer is
3134 * cancelled, but warn the user that something is awry if the buffers
3135 * are not consistent.
3136 *
3137 * XXX: This must be refined to only skip cancelled clusters once we use
3138 * icreate for multiple chunk allocations.
3139 */
3140 ASSERT(!cancel_count || cancel_count == nbufs);
3141 if (cancel_count) {
3142 if (cancel_count != nbufs)
3143 xfs_warn(mp,
3144 "WARNING: partial inode chunk cancellation, skipped icreate.");
Brian Foster78d57e42015-08-19 09:58:48 +10003145 trace_xfs_log_recover_icreate_cancel(log, icl);
Dave Chinner28c8e412013-06-27 16:04:55 +10003146 return 0;
Brian Foster78d57e42015-08-19 09:58:48 +10003147 }
Dave Chinner28c8e412013-06-27 16:04:55 +10003148
Brian Foster78d57e42015-08-19 09:58:48 +10003149 trace_xfs_log_recover_icreate_recover(log, icl);
Brian Fosterfc0d1652015-08-19 09:59:38 +10003150 return xfs_ialloc_inode_init(mp, NULL, buffer_list, count, agno, agbno,
3151 length, be32_to_cpu(icl->icl_gen));
Dave Chinner28c8e412013-06-27 16:04:55 +10003152}
3153
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003154STATIC void
3155xlog_recover_buffer_ra_pass2(
3156 struct xlog *log,
3157 struct xlog_recover_item *item)
3158{
3159 struct xfs_buf_log_format *buf_f = item->ri_buf[0].i_addr;
3160 struct xfs_mount *mp = log->l_mp;
3161
Dave Chinner84a5b732013-08-27 08:10:53 +10003162 if (xlog_peek_buffer_cancelled(log, buf_f->blf_blkno,
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003163 buf_f->blf_len, buf_f->blf_flags)) {
3164 return;
3165 }
3166
3167 xfs_buf_readahead(mp->m_ddev_targp, buf_f->blf_blkno,
3168 buf_f->blf_len, NULL);
3169}
3170
3171STATIC void
3172xlog_recover_inode_ra_pass2(
3173 struct xlog *log,
3174 struct xlog_recover_item *item)
3175{
3176 struct xfs_inode_log_format ilf_buf;
3177 struct xfs_inode_log_format *ilfp;
3178 struct xfs_mount *mp = log->l_mp;
3179 int error;
3180
3181 if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) {
3182 ilfp = item->ri_buf[0].i_addr;
3183 } else {
3184 ilfp = &ilf_buf;
3185 memset(ilfp, 0, sizeof(*ilfp));
3186 error = xfs_inode_item_format_convert(&item->ri_buf[0], ilfp);
3187 if (error)
3188 return;
3189 }
3190
Dave Chinner84a5b732013-08-27 08:10:53 +10003191 if (xlog_peek_buffer_cancelled(log, ilfp->ilf_blkno, ilfp->ilf_len, 0))
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003192 return;
3193
3194 xfs_buf_readahead(mp->m_ddev_targp, ilfp->ilf_blkno,
Dave Chinnerd8914002013-08-27 11:39:37 +10003195 ilfp->ilf_len, &xfs_inode_buf_ra_ops);
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003196}
3197
3198STATIC void
3199xlog_recover_dquot_ra_pass2(
3200 struct xlog *log,
3201 struct xlog_recover_item *item)
3202{
3203 struct xfs_mount *mp = log->l_mp;
3204 struct xfs_disk_dquot *recddq;
3205 struct xfs_dq_logformat *dq_f;
3206 uint type;
3207
3208
3209 if (mp->m_qflags == 0)
3210 return;
3211
3212 recddq = item->ri_buf[1].i_addr;
3213 if (recddq == NULL)
3214 return;
3215 if (item->ri_buf[1].i_len < sizeof(struct xfs_disk_dquot))
3216 return;
3217
3218 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
3219 ASSERT(type);
3220 if (log->l_quotaoffs_flag & type)
3221 return;
3222
3223 dq_f = item->ri_buf[0].i_addr;
3224 ASSERT(dq_f);
3225 ASSERT(dq_f->qlf_len == 1);
3226
3227 xfs_buf_readahead(mp->m_ddev_targp, dq_f->qlf_blkno,
Dave Chinner0f0d3342013-08-27 13:25:43 +10003228 XFS_FSB_TO_BB(mp, dq_f->qlf_len), NULL);
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003229}
3230
3231STATIC void
3232xlog_recover_ra_pass2(
3233 struct xlog *log,
3234 struct xlog_recover_item *item)
3235{
3236 switch (ITEM_TYPE(item)) {
3237 case XFS_LI_BUF:
3238 xlog_recover_buffer_ra_pass2(log, item);
3239 break;
3240 case XFS_LI_INODE:
3241 xlog_recover_inode_ra_pass2(log, item);
3242 break;
3243 case XFS_LI_DQUOT:
3244 xlog_recover_dquot_ra_pass2(log, item);
3245 break;
3246 case XFS_LI_EFI:
3247 case XFS_LI_EFD:
3248 case XFS_LI_QUOTAOFF:
3249 default:
3250 break;
3251 }
3252}
3253
Linus Torvalds1da177e2005-04-16 15:20:36 -07003254STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003255xlog_recover_commit_pass1(
Mark Tinguelyad223e62012-06-14 09:22:15 -05003256 struct xlog *log,
3257 struct xlog_recover *trans,
3258 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003259{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003260 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1);
Christoph Hellwigd0450942010-12-01 22:06:23 +00003261
3262 switch (ITEM_TYPE(item)) {
3263 case XFS_LI_BUF:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003264 return xlog_recover_buffer_pass1(log, item);
Christoph Hellwigd0450942010-12-01 22:06:23 +00003265 case XFS_LI_QUOTAOFF:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003266 return xlog_recover_quotaoff_pass1(log, item);
3267 case XFS_LI_INODE:
3268 case XFS_LI_EFI:
3269 case XFS_LI_EFD:
3270 case XFS_LI_DQUOT:
Dave Chinner28c8e412013-06-27 16:04:55 +10003271 case XFS_LI_ICREATE:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003272 /* nothing to do in pass 1 */
3273 return 0;
Christoph Hellwigd0450942010-12-01 22:06:23 +00003274 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003275 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
3276 __func__, ITEM_TYPE(item));
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003277 ASSERT(0);
Dave Chinner24513372014-06-25 14:58:08 +10003278 return -EIO;
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003279 }
3280}
3281
3282STATIC int
3283xlog_recover_commit_pass2(
Mark Tinguelyad223e62012-06-14 09:22:15 -05003284 struct xlog *log,
3285 struct xlog_recover *trans,
3286 struct list_head *buffer_list,
3287 struct xlog_recover_item *item)
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003288{
3289 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2);
3290
3291 switch (ITEM_TYPE(item)) {
3292 case XFS_LI_BUF:
Dave Chinner50d5c8d2013-08-28 21:22:47 +10003293 return xlog_recover_buffer_pass2(log, buffer_list, item,
3294 trans->r_lsn);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003295 case XFS_LI_INODE:
Dave Chinner50d5c8d2013-08-28 21:22:47 +10003296 return xlog_recover_inode_pass2(log, buffer_list, item,
3297 trans->r_lsn);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003298 case XFS_LI_EFI:
3299 return xlog_recover_efi_pass2(log, item, trans->r_lsn);
3300 case XFS_LI_EFD:
3301 return xlog_recover_efd_pass2(log, item);
3302 case XFS_LI_DQUOT:
Dave Chinner50d5c8d2013-08-28 21:22:47 +10003303 return xlog_recover_dquot_pass2(log, buffer_list, item,
3304 trans->r_lsn);
Dave Chinner28c8e412013-06-27 16:04:55 +10003305 case XFS_LI_ICREATE:
3306 return xlog_recover_do_icreate_pass2(log, buffer_list, item);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003307 case XFS_LI_QUOTAOFF:
3308 /* nothing to do in pass2 */
3309 return 0;
3310 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003311 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
3312 __func__, ITEM_TYPE(item));
Christoph Hellwigd0450942010-12-01 22:06:23 +00003313 ASSERT(0);
Dave Chinner24513372014-06-25 14:58:08 +10003314 return -EIO;
Christoph Hellwigd0450942010-12-01 22:06:23 +00003315 }
3316}
3317
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003318STATIC int
3319xlog_recover_items_pass2(
3320 struct xlog *log,
3321 struct xlog_recover *trans,
3322 struct list_head *buffer_list,
3323 struct list_head *item_list)
3324{
3325 struct xlog_recover_item *item;
3326 int error = 0;
3327
3328 list_for_each_entry(item, item_list, ri_list) {
3329 error = xlog_recover_commit_pass2(log, trans,
3330 buffer_list, item);
3331 if (error)
3332 return error;
3333 }
3334
3335 return error;
3336}
3337
Christoph Hellwigd0450942010-12-01 22:06:23 +00003338/*
3339 * Perform the transaction.
3340 *
3341 * If the transaction modifies a buffer or inode, do it now. Otherwise,
3342 * EFIs and EFDs get queued up by adding entries into the AIL for them.
3343 */
3344STATIC int
3345xlog_recover_commit_trans(
Mark Tinguelyad223e62012-06-14 09:22:15 -05003346 struct xlog *log,
Christoph Hellwigd0450942010-12-01 22:06:23 +00003347 struct xlog_recover *trans,
3348 int pass)
3349{
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003350 int error = 0;
3351 int error2;
3352 int items_queued = 0;
3353 struct xlog_recover_item *item;
3354 struct xlog_recover_item *next;
3355 LIST_HEAD (buffer_list);
3356 LIST_HEAD (ra_list);
3357 LIST_HEAD (done_list);
3358
3359 #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100
Linus Torvalds1da177e2005-04-16 15:20:36 -07003360
Dave Chinnerf0a76952010-01-11 11:49:57 +00003361 hlist_del(&trans->r_list);
Christoph Hellwigd0450942010-12-01 22:06:23 +00003362
3363 error = xlog_recover_reorder_trans(log, trans, pass);
3364 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003365 return error;
Christoph Hellwigd0450942010-12-01 22:06:23 +00003366
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003367 list_for_each_entry_safe(item, next, &trans->r_itemq, ri_list) {
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003368 switch (pass) {
3369 case XLOG_RECOVER_PASS1:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003370 error = xlog_recover_commit_pass1(log, trans, item);
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003371 break;
3372 case XLOG_RECOVER_PASS2:
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003373 xlog_recover_ra_pass2(log, item);
3374 list_move_tail(&item->ri_list, &ra_list);
3375 items_queued++;
3376 if (items_queued >= XLOG_RECOVER_COMMIT_QUEUE_MAX) {
3377 error = xlog_recover_items_pass2(log, trans,
3378 &buffer_list, &ra_list);
3379 list_splice_tail_init(&ra_list, &done_list);
3380 items_queued = 0;
3381 }
3382
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003383 break;
3384 default:
3385 ASSERT(0);
3386 }
3387
Christoph Hellwigd0450942010-12-01 22:06:23 +00003388 if (error)
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003389 goto out;
Christoph Hellwigd0450942010-12-01 22:06:23 +00003390 }
3391
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003392out:
3393 if (!list_empty(&ra_list)) {
3394 if (!error)
3395 error = xlog_recover_items_pass2(log, trans,
3396 &buffer_list, &ra_list);
3397 list_splice_tail_init(&ra_list, &done_list);
3398 }
3399
3400 if (!list_empty(&done_list))
3401 list_splice_init(&done_list, &trans->r_itemq);
3402
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003403 error2 = xfs_buf_delwri_submit(&buffer_list);
3404 return error ? error : error2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003405}
3406
Dave Chinner76560662014-09-29 09:45:42 +10003407STATIC void
3408xlog_recover_add_item(
3409 struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003410{
Dave Chinner76560662014-09-29 09:45:42 +10003411 xlog_recover_item_t *item;
3412
3413 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
3414 INIT_LIST_HEAD(&item->ri_list);
3415 list_add_tail(&item->ri_list, head);
3416}
3417
3418STATIC int
3419xlog_recover_add_to_cont_trans(
3420 struct xlog *log,
3421 struct xlog_recover *trans,
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10003422 char *dp,
Dave Chinner76560662014-09-29 09:45:42 +10003423 int len)
3424{
3425 xlog_recover_item_t *item;
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10003426 char *ptr, *old_ptr;
Dave Chinner76560662014-09-29 09:45:42 +10003427 int old_len;
3428
Brian Foster89cebc842015-07-29 11:51:10 +10003429 /*
3430 * If the transaction is empty, the header was split across this and the
3431 * previous record. Copy the rest of the header.
3432 */
Dave Chinner76560662014-09-29 09:45:42 +10003433 if (list_empty(&trans->r_itemq)) {
Brian Foster89cebc842015-07-29 11:51:10 +10003434 ASSERT(len < sizeof(struct xfs_trans_header));
3435 if (len > sizeof(struct xfs_trans_header)) {
3436 xfs_warn(log->l_mp, "%s: bad header length", __func__);
3437 return -EIO;
3438 }
3439
Dave Chinner76560662014-09-29 09:45:42 +10003440 xlog_recover_add_item(&trans->r_itemq);
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10003441 ptr = (char *)&trans->r_theader +
Brian Foster89cebc842015-07-29 11:51:10 +10003442 sizeof(struct xfs_trans_header) - len;
Dave Chinner76560662014-09-29 09:45:42 +10003443 memcpy(ptr, dp, len);
3444 return 0;
3445 }
Brian Foster89cebc842015-07-29 11:51:10 +10003446
Dave Chinner76560662014-09-29 09:45:42 +10003447 /* take the tail entry */
3448 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
3449
3450 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
3451 old_len = item->ri_buf[item->ri_cnt-1].i_len;
3452
3453 ptr = kmem_realloc(old_ptr, len+old_len, old_len, KM_SLEEP);
3454 memcpy(&ptr[old_len], dp, len);
3455 item->ri_buf[item->ri_cnt-1].i_len += len;
3456 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
3457 trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003458 return 0;
3459}
3460
3461/*
Dave Chinner76560662014-09-29 09:45:42 +10003462 * The next region to add is the start of a new region. It could be
3463 * a whole region or it could be the first part of a new region. Because
3464 * of this, the assumption here is that the type and size fields of all
3465 * format structures fit into the first 32 bits of the structure.
3466 *
3467 * This works because all regions must be 32 bit aligned. Therefore, we
3468 * either have both fields or we have neither field. In the case we have
3469 * neither field, the data part of the region is zero length. We only have
3470 * a log_op_header and can throw away the header since a new one will appear
3471 * later. If we have at least 4 bytes, then we can determine how many regions
3472 * will appear in the current log item.
3473 */
3474STATIC int
3475xlog_recover_add_to_trans(
3476 struct xlog *log,
3477 struct xlog_recover *trans,
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10003478 char *dp,
Dave Chinner76560662014-09-29 09:45:42 +10003479 int len)
3480{
3481 xfs_inode_log_format_t *in_f; /* any will do */
3482 xlog_recover_item_t *item;
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10003483 char *ptr;
Dave Chinner76560662014-09-29 09:45:42 +10003484
3485 if (!len)
3486 return 0;
3487 if (list_empty(&trans->r_itemq)) {
3488 /* we need to catch log corruptions here */
3489 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
3490 xfs_warn(log->l_mp, "%s: bad header magic number",
3491 __func__);
3492 ASSERT(0);
3493 return -EIO;
3494 }
Brian Foster89cebc842015-07-29 11:51:10 +10003495
3496 if (len > sizeof(struct xfs_trans_header)) {
3497 xfs_warn(log->l_mp, "%s: bad header length", __func__);
3498 ASSERT(0);
3499 return -EIO;
3500 }
3501
3502 /*
3503 * The transaction header can be arbitrarily split across op
3504 * records. If we don't have the whole thing here, copy what we
3505 * do have and handle the rest in the next record.
3506 */
3507 if (len == sizeof(struct xfs_trans_header))
Dave Chinner76560662014-09-29 09:45:42 +10003508 xlog_recover_add_item(&trans->r_itemq);
3509 memcpy(&trans->r_theader, dp, len);
3510 return 0;
3511 }
3512
3513 ptr = kmem_alloc(len, KM_SLEEP);
3514 memcpy(ptr, dp, len);
3515 in_f = (xfs_inode_log_format_t *)ptr;
3516
3517 /* take the tail entry */
3518 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
3519 if (item->ri_total != 0 &&
3520 item->ri_total == item->ri_cnt) {
3521 /* tail item is in use, get a new one */
3522 xlog_recover_add_item(&trans->r_itemq);
3523 item = list_entry(trans->r_itemq.prev,
3524 xlog_recover_item_t, ri_list);
3525 }
3526
3527 if (item->ri_total == 0) { /* first region to be added */
3528 if (in_f->ilf_size == 0 ||
3529 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
3530 xfs_warn(log->l_mp,
3531 "bad number of regions (%d) in inode log format",
3532 in_f->ilf_size);
3533 ASSERT(0);
3534 kmem_free(ptr);
3535 return -EIO;
3536 }
3537
3538 item->ri_total = in_f->ilf_size;
3539 item->ri_buf =
3540 kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
3541 KM_SLEEP);
3542 }
3543 ASSERT(item->ri_total > item->ri_cnt);
3544 /* Description region is ri_buf[0] */
3545 item->ri_buf[item->ri_cnt].i_addr = ptr;
3546 item->ri_buf[item->ri_cnt].i_len = len;
3547 item->ri_cnt++;
3548 trace_xfs_log_recover_item_add(log, trans, item, 0);
3549 return 0;
3550}
Dave Chinnerb818cca2014-09-29 09:45:54 +10003551
Dave Chinner76560662014-09-29 09:45:42 +10003552/*
3553 * Free up any resources allocated by the transaction
3554 *
3555 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
3556 */
3557STATIC void
3558xlog_recover_free_trans(
3559 struct xlog_recover *trans)
3560{
3561 xlog_recover_item_t *item, *n;
3562 int i;
3563
3564 list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
3565 /* Free the regions in the item. */
3566 list_del(&item->ri_list);
3567 for (i = 0; i < item->ri_cnt; i++)
3568 kmem_free(item->ri_buf[i].i_addr);
3569 /* Free the item itself */
3570 kmem_free(item->ri_buf);
3571 kmem_free(item);
3572 }
3573 /* Free the transaction recover structure */
3574 kmem_free(trans);
3575}
3576
Dave Chinnere9131e52014-09-29 09:45:18 +10003577/*
3578 * On error or completion, trans is freed.
3579 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003580STATIC int
Dave Chinnereeb11682014-09-29 09:45:03 +10003581xlog_recovery_process_trans(
3582 struct xlog *log,
3583 struct xlog_recover *trans,
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10003584 char *dp,
Dave Chinnereeb11682014-09-29 09:45:03 +10003585 unsigned int len,
3586 unsigned int flags,
3587 int pass)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003588{
Dave Chinnere9131e52014-09-29 09:45:18 +10003589 int error = 0;
3590 bool freeit = false;
Dave Chinnereeb11682014-09-29 09:45:03 +10003591
3592 /* mask off ophdr transaction container flags */
3593 flags &= ~XLOG_END_TRANS;
3594 if (flags & XLOG_WAS_CONT_TRANS)
3595 flags &= ~XLOG_CONTINUE_TRANS;
3596
Dave Chinner88b863d2014-09-29 09:45:32 +10003597 /*
3598 * Callees must not free the trans structure. We'll decide if we need to
3599 * free it or not based on the operation being done and it's result.
3600 */
Dave Chinnereeb11682014-09-29 09:45:03 +10003601 switch (flags) {
3602 /* expected flag values */
3603 case 0:
3604 case XLOG_CONTINUE_TRANS:
3605 error = xlog_recover_add_to_trans(log, trans, dp, len);
3606 break;
3607 case XLOG_WAS_CONT_TRANS:
3608 error = xlog_recover_add_to_cont_trans(log, trans, dp, len);
3609 break;
3610 case XLOG_COMMIT_TRANS:
3611 error = xlog_recover_commit_trans(log, trans, pass);
Dave Chinner88b863d2014-09-29 09:45:32 +10003612 /* success or fail, we are now done with this transaction. */
3613 freeit = true;
Dave Chinnereeb11682014-09-29 09:45:03 +10003614 break;
3615
3616 /* unexpected flag values */
3617 case XLOG_UNMOUNT_TRANS:
Dave Chinnere9131e52014-09-29 09:45:18 +10003618 /* just skip trans */
Dave Chinnereeb11682014-09-29 09:45:03 +10003619 xfs_warn(log->l_mp, "%s: Unmount LR", __func__);
Dave Chinnere9131e52014-09-29 09:45:18 +10003620 freeit = true;
Dave Chinnereeb11682014-09-29 09:45:03 +10003621 break;
3622 case XLOG_START_TRANS:
Dave Chinnereeb11682014-09-29 09:45:03 +10003623 default:
3624 xfs_warn(log->l_mp, "%s: bad flag 0x%x", __func__, flags);
3625 ASSERT(0);
Dave Chinnere9131e52014-09-29 09:45:18 +10003626 error = -EIO;
Dave Chinnereeb11682014-09-29 09:45:03 +10003627 break;
3628 }
Dave Chinnere9131e52014-09-29 09:45:18 +10003629 if (error || freeit)
3630 xlog_recover_free_trans(trans);
Dave Chinnereeb11682014-09-29 09:45:03 +10003631 return error;
3632}
3633
Dave Chinnerb818cca2014-09-29 09:45:54 +10003634/*
3635 * Lookup the transaction recovery structure associated with the ID in the
3636 * current ophdr. If the transaction doesn't exist and the start flag is set in
3637 * the ophdr, then allocate a new transaction for future ID matches to find.
3638 * Either way, return what we found during the lookup - an existing transaction
3639 * or nothing.
3640 */
Dave Chinnereeb11682014-09-29 09:45:03 +10003641STATIC struct xlog_recover *
3642xlog_recover_ophdr_to_trans(
3643 struct hlist_head rhash[],
3644 struct xlog_rec_header *rhead,
3645 struct xlog_op_header *ohead)
3646{
3647 struct xlog_recover *trans;
3648 xlog_tid_t tid;
3649 struct hlist_head *rhp;
3650
3651 tid = be32_to_cpu(ohead->oh_tid);
3652 rhp = &rhash[XLOG_RHASH(tid)];
Dave Chinnerb818cca2014-09-29 09:45:54 +10003653 hlist_for_each_entry(trans, rhp, r_list) {
3654 if (trans->r_log_tid == tid)
3655 return trans;
3656 }
Dave Chinnereeb11682014-09-29 09:45:03 +10003657
3658 /*
Dave Chinnerb818cca2014-09-29 09:45:54 +10003659 * skip over non-start transaction headers - we could be
3660 * processing slack space before the next transaction starts
Dave Chinnereeb11682014-09-29 09:45:03 +10003661 */
Dave Chinnerb818cca2014-09-29 09:45:54 +10003662 if (!(ohead->oh_flags & XLOG_START_TRANS))
3663 return NULL;
3664
3665 ASSERT(be32_to_cpu(ohead->oh_len) == 0);
3666
3667 /*
3668 * This is a new transaction so allocate a new recovery container to
3669 * hold the recovery ops that will follow.
3670 */
3671 trans = kmem_zalloc(sizeof(struct xlog_recover), KM_SLEEP);
3672 trans->r_log_tid = tid;
3673 trans->r_lsn = be64_to_cpu(rhead->h_lsn);
3674 INIT_LIST_HEAD(&trans->r_itemq);
3675 INIT_HLIST_NODE(&trans->r_list);
3676 hlist_add_head(&trans->r_list, rhp);
3677
3678 /*
3679 * Nothing more to do for this ophdr. Items to be added to this new
3680 * transaction will be in subsequent ophdr containers.
3681 */
Dave Chinnereeb11682014-09-29 09:45:03 +10003682 return NULL;
3683}
3684
3685STATIC int
3686xlog_recover_process_ophdr(
3687 struct xlog *log,
3688 struct hlist_head rhash[],
3689 struct xlog_rec_header *rhead,
3690 struct xlog_op_header *ohead,
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10003691 char *dp,
3692 char *end,
Dave Chinnereeb11682014-09-29 09:45:03 +10003693 int pass)
3694{
3695 struct xlog_recover *trans;
Dave Chinnereeb11682014-09-29 09:45:03 +10003696 unsigned int len;
3697
3698 /* Do we understand who wrote this op? */
3699 if (ohead->oh_clientid != XFS_TRANSACTION &&
3700 ohead->oh_clientid != XFS_LOG) {
3701 xfs_warn(log->l_mp, "%s: bad clientid 0x%x",
3702 __func__, ohead->oh_clientid);
3703 ASSERT(0);
3704 return -EIO;
3705 }
3706
3707 /*
3708 * Check the ophdr contains all the data it is supposed to contain.
3709 */
3710 len = be32_to_cpu(ohead->oh_len);
3711 if (dp + len > end) {
3712 xfs_warn(log->l_mp, "%s: bad length 0x%x", __func__, len);
3713 WARN_ON(1);
3714 return -EIO;
3715 }
3716
3717 trans = xlog_recover_ophdr_to_trans(rhash, rhead, ohead);
3718 if (!trans) {
3719 /* nothing to do, so skip over this ophdr */
3720 return 0;
3721 }
3722
Dave Chinnere9131e52014-09-29 09:45:18 +10003723 return xlog_recovery_process_trans(log, trans, dp, len,
3724 ohead->oh_flags, pass);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003725}
3726
3727/*
3728 * There are two valid states of the r_state field. 0 indicates that the
3729 * transaction structure is in a normal state. We have either seen the
3730 * start of the transaction or the last operation we added was not a partial
3731 * operation. If the last operation we added to the transaction was a
3732 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
3733 *
3734 * NOTE: skip LRs with 0 data length.
3735 */
3736STATIC int
3737xlog_recover_process_data(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003738 struct xlog *log,
Dave Chinnerf0a76952010-01-11 11:49:57 +00003739 struct hlist_head rhash[],
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003740 struct xlog_rec_header *rhead,
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10003741 char *dp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003742 int pass)
3743{
Dave Chinnereeb11682014-09-29 09:45:03 +10003744 struct xlog_op_header *ohead;
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10003745 char *end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003746 int num_logops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003747 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003748
Dave Chinnereeb11682014-09-29 09:45:03 +10003749 end = dp + be32_to_cpu(rhead->h_len);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003750 num_logops = be32_to_cpu(rhead->h_num_logops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003751
3752 /* check the log format matches our own - else we can't recover */
3753 if (xlog_header_check_recover(log->l_mp, rhead))
Dave Chinner24513372014-06-25 14:58:08 +10003754 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003755
Dave Chinnereeb11682014-09-29 09:45:03 +10003756 while ((dp < end) && num_logops) {
3757
3758 ohead = (struct xlog_op_header *)dp;
3759 dp += sizeof(*ohead);
3760 ASSERT(dp <= end);
3761
3762 /* errors will abort recovery */
3763 error = xlog_recover_process_ophdr(log, rhash, rhead, ohead,
3764 dp, end, pass);
3765 if (error)
3766 return error;
3767
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10003768 dp += be32_to_cpu(ohead->oh_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003769 num_logops--;
3770 }
3771 return 0;
3772}
3773
3774/*
3775 * Process an extent free intent item that was recovered from
3776 * the log. We need to free the extents that it describes.
3777 */
David Chinner3c1e2bb2008-04-10 12:21:11 +10003778STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003779xlog_recover_process_efi(
3780 xfs_mount_t *mp,
3781 xfs_efi_log_item_t *efip)
3782{
3783 xfs_efd_log_item_t *efdp;
3784 xfs_trans_t *tp;
3785 int i;
David Chinner3c1e2bb2008-04-10 12:21:11 +10003786 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003787 xfs_extent_t *extp;
3788 xfs_fsblock_t startblock_fsb;
3789
Dave Chinnerb199c8a2010-12-20 11:59:49 +11003790 ASSERT(!test_bit(XFS_EFI_RECOVERED, &efip->efi_flags));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003791
3792 /*
3793 * First check the validity of the extents described by the
3794 * EFI. If any are bad, then assume that all are bad and
3795 * just toss the EFI.
3796 */
3797 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3798 extp = &(efip->efi_format.efi_extents[i]);
3799 startblock_fsb = XFS_BB_TO_FSB(mp,
3800 XFS_FSB_TO_DADDR(mp, extp->ext_start));
3801 if ((startblock_fsb == 0) ||
3802 (extp->ext_len == 0) ||
3803 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
3804 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
3805 /*
3806 * This will pull the EFI from the AIL and
3807 * free the memory associated with it.
3808 */
Dave Chinner666d6442013-04-03 14:09:21 +11003809 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
Brian Foster5e4b5382015-08-19 09:50:12 +10003810 xfs_efi_release(efip);
Dave Chinner24513372014-06-25 14:58:08 +10003811 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003812 }
3813 }
3814
3815 tp = xfs_trans_alloc(mp, 0);
Jie Liu3d3c8b52013-08-12 20:49:59 +10003816 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
David Chinnerfc6149d2008-04-10 12:21:53 +10003817 if (error)
3818 goto abort_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003819 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
3820
3821 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3822 extp = &(efip->efi_format.efi_extents[i]);
Brian Foster6bc43af2015-08-19 09:51:43 +10003823 error = xfs_trans_free_extent(tp, efdp, extp->ext_start,
3824 extp->ext_len);
David Chinnerfc6149d2008-04-10 12:21:53 +10003825 if (error)
3826 goto abort_error;
Brian Foster6bc43af2015-08-19 09:51:43 +10003827
Linus Torvalds1da177e2005-04-16 15:20:36 -07003828 }
3829
Dave Chinnerb199c8a2010-12-20 11:59:49 +11003830 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
Christoph Hellwig70393312015-06-04 13:48:08 +10003831 error = xfs_trans_commit(tp);
David Chinner3c1e2bb2008-04-10 12:21:11 +10003832 return error;
David Chinnerfc6149d2008-04-10 12:21:53 +10003833
3834abort_error:
Christoph Hellwig4906e212015-06-04 13:47:56 +10003835 xfs_trans_cancel(tp);
David Chinnerfc6149d2008-04-10 12:21:53 +10003836 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003837}
3838
3839/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003840 * When this is called, all of the EFIs which did not have
3841 * corresponding EFDs should be in the AIL. What we do now
3842 * is free the extents associated with each one.
3843 *
3844 * Since we process the EFIs in normal transactions, they
3845 * will be removed at some point after the commit. This prevents
3846 * us from just walking down the list processing each one.
3847 * We'll use a flag in the EFI to skip those that we've already
3848 * processed and use the AIL iteration mechanism's generation
3849 * count to try to speed this up at least a bit.
3850 *
3851 * When we start, we know that the EFIs are the only things in
3852 * the AIL. As we process them, however, other items are added
3853 * to the AIL. Since everything added to the AIL must come after
3854 * everything already in the AIL, we stop processing as soon as
3855 * we see something other than an EFI in the AIL.
3856 */
David Chinner3c1e2bb2008-04-10 12:21:11 +10003857STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003858xlog_recover_process_efis(
Brian Fosterf0b2efa2015-08-19 09:58:36 +10003859 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003860{
Brian Fosterf0b2efa2015-08-19 09:58:36 +10003861 struct xfs_log_item *lip;
3862 struct xfs_efi_log_item *efip;
David Chinner3c1e2bb2008-04-10 12:21:11 +10003863 int error = 0;
David Chinner27d8d5f2008-10-30 17:38:39 +11003864 struct xfs_ail_cursor cur;
David Chinnera9c21c12008-10-30 17:39:35 +11003865 struct xfs_ail *ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003866
David Chinnera9c21c12008-10-30 17:39:35 +11003867 ailp = log->l_ailp;
3868 spin_lock(&ailp->xa_lock);
3869 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003870 while (lip != NULL) {
3871 /*
3872 * We're done when we see something other than an EFI.
David Chinner27d8d5f2008-10-30 17:38:39 +11003873 * There should be no EFIs left in the AIL now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003874 */
3875 if (lip->li_type != XFS_LI_EFI) {
David Chinner27d8d5f2008-10-30 17:38:39 +11003876#ifdef DEBUG
David Chinnera9c21c12008-10-30 17:39:35 +11003877 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
David Chinner27d8d5f2008-10-30 17:38:39 +11003878 ASSERT(lip->li_type != XFS_LI_EFI);
3879#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003880 break;
3881 }
3882
3883 /*
3884 * Skip EFIs that we've already processed.
3885 */
Brian Fosterf0b2efa2015-08-19 09:58:36 +10003886 efip = container_of(lip, struct xfs_efi_log_item, efi_item);
Dave Chinnerb199c8a2010-12-20 11:59:49 +11003887 if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) {
David Chinnera9c21c12008-10-30 17:39:35 +11003888 lip = xfs_trans_ail_cursor_next(ailp, &cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003889 continue;
3890 }
3891
David Chinnera9c21c12008-10-30 17:39:35 +11003892 spin_unlock(&ailp->xa_lock);
3893 error = xlog_recover_process_efi(log->l_mp, efip);
3894 spin_lock(&ailp->xa_lock);
David Chinner27d8d5f2008-10-30 17:38:39 +11003895 if (error)
3896 goto out;
David Chinnera9c21c12008-10-30 17:39:35 +11003897 lip = xfs_trans_ail_cursor_next(ailp, &cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003898 }
David Chinner27d8d5f2008-10-30 17:38:39 +11003899out:
Eric Sandeene4a1e292014-04-14 19:06:05 +10003900 xfs_trans_ail_cursor_done(&cur);
David Chinnera9c21c12008-10-30 17:39:35 +11003901 spin_unlock(&ailp->xa_lock);
David Chinner3c1e2bb2008-04-10 12:21:11 +10003902 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003903}
3904
3905/*
Brian Fosterf0b2efa2015-08-19 09:58:36 +10003906 * A cancel occurs when the mount has failed and we're bailing out. Release all
3907 * pending EFIs so they don't pin the AIL.
3908 */
3909STATIC int
3910xlog_recover_cancel_efis(
3911 struct xlog *log)
3912{
3913 struct xfs_log_item *lip;
3914 struct xfs_efi_log_item *efip;
3915 int error = 0;
3916 struct xfs_ail_cursor cur;
3917 struct xfs_ail *ailp;
3918
3919 ailp = log->l_ailp;
3920 spin_lock(&ailp->xa_lock);
3921 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3922 while (lip != NULL) {
3923 /*
3924 * We're done when we see something other than an EFI.
3925 * There should be no EFIs left in the AIL now.
3926 */
3927 if (lip->li_type != XFS_LI_EFI) {
3928#ifdef DEBUG
3929 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
3930 ASSERT(lip->li_type != XFS_LI_EFI);
3931#endif
3932 break;
3933 }
3934
3935 efip = container_of(lip, struct xfs_efi_log_item, efi_item);
3936
3937 spin_unlock(&ailp->xa_lock);
3938 xfs_efi_release(efip);
3939 spin_lock(&ailp->xa_lock);
3940
3941 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3942 }
3943
3944 xfs_trans_ail_cursor_done(&cur);
3945 spin_unlock(&ailp->xa_lock);
3946 return error;
3947}
3948
3949/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003950 * This routine performs a transaction to null out a bad inode pointer
3951 * in an agi unlinked inode hash bucket.
3952 */
3953STATIC void
3954xlog_recover_clear_agi_bucket(
3955 xfs_mount_t *mp,
3956 xfs_agnumber_t agno,
3957 int bucket)
3958{
3959 xfs_trans_t *tp;
3960 xfs_agi_t *agi;
3961 xfs_buf_t *agibp;
3962 int offset;
3963 int error;
3964
3965 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
Jie Liu3d3c8b52013-08-12 20:49:59 +10003966 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_clearagi, 0, 0);
David Chinnere5720ee2008-04-10 12:21:18 +10003967 if (error)
3968 goto out_abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003969
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003970 error = xfs_read_agi(mp, tp, agno, &agibp);
3971 if (error)
David Chinnere5720ee2008-04-10 12:21:18 +10003972 goto out_abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003973
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003974 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11003975 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003976 offset = offsetof(xfs_agi_t, agi_unlinked) +
3977 (sizeof(xfs_agino_t) * bucket);
3978 xfs_trans_log_buf(tp, agibp, offset,
3979 (offset + sizeof(xfs_agino_t) - 1));
3980
Christoph Hellwig70393312015-06-04 13:48:08 +10003981 error = xfs_trans_commit(tp);
David Chinnere5720ee2008-04-10 12:21:18 +10003982 if (error)
3983 goto out_error;
3984 return;
3985
3986out_abort:
Christoph Hellwig4906e212015-06-04 13:47:56 +10003987 xfs_trans_cancel(tp);
David Chinnere5720ee2008-04-10 12:21:18 +10003988out_error:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003989 xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno);
David Chinnere5720ee2008-04-10 12:21:18 +10003990 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003991}
3992
Christoph Hellwig23fac502008-11-28 14:23:40 +11003993STATIC xfs_agino_t
3994xlog_recover_process_one_iunlink(
3995 struct xfs_mount *mp,
3996 xfs_agnumber_t agno,
3997 xfs_agino_t agino,
3998 int bucket)
3999{
4000 struct xfs_buf *ibp;
4001 struct xfs_dinode *dip;
4002 struct xfs_inode *ip;
4003 xfs_ino_t ino;
4004 int error;
4005
4006 ino = XFS_AGINO_TO_INO(mp, agno, agino);
Dave Chinner7b6259e2010-06-24 11:35:17 +10004007 error = xfs_iget(mp, NULL, ino, 0, 0, &ip);
Christoph Hellwig23fac502008-11-28 14:23:40 +11004008 if (error)
4009 goto fail;
4010
4011 /*
4012 * Get the on disk inode to find the next inode in the bucket.
4013 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04004014 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &ibp, 0, 0);
Christoph Hellwig23fac502008-11-28 14:23:40 +11004015 if (error)
Christoph Hellwig0e446672008-11-28 14:23:42 +11004016 goto fail_iput;
Christoph Hellwig23fac502008-11-28 14:23:40 +11004017
Christoph Hellwig23fac502008-11-28 14:23:40 +11004018 ASSERT(ip->i_d.di_nlink == 0);
Christoph Hellwig0e446672008-11-28 14:23:42 +11004019 ASSERT(ip->i_d.di_mode != 0);
Christoph Hellwig23fac502008-11-28 14:23:40 +11004020
4021 /* setup for the next pass */
4022 agino = be32_to_cpu(dip->di_next_unlinked);
4023 xfs_buf_relse(ibp);
4024
4025 /*
4026 * Prevent any DMAPI event from being sent when the reference on
4027 * the inode is dropped.
4028 */
4029 ip->i_d.di_dmevmask = 0;
4030
Christoph Hellwig0e446672008-11-28 14:23:42 +11004031 IRELE(ip);
Christoph Hellwig23fac502008-11-28 14:23:40 +11004032 return agino;
4033
Christoph Hellwig0e446672008-11-28 14:23:42 +11004034 fail_iput:
4035 IRELE(ip);
Christoph Hellwig23fac502008-11-28 14:23:40 +11004036 fail:
4037 /*
4038 * We can't read in the inode this bucket points to, or this inode
4039 * is messed up. Just ditch this bucket of inodes. We will lose
4040 * some inodes and space, but at least we won't hang.
4041 *
4042 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
4043 * clear the inode pointer in the bucket.
4044 */
4045 xlog_recover_clear_agi_bucket(mp, agno, bucket);
4046 return NULLAGINO;
4047}
4048
Linus Torvalds1da177e2005-04-16 15:20:36 -07004049/*
4050 * xlog_iunlink_recover
4051 *
4052 * This is called during recovery to process any inodes which
4053 * we unlinked but not freed when the system crashed. These
4054 * inodes will be on the lists in the AGI blocks. What we do
4055 * here is scan all the AGIs and fully truncate and free any
4056 * inodes found on the lists. Each inode is removed from the
4057 * lists when it has been fully truncated and is freed. The
4058 * freeing of the inode and its removal from the list must be
4059 * atomic.
4060 */
Eric Sandeend96f8f82009-07-02 00:09:33 -05004061STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07004062xlog_recover_process_iunlinks(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004063 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004064{
4065 xfs_mount_t *mp;
4066 xfs_agnumber_t agno;
4067 xfs_agi_t *agi;
4068 xfs_buf_t *agibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004069 xfs_agino_t agino;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004070 int bucket;
4071 int error;
4072 uint mp_dmevmask;
4073
4074 mp = log->l_mp;
4075
4076 /*
4077 * Prevent any DMAPI event from being sent while in this function.
4078 */
4079 mp_dmevmask = mp->m_dmevmask;
4080 mp->m_dmevmask = 0;
4081
4082 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
4083 /*
4084 * Find the agi for this ag.
4085 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11004086 error = xfs_read_agi(mp, NULL, agno, &agibp);
4087 if (error) {
4088 /*
4089 * AGI is b0rked. Don't process it.
4090 *
4091 * We should probably mark the filesystem as corrupt
4092 * after we've recovered all the ag's we can....
4093 */
4094 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004095 }
Jan Karad97d32e2012-03-15 09:34:02 +00004096 /*
4097 * Unlock the buffer so that it can be acquired in the normal
4098 * course of the transaction to truncate and free each inode.
4099 * Because we are not racing with anyone else here for the AGI
4100 * buffer, we don't even need to hold it locked to read the
4101 * initial unlinked bucket entries out of the buffer. We keep
4102 * buffer reference though, so that it stays pinned in memory
4103 * while we need the buffer.
4104 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004105 agi = XFS_BUF_TO_AGI(agibp);
Jan Karad97d32e2012-03-15 09:34:02 +00004106 xfs_buf_unlock(agibp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004107
4108 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
Christoph Hellwig16259e72005-11-02 15:11:25 +11004109 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004110 while (agino != NULLAGINO) {
Christoph Hellwig23fac502008-11-28 14:23:40 +11004111 agino = xlog_recover_process_one_iunlink(mp,
4112 agno, agino, bucket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004113 }
4114 }
Jan Karad97d32e2012-03-15 09:34:02 +00004115 xfs_buf_rele(agibp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004116 }
4117
4118 mp->m_dmevmask = mp_dmevmask;
4119}
4120
Linus Torvalds1da177e2005-04-16 15:20:36 -07004121/*
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004122 * Upack the log buffer data and crc check it. If the check fails, issue a
4123 * warning if and only if the CRC in the header is non-zero. This makes the
4124 * check an advisory warning, and the zero CRC check will prevent failure
4125 * warnings from being emitted when upgrading the kernel from one that does not
4126 * add CRCs by default.
4127 *
4128 * When filesystems are CRC enabled, this CRC mismatch becomes a fatal log
4129 * corruption failure
Linus Torvalds1da177e2005-04-16 15:20:36 -07004130 */
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004131STATIC int
4132xlog_unpack_data_crc(
4133 struct xlog_rec_header *rhead,
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10004134 char *dp,
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004135 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004136{
Dave Chinnerf9668a02012-11-28 13:01:03 +11004137 __le32 crc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004138
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004139 crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len));
4140 if (crc != rhead->h_crc) {
4141 if (rhead->h_crc || xfs_sb_version_hascrc(&log->l_mp->m_sb)) {
4142 xfs_alert(log->l_mp,
Eric Sandeen08e96e12013-10-11 20:59:05 -05004143 "log record CRC mismatch: found 0x%x, expected 0x%x.",
Dave Chinnerf9668a02012-11-28 13:01:03 +11004144 le32_to_cpu(rhead->h_crc),
4145 le32_to_cpu(crc));
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004146 xfs_hex_dump(dp, 32);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004147 }
4148
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004149 /*
4150 * If we've detected a log record corruption, then we can't
4151 * recover past this point. Abort recovery if we are enforcing
4152 * CRC protection by punting an error back up the stack.
4153 */
4154 if (xfs_sb_version_hascrc(&log->l_mp->m_sb))
Dave Chinner24513372014-06-25 14:58:08 +10004155 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004156 }
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004157
4158 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004159}
4160
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004161STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07004162xlog_unpack_data(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004163 struct xlog_rec_header *rhead,
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10004164 char *dp,
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004165 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004166{
4167 int i, j, k;
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004168 int error;
4169
4170 error = xlog_unpack_data_crc(rhead, dp, log);
4171 if (error)
4172 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004173
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004174 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07004175 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004176 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07004177 dp += BBSIZE;
4178 }
4179
Eric Sandeen62118702008-03-06 13:44:28 +11004180 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Christoph Hellwigb28708d2008-11-28 14:23:38 +11004181 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004182 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004183 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
4184 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004185 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
Linus Torvalds1da177e2005-04-16 15:20:36 -07004186 dp += BBSIZE;
4187 }
4188 }
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004189
4190 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004191}
4192
4193STATIC int
4194xlog_valid_rec_header(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004195 struct xlog *log,
4196 struct xlog_rec_header *rhead,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004197 xfs_daddr_t blkno)
4198{
4199 int hlen;
4200
Christoph Hellwig69ef9212011-07-08 14:36:05 +02004201 if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004202 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
4203 XFS_ERRLEVEL_LOW, log->l_mp);
Dave Chinner24513372014-06-25 14:58:08 +10004204 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004205 }
4206 if (unlikely(
4207 (!rhead->h_version ||
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004208 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004209 xfs_warn(log->l_mp, "%s: unrecognised log version (%d).",
Harvey Harrison34a622b2008-04-10 12:19:21 +10004210 __func__, be32_to_cpu(rhead->h_version));
Dave Chinner24513372014-06-25 14:58:08 +10004211 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004212 }
4213
4214 /* LR body must have data or it wouldn't have been written */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004215 hlen = be32_to_cpu(rhead->h_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004216 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
4217 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
4218 XFS_ERRLEVEL_LOW, log->l_mp);
Dave Chinner24513372014-06-25 14:58:08 +10004219 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004220 }
4221 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
4222 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
4223 XFS_ERRLEVEL_LOW, log->l_mp);
Dave Chinner24513372014-06-25 14:58:08 +10004224 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004225 }
4226 return 0;
4227}
4228
4229/*
4230 * Read the log from tail to head and process the log records found.
4231 * Handle the two cases where the tail and head are in the same cycle
4232 * and where the active portion of the log wraps around the end of
4233 * the physical log separately. The pass parameter is passed through
4234 * to the routines called to process the data and is not looked at
4235 * here.
4236 */
4237STATIC int
4238xlog_do_recovery_pass(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004239 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004240 xfs_daddr_t head_blk,
4241 xfs_daddr_t tail_blk,
4242 int pass)
4243{
4244 xlog_rec_header_t *rhead;
4245 xfs_daddr_t blk_no;
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10004246 char *offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004247 xfs_buf_t *hbp, *dbp;
4248 int error = 0, h_size;
4249 int bblks, split_bblks;
4250 int hblks, split_hblks, wrapped_hblks;
Dave Chinnerf0a76952010-01-11 11:49:57 +00004251 struct hlist_head rhash[XLOG_RHASH_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -07004252
4253 ASSERT(head_blk != tail_blk);
4254
4255 /*
4256 * Read the header of the tail block and get the iclog buffer size from
4257 * h_size. Use this to tell how many sectors make up the log header.
4258 */
Eric Sandeen62118702008-03-06 13:44:28 +11004259 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004260 /*
4261 * When using variable length iclogs, read first sector of
4262 * iclog header and extract the header size from it. Get a
4263 * new hbp that is the correct size.
4264 */
4265 hbp = xlog_get_bp(log, 1);
4266 if (!hbp)
Dave Chinner24513372014-06-25 14:58:08 +10004267 return -ENOMEM;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004268
4269 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
4270 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004271 goto bread_err1;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004272
Linus Torvalds1da177e2005-04-16 15:20:36 -07004273 rhead = (xlog_rec_header_t *)offset;
4274 error = xlog_valid_rec_header(log, rhead, tail_blk);
4275 if (error)
4276 goto bread_err1;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004277 h_size = be32_to_cpu(rhead->h_size);
4278 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07004279 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
4280 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
4281 if (h_size % XLOG_HEADER_CYCLE_SIZE)
4282 hblks++;
4283 xlog_put_bp(hbp);
4284 hbp = xlog_get_bp(log, hblks);
4285 } else {
4286 hblks = 1;
4287 }
4288 } else {
Alex Elder69ce58f2010-04-20 17:09:59 +10004289 ASSERT(log->l_sectBBsize == 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004290 hblks = 1;
4291 hbp = xlog_get_bp(log, 1);
4292 h_size = XLOG_BIG_RECORD_BSIZE;
4293 }
4294
4295 if (!hbp)
Dave Chinner24513372014-06-25 14:58:08 +10004296 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004297 dbp = xlog_get_bp(log, BTOBB(h_size));
4298 if (!dbp) {
4299 xlog_put_bp(hbp);
Dave Chinner24513372014-06-25 14:58:08 +10004300 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004301 }
4302
4303 memset(rhash, 0, sizeof(rhash));
Eric Sandeen970fd3f2014-09-09 11:57:29 +10004304 blk_no = tail_blk;
4305 if (tail_blk > head_blk) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004306 /*
4307 * Perform recovery around the end of the physical log.
4308 * When the head is not on the same cycle number as the tail,
Eric Sandeen970fd3f2014-09-09 11:57:29 +10004309 * we can't do a sequential recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004310 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004311 while (blk_no < log->l_logBBsize) {
4312 /*
4313 * Check for header wrapping around physical end-of-log
4314 */
Chandra Seetharaman62926042011-07-22 23:40:15 +00004315 offset = hbp->b_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004316 split_hblks = 0;
4317 wrapped_hblks = 0;
4318 if (blk_no + hblks <= log->l_logBBsize) {
4319 /* Read header in one read */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004320 error = xlog_bread(log, blk_no, hblks, hbp,
4321 &offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004322 if (error)
4323 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004324 } else {
4325 /* This LR is split across physical log end */
4326 if (blk_no != log->l_logBBsize) {
4327 /* some data before physical log end */
4328 ASSERT(blk_no <= INT_MAX);
4329 split_hblks = log->l_logBBsize - (int)blk_no;
4330 ASSERT(split_hblks > 0);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004331 error = xlog_bread(log, blk_no,
4332 split_hblks, hbp,
4333 &offset);
4334 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004335 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004336 }
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004337
Linus Torvalds1da177e2005-04-16 15:20:36 -07004338 /*
4339 * Note: this black magic still works with
4340 * large sector sizes (non-512) only because:
4341 * - we increased the buffer size originally
4342 * by 1 sector giving us enough extra space
4343 * for the second read;
4344 * - the log start is guaranteed to be sector
4345 * aligned;
4346 * - we read the log end (LR header start)
4347 * _first_, then the log start (LR header end)
4348 * - order is important.
4349 */
David Chinner234f56a2008-04-10 12:24:24 +10004350 wrapped_hblks = hblks - split_hblks;
Dave Chinner44396472011-04-21 09:34:27 +00004351 error = xlog_bread_offset(log, 0,
4352 wrapped_hblks, hbp,
4353 offset + BBTOB(split_hblks));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004354 if (error)
4355 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004356 }
4357 rhead = (xlog_rec_header_t *)offset;
4358 error = xlog_valid_rec_header(log, rhead,
4359 split_hblks ? blk_no : 0);
4360 if (error)
4361 goto bread_err2;
4362
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004363 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004364 blk_no += hblks;
4365
4366 /* Read in data for log record */
4367 if (blk_no + bblks <= log->l_logBBsize) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004368 error = xlog_bread(log, blk_no, bblks, dbp,
4369 &offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004370 if (error)
4371 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004372 } else {
4373 /* This log record is split across the
4374 * physical end of log */
Chandra Seetharaman62926042011-07-22 23:40:15 +00004375 offset = dbp->b_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004376 split_bblks = 0;
4377 if (blk_no != log->l_logBBsize) {
4378 /* some data is before the physical
4379 * end of log */
4380 ASSERT(!wrapped_hblks);
4381 ASSERT(blk_no <= INT_MAX);
4382 split_bblks =
4383 log->l_logBBsize - (int)blk_no;
4384 ASSERT(split_bblks > 0);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004385 error = xlog_bread(log, blk_no,
4386 split_bblks, dbp,
4387 &offset);
4388 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004389 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004390 }
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004391
Linus Torvalds1da177e2005-04-16 15:20:36 -07004392 /*
4393 * Note: this black magic still works with
4394 * large sector sizes (non-512) only because:
4395 * - we increased the buffer size originally
4396 * by 1 sector giving us enough extra space
4397 * for the second read;
4398 * - the log start is guaranteed to be sector
4399 * aligned;
4400 * - we read the log end (LR header start)
4401 * _first_, then the log start (LR header end)
4402 * - order is important.
4403 */
Dave Chinner44396472011-04-21 09:34:27 +00004404 error = xlog_bread_offset(log, 0,
Dave Chinner009507b2012-11-02 11:38:44 +11004405 bblks - split_bblks, dbp,
Dave Chinner44396472011-04-21 09:34:27 +00004406 offset + BBTOB(split_bblks));
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004407 if (error)
4408 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004409 }
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004410
4411 error = xlog_unpack_data(rhead, offset, log);
4412 if (error)
4413 goto bread_err2;
4414
4415 error = xlog_recover_process_data(log, rhash,
4416 rhead, offset, pass);
4417 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004418 goto bread_err2;
4419 blk_no += bblks;
4420 }
4421
4422 ASSERT(blk_no >= log->l_logBBsize);
4423 blk_no -= log->l_logBBsize;
Eric Sandeen970fd3f2014-09-09 11:57:29 +10004424 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004425
Eric Sandeen970fd3f2014-09-09 11:57:29 +10004426 /* read first part of physical log */
4427 while (blk_no < head_blk) {
4428 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
4429 if (error)
4430 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004431
Eric Sandeen970fd3f2014-09-09 11:57:29 +10004432 rhead = (xlog_rec_header_t *)offset;
4433 error = xlog_valid_rec_header(log, rhead, blk_no);
4434 if (error)
4435 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004436
Eric Sandeen970fd3f2014-09-09 11:57:29 +10004437 /* blocks in data section */
4438 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
4439 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
4440 &offset);
4441 if (error)
4442 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004443
Eric Sandeen970fd3f2014-09-09 11:57:29 +10004444 error = xlog_unpack_data(rhead, offset, log);
4445 if (error)
4446 goto bread_err2;
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004447
Eric Sandeen970fd3f2014-09-09 11:57:29 +10004448 error = xlog_recover_process_data(log, rhash,
4449 rhead, offset, pass);
4450 if (error)
4451 goto bread_err2;
4452 blk_no += bblks + hblks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004453 }
4454
4455 bread_err2:
4456 xlog_put_bp(dbp);
4457 bread_err1:
4458 xlog_put_bp(hbp);
4459 return error;
4460}
4461
4462/*
4463 * Do the recovery of the log. We actually do this in two phases.
4464 * The two passes are necessary in order to implement the function
4465 * of cancelling a record written into the log. The first pass
4466 * determines those things which have been cancelled, and the
4467 * second pass replays log items normally except for those which
4468 * have been cancelled. The handling of the replay and cancellations
4469 * takes place in the log item type specific routines.
4470 *
4471 * The table of items which have cancel records in the log is allocated
4472 * and freed at this level, since only here do we know when all of
4473 * the log recovery has been completed.
4474 */
4475STATIC int
4476xlog_do_log_recovery(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004477 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004478 xfs_daddr_t head_blk,
4479 xfs_daddr_t tail_blk)
4480{
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00004481 int error, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004482
4483 ASSERT(head_blk != tail_blk);
4484
4485 /*
4486 * First do a pass to find all of the cancelled buf log items.
4487 * Store them in the buf_cancel_table for use in the second pass.
4488 */
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00004489 log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE *
4490 sizeof(struct list_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07004491 KM_SLEEP);
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00004492 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
4493 INIT_LIST_HEAD(&log->l_buf_cancel_table[i]);
4494
Linus Torvalds1da177e2005-04-16 15:20:36 -07004495 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
4496 XLOG_RECOVER_PASS1);
4497 if (error != 0) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10004498 kmem_free(log->l_buf_cancel_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004499 log->l_buf_cancel_table = NULL;
4500 return error;
4501 }
4502 /*
4503 * Then do a second pass to actually recover the items in the log.
4504 * When it is complete free the table of buf cancel items.
4505 */
4506 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
4507 XLOG_RECOVER_PASS2);
4508#ifdef DEBUG
Tim Shimmin6d192a92006-06-09 14:55:38 +10004509 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004510 int i;
4511
4512 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00004513 ASSERT(list_empty(&log->l_buf_cancel_table[i]));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004514 }
4515#endif /* DEBUG */
4516
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10004517 kmem_free(log->l_buf_cancel_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004518 log->l_buf_cancel_table = NULL;
4519
4520 return error;
4521}
4522
4523/*
4524 * Do the actual recovery
4525 */
4526STATIC int
4527xlog_do_recover(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004528 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004529 xfs_daddr_t head_blk,
4530 xfs_daddr_t tail_blk)
4531{
4532 int error;
4533 xfs_buf_t *bp;
4534 xfs_sb_t *sbp;
4535
4536 /*
4537 * First replay the images in the log.
4538 */
4539 error = xlog_do_log_recovery(log, head_blk, tail_blk);
Christoph Hellwig43ff2122012-04-23 15:58:39 +10004540 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004541 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004542
4543 /*
4544 * If IO errors happened during recovery, bail out.
4545 */
4546 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
Dave Chinner24513372014-06-25 14:58:08 +10004547 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004548 }
4549
4550 /*
4551 * We now update the tail_lsn since much of the recovery has completed
4552 * and there may be space available to use. If there were no extent
4553 * or iunlinks, we can free up the entire log and set the tail_lsn to
4554 * be the last_sync_lsn. This was set in xlog_find_tail to be the
4555 * lsn of the last known good LR on disk. If there are extent frees
4556 * or iunlinks they will have some entries in the AIL; so we look at
4557 * the AIL to determine how to set the tail_lsn.
4558 */
4559 xlog_assign_tail_lsn(log->l_mp);
4560
4561 /*
4562 * Now that we've finished replaying all buffer and inode
Dave Chinner98021822012-11-12 22:54:03 +11004563 * updates, re-read in the superblock and reverify it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004564 */
4565 bp = xfs_getsb(log->l_mp, 0);
4566 XFS_BUF_UNDONE(bp);
Lachlan McIlroybebf9632007-10-15 13:18:02 +10004567 ASSERT(!(XFS_BUF_ISWRITE(bp)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004568 XFS_BUF_READ(bp);
Lachlan McIlroybebf9632007-10-15 13:18:02 +10004569 XFS_BUF_UNASYNC(bp);
Dave Chinner1813dd62012-11-14 17:54:40 +11004570 bp->b_ops = &xfs_sb_buf_ops;
Christoph Hellwig83a0adc2013-12-17 00:03:52 -08004571
Dave Chinner595bff72014-10-02 09:05:14 +10004572 error = xfs_buf_submit_wait(bp);
David Chinnerd64e31a2008-04-10 12:22:17 +10004573 if (error) {
Dave Chinner595bff72014-10-02 09:05:14 +10004574 if (!XFS_FORCED_SHUTDOWN(log->l_mp)) {
4575 xfs_buf_ioerror_alert(bp, __func__);
4576 ASSERT(0);
4577 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004578 xfs_buf_relse(bp);
4579 return error;
4580 }
4581
4582 /* Convert superblock from on-disk format */
4583 sbp = &log->l_mp->m_sb;
Dave Chinner98021822012-11-12 22:54:03 +11004584 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004585 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
Eric Sandeen62118702008-03-06 13:44:28 +11004586 ASSERT(xfs_sb_good_version(sbp));
Dave Chinner5681ca42015-02-23 21:22:31 +11004587 xfs_reinit_percpu_counters(log->l_mp);
4588
Linus Torvalds1da177e2005-04-16 15:20:36 -07004589 xfs_buf_relse(bp);
4590
Lachlan McIlroy5478eea2007-02-10 18:36:29 +11004591
Linus Torvalds1da177e2005-04-16 15:20:36 -07004592 xlog_recover_check_summary(log);
4593
4594 /* Normal transactions can now occur */
4595 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
4596 return 0;
4597}
4598
4599/*
4600 * Perform recovery and re-initialize some log variables in xlog_find_tail.
4601 *
4602 * Return error or zero.
4603 */
4604int
4605xlog_recover(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004606 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004607{
4608 xfs_daddr_t head_blk, tail_blk;
4609 int error;
4610
4611 /* find the tail of the log */
Eric Sandeen65be6052006-01-11 15:34:19 +11004612 if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004613 return error;
4614
4615 if (tail_blk != head_blk) {
4616 /* There used to be a comment here:
4617 *
4618 * disallow recovery on read-only mounts. note -- mount
4619 * checks for ENOSPC and turns it into an intelligent
4620 * error message.
4621 * ...but this is no longer true. Now, unless you specify
4622 * NORECOVERY (in which case this function would never be
4623 * called), we just go ahead and recover. We do this all
4624 * under the vfs layer, so we can get away with it unless
4625 * the device itself is read-only, in which case we fail.
4626 */
Utako Kusaka3a02ee12007-05-08 13:50:06 +10004627 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004628 return error;
4629 }
4630
Dave Chinnere721f502013-04-03 16:11:32 +11004631 /*
4632 * Version 5 superblock log feature mask validation. We know the
4633 * log is dirty so check if there are any unknown log features
4634 * in what we need to recover. If there are unknown features
4635 * (e.g. unsupported transactions, then simply reject the
4636 * attempt at recovery before touching anything.
4637 */
4638 if (XFS_SB_VERSION_NUM(&log->l_mp->m_sb) == XFS_SB_VERSION_5 &&
4639 xfs_sb_has_incompat_log_feature(&log->l_mp->m_sb,
4640 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
4641 xfs_warn(log->l_mp,
Joe Perchesf41febd2015-07-29 11:52:04 +10004642"Superblock has unknown incompatible log features (0x%x) enabled.",
Dave Chinnere721f502013-04-03 16:11:32 +11004643 (log->l_mp->m_sb.sb_features_log_incompat &
4644 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
Joe Perchesf41febd2015-07-29 11:52:04 +10004645 xfs_warn(log->l_mp,
4646"The log can not be fully and/or safely recovered by this kernel.");
4647 xfs_warn(log->l_mp,
4648"Please recover the log on a kernel that supports the unknown features.");
Dave Chinner24513372014-06-25 14:58:08 +10004649 return -EINVAL;
Dave Chinnere721f502013-04-03 16:11:32 +11004650 }
4651
Brian Foster2e227172014-09-09 11:56:13 +10004652 /*
4653 * Delay log recovery if the debug hook is set. This is debug
4654 * instrumention to coordinate simulation of I/O failures with
4655 * log recovery.
4656 */
4657 if (xfs_globals.log_recovery_delay) {
4658 xfs_notice(log->l_mp,
4659 "Delaying log recovery for %d seconds.",
4660 xfs_globals.log_recovery_delay);
4661 msleep(xfs_globals.log_recovery_delay * 1000);
4662 }
4663
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004664 xfs_notice(log->l_mp, "Starting recovery (logdev: %s)",
4665 log->l_mp->m_logname ? log->l_mp->m_logname
4666 : "internal");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004667
4668 error = xlog_do_recover(log, head_blk, tail_blk);
4669 log->l_flags |= XLOG_RECOVERY_NEEDED;
4670 }
4671 return error;
4672}
4673
4674/*
4675 * In the first part of recovery we replay inodes and buffers and build
4676 * up the list of extent free items which need to be processed. Here
4677 * we process the extent free items and clean up the on disk unlinked
4678 * inode lists. This is separated from the first part of recovery so
4679 * that the root and real-time bitmap inodes can be read in from disk in
4680 * between the two stages. This is necessary so that we can free space
4681 * in the real-time portion of the file system.
4682 */
4683int
4684xlog_recover_finish(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004685 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004686{
4687 /*
4688 * Now we're ready to do the transactions needed for the
4689 * rest of recovery. Start with completing all the extent
4690 * free intent records and then process the unlinked inode
4691 * lists. At this point, we essentially run in normal mode
4692 * except that we're still performing recovery actions
4693 * rather than accepting new requests.
4694 */
4695 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
David Chinner3c1e2bb2008-04-10 12:21:11 +10004696 int error;
4697 error = xlog_recover_process_efis(log);
4698 if (error) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004699 xfs_alert(log->l_mp, "Failed to recover EFIs");
David Chinner3c1e2bb2008-04-10 12:21:11 +10004700 return error;
4701 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004702 /*
4703 * Sync the log to get all the EFIs out of the AIL.
4704 * This isn't absolutely necessary, but it helps in
4705 * case the unlink transactions would have problems
4706 * pushing the EFIs out of the way.
4707 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +00004708 xfs_log_force(log->l_mp, XFS_LOG_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004709
Christoph Hellwig42490232008-08-13 16:49:32 +10004710 xlog_recover_process_iunlinks(log);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004711
4712 xlog_recover_check_summary(log);
4713
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004714 xfs_notice(log->l_mp, "Ending recovery (logdev: %s)",
4715 log->l_mp->m_logname ? log->l_mp->m_logname
4716 : "internal");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004717 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
4718 } else {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004719 xfs_info(log->l_mp, "Ending clean mount");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004720 }
4721 return 0;
4722}
4723
Brian Fosterf0b2efa2015-08-19 09:58:36 +10004724int
4725xlog_recover_cancel(
4726 struct xlog *log)
4727{
4728 int error = 0;
4729
4730 if (log->l_flags & XLOG_RECOVERY_NEEDED)
4731 error = xlog_recover_cancel_efis(log);
4732
4733 return error;
4734}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004735
4736#if defined(DEBUG)
4737/*
4738 * Read all of the agf and agi counters and check that they
4739 * are consistent with the superblock counters.
4740 */
4741void
4742xlog_recover_check_summary(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004743 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004744{
4745 xfs_mount_t *mp;
4746 xfs_agf_t *agfp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004747 xfs_buf_t *agfbp;
4748 xfs_buf_t *agibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004749 xfs_agnumber_t agno;
4750 __uint64_t freeblks;
4751 __uint64_t itotal;
4752 __uint64_t ifree;
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11004753 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004754
4755 mp = log->l_mp;
4756
4757 freeblks = 0LL;
4758 itotal = 0LL;
4759 ifree = 0LL;
4760 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
From: Christoph Hellwig48056212008-11-28 14:23:38 +11004761 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
4762 if (error) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004763 xfs_alert(mp, "%s agf read failed agno %d error %d",
4764 __func__, agno, error);
From: Christoph Hellwig48056212008-11-28 14:23:38 +11004765 } else {
4766 agfp = XFS_BUF_TO_AGF(agfbp);
4767 freeblks += be32_to_cpu(agfp->agf_freeblks) +
4768 be32_to_cpu(agfp->agf_flcount);
4769 xfs_buf_relse(agfbp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004770 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004771
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11004772 error = xfs_read_agi(mp, NULL, agno, &agibp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004773 if (error) {
4774 xfs_alert(mp, "%s agi read failed agno %d error %d",
4775 __func__, agno, error);
4776 } else {
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11004777 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004778
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11004779 itotal += be32_to_cpu(agi->agi_count);
4780 ifree += be32_to_cpu(agi->agi_freecount);
4781 xfs_buf_relse(agibp);
4782 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004783 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004784}
4785#endif /* DEBUG */