blob: 4f5784f85a5b22e4bad0946106dcb1302f817ace [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 Hellwig076e6ac2009-03-16 08:24:13 +0100150STATIC xfs_caddr_t
151xlog_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 Hellwig076e6ac2009-03-16 08:24:13 +0100206 xfs_caddr_t *offset)
207{
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,
Dave Chinner44396472011-04-21 09:34:27 +0000228 xfs_caddr_t offset)
229{
Chandra Seetharaman62926042011-07-22 23:40:15 +0000230 xfs_caddr_t 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{
399 xfs_caddr_t offset;
400 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;
446 xfs_caddr_t buf = NULL;
447 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;
512 xfs_caddr_t offset = NULL;
513 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;
619 xfs_caddr_t offset;
620 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;
894 xfs_caddr_t offset = NULL;
895 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;
1102 xfs_caddr_t offset;
1103 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,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 xfs_caddr_t buf,
1203 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{
1230 xfs_caddr_t offset;
1231 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
1792 buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
1793 next_unlinked_offset);
Tim Shimmin87c199c2006-06-09 14:56:16 +10001794 *buffer_nextp = *logged_nextp;
Dave Chinner0a32c262013-06-05 12:09:08 +10001795
1796 /*
1797 * If necessary, recalculate the CRC in the on-disk inode. We
1798 * have to leave the inode in a consistent state for whoever
1799 * reads it next....
1800 */
1801 xfs_dinode_calc_crc(mp, (struct xfs_dinode *)
1802 xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize));
1803
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 }
1805
1806 return 0;
1807}
1808
1809/*
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001810 * V5 filesystems know the age of the buffer on disk being recovered. We can
1811 * have newer objects on disk than we are replaying, and so for these cases we
1812 * don't want to replay the current change as that will make the buffer contents
1813 * temporarily invalid on disk.
1814 *
1815 * The magic number might not match the buffer type we are going to recover
1816 * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags. Hence
1817 * extract the LSN of the existing object in the buffer based on it's current
1818 * magic number. If we don't recognise the magic number in the buffer, then
1819 * return a LSN of -1 so that the caller knows it was an unrecognised block and
1820 * so can recover the buffer.
Dave Chinner566055d2013-09-24 16:01:16 +10001821 *
1822 * Note: we cannot rely solely on magic number matches to determine that the
1823 * buffer has a valid LSN - we also need to verify that it belongs to this
1824 * filesystem, so we need to extract the object's LSN and compare it to that
1825 * which we read from the superblock. If the UUIDs don't match, then we've got a
1826 * stale metadata block from an old filesystem instance that we need to recover
1827 * over the top of.
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001828 */
1829static xfs_lsn_t
1830xlog_recover_get_buf_lsn(
1831 struct xfs_mount *mp,
1832 struct xfs_buf *bp)
1833{
1834 __uint32_t magic32;
1835 __uint16_t magic16;
1836 __uint16_t magicda;
1837 void *blk = bp->b_addr;
Dave Chinner566055d2013-09-24 16:01:16 +10001838 uuid_t *uuid;
1839 xfs_lsn_t lsn = -1;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001840
1841 /* v4 filesystems always recover immediately */
1842 if (!xfs_sb_version_hascrc(&mp->m_sb))
1843 goto recover_immediately;
1844
1845 magic32 = be32_to_cpu(*(__be32 *)blk);
1846 switch (magic32) {
1847 case XFS_ABTB_CRC_MAGIC:
1848 case XFS_ABTC_CRC_MAGIC:
1849 case XFS_ABTB_MAGIC:
1850 case XFS_ABTC_MAGIC:
1851 case XFS_IBT_CRC_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10001852 case XFS_IBT_MAGIC: {
1853 struct xfs_btree_block *btb = blk;
1854
1855 lsn = be64_to_cpu(btb->bb_u.s.bb_lsn);
1856 uuid = &btb->bb_u.s.bb_uuid;
1857 break;
1858 }
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001859 case XFS_BMAP_CRC_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10001860 case XFS_BMAP_MAGIC: {
1861 struct xfs_btree_block *btb = blk;
1862
1863 lsn = be64_to_cpu(btb->bb_u.l.bb_lsn);
1864 uuid = &btb->bb_u.l.bb_uuid;
1865 break;
1866 }
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001867 case XFS_AGF_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10001868 lsn = be64_to_cpu(((struct xfs_agf *)blk)->agf_lsn);
1869 uuid = &((struct xfs_agf *)blk)->agf_uuid;
1870 break;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001871 case XFS_AGFL_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10001872 lsn = be64_to_cpu(((struct xfs_agfl *)blk)->agfl_lsn);
1873 uuid = &((struct xfs_agfl *)blk)->agfl_uuid;
1874 break;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001875 case XFS_AGI_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10001876 lsn = be64_to_cpu(((struct xfs_agi *)blk)->agi_lsn);
1877 uuid = &((struct xfs_agi *)blk)->agi_uuid;
1878 break;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001879 case XFS_SYMLINK_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10001880 lsn = be64_to_cpu(((struct xfs_dsymlink_hdr *)blk)->sl_lsn);
1881 uuid = &((struct xfs_dsymlink_hdr *)blk)->sl_uuid;
1882 break;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001883 case XFS_DIR3_BLOCK_MAGIC:
1884 case XFS_DIR3_DATA_MAGIC:
1885 case XFS_DIR3_FREE_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10001886 lsn = be64_to_cpu(((struct xfs_dir3_blk_hdr *)blk)->lsn);
1887 uuid = &((struct xfs_dir3_blk_hdr *)blk)->uuid;
1888 break;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001889 case XFS_ATTR3_RMT_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10001890 lsn = be64_to_cpu(((struct xfs_attr3_rmt_hdr *)blk)->rm_lsn);
1891 uuid = &((struct xfs_attr3_rmt_hdr *)blk)->rm_uuid;
1892 break;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001893 case XFS_SB_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10001894 lsn = be64_to_cpu(((struct xfs_dsb *)blk)->sb_lsn);
1895 uuid = &((struct xfs_dsb *)blk)->sb_uuid;
1896 break;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001897 default:
1898 break;
1899 }
1900
Dave Chinner566055d2013-09-24 16:01:16 +10001901 if (lsn != (xfs_lsn_t)-1) {
1902 if (!uuid_equal(&mp->m_sb.sb_uuid, uuid))
1903 goto recover_immediately;
1904 return lsn;
1905 }
1906
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001907 magicda = be16_to_cpu(((struct xfs_da_blkinfo *)blk)->magic);
1908 switch (magicda) {
1909 case XFS_DIR3_LEAF1_MAGIC:
1910 case XFS_DIR3_LEAFN_MAGIC:
1911 case XFS_DA3_NODE_MAGIC:
Dave Chinner566055d2013-09-24 16:01:16 +10001912 lsn = be64_to_cpu(((struct xfs_da3_blkinfo *)blk)->lsn);
1913 uuid = &((struct xfs_da3_blkinfo *)blk)->uuid;
1914 break;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001915 default:
1916 break;
1917 }
1918
Dave Chinner566055d2013-09-24 16:01:16 +10001919 if (lsn != (xfs_lsn_t)-1) {
1920 if (!uuid_equal(&mp->m_sb.sb_uuid, uuid))
1921 goto recover_immediately;
1922 return lsn;
1923 }
1924
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001925 /*
1926 * We do individual object checks on dquot and inode buffers as they
1927 * have their own individual LSN records. Also, we could have a stale
1928 * buffer here, so we have to at least recognise these buffer types.
1929 *
1930 * A notd complexity here is inode unlinked list processing - it logs
1931 * the inode directly in the buffer, but we don't know which inodes have
1932 * been modified, and there is no global buffer LSN. Hence we need to
1933 * recover all inode buffer types immediately. This problem will be
1934 * fixed by logical logging of the unlinked list modifications.
1935 */
1936 magic16 = be16_to_cpu(*(__be16 *)blk);
1937 switch (magic16) {
1938 case XFS_DQUOT_MAGIC:
1939 case XFS_DINODE_MAGIC:
1940 goto recover_immediately;
1941 default:
1942 break;
1943 }
1944
1945 /* unknown buffer contents, recover immediately */
1946
1947recover_immediately:
1948 return (xfs_lsn_t)-1;
1949
1950}
1951
1952/*
Dave Chinnerd75afeb2013-04-03 16:11:29 +11001953 * Validate the recovered buffer is of the correct type and attach the
1954 * appropriate buffer operations to them for writeback. Magic numbers are in a
1955 * few places:
1956 * the first 16 bits of the buffer (inode buffer, dquot buffer),
1957 * the first 32 bits of the buffer (most blocks),
1958 * inside a struct xfs_da_blkinfo at the start of the buffer.
1959 */
1960static void
Dave Chinner50d5c8d2013-08-28 21:22:47 +10001961xlog_recover_validate_buf_type(
Dave Chinnerd75afeb2013-04-03 16:11:29 +11001962 struct xfs_mount *mp,
1963 struct xfs_buf *bp,
1964 xfs_buf_log_format_t *buf_f)
1965{
1966 struct xfs_da_blkinfo *info = bp->b_addr;
1967 __uint32_t magic32;
1968 __uint16_t magic16;
1969 __uint16_t magicda;
1970
Dave Chinner67dc2882014-08-04 12:43:06 +10001971 /*
1972 * We can only do post recovery validation on items on CRC enabled
1973 * fielsystems as we need to know when the buffer was written to be able
1974 * to determine if we should have replayed the item. If we replay old
1975 * metadata over a newer buffer, then it will enter a temporarily
1976 * inconsistent state resulting in verification failures. Hence for now
1977 * just avoid the verification stage for non-crc filesystems
1978 */
1979 if (!xfs_sb_version_hascrc(&mp->m_sb))
1980 return;
1981
Dave Chinnerd75afeb2013-04-03 16:11:29 +11001982 magic32 = be32_to_cpu(*(__be32 *)bp->b_addr);
1983 magic16 = be16_to_cpu(*(__be16*)bp->b_addr);
1984 magicda = be16_to_cpu(info->magic);
Dave Chinner61fe1352013-04-03 16:11:30 +11001985 switch (xfs_blft_from_flags(buf_f)) {
1986 case XFS_BLFT_BTREE_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11001987 switch (magic32) {
1988 case XFS_ABTB_CRC_MAGIC:
1989 case XFS_ABTC_CRC_MAGIC:
1990 case XFS_ABTB_MAGIC:
1991 case XFS_ABTC_MAGIC:
1992 bp->b_ops = &xfs_allocbt_buf_ops;
1993 break;
1994 case XFS_IBT_CRC_MAGIC:
Brian Fosteraafc3c22014-04-24 16:00:52 +10001995 case XFS_FIBT_CRC_MAGIC:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11001996 case XFS_IBT_MAGIC:
Brian Fosteraafc3c22014-04-24 16:00:52 +10001997 case XFS_FIBT_MAGIC:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11001998 bp->b_ops = &xfs_inobt_buf_ops;
1999 break;
2000 case XFS_BMAP_CRC_MAGIC:
2001 case XFS_BMAP_MAGIC:
2002 bp->b_ops = &xfs_bmbt_buf_ops;
2003 break;
2004 default:
2005 xfs_warn(mp, "Bad btree block magic!");
2006 ASSERT(0);
2007 break;
2008 }
2009 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002010 case XFS_BLFT_AGF_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002011 if (magic32 != XFS_AGF_MAGIC) {
2012 xfs_warn(mp, "Bad AGF block magic!");
2013 ASSERT(0);
2014 break;
2015 }
2016 bp->b_ops = &xfs_agf_buf_ops;
2017 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002018 case XFS_BLFT_AGFL_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002019 if (magic32 != XFS_AGFL_MAGIC) {
2020 xfs_warn(mp, "Bad AGFL block magic!");
2021 ASSERT(0);
2022 break;
2023 }
2024 bp->b_ops = &xfs_agfl_buf_ops;
2025 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002026 case XFS_BLFT_AGI_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002027 if (magic32 != XFS_AGI_MAGIC) {
2028 xfs_warn(mp, "Bad AGI block magic!");
2029 ASSERT(0);
2030 break;
2031 }
2032 bp->b_ops = &xfs_agi_buf_ops;
2033 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002034 case XFS_BLFT_UDQUOT_BUF:
2035 case XFS_BLFT_PDQUOT_BUF:
2036 case XFS_BLFT_GDQUOT_BUF:
Dave Chinner123887e2013-04-30 21:39:33 +10002037#ifdef CONFIG_XFS_QUOTA
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002038 if (magic16 != XFS_DQUOT_MAGIC) {
2039 xfs_warn(mp, "Bad DQUOT block magic!");
2040 ASSERT(0);
2041 break;
2042 }
2043 bp->b_ops = &xfs_dquot_buf_ops;
Dave Chinner123887e2013-04-30 21:39:33 +10002044#else
2045 xfs_alert(mp,
2046 "Trying to recover dquots without QUOTA support built in!");
2047 ASSERT(0);
2048#endif
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002049 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002050 case XFS_BLFT_DINO_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002051 if (magic16 != XFS_DINODE_MAGIC) {
2052 xfs_warn(mp, "Bad INODE block magic!");
2053 ASSERT(0);
2054 break;
2055 }
2056 bp->b_ops = &xfs_inode_buf_ops;
2057 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002058 case XFS_BLFT_SYMLINK_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002059 if (magic32 != XFS_SYMLINK_MAGIC) {
2060 xfs_warn(mp, "Bad symlink block magic!");
2061 ASSERT(0);
2062 break;
2063 }
2064 bp->b_ops = &xfs_symlink_buf_ops;
2065 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002066 case XFS_BLFT_DIR_BLOCK_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002067 if (magic32 != XFS_DIR2_BLOCK_MAGIC &&
2068 magic32 != XFS_DIR3_BLOCK_MAGIC) {
2069 xfs_warn(mp, "Bad dir block magic!");
2070 ASSERT(0);
2071 break;
2072 }
2073 bp->b_ops = &xfs_dir3_block_buf_ops;
2074 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002075 case XFS_BLFT_DIR_DATA_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002076 if (magic32 != XFS_DIR2_DATA_MAGIC &&
2077 magic32 != XFS_DIR3_DATA_MAGIC) {
2078 xfs_warn(mp, "Bad dir data magic!");
2079 ASSERT(0);
2080 break;
2081 }
2082 bp->b_ops = &xfs_dir3_data_buf_ops;
2083 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002084 case XFS_BLFT_DIR_FREE_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002085 if (magic32 != XFS_DIR2_FREE_MAGIC &&
2086 magic32 != XFS_DIR3_FREE_MAGIC) {
2087 xfs_warn(mp, "Bad dir3 free magic!");
2088 ASSERT(0);
2089 break;
2090 }
2091 bp->b_ops = &xfs_dir3_free_buf_ops;
2092 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002093 case XFS_BLFT_DIR_LEAF1_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002094 if (magicda != XFS_DIR2_LEAF1_MAGIC &&
2095 magicda != XFS_DIR3_LEAF1_MAGIC) {
2096 xfs_warn(mp, "Bad dir leaf1 magic!");
2097 ASSERT(0);
2098 break;
2099 }
2100 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
2101 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002102 case XFS_BLFT_DIR_LEAFN_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002103 if (magicda != XFS_DIR2_LEAFN_MAGIC &&
2104 magicda != XFS_DIR3_LEAFN_MAGIC) {
2105 xfs_warn(mp, "Bad dir leafn magic!");
2106 ASSERT(0);
2107 break;
2108 }
2109 bp->b_ops = &xfs_dir3_leafn_buf_ops;
2110 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002111 case XFS_BLFT_DA_NODE_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002112 if (magicda != XFS_DA_NODE_MAGIC &&
2113 magicda != XFS_DA3_NODE_MAGIC) {
2114 xfs_warn(mp, "Bad da node magic!");
2115 ASSERT(0);
2116 break;
2117 }
2118 bp->b_ops = &xfs_da3_node_buf_ops;
2119 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002120 case XFS_BLFT_ATTR_LEAF_BUF:
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002121 if (magicda != XFS_ATTR_LEAF_MAGIC &&
2122 magicda != XFS_ATTR3_LEAF_MAGIC) {
2123 xfs_warn(mp, "Bad attr leaf magic!");
2124 ASSERT(0);
2125 break;
2126 }
2127 bp->b_ops = &xfs_attr3_leaf_buf_ops;
2128 break;
Dave Chinner61fe1352013-04-03 16:11:30 +11002129 case XFS_BLFT_ATTR_RMT_BUF:
Dave Chinnercab09a82013-04-30 21:39:36 +10002130 if (magic32 != XFS_ATTR3_RMT_MAGIC) {
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002131 xfs_warn(mp, "Bad attr remote magic!");
2132 ASSERT(0);
2133 break;
2134 }
2135 bp->b_ops = &xfs_attr3_rmt_buf_ops;
2136 break;
Dave Chinner04a1e6c2013-04-03 16:11:31 +11002137 case XFS_BLFT_SB_BUF:
2138 if (magic32 != XFS_SB_MAGIC) {
2139 xfs_warn(mp, "Bad SB block magic!");
2140 ASSERT(0);
2141 break;
2142 }
2143 bp->b_ops = &xfs_sb_buf_ops;
2144 break;
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002145 default:
Dave Chinner61fe1352013-04-03 16:11:30 +11002146 xfs_warn(mp, "Unknown buffer type %d!",
2147 xfs_blft_from_flags(buf_f));
Dave Chinnerd75afeb2013-04-03 16:11:29 +11002148 break;
2149 }
2150}
2151
2152/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153 * Perform a 'normal' buffer recovery. Each logged region of the
2154 * buffer should be copied over the corresponding region in the
2155 * given buffer. The bitmap in the buf log format structure indicates
2156 * where to place the logged data.
2157 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158STATIC void
2159xlog_recover_do_reg_buffer(
Dave Chinner9abbc532010-04-13 15:06:46 +10002160 struct xfs_mount *mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 xlog_recover_item_t *item,
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002162 struct xfs_buf *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163 xfs_buf_log_format_t *buf_f)
2164{
2165 int i;
2166 int bit;
2167 int nbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 int error;
2169
Dave Chinner9abbc532010-04-13 15:06:46 +10002170 trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f);
2171
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172 bit = 0;
2173 i = 1; /* 0 is the buf format structure */
2174 while (1) {
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002175 bit = xfs_next_bit(buf_f->blf_data_map,
2176 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177 if (bit == -1)
2178 break;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002179 nbits = xfs_contig_bits(buf_f->blf_data_map,
2180 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181 ASSERT(nbits > 0);
Christoph Hellwig4b809162007-08-16 15:37:36 +10002182 ASSERT(item->ri_buf[i].i_addr != NULL);
Dave Chinnerc1155412010-05-07 11:05:19 +10002183 ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0);
Dave Chinneraa0e8832012-04-23 15:58:52 +10002184 ASSERT(BBTOB(bp->b_io_length) >=
2185 ((uint)bit << XFS_BLF_SHIFT) + (nbits << XFS_BLF_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186
2187 /*
Dave Chinner709da6a2013-05-27 16:38:23 +10002188 * The dirty regions logged in the buffer, even though
2189 * contiguous, may span multiple chunks. This is because the
2190 * dirty region may span a physical page boundary in a buffer
2191 * and hence be split into two separate vectors for writing into
2192 * the log. Hence we need to trim nbits back to the length of
2193 * the current region being copied out of the log.
2194 */
2195 if (item->ri_buf[i].i_len < (nbits << XFS_BLF_SHIFT))
2196 nbits = item->ri_buf[i].i_len >> XFS_BLF_SHIFT;
2197
2198 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199 * Do a sanity check if this is a dquot buffer. Just checking
2200 * the first dquot in the buffer should do. XXXThis is
2201 * probably a good thing to do for other buf types also.
2202 */
2203 error = 0;
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002204 if (buf_f->blf_flags &
Dave Chinnerc1155412010-05-07 11:05:19 +10002205 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002206 if (item->ri_buf[i].i_addr == NULL) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002207 xfs_alert(mp,
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002208 "XFS: NULL dquot in %s.", __func__);
2209 goto next;
2210 }
Jan Rekorajski8ec6dba2009-11-16 11:57:02 +00002211 if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002212 xfs_alert(mp,
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002213 "XFS: dquot too small (%d) in %s.",
2214 item->ri_buf[i].i_len, __func__);
2215 goto next;
2216 }
Dave Chinner9aede1d2013-10-15 09:17:52 +11002217 error = xfs_dqcheck(mp, item->ri_buf[i].i_addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 -1, 0, XFS_QMOPT_DOWARN,
2219 "dquot_buf_recover");
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002220 if (error)
2221 goto next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 }
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002223
2224 memcpy(xfs_buf_offset(bp,
Dave Chinnerc1155412010-05-07 11:05:19 +10002225 (uint)bit << XFS_BLF_SHIFT), /* dest */
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002226 item->ri_buf[i].i_addr, /* source */
Dave Chinnerc1155412010-05-07 11:05:19 +10002227 nbits<<XFS_BLF_SHIFT); /* length */
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002228 next:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229 i++;
2230 bit += nbits;
2231 }
2232
2233 /* Shouldn't be any more regions */
2234 ASSERT(i == item->ri_total);
Christoph Hellwigee1a47a2013-04-21 14:53:46 -05002235
Dave Chinner67dc2882014-08-04 12:43:06 +10002236 xlog_recover_validate_buf_type(mp, bp, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237}
2238
2239/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 * Perform a dquot buffer recovery.
Zhi Yong Wu8ba701e2013-08-12 03:15:01 +00002241 * Simple algorithm: if we have found a QUOTAOFF log item of the same type
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2243 * Else, treat it as a regular buffer and do recovery.
Dave Chinnerad3714b2014-08-04 12:59:31 +10002244 *
2245 * Return false if the buffer was tossed and true if we recovered the buffer to
2246 * indicate to the caller if the buffer needs writing.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 */
Dave Chinnerad3714b2014-08-04 12:59:31 +10002248STATIC bool
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249xlog_recover_do_dquot_buffer(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002250 struct xfs_mount *mp,
2251 struct xlog *log,
2252 struct xlog_recover_item *item,
2253 struct xfs_buf *bp,
2254 struct xfs_buf_log_format *buf_f)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255{
2256 uint type;
2257
Dave Chinner9abbc532010-04-13 15:06:46 +10002258 trace_xfs_log_recover_buf_dquot_buf(log, buf_f);
2259
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 /*
2261 * Filesystems are required to send in quota flags at mount time.
2262 */
Dave Chinnerad3714b2014-08-04 12:59:31 +10002263 if (!mp->m_qflags)
2264 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265
2266 type = 0;
Dave Chinnerc1155412010-05-07 11:05:19 +10002267 if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268 type |= XFS_DQ_USER;
Dave Chinnerc1155412010-05-07 11:05:19 +10002269 if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF)
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002270 type |= XFS_DQ_PROJ;
Dave Chinnerc1155412010-05-07 11:05:19 +10002271 if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 type |= XFS_DQ_GROUP;
2273 /*
2274 * This type of quotas was turned off, so ignore this buffer
2275 */
2276 if (log->l_quotaoffs_flag & type)
Dave Chinnerad3714b2014-08-04 12:59:31 +10002277 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278
Dave Chinner9abbc532010-04-13 15:06:46 +10002279 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
Dave Chinnerad3714b2014-08-04 12:59:31 +10002280 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281}
2282
2283/*
2284 * This routine replays a modification made to a buffer at runtime.
2285 * There are actually two types of buffer, regular and inode, which
2286 * are handled differently. Inode buffers are handled differently
2287 * in that we only recover a specific set of data from them, namely
2288 * the inode di_next_unlinked fields. This is because all other inode
2289 * data is actually logged via inode records and any data we replay
2290 * here which overlaps that may be stale.
2291 *
2292 * When meta-data buffers are freed at run time we log a buffer item
Dave Chinnerc1155412010-05-07 11:05:19 +10002293 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 * of the buffer in the log should not be replayed at recovery time.
2295 * This is so that if the blocks covered by the buffer are reused for
2296 * file data before we crash we don't end up replaying old, freed
2297 * meta-data into a user's file.
2298 *
2299 * To handle the cancellation of buffer log items, we make two passes
2300 * over the log during recovery. During the first we build a table of
2301 * those buffers which have been cancelled, and during the second we
2302 * only replay those buffers which do not have corresponding cancel
Zhi Yong Wu34be5ff2013-08-07 10:11:07 +00002303 * records in the table. See xlog_recover_buffer_pass[1,2] above
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 * for more details on the implementation of the table of cancel records.
2305 */
2306STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002307xlog_recover_buffer_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002308 struct xlog *log,
2309 struct list_head *buffer_list,
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002310 struct xlog_recover_item *item,
2311 xfs_lsn_t current_lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312{
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002313 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002314 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 xfs_buf_t *bp;
2316 int error;
Christoph Hellwig6ad112b2009-11-24 18:02:23 +00002317 uint buf_flags;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002318 xfs_lsn_t lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002320 /*
2321 * In this pass we only want to recover all the buffers which have
2322 * not been cancelled and are not cancellation buffers themselves.
2323 */
2324 if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno,
2325 buf_f->blf_len, buf_f->blf_flags)) {
2326 trace_xfs_log_recover_buf_cancel(log, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328 }
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002329
Dave Chinner9abbc532010-04-13 15:06:46 +10002330 trace_xfs_log_recover_buf_recover(log, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331
Dave Chinnera8acad72012-04-23 15:58:54 +10002332 buf_flags = 0;
Dave Chinner611c9942012-04-23 15:59:07 +10002333 if (buf_f->blf_flags & XFS_BLF_INODE_BUF)
2334 buf_flags |= XBF_UNMAPPED;
Christoph Hellwig6ad112b2009-11-24 18:02:23 +00002335
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002336 bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len,
Dave Chinnerc3f8fc72012-11-12 22:54:01 +11002337 buf_flags, NULL);
Chandra Seetharamanac4d6882011-08-03 02:18:29 +00002338 if (!bp)
Dave Chinner24513372014-06-25 14:58:08 +10002339 return -ENOMEM;
Chandra Seetharamane5702802011-08-03 02:18:34 +00002340 error = bp->b_error;
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +00002341 if (error) {
Christoph Hellwig901796a2011-10-10 16:52:49 +00002342 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)");
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002343 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344 }
2345
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002346 /*
Dave Chinner67dc2882014-08-04 12:43:06 +10002347 * Recover the buffer only if we get an LSN from it and it's less than
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002348 * the lsn of the transaction we are replaying.
Dave Chinner67dc2882014-08-04 12:43:06 +10002349 *
2350 * Note that we have to be extremely careful of readahead here.
2351 * Readahead does not attach verfiers to the buffers so if we don't
2352 * actually do any replay after readahead because of the LSN we found
2353 * in the buffer if more recent than that current transaction then we
2354 * need to attach the verifier directly. Failure to do so can lead to
2355 * future recovery actions (e.g. EFI and unlinked list recovery) can
2356 * operate on the buffers and they won't get the verifier attached. This
2357 * can lead to blocks on disk having the correct content but a stale
2358 * CRC.
2359 *
2360 * It is safe to assume these clean buffers are currently up to date.
2361 * If the buffer is dirtied by a later transaction being replayed, then
2362 * the verifier will be reset to match whatever recover turns that
2363 * buffer into.
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002364 */
2365 lsn = xlog_recover_get_buf_lsn(mp, bp);
Dave Chinner67dc2882014-08-04 12:43:06 +10002366 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
2367 xlog_recover_validate_buf_type(mp, bp, buf_f);
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002368 goto out_release;
Dave Chinner67dc2882014-08-04 12:43:06 +10002369 }
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002370
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002371 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
Dave Chinnerad3714b2014-08-04 12:59:31 +10002373 if (error)
2374 goto out_release;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002375 } else if (buf_f->blf_flags &
Dave Chinnerc1155412010-05-07 11:05:19 +10002376 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
Dave Chinnerad3714b2014-08-04 12:59:31 +10002377 bool dirty;
2378
2379 dirty = xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2380 if (!dirty)
2381 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 } else {
Dave Chinner9abbc532010-04-13 15:06:46 +10002383 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385
2386 /*
2387 * Perform delayed write on the buffer. Asynchronous writes will be
2388 * slower when taking into account all the buffers to be flushed.
2389 *
2390 * Also make sure that only inode buffers with good sizes stay in
2391 * the buffer cache. The kernel moves inodes in buffers of 1 block
Jie Liu0f49efd2013-12-13 15:51:48 +11002392 * or mp->m_inode_cluster_size bytes, whichever is bigger. The inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393 * buffers in the log can be a different size if the log was generated
2394 * by an older kernel using unclustered inode buffers or a newer kernel
2395 * running with a different inode cluster size. Regardless, if the
Jie Liu0f49efd2013-12-13 15:51:48 +11002396 * the inode buffer size isn't MAX(blocksize, mp->m_inode_cluster_size)
2397 * for *our* value of mp->m_inode_cluster_size, then we need to keep
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398 * the buffer out of the buffer cache so that the buffer won't
2399 * overlap with future reads of those inodes.
2400 */
2401 if (XFS_DINODE_MAGIC ==
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002402 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
Dave Chinneraa0e8832012-04-23 15:58:52 +10002403 (BBTOB(bp->b_io_length) != MAX(log->l_mp->m_sb.sb_blocksize,
Jie Liu0f49efd2013-12-13 15:51:48 +11002404 (__uint32_t)log->l_mp->m_inode_cluster_size))) {
Christoph Hellwigc867cb62011-10-10 16:52:46 +00002405 xfs_buf_stale(bp);
Christoph Hellwigc2b006c2011-08-23 08:28:07 +00002406 error = xfs_bwrite(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407 } else {
Dave Chinnerebad8612010-09-22 10:47:20 +10002408 ASSERT(bp->b_target->bt_mount == mp);
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02002409 bp->b_iodone = xlog_recover_iodone;
Christoph Hellwig43ff2122012-04-23 15:58:39 +10002410 xfs_buf_delwri_queue(bp, buffer_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411 }
2412
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002413out_release:
Christoph Hellwigc2b006c2011-08-23 08:28:07 +00002414 xfs_buf_relse(bp);
2415 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416}
2417
Dave Chinner638f44162013-08-30 10:23:45 +10002418/*
2419 * Inode fork owner changes
2420 *
2421 * If we have been told that we have to reparent the inode fork, it's because an
2422 * extent swap operation on a CRC enabled filesystem has been done and we are
2423 * replaying it. We need to walk the BMBT of the appropriate fork and change the
2424 * owners of it.
2425 *
2426 * The complexity here is that we don't have an inode context to work with, so
2427 * after we've replayed the inode we need to instantiate one. This is where the
2428 * fun begins.
2429 *
2430 * We are in the middle of log recovery, so we can't run transactions. That
2431 * means we cannot use cache coherent inode instantiation via xfs_iget(), as
2432 * that will result in the corresponding iput() running the inode through
2433 * xfs_inactive(). If we've just replayed an inode core that changes the link
2434 * count to zero (i.e. it's been unlinked), then xfs_inactive() will run
2435 * transactions (bad!).
2436 *
2437 * So, to avoid this, we instantiate an inode directly from the inode core we've
2438 * just recovered. We have the buffer still locked, and all we really need to
2439 * instantiate is the inode core and the forks being modified. We can do this
2440 * manually, then run the inode btree owner change, and then tear down the
2441 * xfs_inode without having to run any transactions at all.
2442 *
2443 * Also, because we don't have a transaction context available here but need to
2444 * gather all the buffers we modify for writeback so we pass the buffer_list
2445 * instead for the operation to use.
2446 */
2447
2448STATIC int
2449xfs_recover_inode_owner_change(
2450 struct xfs_mount *mp,
2451 struct xfs_dinode *dip,
2452 struct xfs_inode_log_format *in_f,
2453 struct list_head *buffer_list)
2454{
2455 struct xfs_inode *ip;
2456 int error;
2457
2458 ASSERT(in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER));
2459
2460 ip = xfs_inode_alloc(mp, in_f->ilf_ino);
2461 if (!ip)
Dave Chinner24513372014-06-25 14:58:08 +10002462 return -ENOMEM;
Dave Chinner638f44162013-08-30 10:23:45 +10002463
2464 /* instantiate the inode */
2465 xfs_dinode_from_disk(&ip->i_d, dip);
2466 ASSERT(ip->i_d.di_version >= 3);
2467
2468 error = xfs_iformat_fork(ip, dip);
2469 if (error)
2470 goto out_free_ip;
2471
2472
2473 if (in_f->ilf_fields & XFS_ILOG_DOWNER) {
2474 ASSERT(in_f->ilf_fields & XFS_ILOG_DBROOT);
2475 error = xfs_bmbt_change_owner(NULL, ip, XFS_DATA_FORK,
2476 ip->i_ino, buffer_list);
2477 if (error)
2478 goto out_free_ip;
2479 }
2480
2481 if (in_f->ilf_fields & XFS_ILOG_AOWNER) {
2482 ASSERT(in_f->ilf_fields & XFS_ILOG_ABROOT);
2483 error = xfs_bmbt_change_owner(NULL, ip, XFS_ATTR_FORK,
2484 ip->i_ino, buffer_list);
2485 if (error)
2486 goto out_free_ip;
2487 }
2488
2489out_free_ip:
2490 xfs_inode_free(ip);
2491 return error;
2492}
2493
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002495xlog_recover_inode_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002496 struct xlog *log,
2497 struct list_head *buffer_list,
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002498 struct xlog_recover_item *item,
2499 xfs_lsn_t current_lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002500{
2501 xfs_inode_log_format_t *in_f;
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002502 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503 xfs_buf_t *bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504 xfs_dinode_t *dip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505 int len;
2506 xfs_caddr_t src;
2507 xfs_caddr_t dest;
2508 int error;
2509 int attr_index;
2510 uint fields;
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002511 xfs_icdinode_t *dicp;
Christoph Hellwig93848a92013-04-03 16:11:17 +11002512 uint isize;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002513 int need_free = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514
Tim Shimmin6d192a92006-06-09 14:55:38 +10002515 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002516 in_f = item->ri_buf[0].i_addr;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002517 } else {
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002518 in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002519 need_free = 1;
2520 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2521 if (error)
2522 goto error;
2523 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524
2525 /*
2526 * Inode buffers can be freed, look out for it,
2527 * and do not replay the inode.
2528 */
Christoph Hellwiga1941892008-11-28 14:23:40 +11002529 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
2530 in_f->ilf_len, 0)) {
Tim Shimmin6d192a92006-06-09 14:55:38 +10002531 error = 0;
Dave Chinner9abbc532010-04-13 15:06:46 +10002532 trace_xfs_log_recover_inode_cancel(log, in_f);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002533 goto error;
2534 }
Dave Chinner9abbc532010-04-13 15:06:46 +10002535 trace_xfs_log_recover_inode_recover(log, in_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536
Dave Chinnerc3f8fc72012-11-12 22:54:01 +11002537 bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len, 0,
Christoph Hellwig93848a92013-04-03 16:11:17 +11002538 &xfs_inode_buf_ops);
Chandra Seetharamanac4d6882011-08-03 02:18:29 +00002539 if (!bp) {
Dave Chinner24513372014-06-25 14:58:08 +10002540 error = -ENOMEM;
Chandra Seetharamanac4d6882011-08-03 02:18:29 +00002541 goto error;
2542 }
Chandra Seetharamane5702802011-08-03 02:18:34 +00002543 error = bp->b_error;
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +00002544 if (error) {
Christoph Hellwig901796a2011-10-10 16:52:49 +00002545 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)");
Dave Chinner638f44162013-08-30 10:23:45 +10002546 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
Christoph Hellwiga1941892008-11-28 14:23:40 +11002549 dip = (xfs_dinode_t *)xfs_buf_offset(bp, in_f->ilf_boffset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550
2551 /*
2552 * Make sure the place we're flushing out to really looks
2553 * like an inode!
2554 */
Christoph Hellwig69ef9212011-07-08 14:36:05 +02002555 if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002556 xfs_alert(mp,
2557 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
2558 __func__, dip, bp, in_f->ilf_ino);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002559 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560 XFS_ERRLEVEL_LOW, mp);
Dave Chinner24513372014-06-25 14:58:08 +10002561 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002562 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563 }
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002564 dicp = item->ri_buf[1].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002566 xfs_alert(mp,
2567 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
2568 __func__, item, in_f->ilf_ino);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002569 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570 XFS_ERRLEVEL_LOW, mp);
Dave Chinner24513372014-06-25 14:58:08 +10002571 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002572 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573 }
2574
Dave Chinnere60896d2013-07-24 15:47:30 +10002575 /*
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002576 * If the inode has an LSN in it, recover the inode only if it's less
Dave Chinner638f44162013-08-30 10:23:45 +10002577 * than the lsn of the transaction we are replaying. Note: we still
2578 * need to replay an owner change even though the inode is more recent
2579 * than the transaction as there is no guarantee that all the btree
2580 * blocks are more recent than this transaction, too.
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002581 */
2582 if (dip->di_version >= 3) {
2583 xfs_lsn_t lsn = be64_to_cpu(dip->di_lsn);
2584
2585 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
2586 trace_xfs_log_recover_inode_skip(log, in_f);
2587 error = 0;
Dave Chinner638f44162013-08-30 10:23:45 +10002588 goto out_owner_change;
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002589 }
2590 }
2591
2592 /*
Dave Chinnere60896d2013-07-24 15:47:30 +10002593 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
2594 * are transactional and if ordering is necessary we can determine that
2595 * more accurately by the LSN field in the V3 inode core. Don't trust
2596 * the inode versions we might be changing them here - use the
2597 * superblock flag to determine whether we need to look at di_flushiter
2598 * to skip replay when the on disk inode is newer than the log one
2599 */
2600 if (!xfs_sb_version_hascrc(&mp->m_sb) &&
2601 dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602 /*
2603 * Deal with the wrap case, DI_MAX_FLUSH is less
2604 * than smaller numbers
2605 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002606 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002607 dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608 /* do nothing */
2609 } else {
Dave Chinner9abbc532010-04-13 15:06:46 +10002610 trace_xfs_log_recover_inode_skip(log, in_f);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002611 error = 0;
Dave Chinner638f44162013-08-30 10:23:45 +10002612 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613 }
2614 }
Dave Chinnere60896d2013-07-24 15:47:30 +10002615
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616 /* Take the opportunity to reset the flush iteration count */
2617 dicp->di_flushiter = 0;
2618
Al Viroabbede12011-07-26 02:31:30 -04002619 if (unlikely(S_ISREG(dicp->di_mode))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2621 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002622 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 XFS_ERRLEVEL_LOW, mp, dicp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002624 xfs_alert(mp,
2625 "%s: Bad regular inode log record, rec ptr 0x%p, "
2626 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2627 __func__, item, dip, bp, in_f->ilf_ino);
Dave Chinner24513372014-06-25 14:58:08 +10002628 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002629 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630 }
Al Viroabbede12011-07-26 02:31:30 -04002631 } else if (unlikely(S_ISDIR(dicp->di_mode))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2633 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2634 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002635 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636 XFS_ERRLEVEL_LOW, mp, dicp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002637 xfs_alert(mp,
2638 "%s: Bad dir inode log record, rec ptr 0x%p, "
2639 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2640 __func__, item, dip, bp, in_f->ilf_ino);
Dave Chinner24513372014-06-25 14:58:08 +10002641 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002642 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643 }
2644 }
2645 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002646 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647 XFS_ERRLEVEL_LOW, mp, dicp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002648 xfs_alert(mp,
2649 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2650 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2651 __func__, item, dip, bp, in_f->ilf_ino,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652 dicp->di_nextents + dicp->di_anextents,
2653 dicp->di_nblocks);
Dave Chinner24513372014-06-25 14:58:08 +10002654 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002655 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002656 }
2657 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002658 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659 XFS_ERRLEVEL_LOW, mp, dicp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002660 xfs_alert(mp,
2661 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2662 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__,
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002663 item, dip, bp, in_f->ilf_ino, dicp->di_forkoff);
Dave Chinner24513372014-06-25 14:58:08 +10002664 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002665 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 }
Christoph Hellwig93848a92013-04-03 16:11:17 +11002667 isize = xfs_icdinode_size(dicp->di_version);
2668 if (unlikely(item->ri_buf[1].i_len > isize)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002669 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670 XFS_ERRLEVEL_LOW, mp, dicp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002671 xfs_alert(mp,
2672 "%s: Bad inode log record length %d, rec ptr 0x%p",
2673 __func__, item->ri_buf[1].i_len, item);
Dave Chinner24513372014-06-25 14:58:08 +10002674 error = -EFSCORRUPTED;
Dave Chinner638f44162013-08-30 10:23:45 +10002675 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676 }
2677
2678 /* The core is in in-core format */
Christoph Hellwig93848a92013-04-03 16:11:17 +11002679 xfs_dinode_to_disk(dip, dicp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680
2681 /* the rest is in on-disk format */
Christoph Hellwig93848a92013-04-03 16:11:17 +11002682 if (item->ri_buf[1].i_len > isize) {
2683 memcpy((char *)dip + isize,
2684 item->ri_buf[1].i_addr + isize,
2685 item->ri_buf[1].i_len - isize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686 }
2687
2688 fields = in_f->ilf_fields;
2689 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2690 case XFS_ILOG_DEV:
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002691 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692 break;
2693 case XFS_ILOG_UUID:
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002694 memcpy(XFS_DFORK_DPTR(dip),
2695 &in_f->ilf_u.ilfu_uuid,
2696 sizeof(uuid_t));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697 break;
2698 }
2699
2700 if (in_f->ilf_size == 2)
Dave Chinner638f44162013-08-30 10:23:45 +10002701 goto out_owner_change;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702 len = item->ri_buf[2].i_len;
2703 src = item->ri_buf[2].i_addr;
2704 ASSERT(in_f->ilf_size <= 4);
2705 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2706 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2707 (len == in_f->ilf_dsize));
2708
2709 switch (fields & XFS_ILOG_DFORK) {
2710 case XFS_ILOG_DDATA:
2711 case XFS_ILOG_DEXT:
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002712 memcpy(XFS_DFORK_DPTR(dip), src, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713 break;
2714
2715 case XFS_ILOG_DBROOT:
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002716 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002717 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718 XFS_DFORK_DSIZE(dip, mp));
2719 break;
2720
2721 default:
2722 /*
2723 * There are no data fork flags set.
2724 */
2725 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2726 break;
2727 }
2728
2729 /*
2730 * If we logged any attribute data, recover it. There may or
2731 * may not have been any other non-core data logged in this
2732 * transaction.
2733 */
2734 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2735 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2736 attr_index = 3;
2737 } else {
2738 attr_index = 2;
2739 }
2740 len = item->ri_buf[attr_index].i_len;
2741 src = item->ri_buf[attr_index].i_addr;
2742 ASSERT(len == in_f->ilf_asize);
2743
2744 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2745 case XFS_ILOG_ADATA:
2746 case XFS_ILOG_AEXT:
2747 dest = XFS_DFORK_APTR(dip);
2748 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2749 memcpy(dest, src, len);
2750 break;
2751
2752 case XFS_ILOG_ABROOT:
2753 dest = XFS_DFORK_APTR(dip);
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002754 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
2755 len, (xfs_bmdr_block_t*)dest,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002756 XFS_DFORK_ASIZE(dip, mp));
2757 break;
2758
2759 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002760 xfs_warn(log->l_mp, "%s: Invalid flag", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761 ASSERT(0);
Dave Chinner24513372014-06-25 14:58:08 +10002762 error = -EIO;
Dave Chinner638f44162013-08-30 10:23:45 +10002763 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002764 }
2765 }
2766
Dave Chinner638f44162013-08-30 10:23:45 +10002767out_owner_change:
2768 if (in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER))
2769 error = xfs_recover_inode_owner_change(mp, dip, in_f,
2770 buffer_list);
Christoph Hellwig93848a92013-04-03 16:11:17 +11002771 /* re-generate the checksum. */
2772 xfs_dinode_calc_crc(log->l_mp, dip);
2773
Dave Chinnerebad8612010-09-22 10:47:20 +10002774 ASSERT(bp->b_target->bt_mount == mp);
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02002775 bp->b_iodone = xlog_recover_iodone;
Christoph Hellwig43ff2122012-04-23 15:58:39 +10002776 xfs_buf_delwri_queue(bp, buffer_list);
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002777
2778out_release:
Christoph Hellwig61551f12011-08-23 08:28:06 +00002779 xfs_buf_relse(bp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002780error:
2781 if (need_free)
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002782 kmem_free(in_f);
Eric Sandeenb474c7a2014-06-22 15:04:54 +10002783 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784}
2785
2786/*
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002787 * Recover QUOTAOFF records. We simply make a note of it in the xlog
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2789 * of that type.
2790 */
2791STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002792xlog_recover_quotaoff_pass1(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002793 struct xlog *log,
2794 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002795{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002796 xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797 ASSERT(qoff_f);
2798
2799 /*
2800 * The logitem format's flag tells us if this was user quotaoff,
Nathan Scott77a7cce2006-01-11 15:35:57 +11002801 * group/project quotaoff or both.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802 */
2803 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2804 log->l_quotaoffs_flag |= XFS_DQ_USER;
Nathan Scott77a7cce2006-01-11 15:35:57 +11002805 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2806 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002807 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2808 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2809
Eric Sandeend99831f2014-06-22 15:03:54 +10002810 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811}
2812
2813/*
2814 * Recover a dquot record
2815 */
2816STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002817xlog_recover_dquot_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002818 struct xlog *log,
2819 struct list_head *buffer_list,
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002820 struct xlog_recover_item *item,
2821 xfs_lsn_t current_lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002823 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002824 xfs_buf_t *bp;
2825 struct xfs_disk_dquot *ddq, *recddq;
2826 int error;
2827 xfs_dq_logformat_t *dq_f;
2828 uint type;
2829
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830
2831 /*
2832 * Filesystems are required to send in quota flags at mount time.
2833 */
2834 if (mp->m_qflags == 0)
Eric Sandeend99831f2014-06-22 15:03:54 +10002835 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002837 recddq = item->ri_buf[1].i_addr;
2838 if (recddq == NULL) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002839 xfs_alert(log->l_mp, "NULL dquot in %s.", __func__);
Dave Chinner24513372014-06-25 14:58:08 +10002840 return -EIO;
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002841 }
Jan Rekorajski8ec6dba2009-11-16 11:57:02 +00002842 if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002843 xfs_alert(log->l_mp, "dquot too small (%d) in %s.",
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002844 item->ri_buf[1].i_len, __func__);
Dave Chinner24513372014-06-25 14:58:08 +10002845 return -EIO;
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002846 }
2847
Linus Torvalds1da177e2005-04-16 15:20:36 -07002848 /*
2849 * This type of quotas was turned off, so ignore this record.
2850 */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002851 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852 ASSERT(type);
2853 if (log->l_quotaoffs_flag & type)
Eric Sandeend99831f2014-06-22 15:03:54 +10002854 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855
2856 /*
2857 * At this point we know that quota was _not_ turned off.
2858 * Since the mount flags are not indicating to us otherwise, this
2859 * must mean that quota is on, and the dquot needs to be replayed.
2860 * Remember that we may not have fully recovered the superblock yet,
2861 * so we can't do the usual trick of looking at the SB quota bits.
2862 *
2863 * The other possibility, of course, is that the quota subsystem was
2864 * removed since the last mount - ENOSYS.
2865 */
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002866 dq_f = item->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867 ASSERT(dq_f);
Dave Chinner9aede1d2013-10-15 09:17:52 +11002868 error = xfs_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002869 "xlog_recover_dquot_pass2 (log copy)");
2870 if (error)
Dave Chinner24513372014-06-25 14:58:08 +10002871 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872 ASSERT(dq_f->qlf_len == 1);
2873
Dave Chinnerad3714b2014-08-04 12:59:31 +10002874 /*
2875 * At this point we are assuming that the dquots have been allocated
2876 * and hence the buffer has valid dquots stamped in it. It should,
2877 * therefore, pass verifier validation. If the dquot is bad, then the
2878 * we'll return an error here, so we don't need to specifically check
2879 * the dquot in the buffer after the verifier has run.
2880 */
Dave Chinner7ca790a2012-04-23 15:58:55 +10002881 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dq_f->qlf_blkno,
Dave Chinnerc3f8fc72012-11-12 22:54:01 +11002882 XFS_FSB_TO_BB(mp, dq_f->qlf_len), 0, &bp,
Dave Chinnerad3714b2014-08-04 12:59:31 +10002883 &xfs_dquot_buf_ops);
Dave Chinner7ca790a2012-04-23 15:58:55 +10002884 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885 return error;
Dave Chinner7ca790a2012-04-23 15:58:55 +10002886
Linus Torvalds1da177e2005-04-16 15:20:36 -07002887 ASSERT(bp);
2888 ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
2889
2890 /*
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002891 * If the dquot has an LSN in it, recover the dquot only if it's less
2892 * than the lsn of the transaction we are replaying.
2893 */
2894 if (xfs_sb_version_hascrc(&mp->m_sb)) {
2895 struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddq;
2896 xfs_lsn_t lsn = be64_to_cpu(dqb->dd_lsn);
2897
2898 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
2899 goto out_release;
2900 }
2901 }
2902
Linus Torvalds1da177e2005-04-16 15:20:36 -07002903 memcpy(ddq, recddq, item->ri_buf[1].i_len);
Dave Chinner6fcdc592013-06-03 15:28:46 +10002904 if (xfs_sb_version_hascrc(&mp->m_sb)) {
2905 xfs_update_cksum((char *)ddq, sizeof(struct xfs_dqblk),
2906 XFS_DQUOT_CRC_OFF);
2907 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002908
2909 ASSERT(dq_f->qlf_size == 2);
Dave Chinnerebad8612010-09-22 10:47:20 +10002910 ASSERT(bp->b_target->bt_mount == mp);
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02002911 bp->b_iodone = xlog_recover_iodone;
Christoph Hellwig43ff2122012-04-23 15:58:39 +10002912 xfs_buf_delwri_queue(bp, buffer_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002913
Dave Chinner50d5c8d2013-08-28 21:22:47 +10002914out_release:
2915 xfs_buf_relse(bp);
2916 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002917}
2918
2919/*
2920 * This routine is called to create an in-core extent free intent
2921 * item from the efi format structure which was logged on disk.
2922 * It allocates an in-core efi, copies the extents from the format
2923 * structure into it, and adds the efi to the AIL with the given
2924 * LSN.
2925 */
Tim Shimmin6d192a92006-06-09 14:55:38 +10002926STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002927xlog_recover_efi_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002928 struct xlog *log,
2929 struct xlog_recover_item *item,
2930 xfs_lsn_t lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931{
Tim Shimmin6d192a92006-06-09 14:55:38 +10002932 int error;
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002933 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934 xfs_efi_log_item_t *efip;
2935 xfs_efi_log_format_t *efi_formatp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002936
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002937 efi_formatp = item->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002938
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002940 if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
2941 &(efip->efi_format)))) {
2942 xfs_efi_item_free(efip);
2943 return error;
2944 }
Dave Chinnerb199c8a2010-12-20 11:59:49 +11002945 atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002946
David Chinnera9c21c12008-10-30 17:39:35 +11002947 spin_lock(&log->l_ailp->xa_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002948 /*
David Chinner783a2f62008-10-30 17:39:58 +11002949 * xfs_trans_ail_update() drops the AIL lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002950 */
Dave Chinnere6059942010-12-20 12:34:26 +11002951 xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002952 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002953}
2954
2955
2956/*
2957 * This routine is called when an efd format structure is found in
2958 * a committed transaction in the log. It's purpose is to cancel
2959 * the corresponding efi if it was still in the log. To do this
2960 * it searches the AIL for the efi with an id equal to that in the
2961 * efd format structure. If we find it, we remove the efi from the
2962 * AIL and free it.
2963 */
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002964STATIC int
2965xlog_recover_efd_pass2(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002966 struct xlog *log,
2967 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002968{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002969 xfs_efd_log_format_t *efd_formatp;
2970 xfs_efi_log_item_t *efip = NULL;
2971 xfs_log_item_t *lip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002972 __uint64_t efi_id;
David Chinner27d8d5f2008-10-30 17:38:39 +11002973 struct xfs_ail_cursor cur;
David Chinner783a2f62008-10-30 17:39:58 +11002974 struct xfs_ail *ailp = log->l_ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002976 efd_formatp = item->ri_buf[0].i_addr;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002977 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
2978 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
2979 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
2980 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002981 efi_id = efd_formatp->efd_efi_id;
2982
2983 /*
2984 * Search for the efi with the id in the efd format structure
2985 * in the AIL.
2986 */
David Chinnera9c21c12008-10-30 17:39:35 +11002987 spin_lock(&ailp->xa_lock);
2988 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989 while (lip != NULL) {
2990 if (lip->li_type == XFS_LI_EFI) {
2991 efip = (xfs_efi_log_item_t *)lip;
2992 if (efip->efi_format.efi_id == efi_id) {
2993 /*
David Chinner783a2f62008-10-30 17:39:58 +11002994 * xfs_trans_ail_delete() drops the
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995 * AIL lock.
2996 */
Dave Chinner04913fd2012-04-23 15:58:41 +10002997 xfs_trans_ail_delete(ailp, lip,
2998 SHUTDOWN_CORRUPT_INCORE);
David Chinner8ae2c0f2007-11-23 16:28:17 +11002999 xfs_efi_item_free(efip);
David Chinnera9c21c12008-10-30 17:39:35 +11003000 spin_lock(&ailp->xa_lock);
David Chinner27d8d5f2008-10-30 17:38:39 +11003001 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003002 }
3003 }
David Chinnera9c21c12008-10-30 17:39:35 +11003004 lip = xfs_trans_ail_cursor_next(ailp, &cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005 }
Eric Sandeene4a1e292014-04-14 19:06:05 +10003006 xfs_trans_ail_cursor_done(&cur);
David Chinnera9c21c12008-10-30 17:39:35 +11003007 spin_unlock(&ailp->xa_lock);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003008
3009 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003010}
3011
3012/*
Dave Chinner28c8e412013-06-27 16:04:55 +10003013 * This routine is called when an inode create format structure is found in a
3014 * committed transaction in the log. It's purpose is to initialise the inodes
3015 * being allocated on disk. This requires us to get inode cluster buffers that
3016 * match the range to be intialised, stamped with inode templates and written
3017 * by delayed write so that subsequent modifications will hit the cached buffer
3018 * and only need writing out at the end of recovery.
3019 */
3020STATIC int
3021xlog_recover_do_icreate_pass2(
3022 struct xlog *log,
3023 struct list_head *buffer_list,
3024 xlog_recover_item_t *item)
3025{
3026 struct xfs_mount *mp = log->l_mp;
3027 struct xfs_icreate_log *icl;
3028 xfs_agnumber_t agno;
3029 xfs_agblock_t agbno;
3030 unsigned int count;
3031 unsigned int isize;
3032 xfs_agblock_t length;
3033
3034 icl = (struct xfs_icreate_log *)item->ri_buf[0].i_addr;
3035 if (icl->icl_type != XFS_LI_ICREATE) {
3036 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad type");
Dave Chinner24513372014-06-25 14:58:08 +10003037 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003038 }
3039
3040 if (icl->icl_size != 1) {
3041 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad icl size");
Dave Chinner24513372014-06-25 14:58:08 +10003042 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003043 }
3044
3045 agno = be32_to_cpu(icl->icl_ag);
3046 if (agno >= mp->m_sb.sb_agcount) {
3047 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agno");
Dave Chinner24513372014-06-25 14:58:08 +10003048 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003049 }
3050 agbno = be32_to_cpu(icl->icl_agbno);
3051 if (!agbno || agbno == NULLAGBLOCK || agbno >= mp->m_sb.sb_agblocks) {
3052 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agbno");
Dave Chinner24513372014-06-25 14:58:08 +10003053 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003054 }
3055 isize = be32_to_cpu(icl->icl_isize);
3056 if (isize != mp->m_sb.sb_inodesize) {
3057 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad isize");
Dave Chinner24513372014-06-25 14:58:08 +10003058 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003059 }
3060 count = be32_to_cpu(icl->icl_count);
3061 if (!count) {
3062 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count");
Dave Chinner24513372014-06-25 14:58:08 +10003063 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003064 }
3065 length = be32_to_cpu(icl->icl_length);
3066 if (!length || length >= mp->m_sb.sb_agblocks) {
3067 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad length");
Dave Chinner24513372014-06-25 14:58:08 +10003068 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003069 }
3070
3071 /* existing allocation is fixed value */
Jie Liu71783432013-12-13 15:51:46 +11003072 ASSERT(count == mp->m_ialloc_inos);
Jie Liu126cd102013-12-13 15:51:48 +11003073 ASSERT(length == mp->m_ialloc_blks);
Jie Liu71783432013-12-13 15:51:46 +11003074 if (count != mp->m_ialloc_inos ||
Jie Liu126cd102013-12-13 15:51:48 +11003075 length != mp->m_ialloc_blks) {
Dave Chinner28c8e412013-06-27 16:04:55 +10003076 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count 2");
Dave Chinner24513372014-06-25 14:58:08 +10003077 return -EINVAL;
Dave Chinner28c8e412013-06-27 16:04:55 +10003078 }
3079
3080 /*
3081 * Inode buffers can be freed. Do not replay the inode initialisation as
3082 * we could be overwriting something written after this inode buffer was
3083 * cancelled.
3084 *
3085 * XXX: we need to iterate all buffers and only init those that are not
3086 * cancelled. I think that a more fine grained factoring of
3087 * xfs_ialloc_inode_init may be appropriate here to enable this to be
3088 * done easily.
3089 */
3090 if (xlog_check_buffer_cancelled(log,
3091 XFS_AGB_TO_DADDR(mp, agno, agbno), length, 0))
3092 return 0;
3093
3094 xfs_ialloc_inode_init(mp, NULL, buffer_list, agno, agbno, length,
3095 be32_to_cpu(icl->icl_gen));
3096 return 0;
3097}
3098
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003099STATIC void
3100xlog_recover_buffer_ra_pass2(
3101 struct xlog *log,
3102 struct xlog_recover_item *item)
3103{
3104 struct xfs_buf_log_format *buf_f = item->ri_buf[0].i_addr;
3105 struct xfs_mount *mp = log->l_mp;
3106
Dave Chinner84a5b732013-08-27 08:10:53 +10003107 if (xlog_peek_buffer_cancelled(log, buf_f->blf_blkno,
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003108 buf_f->blf_len, buf_f->blf_flags)) {
3109 return;
3110 }
3111
3112 xfs_buf_readahead(mp->m_ddev_targp, buf_f->blf_blkno,
3113 buf_f->blf_len, NULL);
3114}
3115
3116STATIC void
3117xlog_recover_inode_ra_pass2(
3118 struct xlog *log,
3119 struct xlog_recover_item *item)
3120{
3121 struct xfs_inode_log_format ilf_buf;
3122 struct xfs_inode_log_format *ilfp;
3123 struct xfs_mount *mp = log->l_mp;
3124 int error;
3125
3126 if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) {
3127 ilfp = item->ri_buf[0].i_addr;
3128 } else {
3129 ilfp = &ilf_buf;
3130 memset(ilfp, 0, sizeof(*ilfp));
3131 error = xfs_inode_item_format_convert(&item->ri_buf[0], ilfp);
3132 if (error)
3133 return;
3134 }
3135
Dave Chinner84a5b732013-08-27 08:10:53 +10003136 if (xlog_peek_buffer_cancelled(log, ilfp->ilf_blkno, ilfp->ilf_len, 0))
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003137 return;
3138
3139 xfs_buf_readahead(mp->m_ddev_targp, ilfp->ilf_blkno,
Dave Chinnerd8914002013-08-27 11:39:37 +10003140 ilfp->ilf_len, &xfs_inode_buf_ra_ops);
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003141}
3142
3143STATIC void
3144xlog_recover_dquot_ra_pass2(
3145 struct xlog *log,
3146 struct xlog_recover_item *item)
3147{
3148 struct xfs_mount *mp = log->l_mp;
3149 struct xfs_disk_dquot *recddq;
3150 struct xfs_dq_logformat *dq_f;
3151 uint type;
3152
3153
3154 if (mp->m_qflags == 0)
3155 return;
3156
3157 recddq = item->ri_buf[1].i_addr;
3158 if (recddq == NULL)
3159 return;
3160 if (item->ri_buf[1].i_len < sizeof(struct xfs_disk_dquot))
3161 return;
3162
3163 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
3164 ASSERT(type);
3165 if (log->l_quotaoffs_flag & type)
3166 return;
3167
3168 dq_f = item->ri_buf[0].i_addr;
3169 ASSERT(dq_f);
3170 ASSERT(dq_f->qlf_len == 1);
3171
3172 xfs_buf_readahead(mp->m_ddev_targp, dq_f->qlf_blkno,
Dave Chinner0f0d3342013-08-27 13:25:43 +10003173 XFS_FSB_TO_BB(mp, dq_f->qlf_len), NULL);
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003174}
3175
3176STATIC void
3177xlog_recover_ra_pass2(
3178 struct xlog *log,
3179 struct xlog_recover_item *item)
3180{
3181 switch (ITEM_TYPE(item)) {
3182 case XFS_LI_BUF:
3183 xlog_recover_buffer_ra_pass2(log, item);
3184 break;
3185 case XFS_LI_INODE:
3186 xlog_recover_inode_ra_pass2(log, item);
3187 break;
3188 case XFS_LI_DQUOT:
3189 xlog_recover_dquot_ra_pass2(log, item);
3190 break;
3191 case XFS_LI_EFI:
3192 case XFS_LI_EFD:
3193 case XFS_LI_QUOTAOFF:
3194 default:
3195 break;
3196 }
3197}
3198
Linus Torvalds1da177e2005-04-16 15:20:36 -07003199STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003200xlog_recover_commit_pass1(
Mark Tinguelyad223e62012-06-14 09:22:15 -05003201 struct xlog *log,
3202 struct xlog_recover *trans,
3203 struct xlog_recover_item *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003204{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003205 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1);
Christoph Hellwigd0450942010-12-01 22:06:23 +00003206
3207 switch (ITEM_TYPE(item)) {
3208 case XFS_LI_BUF:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003209 return xlog_recover_buffer_pass1(log, item);
Christoph Hellwigd0450942010-12-01 22:06:23 +00003210 case XFS_LI_QUOTAOFF:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003211 return xlog_recover_quotaoff_pass1(log, item);
3212 case XFS_LI_INODE:
3213 case XFS_LI_EFI:
3214 case XFS_LI_EFD:
3215 case XFS_LI_DQUOT:
Dave Chinner28c8e412013-06-27 16:04:55 +10003216 case XFS_LI_ICREATE:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003217 /* nothing to do in pass 1 */
3218 return 0;
Christoph Hellwigd0450942010-12-01 22:06:23 +00003219 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003220 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
3221 __func__, ITEM_TYPE(item));
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003222 ASSERT(0);
Dave Chinner24513372014-06-25 14:58:08 +10003223 return -EIO;
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003224 }
3225}
3226
3227STATIC int
3228xlog_recover_commit_pass2(
Mark Tinguelyad223e62012-06-14 09:22:15 -05003229 struct xlog *log,
3230 struct xlog_recover *trans,
3231 struct list_head *buffer_list,
3232 struct xlog_recover_item *item)
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003233{
3234 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2);
3235
3236 switch (ITEM_TYPE(item)) {
3237 case XFS_LI_BUF:
Dave Chinner50d5c8d2013-08-28 21:22:47 +10003238 return xlog_recover_buffer_pass2(log, buffer_list, item,
3239 trans->r_lsn);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003240 case XFS_LI_INODE:
Dave Chinner50d5c8d2013-08-28 21:22:47 +10003241 return xlog_recover_inode_pass2(log, buffer_list, item,
3242 trans->r_lsn);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003243 case XFS_LI_EFI:
3244 return xlog_recover_efi_pass2(log, item, trans->r_lsn);
3245 case XFS_LI_EFD:
3246 return xlog_recover_efd_pass2(log, item);
3247 case XFS_LI_DQUOT:
Dave Chinner50d5c8d2013-08-28 21:22:47 +10003248 return xlog_recover_dquot_pass2(log, buffer_list, item,
3249 trans->r_lsn);
Dave Chinner28c8e412013-06-27 16:04:55 +10003250 case XFS_LI_ICREATE:
3251 return xlog_recover_do_icreate_pass2(log, buffer_list, item);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003252 case XFS_LI_QUOTAOFF:
3253 /* nothing to do in pass2 */
3254 return 0;
3255 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003256 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
3257 __func__, ITEM_TYPE(item));
Christoph Hellwigd0450942010-12-01 22:06:23 +00003258 ASSERT(0);
Dave Chinner24513372014-06-25 14:58:08 +10003259 return -EIO;
Christoph Hellwigd0450942010-12-01 22:06:23 +00003260 }
3261}
3262
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003263STATIC int
3264xlog_recover_items_pass2(
3265 struct xlog *log,
3266 struct xlog_recover *trans,
3267 struct list_head *buffer_list,
3268 struct list_head *item_list)
3269{
3270 struct xlog_recover_item *item;
3271 int error = 0;
3272
3273 list_for_each_entry(item, item_list, ri_list) {
3274 error = xlog_recover_commit_pass2(log, trans,
3275 buffer_list, item);
3276 if (error)
3277 return error;
3278 }
3279
3280 return error;
3281}
3282
Christoph Hellwigd0450942010-12-01 22:06:23 +00003283/*
3284 * Perform the transaction.
3285 *
3286 * If the transaction modifies a buffer or inode, do it now. Otherwise,
3287 * EFIs and EFDs get queued up by adding entries into the AIL for them.
3288 */
3289STATIC int
3290xlog_recover_commit_trans(
Mark Tinguelyad223e62012-06-14 09:22:15 -05003291 struct xlog *log,
Christoph Hellwigd0450942010-12-01 22:06:23 +00003292 struct xlog_recover *trans,
3293 int pass)
3294{
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003295 int error = 0;
3296 int error2;
3297 int items_queued = 0;
3298 struct xlog_recover_item *item;
3299 struct xlog_recover_item *next;
3300 LIST_HEAD (buffer_list);
3301 LIST_HEAD (ra_list);
3302 LIST_HEAD (done_list);
3303
3304 #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100
Linus Torvalds1da177e2005-04-16 15:20:36 -07003305
Dave Chinnerf0a76952010-01-11 11:49:57 +00003306 hlist_del(&trans->r_list);
Christoph Hellwigd0450942010-12-01 22:06:23 +00003307
3308 error = xlog_recover_reorder_trans(log, trans, pass);
3309 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003310 return error;
Christoph Hellwigd0450942010-12-01 22:06:23 +00003311
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003312 list_for_each_entry_safe(item, next, &trans->r_itemq, ri_list) {
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003313 switch (pass) {
3314 case XLOG_RECOVER_PASS1:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00003315 error = xlog_recover_commit_pass1(log, trans, item);
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003316 break;
3317 case XLOG_RECOVER_PASS2:
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003318 xlog_recover_ra_pass2(log, item);
3319 list_move_tail(&item->ri_list, &ra_list);
3320 items_queued++;
3321 if (items_queued >= XLOG_RECOVER_COMMIT_QUEUE_MAX) {
3322 error = xlog_recover_items_pass2(log, trans,
3323 &buffer_list, &ra_list);
3324 list_splice_tail_init(&ra_list, &done_list);
3325 items_queued = 0;
3326 }
3327
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003328 break;
3329 default:
3330 ASSERT(0);
3331 }
3332
Christoph Hellwigd0450942010-12-01 22:06:23 +00003333 if (error)
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003334 goto out;
Christoph Hellwigd0450942010-12-01 22:06:23 +00003335 }
3336
Zhi Yong Wu00574da2013-08-14 15:16:03 +08003337out:
3338 if (!list_empty(&ra_list)) {
3339 if (!error)
3340 error = xlog_recover_items_pass2(log, trans,
3341 &buffer_list, &ra_list);
3342 list_splice_tail_init(&ra_list, &done_list);
3343 }
3344
3345 if (!list_empty(&done_list))
3346 list_splice_init(&done_list, &trans->r_itemq);
3347
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003348 error2 = xfs_buf_delwri_submit(&buffer_list);
3349 return error ? error : error2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003350}
3351
Dave Chinner76560662014-09-29 09:45:42 +10003352STATIC void
3353xlog_recover_add_item(
3354 struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003355{
Dave Chinner76560662014-09-29 09:45:42 +10003356 xlog_recover_item_t *item;
3357
3358 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
3359 INIT_LIST_HEAD(&item->ri_list);
3360 list_add_tail(&item->ri_list, head);
3361}
3362
3363STATIC int
3364xlog_recover_add_to_cont_trans(
3365 struct xlog *log,
3366 struct xlog_recover *trans,
3367 xfs_caddr_t dp,
3368 int len)
3369{
3370 xlog_recover_item_t *item;
3371 xfs_caddr_t ptr, old_ptr;
3372 int old_len;
3373
3374 if (list_empty(&trans->r_itemq)) {
3375 /* finish copying rest of trans header */
3376 xlog_recover_add_item(&trans->r_itemq);
3377 ptr = (xfs_caddr_t) &trans->r_theader +
3378 sizeof(xfs_trans_header_t) - len;
3379 memcpy(ptr, dp, len);
3380 return 0;
3381 }
3382 /* take the tail entry */
3383 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
3384
3385 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
3386 old_len = item->ri_buf[item->ri_cnt-1].i_len;
3387
3388 ptr = kmem_realloc(old_ptr, len+old_len, old_len, KM_SLEEP);
3389 memcpy(&ptr[old_len], dp, len);
3390 item->ri_buf[item->ri_cnt-1].i_len += len;
3391 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
3392 trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003393 return 0;
3394}
3395
3396/*
Dave Chinner76560662014-09-29 09:45:42 +10003397 * The next region to add is the start of a new region. It could be
3398 * a whole region or it could be the first part of a new region. Because
3399 * of this, the assumption here is that the type and size fields of all
3400 * format structures fit into the first 32 bits of the structure.
3401 *
3402 * This works because all regions must be 32 bit aligned. Therefore, we
3403 * either have both fields or we have neither field. In the case we have
3404 * neither field, the data part of the region is zero length. We only have
3405 * a log_op_header and can throw away the header since a new one will appear
3406 * later. If we have at least 4 bytes, then we can determine how many regions
3407 * will appear in the current log item.
3408 */
3409STATIC int
3410xlog_recover_add_to_trans(
3411 struct xlog *log,
3412 struct xlog_recover *trans,
3413 xfs_caddr_t dp,
3414 int len)
3415{
3416 xfs_inode_log_format_t *in_f; /* any will do */
3417 xlog_recover_item_t *item;
3418 xfs_caddr_t ptr;
3419
3420 if (!len)
3421 return 0;
3422 if (list_empty(&trans->r_itemq)) {
3423 /* we need to catch log corruptions here */
3424 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
3425 xfs_warn(log->l_mp, "%s: bad header magic number",
3426 __func__);
3427 ASSERT(0);
3428 return -EIO;
3429 }
3430 if (len == sizeof(xfs_trans_header_t))
3431 xlog_recover_add_item(&trans->r_itemq);
3432 memcpy(&trans->r_theader, dp, len);
3433 return 0;
3434 }
3435
3436 ptr = kmem_alloc(len, KM_SLEEP);
3437 memcpy(ptr, dp, len);
3438 in_f = (xfs_inode_log_format_t *)ptr;
3439
3440 /* take the tail entry */
3441 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
3442 if (item->ri_total != 0 &&
3443 item->ri_total == item->ri_cnt) {
3444 /* tail item is in use, get a new one */
3445 xlog_recover_add_item(&trans->r_itemq);
3446 item = list_entry(trans->r_itemq.prev,
3447 xlog_recover_item_t, ri_list);
3448 }
3449
3450 if (item->ri_total == 0) { /* first region to be added */
3451 if (in_f->ilf_size == 0 ||
3452 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
3453 xfs_warn(log->l_mp,
3454 "bad number of regions (%d) in inode log format",
3455 in_f->ilf_size);
3456 ASSERT(0);
3457 kmem_free(ptr);
3458 return -EIO;
3459 }
3460
3461 item->ri_total = in_f->ilf_size;
3462 item->ri_buf =
3463 kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
3464 KM_SLEEP);
3465 }
3466 ASSERT(item->ri_total > item->ri_cnt);
3467 /* Description region is ri_buf[0] */
3468 item->ri_buf[item->ri_cnt].i_addr = ptr;
3469 item->ri_buf[item->ri_cnt].i_len = len;
3470 item->ri_cnt++;
3471 trace_xfs_log_recover_item_add(log, trans, item, 0);
3472 return 0;
3473}
Dave Chinnerb818cca2014-09-29 09:45:54 +10003474
Dave Chinner76560662014-09-29 09:45:42 +10003475/*
3476 * Free up any resources allocated by the transaction
3477 *
3478 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
3479 */
3480STATIC void
3481xlog_recover_free_trans(
3482 struct xlog_recover *trans)
3483{
3484 xlog_recover_item_t *item, *n;
3485 int i;
3486
3487 list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
3488 /* Free the regions in the item. */
3489 list_del(&item->ri_list);
3490 for (i = 0; i < item->ri_cnt; i++)
3491 kmem_free(item->ri_buf[i].i_addr);
3492 /* Free the item itself */
3493 kmem_free(item->ri_buf);
3494 kmem_free(item);
3495 }
3496 /* Free the transaction recover structure */
3497 kmem_free(trans);
3498}
3499
Dave Chinnere9131e52014-09-29 09:45:18 +10003500/*
3501 * On error or completion, trans is freed.
3502 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003503STATIC int
Dave Chinnereeb11682014-09-29 09:45:03 +10003504xlog_recovery_process_trans(
3505 struct xlog *log,
3506 struct xlog_recover *trans,
3507 xfs_caddr_t dp,
3508 unsigned int len,
3509 unsigned int flags,
3510 int pass)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003511{
Dave Chinnere9131e52014-09-29 09:45:18 +10003512 int error = 0;
3513 bool freeit = false;
Dave Chinnereeb11682014-09-29 09:45:03 +10003514
3515 /* mask off ophdr transaction container flags */
3516 flags &= ~XLOG_END_TRANS;
3517 if (flags & XLOG_WAS_CONT_TRANS)
3518 flags &= ~XLOG_CONTINUE_TRANS;
3519
Dave Chinner88b863d2014-09-29 09:45:32 +10003520 /*
3521 * Callees must not free the trans structure. We'll decide if we need to
3522 * free it or not based on the operation being done and it's result.
3523 */
Dave Chinnereeb11682014-09-29 09:45:03 +10003524 switch (flags) {
3525 /* expected flag values */
3526 case 0:
3527 case XLOG_CONTINUE_TRANS:
3528 error = xlog_recover_add_to_trans(log, trans, dp, len);
3529 break;
3530 case XLOG_WAS_CONT_TRANS:
3531 error = xlog_recover_add_to_cont_trans(log, trans, dp, len);
3532 break;
3533 case XLOG_COMMIT_TRANS:
3534 error = xlog_recover_commit_trans(log, trans, pass);
Dave Chinner88b863d2014-09-29 09:45:32 +10003535 /* success or fail, we are now done with this transaction. */
3536 freeit = true;
Dave Chinnereeb11682014-09-29 09:45:03 +10003537 break;
3538
3539 /* unexpected flag values */
3540 case XLOG_UNMOUNT_TRANS:
Dave Chinnere9131e52014-09-29 09:45:18 +10003541 /* just skip trans */
Dave Chinnereeb11682014-09-29 09:45:03 +10003542 xfs_warn(log->l_mp, "%s: Unmount LR", __func__);
Dave Chinnere9131e52014-09-29 09:45:18 +10003543 freeit = true;
Dave Chinnereeb11682014-09-29 09:45:03 +10003544 break;
3545 case XLOG_START_TRANS:
Dave Chinnereeb11682014-09-29 09:45:03 +10003546 default:
3547 xfs_warn(log->l_mp, "%s: bad flag 0x%x", __func__, flags);
3548 ASSERT(0);
Dave Chinnere9131e52014-09-29 09:45:18 +10003549 error = -EIO;
Dave Chinnereeb11682014-09-29 09:45:03 +10003550 break;
3551 }
Dave Chinnere9131e52014-09-29 09:45:18 +10003552 if (error || freeit)
3553 xlog_recover_free_trans(trans);
Dave Chinnereeb11682014-09-29 09:45:03 +10003554 return error;
3555}
3556
Dave Chinnerb818cca2014-09-29 09:45:54 +10003557/*
3558 * Lookup the transaction recovery structure associated with the ID in the
3559 * current ophdr. If the transaction doesn't exist and the start flag is set in
3560 * the ophdr, then allocate a new transaction for future ID matches to find.
3561 * Either way, return what we found during the lookup - an existing transaction
3562 * or nothing.
3563 */
Dave Chinnereeb11682014-09-29 09:45:03 +10003564STATIC struct xlog_recover *
3565xlog_recover_ophdr_to_trans(
3566 struct hlist_head rhash[],
3567 struct xlog_rec_header *rhead,
3568 struct xlog_op_header *ohead)
3569{
3570 struct xlog_recover *trans;
3571 xlog_tid_t tid;
3572 struct hlist_head *rhp;
3573
3574 tid = be32_to_cpu(ohead->oh_tid);
3575 rhp = &rhash[XLOG_RHASH(tid)];
Dave Chinnerb818cca2014-09-29 09:45:54 +10003576 hlist_for_each_entry(trans, rhp, r_list) {
3577 if (trans->r_log_tid == tid)
3578 return trans;
3579 }
Dave Chinnereeb11682014-09-29 09:45:03 +10003580
3581 /*
Dave Chinnerb818cca2014-09-29 09:45:54 +10003582 * skip over non-start transaction headers - we could be
3583 * processing slack space before the next transaction starts
Dave Chinnereeb11682014-09-29 09:45:03 +10003584 */
Dave Chinnerb818cca2014-09-29 09:45:54 +10003585 if (!(ohead->oh_flags & XLOG_START_TRANS))
3586 return NULL;
3587
3588 ASSERT(be32_to_cpu(ohead->oh_len) == 0);
3589
3590 /*
3591 * This is a new transaction so allocate a new recovery container to
3592 * hold the recovery ops that will follow.
3593 */
3594 trans = kmem_zalloc(sizeof(struct xlog_recover), KM_SLEEP);
3595 trans->r_log_tid = tid;
3596 trans->r_lsn = be64_to_cpu(rhead->h_lsn);
3597 INIT_LIST_HEAD(&trans->r_itemq);
3598 INIT_HLIST_NODE(&trans->r_list);
3599 hlist_add_head(&trans->r_list, rhp);
3600
3601 /*
3602 * Nothing more to do for this ophdr. Items to be added to this new
3603 * transaction will be in subsequent ophdr containers.
3604 */
Dave Chinnereeb11682014-09-29 09:45:03 +10003605 return NULL;
3606}
3607
3608STATIC int
3609xlog_recover_process_ophdr(
3610 struct xlog *log,
3611 struct hlist_head rhash[],
3612 struct xlog_rec_header *rhead,
3613 struct xlog_op_header *ohead,
3614 xfs_caddr_t dp,
3615 xfs_caddr_t end,
3616 int pass)
3617{
3618 struct xlog_recover *trans;
Dave Chinnereeb11682014-09-29 09:45:03 +10003619 unsigned int len;
3620
3621 /* Do we understand who wrote this op? */
3622 if (ohead->oh_clientid != XFS_TRANSACTION &&
3623 ohead->oh_clientid != XFS_LOG) {
3624 xfs_warn(log->l_mp, "%s: bad clientid 0x%x",
3625 __func__, ohead->oh_clientid);
3626 ASSERT(0);
3627 return -EIO;
3628 }
3629
3630 /*
3631 * Check the ophdr contains all the data it is supposed to contain.
3632 */
3633 len = be32_to_cpu(ohead->oh_len);
3634 if (dp + len > end) {
3635 xfs_warn(log->l_mp, "%s: bad length 0x%x", __func__, len);
3636 WARN_ON(1);
3637 return -EIO;
3638 }
3639
3640 trans = xlog_recover_ophdr_to_trans(rhash, rhead, ohead);
3641 if (!trans) {
3642 /* nothing to do, so skip over this ophdr */
3643 return 0;
3644 }
3645
Dave Chinnere9131e52014-09-29 09:45:18 +10003646 return xlog_recovery_process_trans(log, trans, dp, len,
3647 ohead->oh_flags, pass);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003648}
3649
3650/*
3651 * There are two valid states of the r_state field. 0 indicates that the
3652 * transaction structure is in a normal state. We have either seen the
3653 * start of the transaction or the last operation we added was not a partial
3654 * operation. If the last operation we added to the transaction was a
3655 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
3656 *
3657 * NOTE: skip LRs with 0 data length.
3658 */
3659STATIC int
3660xlog_recover_process_data(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003661 struct xlog *log,
Dave Chinnerf0a76952010-01-11 11:49:57 +00003662 struct hlist_head rhash[],
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003663 struct xlog_rec_header *rhead,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003664 xfs_caddr_t dp,
3665 int pass)
3666{
Dave Chinnereeb11682014-09-29 09:45:03 +10003667 struct xlog_op_header *ohead;
3668 xfs_caddr_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003669 int num_logops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003670 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003671
Dave Chinnereeb11682014-09-29 09:45:03 +10003672 end = dp + be32_to_cpu(rhead->h_len);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003673 num_logops = be32_to_cpu(rhead->h_num_logops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003674
3675 /* check the log format matches our own - else we can't recover */
3676 if (xlog_header_check_recover(log->l_mp, rhead))
Dave Chinner24513372014-06-25 14:58:08 +10003677 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003678
Dave Chinnereeb11682014-09-29 09:45:03 +10003679 while ((dp < end) && num_logops) {
3680
3681 ohead = (struct xlog_op_header *)dp;
3682 dp += sizeof(*ohead);
3683 ASSERT(dp <= end);
3684
3685 /* errors will abort recovery */
3686 error = xlog_recover_process_ophdr(log, rhash, rhead, ohead,
3687 dp, end, pass);
3688 if (error)
3689 return error;
3690
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10003691 dp += be32_to_cpu(ohead->oh_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003692 num_logops--;
3693 }
3694 return 0;
3695}
3696
3697/*
3698 * Process an extent free intent item that was recovered from
3699 * the log. We need to free the extents that it describes.
3700 */
David Chinner3c1e2bb2008-04-10 12:21:11 +10003701STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003702xlog_recover_process_efi(
3703 xfs_mount_t *mp,
3704 xfs_efi_log_item_t *efip)
3705{
3706 xfs_efd_log_item_t *efdp;
3707 xfs_trans_t *tp;
3708 int i;
David Chinner3c1e2bb2008-04-10 12:21:11 +10003709 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003710 xfs_extent_t *extp;
3711 xfs_fsblock_t startblock_fsb;
3712
Dave Chinnerb199c8a2010-12-20 11:59:49 +11003713 ASSERT(!test_bit(XFS_EFI_RECOVERED, &efip->efi_flags));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003714
3715 /*
3716 * First check the validity of the extents described by the
3717 * EFI. If any are bad, then assume that all are bad and
3718 * just toss the EFI.
3719 */
3720 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3721 extp = &(efip->efi_format.efi_extents[i]);
3722 startblock_fsb = XFS_BB_TO_FSB(mp,
3723 XFS_FSB_TO_DADDR(mp, extp->ext_start));
3724 if ((startblock_fsb == 0) ||
3725 (extp->ext_len == 0) ||
3726 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
3727 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
3728 /*
3729 * This will pull the EFI from the AIL and
3730 * free the memory associated with it.
3731 */
Dave Chinner666d6442013-04-03 14:09:21 +11003732 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003733 xfs_efi_release(efip, efip->efi_format.efi_nextents);
Dave Chinner24513372014-06-25 14:58:08 +10003734 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003735 }
3736 }
3737
3738 tp = xfs_trans_alloc(mp, 0);
Jie Liu3d3c8b52013-08-12 20:49:59 +10003739 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
David Chinnerfc6149d2008-04-10 12:21:53 +10003740 if (error)
3741 goto abort_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003742 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
3743
3744 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3745 extp = &(efip->efi_format.efi_extents[i]);
David Chinnerfc6149d2008-04-10 12:21:53 +10003746 error = xfs_free_extent(tp, extp->ext_start, extp->ext_len);
3747 if (error)
3748 goto abort_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003749 xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
3750 extp->ext_len);
3751 }
3752
Dave Chinnerb199c8a2010-12-20 11:59:49 +11003753 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
David Chinnere5720ee2008-04-10 12:21:18 +10003754 error = xfs_trans_commit(tp, 0);
David Chinner3c1e2bb2008-04-10 12:21:11 +10003755 return error;
David Chinnerfc6149d2008-04-10 12:21:53 +10003756
3757abort_error:
3758 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3759 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003760}
3761
3762/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003763 * When this is called, all of the EFIs which did not have
3764 * corresponding EFDs should be in the AIL. What we do now
3765 * is free the extents associated with each one.
3766 *
3767 * Since we process the EFIs in normal transactions, they
3768 * will be removed at some point after the commit. This prevents
3769 * us from just walking down the list processing each one.
3770 * We'll use a flag in the EFI to skip those that we've already
3771 * processed and use the AIL iteration mechanism's generation
3772 * count to try to speed this up at least a bit.
3773 *
3774 * When we start, we know that the EFIs are the only things in
3775 * the AIL. As we process them, however, other items are added
3776 * to the AIL. Since everything added to the AIL must come after
3777 * everything already in the AIL, we stop processing as soon as
3778 * we see something other than an EFI in the AIL.
3779 */
David Chinner3c1e2bb2008-04-10 12:21:11 +10003780STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003781xlog_recover_process_efis(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003782 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003783{
3784 xfs_log_item_t *lip;
3785 xfs_efi_log_item_t *efip;
David Chinner3c1e2bb2008-04-10 12:21:11 +10003786 int error = 0;
David Chinner27d8d5f2008-10-30 17:38:39 +11003787 struct xfs_ail_cursor cur;
David Chinnera9c21c12008-10-30 17:39:35 +11003788 struct xfs_ail *ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003789
David Chinnera9c21c12008-10-30 17:39:35 +11003790 ailp = log->l_ailp;
3791 spin_lock(&ailp->xa_lock);
3792 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003793 while (lip != NULL) {
3794 /*
3795 * We're done when we see something other than an EFI.
David Chinner27d8d5f2008-10-30 17:38:39 +11003796 * There should be no EFIs left in the AIL now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003797 */
3798 if (lip->li_type != XFS_LI_EFI) {
David Chinner27d8d5f2008-10-30 17:38:39 +11003799#ifdef DEBUG
David Chinnera9c21c12008-10-30 17:39:35 +11003800 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
David Chinner27d8d5f2008-10-30 17:38:39 +11003801 ASSERT(lip->li_type != XFS_LI_EFI);
3802#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003803 break;
3804 }
3805
3806 /*
3807 * Skip EFIs that we've already processed.
3808 */
3809 efip = (xfs_efi_log_item_t *)lip;
Dave Chinnerb199c8a2010-12-20 11:59:49 +11003810 if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) {
David Chinnera9c21c12008-10-30 17:39:35 +11003811 lip = xfs_trans_ail_cursor_next(ailp, &cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003812 continue;
3813 }
3814
David Chinnera9c21c12008-10-30 17:39:35 +11003815 spin_unlock(&ailp->xa_lock);
3816 error = xlog_recover_process_efi(log->l_mp, efip);
3817 spin_lock(&ailp->xa_lock);
David Chinner27d8d5f2008-10-30 17:38:39 +11003818 if (error)
3819 goto out;
David Chinnera9c21c12008-10-30 17:39:35 +11003820 lip = xfs_trans_ail_cursor_next(ailp, &cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003821 }
David Chinner27d8d5f2008-10-30 17:38:39 +11003822out:
Eric Sandeene4a1e292014-04-14 19:06:05 +10003823 xfs_trans_ail_cursor_done(&cur);
David Chinnera9c21c12008-10-30 17:39:35 +11003824 spin_unlock(&ailp->xa_lock);
David Chinner3c1e2bb2008-04-10 12:21:11 +10003825 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003826}
3827
3828/*
3829 * This routine performs a transaction to null out a bad inode pointer
3830 * in an agi unlinked inode hash bucket.
3831 */
3832STATIC void
3833xlog_recover_clear_agi_bucket(
3834 xfs_mount_t *mp,
3835 xfs_agnumber_t agno,
3836 int bucket)
3837{
3838 xfs_trans_t *tp;
3839 xfs_agi_t *agi;
3840 xfs_buf_t *agibp;
3841 int offset;
3842 int error;
3843
3844 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
Jie Liu3d3c8b52013-08-12 20:49:59 +10003845 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_clearagi, 0, 0);
David Chinnere5720ee2008-04-10 12:21:18 +10003846 if (error)
3847 goto out_abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003848
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003849 error = xfs_read_agi(mp, tp, agno, &agibp);
3850 if (error)
David Chinnere5720ee2008-04-10 12:21:18 +10003851 goto out_abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003852
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003853 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11003854 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003855 offset = offsetof(xfs_agi_t, agi_unlinked) +
3856 (sizeof(xfs_agino_t) * bucket);
3857 xfs_trans_log_buf(tp, agibp, offset,
3858 (offset + sizeof(xfs_agino_t) - 1));
3859
David Chinnere5720ee2008-04-10 12:21:18 +10003860 error = xfs_trans_commit(tp, 0);
3861 if (error)
3862 goto out_error;
3863 return;
3864
3865out_abort:
3866 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3867out_error:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003868 xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno);
David Chinnere5720ee2008-04-10 12:21:18 +10003869 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003870}
3871
Christoph Hellwig23fac502008-11-28 14:23:40 +11003872STATIC xfs_agino_t
3873xlog_recover_process_one_iunlink(
3874 struct xfs_mount *mp,
3875 xfs_agnumber_t agno,
3876 xfs_agino_t agino,
3877 int bucket)
3878{
3879 struct xfs_buf *ibp;
3880 struct xfs_dinode *dip;
3881 struct xfs_inode *ip;
3882 xfs_ino_t ino;
3883 int error;
3884
3885 ino = XFS_AGINO_TO_INO(mp, agno, agino);
Dave Chinner7b6259e2010-06-24 11:35:17 +10003886 error = xfs_iget(mp, NULL, ino, 0, 0, &ip);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003887 if (error)
3888 goto fail;
3889
3890 /*
3891 * Get the on disk inode to find the next inode in the bucket.
3892 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04003893 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &ibp, 0, 0);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003894 if (error)
Christoph Hellwig0e446672008-11-28 14:23:42 +11003895 goto fail_iput;
Christoph Hellwig23fac502008-11-28 14:23:40 +11003896
Christoph Hellwig23fac502008-11-28 14:23:40 +11003897 ASSERT(ip->i_d.di_nlink == 0);
Christoph Hellwig0e446672008-11-28 14:23:42 +11003898 ASSERT(ip->i_d.di_mode != 0);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003899
3900 /* setup for the next pass */
3901 agino = be32_to_cpu(dip->di_next_unlinked);
3902 xfs_buf_relse(ibp);
3903
3904 /*
3905 * Prevent any DMAPI event from being sent when the reference on
3906 * the inode is dropped.
3907 */
3908 ip->i_d.di_dmevmask = 0;
3909
Christoph Hellwig0e446672008-11-28 14:23:42 +11003910 IRELE(ip);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003911 return agino;
3912
Christoph Hellwig0e446672008-11-28 14:23:42 +11003913 fail_iput:
3914 IRELE(ip);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003915 fail:
3916 /*
3917 * We can't read in the inode this bucket points to, or this inode
3918 * is messed up. Just ditch this bucket of inodes. We will lose
3919 * some inodes and space, but at least we won't hang.
3920 *
3921 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3922 * clear the inode pointer in the bucket.
3923 */
3924 xlog_recover_clear_agi_bucket(mp, agno, bucket);
3925 return NULLAGINO;
3926}
3927
Linus Torvalds1da177e2005-04-16 15:20:36 -07003928/*
3929 * xlog_iunlink_recover
3930 *
3931 * This is called during recovery to process any inodes which
3932 * we unlinked but not freed when the system crashed. These
3933 * inodes will be on the lists in the AGI blocks. What we do
3934 * here is scan all the AGIs and fully truncate and free any
3935 * inodes found on the lists. Each inode is removed from the
3936 * lists when it has been fully truncated and is freed. The
3937 * freeing of the inode and its removal from the list must be
3938 * atomic.
3939 */
Eric Sandeend96f8f82009-07-02 00:09:33 -05003940STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07003941xlog_recover_process_iunlinks(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003942 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003943{
3944 xfs_mount_t *mp;
3945 xfs_agnumber_t agno;
3946 xfs_agi_t *agi;
3947 xfs_buf_t *agibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003948 xfs_agino_t agino;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003949 int bucket;
3950 int error;
3951 uint mp_dmevmask;
3952
3953 mp = log->l_mp;
3954
3955 /*
3956 * Prevent any DMAPI event from being sent while in this function.
3957 */
3958 mp_dmevmask = mp->m_dmevmask;
3959 mp->m_dmevmask = 0;
3960
3961 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3962 /*
3963 * Find the agi for this ag.
3964 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003965 error = xfs_read_agi(mp, NULL, agno, &agibp);
3966 if (error) {
3967 /*
3968 * AGI is b0rked. Don't process it.
3969 *
3970 * We should probably mark the filesystem as corrupt
3971 * after we've recovered all the ag's we can....
3972 */
3973 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003974 }
Jan Karad97d32e2012-03-15 09:34:02 +00003975 /*
3976 * Unlock the buffer so that it can be acquired in the normal
3977 * course of the transaction to truncate and free each inode.
3978 * Because we are not racing with anyone else here for the AGI
3979 * buffer, we don't even need to hold it locked to read the
3980 * initial unlinked bucket entries out of the buffer. We keep
3981 * buffer reference though, so that it stays pinned in memory
3982 * while we need the buffer.
3983 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003984 agi = XFS_BUF_TO_AGI(agibp);
Jan Karad97d32e2012-03-15 09:34:02 +00003985 xfs_buf_unlock(agibp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003986
3987 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
Christoph Hellwig16259e72005-11-02 15:11:25 +11003988 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003989 while (agino != NULLAGINO) {
Christoph Hellwig23fac502008-11-28 14:23:40 +11003990 agino = xlog_recover_process_one_iunlink(mp,
3991 agno, agino, bucket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003992 }
3993 }
Jan Karad97d32e2012-03-15 09:34:02 +00003994 xfs_buf_rele(agibp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003995 }
3996
3997 mp->m_dmevmask = mp_dmevmask;
3998}
3999
Linus Torvalds1da177e2005-04-16 15:20:36 -07004000/*
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004001 * Upack the log buffer data and crc check it. If the check fails, issue a
4002 * warning if and only if the CRC in the header is non-zero. This makes the
4003 * check an advisory warning, and the zero CRC check will prevent failure
4004 * warnings from being emitted when upgrading the kernel from one that does not
4005 * add CRCs by default.
4006 *
4007 * When filesystems are CRC enabled, this CRC mismatch becomes a fatal log
4008 * corruption failure
Linus Torvalds1da177e2005-04-16 15:20:36 -07004009 */
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004010STATIC int
4011xlog_unpack_data_crc(
4012 struct xlog_rec_header *rhead,
4013 xfs_caddr_t dp,
4014 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004015{
Dave Chinnerf9668a02012-11-28 13:01:03 +11004016 __le32 crc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004017
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004018 crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len));
4019 if (crc != rhead->h_crc) {
4020 if (rhead->h_crc || xfs_sb_version_hascrc(&log->l_mp->m_sb)) {
4021 xfs_alert(log->l_mp,
Eric Sandeen08e96e12013-10-11 20:59:05 -05004022 "log record CRC mismatch: found 0x%x, expected 0x%x.",
Dave Chinnerf9668a02012-11-28 13:01:03 +11004023 le32_to_cpu(rhead->h_crc),
4024 le32_to_cpu(crc));
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004025 xfs_hex_dump(dp, 32);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004026 }
4027
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004028 /*
4029 * If we've detected a log record corruption, then we can't
4030 * recover past this point. Abort recovery if we are enforcing
4031 * CRC protection by punting an error back up the stack.
4032 */
4033 if (xfs_sb_version_hascrc(&log->l_mp->m_sb))
Dave Chinner24513372014-06-25 14:58:08 +10004034 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004035 }
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004036
4037 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004038}
4039
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004040STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07004041xlog_unpack_data(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004042 struct xlog_rec_header *rhead,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004043 xfs_caddr_t dp,
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004044 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004045{
4046 int i, j, k;
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004047 int error;
4048
4049 error = xlog_unpack_data_crc(rhead, dp, log);
4050 if (error)
4051 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004052
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004053 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07004054 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004055 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07004056 dp += BBSIZE;
4057 }
4058
Eric Sandeen62118702008-03-06 13:44:28 +11004059 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Christoph Hellwigb28708d2008-11-28 14:23:38 +11004060 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004061 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004062 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
4063 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004064 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
Linus Torvalds1da177e2005-04-16 15:20:36 -07004065 dp += BBSIZE;
4066 }
4067 }
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004068
4069 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004070}
4071
4072STATIC int
4073xlog_valid_rec_header(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004074 struct xlog *log,
4075 struct xlog_rec_header *rhead,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004076 xfs_daddr_t blkno)
4077{
4078 int hlen;
4079
Christoph Hellwig69ef9212011-07-08 14:36:05 +02004080 if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004081 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
4082 XFS_ERRLEVEL_LOW, log->l_mp);
Dave Chinner24513372014-06-25 14:58:08 +10004083 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004084 }
4085 if (unlikely(
4086 (!rhead->h_version ||
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004087 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004088 xfs_warn(log->l_mp, "%s: unrecognised log version (%d).",
Harvey Harrison34a622b2008-04-10 12:19:21 +10004089 __func__, be32_to_cpu(rhead->h_version));
Dave Chinner24513372014-06-25 14:58:08 +10004090 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004091 }
4092
4093 /* LR body must have data or it wouldn't have been written */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004094 hlen = be32_to_cpu(rhead->h_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004095 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
4096 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
4097 XFS_ERRLEVEL_LOW, log->l_mp);
Dave Chinner24513372014-06-25 14:58:08 +10004098 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004099 }
4100 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
4101 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
4102 XFS_ERRLEVEL_LOW, log->l_mp);
Dave Chinner24513372014-06-25 14:58:08 +10004103 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004104 }
4105 return 0;
4106}
4107
4108/*
4109 * Read the log from tail to head and process the log records found.
4110 * Handle the two cases where the tail and head are in the same cycle
4111 * and where the active portion of the log wraps around the end of
4112 * the physical log separately. The pass parameter is passed through
4113 * to the routines called to process the data and is not looked at
4114 * here.
4115 */
4116STATIC int
4117xlog_do_recovery_pass(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004118 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004119 xfs_daddr_t head_blk,
4120 xfs_daddr_t tail_blk,
4121 int pass)
4122{
4123 xlog_rec_header_t *rhead;
4124 xfs_daddr_t blk_no;
Andy Polingfc5bc4c2009-11-03 17:26:47 +00004125 xfs_caddr_t offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004126 xfs_buf_t *hbp, *dbp;
4127 int error = 0, h_size;
4128 int bblks, split_bblks;
4129 int hblks, split_hblks, wrapped_hblks;
Dave Chinnerf0a76952010-01-11 11:49:57 +00004130 struct hlist_head rhash[XLOG_RHASH_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -07004131
4132 ASSERT(head_blk != tail_blk);
4133
4134 /*
4135 * Read the header of the tail block and get the iclog buffer size from
4136 * h_size. Use this to tell how many sectors make up the log header.
4137 */
Eric Sandeen62118702008-03-06 13:44:28 +11004138 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004139 /*
4140 * When using variable length iclogs, read first sector of
4141 * iclog header and extract the header size from it. Get a
4142 * new hbp that is the correct size.
4143 */
4144 hbp = xlog_get_bp(log, 1);
4145 if (!hbp)
Dave Chinner24513372014-06-25 14:58:08 +10004146 return -ENOMEM;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004147
4148 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
4149 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004150 goto bread_err1;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004151
Linus Torvalds1da177e2005-04-16 15:20:36 -07004152 rhead = (xlog_rec_header_t *)offset;
4153 error = xlog_valid_rec_header(log, rhead, tail_blk);
4154 if (error)
4155 goto bread_err1;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004156 h_size = be32_to_cpu(rhead->h_size);
4157 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07004158 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
4159 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
4160 if (h_size % XLOG_HEADER_CYCLE_SIZE)
4161 hblks++;
4162 xlog_put_bp(hbp);
4163 hbp = xlog_get_bp(log, hblks);
4164 } else {
4165 hblks = 1;
4166 }
4167 } else {
Alex Elder69ce58f2010-04-20 17:09:59 +10004168 ASSERT(log->l_sectBBsize == 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004169 hblks = 1;
4170 hbp = xlog_get_bp(log, 1);
4171 h_size = XLOG_BIG_RECORD_BSIZE;
4172 }
4173
4174 if (!hbp)
Dave Chinner24513372014-06-25 14:58:08 +10004175 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004176 dbp = xlog_get_bp(log, BTOBB(h_size));
4177 if (!dbp) {
4178 xlog_put_bp(hbp);
Dave Chinner24513372014-06-25 14:58:08 +10004179 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004180 }
4181
4182 memset(rhash, 0, sizeof(rhash));
Eric Sandeen970fd3f2014-09-09 11:57:29 +10004183 blk_no = tail_blk;
4184 if (tail_blk > head_blk) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004185 /*
4186 * Perform recovery around the end of the physical log.
4187 * When the head is not on the same cycle number as the tail,
Eric Sandeen970fd3f2014-09-09 11:57:29 +10004188 * we can't do a sequential recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004189 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004190 while (blk_no < log->l_logBBsize) {
4191 /*
4192 * Check for header wrapping around physical end-of-log
4193 */
Chandra Seetharaman62926042011-07-22 23:40:15 +00004194 offset = hbp->b_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004195 split_hblks = 0;
4196 wrapped_hblks = 0;
4197 if (blk_no + hblks <= log->l_logBBsize) {
4198 /* Read header in one read */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004199 error = xlog_bread(log, blk_no, hblks, hbp,
4200 &offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004201 if (error)
4202 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004203 } else {
4204 /* This LR is split across physical log end */
4205 if (blk_no != log->l_logBBsize) {
4206 /* some data before physical log end */
4207 ASSERT(blk_no <= INT_MAX);
4208 split_hblks = log->l_logBBsize - (int)blk_no;
4209 ASSERT(split_hblks > 0);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004210 error = xlog_bread(log, blk_no,
4211 split_hblks, hbp,
4212 &offset);
4213 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004214 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004215 }
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004216
Linus Torvalds1da177e2005-04-16 15:20:36 -07004217 /*
4218 * Note: this black magic still works with
4219 * large sector sizes (non-512) only because:
4220 * - we increased the buffer size originally
4221 * by 1 sector giving us enough extra space
4222 * for the second read;
4223 * - the log start is guaranteed to be sector
4224 * aligned;
4225 * - we read the log end (LR header start)
4226 * _first_, then the log start (LR header end)
4227 * - order is important.
4228 */
David Chinner234f56a2008-04-10 12:24:24 +10004229 wrapped_hblks = hblks - split_hblks;
Dave Chinner44396472011-04-21 09:34:27 +00004230 error = xlog_bread_offset(log, 0,
4231 wrapped_hblks, hbp,
4232 offset + BBTOB(split_hblks));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004233 if (error)
4234 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004235 }
4236 rhead = (xlog_rec_header_t *)offset;
4237 error = xlog_valid_rec_header(log, rhead,
4238 split_hblks ? blk_no : 0);
4239 if (error)
4240 goto bread_err2;
4241
Christoph Hellwigb53e6752007-10-12 10:59:34 +10004242 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004243 blk_no += hblks;
4244
4245 /* Read in data for log record */
4246 if (blk_no + bblks <= log->l_logBBsize) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004247 error = xlog_bread(log, blk_no, bblks, dbp,
4248 &offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004249 if (error)
4250 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004251 } else {
4252 /* This log record is split across the
4253 * physical end of log */
Chandra Seetharaman62926042011-07-22 23:40:15 +00004254 offset = dbp->b_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004255 split_bblks = 0;
4256 if (blk_no != log->l_logBBsize) {
4257 /* some data is before the physical
4258 * end of log */
4259 ASSERT(!wrapped_hblks);
4260 ASSERT(blk_no <= INT_MAX);
4261 split_bblks =
4262 log->l_logBBsize - (int)blk_no;
4263 ASSERT(split_bblks > 0);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004264 error = xlog_bread(log, blk_no,
4265 split_bblks, dbp,
4266 &offset);
4267 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004268 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004269 }
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004270
Linus Torvalds1da177e2005-04-16 15:20:36 -07004271 /*
4272 * Note: this black magic still works with
4273 * large sector sizes (non-512) only because:
4274 * - we increased the buffer size originally
4275 * by 1 sector giving us enough extra space
4276 * for the second read;
4277 * - the log start is guaranteed to be sector
4278 * aligned;
4279 * - we read the log end (LR header start)
4280 * _first_, then the log start (LR header end)
4281 * - order is important.
4282 */
Dave Chinner44396472011-04-21 09:34:27 +00004283 error = xlog_bread_offset(log, 0,
Dave Chinner009507b2012-11-02 11:38:44 +11004284 bblks - split_bblks, dbp,
Dave Chinner44396472011-04-21 09:34:27 +00004285 offset + BBTOB(split_bblks));
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004286 if (error)
4287 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004288 }
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004289
4290 error = xlog_unpack_data(rhead, offset, log);
4291 if (error)
4292 goto bread_err2;
4293
4294 error = xlog_recover_process_data(log, rhash,
4295 rhead, offset, pass);
4296 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004297 goto bread_err2;
4298 blk_no += bblks;
4299 }
4300
4301 ASSERT(blk_no >= log->l_logBBsize);
4302 blk_no -= log->l_logBBsize;
Eric Sandeen970fd3f2014-09-09 11:57:29 +10004303 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004304
Eric Sandeen970fd3f2014-09-09 11:57:29 +10004305 /* read first part of physical log */
4306 while (blk_no < head_blk) {
4307 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
4308 if (error)
4309 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004310
Eric Sandeen970fd3f2014-09-09 11:57:29 +10004311 rhead = (xlog_rec_header_t *)offset;
4312 error = xlog_valid_rec_header(log, rhead, blk_no);
4313 if (error)
4314 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004315
Eric Sandeen970fd3f2014-09-09 11:57:29 +10004316 /* blocks in data section */
4317 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
4318 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
4319 &offset);
4320 if (error)
4321 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01004322
Eric Sandeen970fd3f2014-09-09 11:57:29 +10004323 error = xlog_unpack_data(rhead, offset, log);
4324 if (error)
4325 goto bread_err2;
Christoph Hellwig0e446be2012-11-12 22:54:24 +11004326
Eric Sandeen970fd3f2014-09-09 11:57:29 +10004327 error = xlog_recover_process_data(log, rhash,
4328 rhead, offset, pass);
4329 if (error)
4330 goto bread_err2;
4331 blk_no += bblks + hblks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004332 }
4333
4334 bread_err2:
4335 xlog_put_bp(dbp);
4336 bread_err1:
4337 xlog_put_bp(hbp);
4338 return error;
4339}
4340
4341/*
4342 * Do the recovery of the log. We actually do this in two phases.
4343 * The two passes are necessary in order to implement the function
4344 * of cancelling a record written into the log. The first pass
4345 * determines those things which have been cancelled, and the
4346 * second pass replays log items normally except for those which
4347 * have been cancelled. The handling of the replay and cancellations
4348 * takes place in the log item type specific routines.
4349 *
4350 * The table of items which have cancel records in the log is allocated
4351 * and freed at this level, since only here do we know when all of
4352 * the log recovery has been completed.
4353 */
4354STATIC int
4355xlog_do_log_recovery(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004356 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004357 xfs_daddr_t head_blk,
4358 xfs_daddr_t tail_blk)
4359{
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00004360 int error, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004361
4362 ASSERT(head_blk != tail_blk);
4363
4364 /*
4365 * First do a pass to find all of the cancelled buf log items.
4366 * Store them in the buf_cancel_table for use in the second pass.
4367 */
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00004368 log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE *
4369 sizeof(struct list_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07004370 KM_SLEEP);
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00004371 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
4372 INIT_LIST_HEAD(&log->l_buf_cancel_table[i]);
4373
Linus Torvalds1da177e2005-04-16 15:20:36 -07004374 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
4375 XLOG_RECOVER_PASS1);
4376 if (error != 0) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10004377 kmem_free(log->l_buf_cancel_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004378 log->l_buf_cancel_table = NULL;
4379 return error;
4380 }
4381 /*
4382 * Then do a second pass to actually recover the items in the log.
4383 * When it is complete free the table of buf cancel items.
4384 */
4385 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
4386 XLOG_RECOVER_PASS2);
4387#ifdef DEBUG
Tim Shimmin6d192a92006-06-09 14:55:38 +10004388 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004389 int i;
4390
4391 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00004392 ASSERT(list_empty(&log->l_buf_cancel_table[i]));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004393 }
4394#endif /* DEBUG */
4395
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10004396 kmem_free(log->l_buf_cancel_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004397 log->l_buf_cancel_table = NULL;
4398
4399 return error;
4400}
4401
4402/*
4403 * Do the actual recovery
4404 */
4405STATIC int
4406xlog_do_recover(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004407 struct xlog *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004408 xfs_daddr_t head_blk,
4409 xfs_daddr_t tail_blk)
4410{
4411 int error;
4412 xfs_buf_t *bp;
4413 xfs_sb_t *sbp;
4414
4415 /*
4416 * First replay the images in the log.
4417 */
4418 error = xlog_do_log_recovery(log, head_blk, tail_blk);
Christoph Hellwig43ff2122012-04-23 15:58:39 +10004419 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004420 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004421
4422 /*
4423 * If IO errors happened during recovery, bail out.
4424 */
4425 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
Dave Chinner24513372014-06-25 14:58:08 +10004426 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004427 }
4428
4429 /*
4430 * We now update the tail_lsn since much of the recovery has completed
4431 * and there may be space available to use. If there were no extent
4432 * or iunlinks, we can free up the entire log and set the tail_lsn to
4433 * be the last_sync_lsn. This was set in xlog_find_tail to be the
4434 * lsn of the last known good LR on disk. If there are extent frees
4435 * or iunlinks they will have some entries in the AIL; so we look at
4436 * the AIL to determine how to set the tail_lsn.
4437 */
4438 xlog_assign_tail_lsn(log->l_mp);
4439
4440 /*
4441 * Now that we've finished replaying all buffer and inode
Dave Chinner98021822012-11-12 22:54:03 +11004442 * updates, re-read in the superblock and reverify it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004443 */
4444 bp = xfs_getsb(log->l_mp, 0);
4445 XFS_BUF_UNDONE(bp);
Lachlan McIlroybebf9632007-10-15 13:18:02 +10004446 ASSERT(!(XFS_BUF_ISWRITE(bp)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004447 XFS_BUF_READ(bp);
Lachlan McIlroybebf9632007-10-15 13:18:02 +10004448 XFS_BUF_UNASYNC(bp);
Dave Chinner1813dd62012-11-14 17:54:40 +11004449 bp->b_ops = &xfs_sb_buf_ops;
Christoph Hellwig83a0adc2013-12-17 00:03:52 -08004450
Dave Chinner595bff72014-10-02 09:05:14 +10004451 error = xfs_buf_submit_wait(bp);
David Chinnerd64e31a2008-04-10 12:22:17 +10004452 if (error) {
Dave Chinner595bff72014-10-02 09:05:14 +10004453 if (!XFS_FORCED_SHUTDOWN(log->l_mp)) {
4454 xfs_buf_ioerror_alert(bp, __func__);
4455 ASSERT(0);
4456 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004457 xfs_buf_relse(bp);
4458 return error;
4459 }
4460
4461 /* Convert superblock from on-disk format */
4462 sbp = &log->l_mp->m_sb;
Dave Chinner98021822012-11-12 22:54:03 +11004463 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004464 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
Eric Sandeen62118702008-03-06 13:44:28 +11004465 ASSERT(xfs_sb_good_version(sbp));
Dave Chinner5681ca42015-02-23 21:22:31 +11004466 xfs_reinit_percpu_counters(log->l_mp);
4467
Linus Torvalds1da177e2005-04-16 15:20:36 -07004468 xfs_buf_relse(bp);
4469
Lachlan McIlroy5478eea2007-02-10 18:36:29 +11004470
Linus Torvalds1da177e2005-04-16 15:20:36 -07004471 xlog_recover_check_summary(log);
4472
4473 /* Normal transactions can now occur */
4474 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
4475 return 0;
4476}
4477
4478/*
4479 * Perform recovery and re-initialize some log variables in xlog_find_tail.
4480 *
4481 * Return error or zero.
4482 */
4483int
4484xlog_recover(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004485 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004486{
4487 xfs_daddr_t head_blk, tail_blk;
4488 int error;
4489
4490 /* find the tail of the log */
Eric Sandeen65be6052006-01-11 15:34:19 +11004491 if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004492 return error;
4493
4494 if (tail_blk != head_blk) {
4495 /* There used to be a comment here:
4496 *
4497 * disallow recovery on read-only mounts. note -- mount
4498 * checks for ENOSPC and turns it into an intelligent
4499 * error message.
4500 * ...but this is no longer true. Now, unless you specify
4501 * NORECOVERY (in which case this function would never be
4502 * called), we just go ahead and recover. We do this all
4503 * under the vfs layer, so we can get away with it unless
4504 * the device itself is read-only, in which case we fail.
4505 */
Utako Kusaka3a02ee12007-05-08 13:50:06 +10004506 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004507 return error;
4508 }
4509
Dave Chinnere721f502013-04-03 16:11:32 +11004510 /*
4511 * Version 5 superblock log feature mask validation. We know the
4512 * log is dirty so check if there are any unknown log features
4513 * in what we need to recover. If there are unknown features
4514 * (e.g. unsupported transactions, then simply reject the
4515 * attempt at recovery before touching anything.
4516 */
4517 if (XFS_SB_VERSION_NUM(&log->l_mp->m_sb) == XFS_SB_VERSION_5 &&
4518 xfs_sb_has_incompat_log_feature(&log->l_mp->m_sb,
4519 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
4520 xfs_warn(log->l_mp,
4521"Superblock has unknown incompatible log features (0x%x) enabled.\n"
4522"The log can not be fully and/or safely recovered by this kernel.\n"
4523"Please recover the log on a kernel that supports the unknown features.",
4524 (log->l_mp->m_sb.sb_features_log_incompat &
4525 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
Dave Chinner24513372014-06-25 14:58:08 +10004526 return -EINVAL;
Dave Chinnere721f502013-04-03 16:11:32 +11004527 }
4528
Brian Foster2e227172014-09-09 11:56:13 +10004529 /*
4530 * Delay log recovery if the debug hook is set. This is debug
4531 * instrumention to coordinate simulation of I/O failures with
4532 * log recovery.
4533 */
4534 if (xfs_globals.log_recovery_delay) {
4535 xfs_notice(log->l_mp,
4536 "Delaying log recovery for %d seconds.",
4537 xfs_globals.log_recovery_delay);
4538 msleep(xfs_globals.log_recovery_delay * 1000);
4539 }
4540
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004541 xfs_notice(log->l_mp, "Starting recovery (logdev: %s)",
4542 log->l_mp->m_logname ? log->l_mp->m_logname
4543 : "internal");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004544
4545 error = xlog_do_recover(log, head_blk, tail_blk);
4546 log->l_flags |= XLOG_RECOVERY_NEEDED;
4547 }
4548 return error;
4549}
4550
4551/*
4552 * In the first part of recovery we replay inodes and buffers and build
4553 * up the list of extent free items which need to be processed. Here
4554 * we process the extent free items and clean up the on disk unlinked
4555 * inode lists. This is separated from the first part of recovery so
4556 * that the root and real-time bitmap inodes can be read in from disk in
4557 * between the two stages. This is necessary so that we can free space
4558 * in the real-time portion of the file system.
4559 */
4560int
4561xlog_recover_finish(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004562 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004563{
4564 /*
4565 * Now we're ready to do the transactions needed for the
4566 * rest of recovery. Start with completing all the extent
4567 * free intent records and then process the unlinked inode
4568 * lists. At this point, we essentially run in normal mode
4569 * except that we're still performing recovery actions
4570 * rather than accepting new requests.
4571 */
4572 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
David Chinner3c1e2bb2008-04-10 12:21:11 +10004573 int error;
4574 error = xlog_recover_process_efis(log);
4575 if (error) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004576 xfs_alert(log->l_mp, "Failed to recover EFIs");
David Chinner3c1e2bb2008-04-10 12:21:11 +10004577 return error;
4578 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004579 /*
4580 * Sync the log to get all the EFIs out of the AIL.
4581 * This isn't absolutely necessary, but it helps in
4582 * case the unlink transactions would have problems
4583 * pushing the EFIs out of the way.
4584 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +00004585 xfs_log_force(log->l_mp, XFS_LOG_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004586
Christoph Hellwig42490232008-08-13 16:49:32 +10004587 xlog_recover_process_iunlinks(log);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004588
4589 xlog_recover_check_summary(log);
4590
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004591 xfs_notice(log->l_mp, "Ending recovery (logdev: %s)",
4592 log->l_mp->m_logname ? log->l_mp->m_logname
4593 : "internal");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004594 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
4595 } else {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004596 xfs_info(log->l_mp, "Ending clean mount");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004597 }
4598 return 0;
4599}
4600
4601
4602#if defined(DEBUG)
4603/*
4604 * Read all of the agf and agi counters and check that they
4605 * are consistent with the superblock counters.
4606 */
4607void
4608xlog_recover_check_summary(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004609 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004610{
4611 xfs_mount_t *mp;
4612 xfs_agf_t *agfp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004613 xfs_buf_t *agfbp;
4614 xfs_buf_t *agibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004615 xfs_agnumber_t agno;
4616 __uint64_t freeblks;
4617 __uint64_t itotal;
4618 __uint64_t ifree;
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11004619 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004620
4621 mp = log->l_mp;
4622
4623 freeblks = 0LL;
4624 itotal = 0LL;
4625 ifree = 0LL;
4626 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
From: Christoph Hellwig48056212008-11-28 14:23:38 +11004627 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
4628 if (error) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004629 xfs_alert(mp, "%s agf read failed agno %d error %d",
4630 __func__, agno, error);
From: Christoph Hellwig48056212008-11-28 14:23:38 +11004631 } else {
4632 agfp = XFS_BUF_TO_AGF(agfbp);
4633 freeblks += be32_to_cpu(agfp->agf_freeblks) +
4634 be32_to_cpu(agfp->agf_flcount);
4635 xfs_buf_relse(agfbp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004636 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004637
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11004638 error = xfs_read_agi(mp, NULL, agno, &agibp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11004639 if (error) {
4640 xfs_alert(mp, "%s agi read failed agno %d error %d",
4641 __func__, agno, error);
4642 } else {
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11004643 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004644
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11004645 itotal += be32_to_cpu(agi->agi_count);
4646 ifree += be32_to_cpu(agi->agi_freecount);
4647 xfs_buf_relse(agibp);
4648 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004649 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004650}
4651#endif /* DEBUG */