blob: 04142caedb2bad62c158b08e6799e5f75bce5b4e [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Tim Shimmin87c199c2006-06-09 14:56:16 +10002 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
Nathan Scott7b718762005-11-02 14:58:39 +11003 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110019#include "xfs_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include "xfs_types.h"
Nathan Scotta844f452005-11-02 14:38:42 +110021#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110023#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_trans.h"
Nathan Scotta844f452005-11-02 14:38:42 +110025#include "xfs_sb.h"
26#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_mount.h"
28#include "xfs_error.h"
29#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110030#include "xfs_alloc_btree.h"
31#include "xfs_ialloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include "xfs_dinode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_inode.h"
Nathan Scotta844f452005-11-02 14:38:42 +110034#include "xfs_inode_item.h"
Nathan Scotta844f452005-11-02 14:38:42 +110035#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include "xfs_ialloc.h"
37#include "xfs_log_priv.h"
38#include "xfs_buf_item.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include "xfs_log_recover.h"
40#include "xfs_extfree_item.h"
41#include "xfs_trans_priv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#include "xfs_quota.h"
43#include "xfs_rw.h"
Christoph Hellwig43355092008-03-27 18:01:08 +110044#include "xfs_utils.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000045#include "xfs_trace.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
47STATIC int xlog_find_zeroed(xlog_t *, xfs_daddr_t *);
48STATIC int xlog_clear_stale_blocks(xlog_t *, xfs_lsn_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#if defined(DEBUG)
50STATIC void xlog_recover_check_summary(xlog_t *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070051#else
52#define xlog_recover_check_summary(log)
Linus Torvalds1da177e2005-04-16 15:20:36 -070053#endif
54
Linus Torvalds1da177e2005-04-16 15:20:36 -070055/*
Christoph Hellwigd5689ea2010-12-01 22:06:22 +000056 * This structure is used during recovery to record the buf log items which
57 * have been canceled and should not be replayed.
58 */
59struct xfs_buf_cancel {
60 xfs_daddr_t bc_blkno;
61 uint bc_len;
62 int bc_refcount;
63 struct list_head bc_list;
64};
65
66/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 * Sector aligned buffer routines for buffer create/read/write/access
68 */
69
Alex Elderff30a622010-04-13 15:22:58 +100070/*
71 * Verify the given count of basic blocks is valid number of blocks
72 * to specify for an operation involving the given XFS log buffer.
73 * Returns nonzero if the count is valid, 0 otherwise.
74 */
75
76static inline int
77xlog_buf_bbcount_valid(
78 xlog_t *log,
79 int bbcount)
80{
81 return bbcount > 0 && bbcount <= log->l_logBBsize;
82}
83
Alex Elder36adecf2010-04-13 15:21:13 +100084/*
85 * Allocate a buffer to hold log data. The buffer needs to be able
86 * to map to a range of nbblks basic blocks at any valid (basic
87 * block) offset within the log.
88 */
Eric Sandeen5d77c0d2009-11-19 15:52:00 +000089STATIC xfs_buf_t *
Linus Torvalds1da177e2005-04-16 15:20:36 -070090xlog_get_bp(
91 xlog_t *log,
Dave Chinner32281492009-01-22 15:37:47 +110092 int nbblks)
Linus Torvalds1da177e2005-04-16 15:20:36 -070093{
Alex Elderff30a622010-04-13 15:22:58 +100094 if (!xlog_buf_bbcount_valid(log, nbblks)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +110095 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
Alex Elderff30a622010-04-13 15:22:58 +100096 nbblks);
97 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
Dave Chinner32281492009-01-22 15:37:47 +110098 return NULL;
99 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100
Alex Elder36adecf2010-04-13 15:21:13 +1000101 /*
102 * We do log I/O in units of log sectors (a power-of-2
103 * multiple of the basic block size), so we round up the
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300104 * requested size to accommodate the basic blocks required
Alex Elder36adecf2010-04-13 15:21:13 +1000105 * for complete log sectors.
106 *
107 * In addition, the buffer may be used for a non-sector-
108 * aligned block offset, in which case an I/O of the
109 * requested size could extend beyond the end of the
110 * buffer. If the requested size is only 1 basic block it
111 * will never straddle a sector boundary, so this won't be
112 * an issue. Nor will this be a problem if the log I/O is
113 * done in basic blocks (sector size 1). But otherwise we
114 * extend the buffer by one extra log sector to ensure
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300115 * there's space to accommodate this possibility.
Alex Elder36adecf2010-04-13 15:21:13 +1000116 */
Alex Elder69ce58f2010-04-20 17:09:59 +1000117 if (nbblks > 1 && log->l_sectBBsize > 1)
118 nbblks += log->l_sectBBsize;
119 nbblks = round_up(nbblks, log->l_sectBBsize);
Alex Elder36adecf2010-04-13 15:21:13 +1000120
Dave Chinner686865f2010-09-24 20:07:47 +1000121 return xfs_buf_get_uncached(log->l_mp->m_logdev_targp,
122 BBTOB(nbblks), 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123}
124
Eric Sandeen5d77c0d2009-11-19 15:52:00 +0000125STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126xlog_put_bp(
127 xfs_buf_t *bp)
128{
129 xfs_buf_free(bp);
130}
131
Alex Elder48389ef2010-04-20 17:10:21 +1000132/*
133 * Return the address of the start of the given block number's data
134 * in a log buffer. The buffer covers a log sector-aligned region.
135 */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100136STATIC xfs_caddr_t
137xlog_align(
138 xlog_t *log,
139 xfs_daddr_t blk_no,
140 int nbblks,
141 xfs_buf_t *bp)
142{
Christoph Hellwigfdc07f42010-05-10 17:28:14 +0000143 xfs_daddr_t offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100144
Christoph Hellwigfdc07f42010-05-10 17:28:14 +0000145 ASSERT(BBTOB(offset + nbblks) <= XFS_BUF_SIZE(bp));
146 return XFS_BUF_PTR(bp) + BBTOB(offset);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100147}
148
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149
150/*
151 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
152 */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100153STATIC int
154xlog_bread_noalign(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 xlog_t *log,
156 xfs_daddr_t blk_no,
157 int nbblks,
158 xfs_buf_t *bp)
159{
160 int error;
161
Alex Elderff30a622010-04-13 15:22:58 +1000162 if (!xlog_buf_bbcount_valid(log, nbblks)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100163 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
Alex Elderff30a622010-04-13 15:22:58 +1000164 nbblks);
165 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
Dave Chinner32281492009-01-22 15:37:47 +1100166 return EFSCORRUPTED;
167 }
168
Alex Elder69ce58f2010-04-20 17:09:59 +1000169 blk_no = round_down(blk_no, log->l_sectBBsize);
170 nbblks = round_up(nbblks, log->l_sectBBsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171
172 ASSERT(nbblks > 0);
173 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174
175 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
176 XFS_BUF_READ(bp);
177 XFS_BUF_BUSY(bp);
178 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
179 XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
180
181 xfsbdstrat(log->l_mp, bp);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +0000182 error = xfs_buf_iowait(bp);
David Chinnerd64e31a2008-04-10 12:22:17 +1000183 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 xfs_ioerror_alert("xlog_bread", log->l_mp,
185 bp, XFS_BUF_ADDR(bp));
186 return error;
187}
188
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100189STATIC int
190xlog_bread(
191 xlog_t *log,
192 xfs_daddr_t blk_no,
193 int nbblks,
194 xfs_buf_t *bp,
195 xfs_caddr_t *offset)
196{
197 int error;
198
199 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
200 if (error)
201 return error;
202
203 *offset = xlog_align(log, blk_no, nbblks, bp);
204 return 0;
205}
206
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207/*
Dave Chinner44396472011-04-21 09:34:27 +0000208 * Read at an offset into the buffer. Returns with the buffer in it's original
209 * state regardless of the result of the read.
210 */
211STATIC int
212xlog_bread_offset(
213 xlog_t *log,
214 xfs_daddr_t blk_no, /* block to read from */
215 int nbblks, /* blocks to read */
216 xfs_buf_t *bp,
217 xfs_caddr_t offset)
218{
219 xfs_caddr_t orig_offset = XFS_BUF_PTR(bp);
220 int orig_len = bp->b_buffer_length;
221 int error, error2;
222
223 error = XFS_BUF_SET_PTR(bp, offset, BBTOB(nbblks));
224 if (error)
225 return error;
226
227 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
228
229 /* must reset buffer pointer even on error */
230 error2 = XFS_BUF_SET_PTR(bp, orig_offset, orig_len);
231 if (error)
232 return error;
233 return error2;
234}
235
236/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 * Write out the buffer at the given block for the given number of blocks.
238 * The buffer is kept locked across the write and is returned locked.
239 * This can only be used for synchronous log writes.
240 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000241STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242xlog_bwrite(
243 xlog_t *log,
244 xfs_daddr_t blk_no,
245 int nbblks,
246 xfs_buf_t *bp)
247{
248 int error;
249
Alex Elderff30a622010-04-13 15:22:58 +1000250 if (!xlog_buf_bbcount_valid(log, nbblks)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100251 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
Alex Elderff30a622010-04-13 15:22:58 +1000252 nbblks);
253 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
Dave Chinner32281492009-01-22 15:37:47 +1100254 return EFSCORRUPTED;
255 }
256
Alex Elder69ce58f2010-04-20 17:09:59 +1000257 blk_no = round_down(blk_no, log->l_sectBBsize);
258 nbblks = round_up(nbblks, log->l_sectBBsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259
260 ASSERT(nbblks > 0);
261 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
262
263 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
264 XFS_BUF_ZEROFLAGS(bp);
265 XFS_BUF_BUSY(bp);
266 XFS_BUF_HOLD(bp);
267 XFS_BUF_PSEMA(bp, PRIBIO);
268 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
269 XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
270
271 if ((error = xfs_bwrite(log->l_mp, bp)))
272 xfs_ioerror_alert("xlog_bwrite", log->l_mp,
273 bp, XFS_BUF_ADDR(bp));
274 return error;
275}
276
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277#ifdef DEBUG
278/*
279 * dump debug superblock and log record information
280 */
281STATIC void
282xlog_header_check_dump(
283 xfs_mount_t *mp,
284 xlog_rec_header_t *head)
285{
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100286 xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d\n",
Joe Perches03daa572009-12-14 18:01:10 -0800287 __func__, &mp->m_sb.sb_uuid, XLOG_FMT);
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100288 xfs_debug(mp, " log : uuid = %pU, fmt = %d\n",
Joe Perches03daa572009-12-14 18:01:10 -0800289 &head->h_fs_uuid, be32_to_cpu(head->h_fmt));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290}
291#else
292#define xlog_header_check_dump(mp, head)
293#endif
294
295/*
296 * check log record header for recovery
297 */
298STATIC int
299xlog_header_check_recover(
300 xfs_mount_t *mp,
301 xlog_rec_header_t *head)
302{
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000303 ASSERT(be32_to_cpu(head->h_magicno) == XLOG_HEADER_MAGIC_NUM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304
305 /*
306 * IRIX doesn't write the h_fmt field and leaves it zeroed
307 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
308 * a dirty log created in IRIX.
309 */
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000310 if (unlikely(be32_to_cpu(head->h_fmt) != XLOG_FMT)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100311 xfs_warn(mp,
312 "dirty log written in incompatible format - can't recover");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 xlog_header_check_dump(mp, head);
314 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
315 XFS_ERRLEVEL_HIGH, mp);
316 return XFS_ERROR(EFSCORRUPTED);
317 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100318 xfs_warn(mp,
319 "dirty log entry has mismatched uuid - can't recover");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 xlog_header_check_dump(mp, head);
321 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
322 XFS_ERRLEVEL_HIGH, mp);
323 return XFS_ERROR(EFSCORRUPTED);
324 }
325 return 0;
326}
327
328/*
329 * read the head block of the log and check the header
330 */
331STATIC int
332xlog_header_check_mount(
333 xfs_mount_t *mp,
334 xlog_rec_header_t *head)
335{
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000336 ASSERT(be32_to_cpu(head->h_magicno) == XLOG_HEADER_MAGIC_NUM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337
338 if (uuid_is_nil(&head->h_fs_uuid)) {
339 /*
340 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
341 * h_fs_uuid is nil, we assume this log was last mounted
342 * by IRIX and continue.
343 */
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100344 xfs_warn(mp, "nil uuid in log - IRIX style log");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100346 xfs_warn(mp, "log has mismatched uuid - can't recover");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 xlog_header_check_dump(mp, head);
348 XFS_ERROR_REPORT("xlog_header_check_mount",
349 XFS_ERRLEVEL_HIGH, mp);
350 return XFS_ERROR(EFSCORRUPTED);
351 }
352 return 0;
353}
354
355STATIC void
356xlog_recover_iodone(
357 struct xfs_buf *bp)
358{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 if (XFS_BUF_GETERROR(bp)) {
360 /*
361 * We're not going to bother about retrying
362 * this during recovery. One strike!
363 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 xfs_ioerror_alert("xlog_recover_iodone",
Dave Chinnerebad8612010-09-22 10:47:20 +1000365 bp->b_target->bt_mount, bp,
366 XFS_BUF_ADDR(bp));
367 xfs_force_shutdown(bp->b_target->bt_mount,
368 SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 XFS_BUF_CLR_IODONE_FUNC(bp);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +0000371 xfs_buf_ioend(bp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372}
373
374/*
375 * This routine finds (to an approximation) the first block in the physical
376 * log which contains the given cycle. It uses a binary search algorithm.
377 * Note that the algorithm can not be perfect because the disk will not
378 * necessarily be perfect.
379 */
David Chinnera8272ce2007-11-23 16:28:09 +1100380STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381xlog_find_cycle_start(
382 xlog_t *log,
383 xfs_buf_t *bp,
384 xfs_daddr_t first_blk,
385 xfs_daddr_t *last_blk,
386 uint cycle)
387{
388 xfs_caddr_t offset;
389 xfs_daddr_t mid_blk;
Alex Eldere3bb2e32010-04-15 18:17:30 +0000390 xfs_daddr_t end_blk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 uint mid_cycle;
392 int error;
393
Alex Eldere3bb2e32010-04-15 18:17:30 +0000394 end_blk = *last_blk;
395 mid_blk = BLK_AVG(first_blk, end_blk);
396 while (mid_blk != first_blk && mid_blk != end_blk) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100397 error = xlog_bread(log, mid_blk, 1, bp, &offset);
398 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 return error;
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000400 mid_cycle = xlog_get_cycle(offset);
Alex Eldere3bb2e32010-04-15 18:17:30 +0000401 if (mid_cycle == cycle)
402 end_blk = mid_blk; /* last_half_cycle == mid_cycle */
403 else
404 first_blk = mid_blk; /* first_half_cycle == mid_cycle */
405 mid_blk = BLK_AVG(first_blk, end_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 }
Alex Eldere3bb2e32010-04-15 18:17:30 +0000407 ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) ||
408 (mid_blk == end_blk && mid_blk-1 == first_blk));
409
410 *last_blk = end_blk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411
412 return 0;
413}
414
415/*
Alex Elder3f943d82010-04-15 18:17:34 +0000416 * Check that a range of blocks does not contain stop_on_cycle_no.
417 * Fill in *new_blk with the block offset where such a block is
418 * found, or with -1 (an invalid block number) if there is no such
419 * block in the range. The scan needs to occur from front to back
420 * and the pointer into the region must be updated since a later
421 * routine will need to perform another test.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 */
423STATIC int
424xlog_find_verify_cycle(
425 xlog_t *log,
426 xfs_daddr_t start_blk,
427 int nbblks,
428 uint stop_on_cycle_no,
429 xfs_daddr_t *new_blk)
430{
431 xfs_daddr_t i, j;
432 uint cycle;
433 xfs_buf_t *bp;
434 xfs_daddr_t bufblks;
435 xfs_caddr_t buf = NULL;
436 int error = 0;
437
Alex Elder6881a222010-04-13 15:22:29 +1000438 /*
439 * Greedily allocate a buffer big enough to handle the full
440 * range of basic blocks we'll be examining. If that fails,
441 * try a smaller size. We need to be able to read at least
442 * a log sector, or we're out of luck.
443 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 bufblks = 1 << ffs(nbblks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 while (!(bp = xlog_get_bp(log, bufblks))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 bufblks >>= 1;
Alex Elder69ce58f2010-04-20 17:09:59 +1000447 if (bufblks < log->l_sectBBsize)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 return ENOMEM;
449 }
450
451 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
452 int bcount;
453
454 bcount = min(bufblks, (start_blk + nbblks - i));
455
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100456 error = xlog_bread(log, i, bcount, bp, &buf);
457 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 goto out;
459
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 for (j = 0; j < bcount; j++) {
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000461 cycle = xlog_get_cycle(buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 if (cycle == stop_on_cycle_no) {
463 *new_blk = i+j;
464 goto out;
465 }
466
467 buf += BBSIZE;
468 }
469 }
470
471 *new_blk = -1;
472
473out:
474 xlog_put_bp(bp);
475 return error;
476}
477
478/*
479 * Potentially backup over partial log record write.
480 *
481 * In the typical case, last_blk is the number of the block directly after
482 * a good log record. Therefore, we subtract one to get the block number
483 * of the last block in the given buffer. extra_bblks contains the number
484 * of blocks we would have read on a previous read. This happens when the
485 * last log record is split over the end of the physical log.
486 *
487 * extra_bblks is the number of blocks potentially verified on a previous
488 * call to this routine.
489 */
490STATIC int
491xlog_find_verify_log_record(
492 xlog_t *log,
493 xfs_daddr_t start_blk,
494 xfs_daddr_t *last_blk,
495 int extra_bblks)
496{
497 xfs_daddr_t i;
498 xfs_buf_t *bp;
499 xfs_caddr_t offset = NULL;
500 xlog_rec_header_t *head = NULL;
501 int error = 0;
502 int smallmem = 0;
503 int num_blks = *last_blk - start_blk;
504 int xhdrs;
505
506 ASSERT(start_blk != 0 || *last_blk != start_blk);
507
508 if (!(bp = xlog_get_bp(log, num_blks))) {
509 if (!(bp = xlog_get_bp(log, 1)))
510 return ENOMEM;
511 smallmem = 1;
512 } else {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100513 error = xlog_bread(log, start_blk, num_blks, bp, &offset);
514 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 offset += ((num_blks - 1) << BBSHIFT);
517 }
518
519 for (i = (*last_blk) - 1; i >= 0; i--) {
520 if (i < start_blk) {
521 /* valid log record not found */
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100522 xfs_warn(log->l_mp,
523 "Log inconsistent (didn't find previous header)");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 ASSERT(0);
525 error = XFS_ERROR(EIO);
526 goto out;
527 }
528
529 if (smallmem) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100530 error = xlog_bread(log, i, 1, bp, &offset);
531 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 }
534
535 head = (xlog_rec_header_t *)offset;
536
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000537 if (XLOG_HEADER_MAGIC_NUM == be32_to_cpu(head->h_magicno))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 break;
539
540 if (!smallmem)
541 offset -= BBSIZE;
542 }
543
544 /*
545 * We hit the beginning of the physical log & still no header. Return
546 * to caller. If caller can handle a return of -1, then this routine
547 * will be called again for the end of the physical log.
548 */
549 if (i == -1) {
550 error = -1;
551 goto out;
552 }
553
554 /*
555 * We have the final block of the good log (the first block
556 * of the log record _before_ the head. So we check the uuid.
557 */
558 if ((error = xlog_header_check_mount(log->l_mp, head)))
559 goto out;
560
561 /*
562 * We may have found a log record header before we expected one.
563 * last_blk will be the 1st block # with a given cycle #. We may end
564 * up reading an entire log record. In this case, we don't want to
565 * reset last_blk. Only when last_blk points in the middle of a log
566 * record do we update last_blk.
567 */
Eric Sandeen62118702008-03-06 13:44:28 +1100568 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000569 uint h_size = be32_to_cpu(head->h_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570
571 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
572 if (h_size % XLOG_HEADER_CYCLE_SIZE)
573 xhdrs++;
574 } else {
575 xhdrs = 1;
576 }
577
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000578 if (*last_blk - i + extra_bblks !=
579 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 *last_blk = i;
581
582out:
583 xlog_put_bp(bp);
584 return error;
585}
586
587/*
588 * Head is defined to be the point of the log where the next log write
589 * write could go. This means that incomplete LR writes at the end are
590 * eliminated when calculating the head. We aren't guaranteed that previous
591 * LR have complete transactions. We only know that a cycle number of
592 * current cycle number -1 won't be present in the log if we start writing
593 * from our current block number.
594 *
595 * last_blk contains the block number of the first block with a given
596 * cycle number.
597 *
598 * Return: zero if normal, non-zero if error.
599 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000600STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601xlog_find_head(
602 xlog_t *log,
603 xfs_daddr_t *return_head_blk)
604{
605 xfs_buf_t *bp;
606 xfs_caddr_t offset;
607 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
608 int num_scan_bblks;
609 uint first_half_cycle, last_half_cycle;
610 uint stop_on_cycle;
611 int error, log_bbnum = log->l_logBBsize;
612
613 /* Is the end of the log device zeroed? */
614 if ((error = xlog_find_zeroed(log, &first_blk)) == -1) {
615 *return_head_blk = first_blk;
616
617 /* Is the whole lot zeroed? */
618 if (!first_blk) {
619 /* Linux XFS shouldn't generate totally zeroed logs -
620 * mkfs etc write a dummy unmount record to a fresh
621 * log so we can store the uuid in there
622 */
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100623 xfs_warn(log->l_mp, "totally zeroed log");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 }
625
626 return 0;
627 } else if (error) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100628 xfs_warn(log->l_mp, "empty log check failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 return error;
630 }
631
632 first_blk = 0; /* get cycle # of 1st block */
633 bp = xlog_get_bp(log, 1);
634 if (!bp)
635 return ENOMEM;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100636
637 error = xlog_bread(log, 0, 1, bp, &offset);
638 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 goto bp_err;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100640
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000641 first_half_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642
643 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100644 error = xlog_bread(log, last_blk, 1, bp, &offset);
645 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 goto bp_err;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100647
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000648 last_half_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 ASSERT(last_half_cycle != 0);
650
651 /*
652 * If the 1st half cycle number is equal to the last half cycle number,
653 * then the entire log is stamped with the same cycle number. In this
654 * case, head_blk can't be set to zero (which makes sense). The below
655 * math doesn't work out properly with head_blk equal to zero. Instead,
656 * we set it to log_bbnum which is an invalid block number, but this
657 * value makes the math correct. If head_blk doesn't changed through
658 * all the tests below, *head_blk is set to zero at the very end rather
659 * than log_bbnum. In a sense, log_bbnum and zero are the same block
660 * in a circular file.
661 */
662 if (first_half_cycle == last_half_cycle) {
663 /*
664 * In this case we believe that the entire log should have
665 * cycle number last_half_cycle. We need to scan backwards
666 * from the end verifying that there are no holes still
667 * containing last_half_cycle - 1. If we find such a hole,
668 * then the start of that hole will be the new head. The
669 * simple case looks like
670 * x | x ... | x - 1 | x
671 * Another case that fits this picture would be
672 * x | x + 1 | x ... | x
Nathan Scottc41564b2006-03-29 08:55:14 +1000673 * In this case the head really is somewhere at the end of the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 * log, as one of the latest writes at the beginning was
675 * incomplete.
676 * One more case is
677 * x | x + 1 | x ... | x - 1 | x
678 * This is really the combination of the above two cases, and
679 * the head has to end up at the start of the x-1 hole at the
680 * end of the log.
681 *
682 * In the 256k log case, we will read from the beginning to the
683 * end of the log and search for cycle numbers equal to x-1.
684 * We don't worry about the x+1 blocks that we encounter,
685 * because we know that they cannot be the head since the log
686 * started with x.
687 */
688 head_blk = log_bbnum;
689 stop_on_cycle = last_half_cycle - 1;
690 } else {
691 /*
692 * In this case we want to find the first block with cycle
693 * number matching last_half_cycle. We expect the log to be
694 * some variation on
Alex Elder3f943d82010-04-15 18:17:34 +0000695 * x + 1 ... | x ... | x
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 * The first block with cycle number x (last_half_cycle) will
697 * be where the new head belongs. First we do a binary search
698 * for the first occurrence of last_half_cycle. The binary
699 * search may not be totally accurate, so then we scan back
700 * from there looking for occurrences of last_half_cycle before
701 * us. If that backwards scan wraps around the beginning of
702 * the log, then we look for occurrences of last_half_cycle - 1
703 * at the end of the log. The cases we're looking for look
704 * like
Alex Elder3f943d82010-04-15 18:17:34 +0000705 * v binary search stopped here
706 * x + 1 ... | x | x + 1 | x ... | x
707 * ^ but we want to locate this spot
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 * or
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 * <---------> less than scan distance
Alex Elder3f943d82010-04-15 18:17:34 +0000710 * x + 1 ... | x ... | x - 1 | x
711 * ^ we want to locate this spot
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 */
713 stop_on_cycle = last_half_cycle;
714 if ((error = xlog_find_cycle_start(log, bp, first_blk,
715 &head_blk, last_half_cycle)))
716 goto bp_err;
717 }
718
719 /*
720 * Now validate the answer. Scan back some number of maximum possible
721 * blocks and make sure each one has the expected cycle number. The
722 * maximum is determined by the total possible amount of buffering
723 * in the in-core log. The following number can be made tighter if
724 * we actually look at the block size of the filesystem.
725 */
726 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
727 if (head_blk >= num_scan_bblks) {
728 /*
729 * We are guaranteed that the entire check can be performed
730 * in one buffer.
731 */
732 start_blk = head_blk - num_scan_bblks;
733 if ((error = xlog_find_verify_cycle(log,
734 start_blk, num_scan_bblks,
735 stop_on_cycle, &new_blk)))
736 goto bp_err;
737 if (new_blk != -1)
738 head_blk = new_blk;
739 } else { /* need to read 2 parts of log */
740 /*
741 * We are going to scan backwards in the log in two parts.
742 * First we scan the physical end of the log. In this part
743 * of the log, we are looking for blocks with cycle number
744 * last_half_cycle - 1.
745 * If we find one, then we know that the log starts there, as
746 * we've found a hole that didn't get written in going around
747 * the end of the physical log. The simple case for this is
748 * x + 1 ... | x ... | x - 1 | x
749 * <---------> less than scan distance
750 * If all of the blocks at the end of the log have cycle number
751 * last_half_cycle, then we check the blocks at the start of
752 * the log looking for occurrences of last_half_cycle. If we
753 * find one, then our current estimate for the location of the
754 * first occurrence of last_half_cycle is wrong and we move
755 * back to the hole we've found. This case looks like
756 * x + 1 ... | x | x + 1 | x ...
757 * ^ binary search stopped here
758 * Another case we need to handle that only occurs in 256k
759 * logs is
760 * x + 1 ... | x ... | x+1 | x ...
761 * ^ binary search stops here
762 * In a 256k log, the scan at the end of the log will see the
763 * x + 1 blocks. We need to skip past those since that is
764 * certainly not the head of the log. By searching for
765 * last_half_cycle-1 we accomplish that.
766 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 ASSERT(head_blk <= INT_MAX &&
Alex Elder3f943d82010-04-15 18:17:34 +0000768 (xfs_daddr_t) num_scan_bblks >= head_blk);
769 start_blk = log_bbnum - (num_scan_bblks - head_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 if ((error = xlog_find_verify_cycle(log, start_blk,
771 num_scan_bblks - (int)head_blk,
772 (stop_on_cycle - 1), &new_blk)))
773 goto bp_err;
774 if (new_blk != -1) {
775 head_blk = new_blk;
Alex Elder9db127e2010-04-15 18:17:26 +0000776 goto validate_head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 }
778
779 /*
780 * Scan beginning of log now. The last part of the physical
781 * log is good. This scan needs to verify that it doesn't find
782 * the last_half_cycle.
783 */
784 start_blk = 0;
785 ASSERT(head_blk <= INT_MAX);
786 if ((error = xlog_find_verify_cycle(log,
787 start_blk, (int)head_blk,
788 stop_on_cycle, &new_blk)))
789 goto bp_err;
790 if (new_blk != -1)
791 head_blk = new_blk;
792 }
793
Alex Elder9db127e2010-04-15 18:17:26 +0000794validate_head:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 /*
796 * Now we need to make sure head_blk is not pointing to a block in
797 * the middle of a log record.
798 */
799 num_scan_bblks = XLOG_REC_SHIFT(log);
800 if (head_blk >= num_scan_bblks) {
801 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
802
803 /* start ptr at last block ptr before head_blk */
804 if ((error = xlog_find_verify_log_record(log, start_blk,
805 &head_blk, 0)) == -1) {
806 error = XFS_ERROR(EIO);
807 goto bp_err;
808 } else if (error)
809 goto bp_err;
810 } else {
811 start_blk = 0;
812 ASSERT(head_blk <= INT_MAX);
813 if ((error = xlog_find_verify_log_record(log, start_blk,
814 &head_blk, 0)) == -1) {
815 /* We hit the beginning of the log during our search */
Alex Elder3f943d82010-04-15 18:17:34 +0000816 start_blk = log_bbnum - (num_scan_bblks - head_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 new_blk = log_bbnum;
818 ASSERT(start_blk <= INT_MAX &&
819 (xfs_daddr_t) log_bbnum-start_blk >= 0);
820 ASSERT(head_blk <= INT_MAX);
821 if ((error = xlog_find_verify_log_record(log,
822 start_blk, &new_blk,
823 (int)head_blk)) == -1) {
824 error = XFS_ERROR(EIO);
825 goto bp_err;
826 } else if (error)
827 goto bp_err;
828 if (new_blk != log_bbnum)
829 head_blk = new_blk;
830 } else if (error)
831 goto bp_err;
832 }
833
834 xlog_put_bp(bp);
835 if (head_blk == log_bbnum)
836 *return_head_blk = 0;
837 else
838 *return_head_blk = head_blk;
839 /*
840 * When returning here, we have a good block number. Bad block
841 * means that during a previous crash, we didn't have a clean break
842 * from cycle number N to cycle number N-1. In this case, we need
843 * to find the first block with cycle number N-1.
844 */
845 return 0;
846
847 bp_err:
848 xlog_put_bp(bp);
849
850 if (error)
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100851 xfs_warn(log->l_mp, "failed to find log head");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 return error;
853}
854
855/*
856 * Find the sync block number or the tail of the log.
857 *
858 * This will be the block number of the last record to have its
859 * associated buffers synced to disk. Every log record header has
860 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
861 * to get a sync block number. The only concern is to figure out which
862 * log record header to believe.
863 *
864 * The following algorithm uses the log record header with the largest
865 * lsn. The entire log record does not need to be valid. We only care
866 * that the header is valid.
867 *
868 * We could speed up search by using current head_blk buffer, but it is not
869 * available.
870 */
Eric Sandeen5d77c0d2009-11-19 15:52:00 +0000871STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872xlog_find_tail(
873 xlog_t *log,
874 xfs_daddr_t *head_blk,
Eric Sandeen65be6052006-01-11 15:34:19 +1100875 xfs_daddr_t *tail_blk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876{
877 xlog_rec_header_t *rhead;
878 xlog_op_header_t *op_head;
879 xfs_caddr_t offset = NULL;
880 xfs_buf_t *bp;
881 int error, i, found;
882 xfs_daddr_t umount_data_blk;
883 xfs_daddr_t after_umount_blk;
884 xfs_lsn_t tail_lsn;
885 int hblks;
886
887 found = 0;
888
889 /*
890 * Find previous log record
891 */
892 if ((error = xlog_find_head(log, head_blk)))
893 return error;
894
895 bp = xlog_get_bp(log, 1);
896 if (!bp)
897 return ENOMEM;
898 if (*head_blk == 0) { /* special case */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100899 error = xlog_bread(log, 0, 1, bp, &offset);
900 if (error)
Alex Elder9db127e2010-04-15 18:17:26 +0000901 goto done;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100902
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000903 if (xlog_get_cycle(offset) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 *tail_blk = 0;
905 /* leave all other log inited values alone */
Alex Elder9db127e2010-04-15 18:17:26 +0000906 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 }
908 }
909
910 /*
911 * Search backwards looking for log record header block
912 */
913 ASSERT(*head_blk < INT_MAX);
914 for (i = (int)(*head_blk) - 1; i >= 0; i--) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100915 error = xlog_bread(log, i, 1, bp, &offset);
916 if (error)
Alex Elder9db127e2010-04-15 18:17:26 +0000917 goto done;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100918
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000919 if (XLOG_HEADER_MAGIC_NUM == be32_to_cpu(*(__be32 *)offset)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 found = 1;
921 break;
922 }
923 }
924 /*
925 * If we haven't found the log record header block, start looking
926 * again from the end of the physical log. XXXmiken: There should be
927 * a check here to make sure we didn't search more than N blocks in
928 * the previous code.
929 */
930 if (!found) {
931 for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100932 error = xlog_bread(log, i, 1, bp, &offset);
933 if (error)
Alex Elder9db127e2010-04-15 18:17:26 +0000934 goto done;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100935
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 if (XLOG_HEADER_MAGIC_NUM ==
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000937 be32_to_cpu(*(__be32 *)offset)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 found = 2;
939 break;
940 }
941 }
942 }
943 if (!found) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100944 xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 ASSERT(0);
946 return XFS_ERROR(EIO);
947 }
948
949 /* find blk_no of tail of log */
950 rhead = (xlog_rec_header_t *)offset;
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000951 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952
953 /*
954 * Reset log values according to the state of the log when we
955 * crashed. In the case where head_blk == 0, we bump curr_cycle
956 * one because the next write starts a new cycle rather than
957 * continuing the cycle of the last good log record. At this
958 * point we have guaranteed that all partial log records have been
959 * accounted for. Therefore, we know that the last good log record
960 * written was complete and ended exactly on the end boundary
961 * of the physical log.
962 */
963 log->l_prev_block = i;
964 log->l_curr_block = (int)*head_blk;
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000965 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 if (found == 2)
967 log->l_curr_cycle++;
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100968 atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn));
Dave Chinner84f3c682010-12-03 22:11:29 +1100969 atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn));
Dave Chinnera69ed032010-12-21 12:08:20 +1100970 xlog_assign_grant_head(&log->l_grant_reserve_head, log->l_curr_cycle,
971 BBTOB(log->l_curr_block));
972 xlog_assign_grant_head(&log->l_grant_write_head, log->l_curr_cycle,
973 BBTOB(log->l_curr_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974
975 /*
976 * Look for unmount record. If we find it, then we know there
977 * was a clean unmount. Since 'i' could be the last block in
978 * the physical log, we convert to a log block before comparing
979 * to the head_blk.
980 *
981 * Save the current tail lsn to use to pass to
982 * xlog_clear_stale_blocks() below. We won't want to clear the
983 * unmount record if there is one, so we pass the lsn of the
984 * unmount record rather than the block after it.
985 */
Eric Sandeen62118702008-03-06 13:44:28 +1100986 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000987 int h_size = be32_to_cpu(rhead->h_size);
988 int h_version = be32_to_cpu(rhead->h_version);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989
990 if ((h_version & XLOG_VERSION_2) &&
991 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
992 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
993 if (h_size % XLOG_HEADER_CYCLE_SIZE)
994 hblks++;
995 } else {
996 hblks = 1;
997 }
998 } else {
999 hblks = 1;
1000 }
1001 after_umount_blk = (i + hblks + (int)
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001002 BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize;
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001003 tail_lsn = atomic64_read(&log->l_tail_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 if (*head_blk == after_umount_blk &&
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001005 be32_to_cpu(rhead->h_num_logops) == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 umount_data_blk = (i + hblks) % log->l_logBBsize;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001007 error = xlog_bread(log, umount_data_blk, 1, bp, &offset);
1008 if (error)
Alex Elder9db127e2010-04-15 18:17:26 +00001009 goto done;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001010
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 op_head = (xlog_op_header_t *)offset;
1012 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
1013 /*
1014 * Set tail and last sync so that newly written
1015 * log records will point recovery to after the
1016 * current unmount record.
1017 */
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001018 xlog_assign_atomic_lsn(&log->l_tail_lsn,
1019 log->l_curr_cycle, after_umount_blk);
1020 xlog_assign_atomic_lsn(&log->l_last_sync_lsn,
1021 log->l_curr_cycle, after_umount_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 *tail_blk = after_umount_blk;
David Chinner92821e22007-05-24 15:26:31 +10001023
1024 /*
1025 * Note that the unmount was clean. If the unmount
1026 * was not clean, we need to know this to rebuild the
1027 * superblock counters from the perag headers if we
1028 * have a filesystem using non-persistent counters.
1029 */
1030 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 }
1032 }
1033
1034 /*
1035 * Make sure that there are no blocks in front of the head
1036 * with the same cycle number as the head. This can happen
1037 * because we allow multiple outstanding log writes concurrently,
1038 * and the later writes might make it out before earlier ones.
1039 *
1040 * We use the lsn from before modifying it so that we'll never
1041 * overwrite the unmount record after a clean unmount.
1042 *
1043 * Do this only if we are going to recover the filesystem
1044 *
1045 * NOTE: This used to say "if (!readonly)"
1046 * However on Linux, we can & do recover a read-only filesystem.
1047 * We only skip recovery if NORECOVERY is specified on mount,
1048 * in which case we would not be here.
1049 *
1050 * But... if the -device- itself is readonly, just skip this.
1051 * We can't recover this device anyway, so it won't matter.
1052 */
Alex Elder9db127e2010-04-15 18:17:26 +00001053 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 error = xlog_clear_stale_blocks(log, tail_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055
Alex Elder9db127e2010-04-15 18:17:26 +00001056done:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 xlog_put_bp(bp);
1058
1059 if (error)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001060 xfs_warn(log->l_mp, "failed to locate log tail");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 return error;
1062}
1063
1064/*
1065 * Is the log zeroed at all?
1066 *
1067 * The last binary search should be changed to perform an X block read
1068 * once X becomes small enough. You can then search linearly through
1069 * the X blocks. This will cut down on the number of reads we need to do.
1070 *
1071 * If the log is partially zeroed, this routine will pass back the blkno
1072 * of the first block with cycle number 0. It won't have a complete LR
1073 * preceding it.
1074 *
1075 * Return:
1076 * 0 => the log is completely written to
1077 * -1 => use *blk_no as the first block of the log
1078 * >0 => error has occurred
1079 */
David Chinnera8272ce2007-11-23 16:28:09 +11001080STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081xlog_find_zeroed(
1082 xlog_t *log,
1083 xfs_daddr_t *blk_no)
1084{
1085 xfs_buf_t *bp;
1086 xfs_caddr_t offset;
1087 uint first_cycle, last_cycle;
1088 xfs_daddr_t new_blk, last_blk, start_blk;
1089 xfs_daddr_t num_scan_bblks;
1090 int error, log_bbnum = log->l_logBBsize;
1091
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10001092 *blk_no = 0;
1093
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 /* check totally zeroed log */
1095 bp = xlog_get_bp(log, 1);
1096 if (!bp)
1097 return ENOMEM;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001098 error = xlog_bread(log, 0, 1, bp, &offset);
1099 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 goto bp_err;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001101
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10001102 first_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 if (first_cycle == 0) { /* completely zeroed log */
1104 *blk_no = 0;
1105 xlog_put_bp(bp);
1106 return -1;
1107 }
1108
1109 /* check partially zeroed log */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001110 error = xlog_bread(log, log_bbnum-1, 1, bp, &offset);
1111 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 goto bp_err;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001113
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10001114 last_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 if (last_cycle != 0) { /* log completely written to */
1116 xlog_put_bp(bp);
1117 return 0;
1118 } else if (first_cycle != 1) {
1119 /*
1120 * If the cycle of the last block is zero, the cycle of
1121 * the first block must be 1. If it's not, maybe we're
1122 * not looking at a log... Bail out.
1123 */
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001124 xfs_warn(log->l_mp,
1125 "Log inconsistent or not a log (last==0, first!=1)");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 return XFS_ERROR(EINVAL);
1127 }
1128
1129 /* we have a partially zeroed log */
1130 last_blk = log_bbnum-1;
1131 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1132 goto bp_err;
1133
1134 /*
1135 * Validate the answer. Because there is no way to guarantee that
1136 * the entire log is made up of log records which are the same size,
1137 * we scan over the defined maximum blocks. At this point, the maximum
1138 * is not chosen to mean anything special. XXXmiken
1139 */
1140 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1141 ASSERT(num_scan_bblks <= INT_MAX);
1142
1143 if (last_blk < num_scan_bblks)
1144 num_scan_bblks = last_blk;
1145 start_blk = last_blk - num_scan_bblks;
1146
1147 /*
1148 * We search for any instances of cycle number 0 that occur before
1149 * our current estimate of the head. What we're trying to detect is
1150 * 1 ... | 0 | 1 | 0...
1151 * ^ binary search ends here
1152 */
1153 if ((error = xlog_find_verify_cycle(log, start_blk,
1154 (int)num_scan_bblks, 0, &new_blk)))
1155 goto bp_err;
1156 if (new_blk != -1)
1157 last_blk = new_blk;
1158
1159 /*
1160 * Potentially backup over partial log record write. We don't need
1161 * to search the end of the log because we know it is zero.
1162 */
1163 if ((error = xlog_find_verify_log_record(log, start_blk,
1164 &last_blk, 0)) == -1) {
1165 error = XFS_ERROR(EIO);
1166 goto bp_err;
1167 } else if (error)
1168 goto bp_err;
1169
1170 *blk_no = last_blk;
1171bp_err:
1172 xlog_put_bp(bp);
1173 if (error)
1174 return error;
1175 return -1;
1176}
1177
1178/*
1179 * These are simple subroutines used by xlog_clear_stale_blocks() below
1180 * to initialize a buffer full of empty log record headers and write
1181 * them into the log.
1182 */
1183STATIC void
1184xlog_add_record(
1185 xlog_t *log,
1186 xfs_caddr_t buf,
1187 int cycle,
1188 int block,
1189 int tail_cycle,
1190 int tail_block)
1191{
1192 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1193
1194 memset(buf, 0, BBSIZE);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001195 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1196 recp->h_cycle = cpu_to_be32(cycle);
1197 recp->h_version = cpu_to_be32(
Eric Sandeen62118702008-03-06 13:44:28 +11001198 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001199 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1200 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1201 recp->h_fmt = cpu_to_be32(XLOG_FMT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1203}
1204
1205STATIC int
1206xlog_write_log_records(
1207 xlog_t *log,
1208 int cycle,
1209 int start_block,
1210 int blocks,
1211 int tail_cycle,
1212 int tail_block)
1213{
1214 xfs_caddr_t offset;
1215 xfs_buf_t *bp;
1216 int balign, ealign;
Alex Elder69ce58f2010-04-20 17:09:59 +10001217 int sectbb = log->l_sectBBsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 int end_block = start_block + blocks;
1219 int bufblks;
1220 int error = 0;
1221 int i, j = 0;
1222
Alex Elder6881a222010-04-13 15:22:29 +10001223 /*
1224 * Greedily allocate a buffer big enough to handle the full
1225 * range of basic blocks to be written. If that fails, try
1226 * a smaller size. We need to be able to write at least a
1227 * log sector, or we're out of luck.
1228 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 bufblks = 1 << ffs(blocks);
1230 while (!(bp = xlog_get_bp(log, bufblks))) {
1231 bufblks >>= 1;
Alex Elder69ce58f2010-04-20 17:09:59 +10001232 if (bufblks < sectbb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 return ENOMEM;
1234 }
1235
1236 /* We may need to do a read at the start to fill in part of
1237 * the buffer in the starting sector not covered by the first
1238 * write below.
1239 */
Alex Elder5c17f532010-04-13 15:22:48 +10001240 balign = round_down(start_block, sectbb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 if (balign != start_block) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001242 error = xlog_bread_noalign(log, start_block, 1, bp);
1243 if (error)
1244 goto out_put_bp;
1245
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 j = start_block - balign;
1247 }
1248
1249 for (i = start_block; i < end_block; i += bufblks) {
1250 int bcount, endcount;
1251
1252 bcount = min(bufblks, end_block - start_block);
1253 endcount = bcount - j;
1254
1255 /* We may need to do a read at the end to fill in part of
1256 * the buffer in the final sector not covered by the write.
1257 * If this is the same sector as the above read, skip it.
1258 */
Alex Elder5c17f532010-04-13 15:22:48 +10001259 ealign = round_down(end_block, sectbb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 if (j == 0 && (start_block + endcount > ealign)) {
Dave Chinner44396472011-04-21 09:34:27 +00001261 offset = XFS_BUF_PTR(bp) + BBTOB(ealign - start_block);
1262 error = xlog_bread_offset(log, ealign, sectbb,
1263 bp, offset);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001264 if (error)
1265 break;
1266
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 }
1268
1269 offset = xlog_align(log, start_block, endcount, bp);
1270 for (; j < endcount; j++) {
1271 xlog_add_record(log, offset, cycle, i+j,
1272 tail_cycle, tail_block);
1273 offset += BBSIZE;
1274 }
1275 error = xlog_bwrite(log, start_block, endcount, bp);
1276 if (error)
1277 break;
1278 start_block += endcount;
1279 j = 0;
1280 }
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001281
1282 out_put_bp:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 xlog_put_bp(bp);
1284 return error;
1285}
1286
1287/*
1288 * This routine is called to blow away any incomplete log writes out
1289 * in front of the log head. We do this so that we won't become confused
1290 * if we come up, write only a little bit more, and then crash again.
1291 * If we leave the partial log records out there, this situation could
1292 * cause us to think those partial writes are valid blocks since they
1293 * have the current cycle number. We get rid of them by overwriting them
1294 * with empty log records with the old cycle number rather than the
1295 * current one.
1296 *
1297 * The tail lsn is passed in rather than taken from
1298 * the log so that we will not write over the unmount record after a
1299 * clean unmount in a 512 block log. Doing so would leave the log without
1300 * any valid log records in it until a new one was written. If we crashed
1301 * during that time we would not be able to recover.
1302 */
1303STATIC int
1304xlog_clear_stale_blocks(
1305 xlog_t *log,
1306 xfs_lsn_t tail_lsn)
1307{
1308 int tail_cycle, head_cycle;
1309 int tail_block, head_block;
1310 int tail_distance, max_distance;
1311 int distance;
1312 int error;
1313
1314 tail_cycle = CYCLE_LSN(tail_lsn);
1315 tail_block = BLOCK_LSN(tail_lsn);
1316 head_cycle = log->l_curr_cycle;
1317 head_block = log->l_curr_block;
1318
1319 /*
1320 * Figure out the distance between the new head of the log
1321 * and the tail. We want to write over any blocks beyond the
1322 * head that we may have written just before the crash, but
1323 * we don't want to overwrite the tail of the log.
1324 */
1325 if (head_cycle == tail_cycle) {
1326 /*
1327 * The tail is behind the head in the physical log,
1328 * so the distance from the head to the tail is the
1329 * distance from the head to the end of the log plus
1330 * the distance from the beginning of the log to the
1331 * tail.
1332 */
1333 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1334 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1335 XFS_ERRLEVEL_LOW, log->l_mp);
1336 return XFS_ERROR(EFSCORRUPTED);
1337 }
1338 tail_distance = tail_block + (log->l_logBBsize - head_block);
1339 } else {
1340 /*
1341 * The head is behind the tail in the physical log,
1342 * so the distance from the head to the tail is just
1343 * the tail block minus the head block.
1344 */
1345 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1346 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1347 XFS_ERRLEVEL_LOW, log->l_mp);
1348 return XFS_ERROR(EFSCORRUPTED);
1349 }
1350 tail_distance = tail_block - head_block;
1351 }
1352
1353 /*
1354 * If the head is right up against the tail, we can't clear
1355 * anything.
1356 */
1357 if (tail_distance <= 0) {
1358 ASSERT(tail_distance == 0);
1359 return 0;
1360 }
1361
1362 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1363 /*
1364 * Take the smaller of the maximum amount of outstanding I/O
1365 * we could have and the distance to the tail to clear out.
1366 * We take the smaller so that we don't overwrite the tail and
1367 * we don't waste all day writing from the head to the tail
1368 * for no reason.
1369 */
1370 max_distance = MIN(max_distance, tail_distance);
1371
1372 if ((head_block + max_distance) <= log->l_logBBsize) {
1373 /*
1374 * We can stomp all the blocks we need to without
1375 * wrapping around the end of the log. Just do it
1376 * in a single write. Use the cycle number of the
1377 * current cycle minus one so that the log will look like:
1378 * n ... | n - 1 ...
1379 */
1380 error = xlog_write_log_records(log, (head_cycle - 1),
1381 head_block, max_distance, tail_cycle,
1382 tail_block);
1383 if (error)
1384 return error;
1385 } else {
1386 /*
1387 * We need to wrap around the end of the physical log in
1388 * order to clear all the blocks. Do it in two separate
1389 * I/Os. The first write should be from the head to the
1390 * end of the physical log, and it should use the current
1391 * cycle number minus one just like above.
1392 */
1393 distance = log->l_logBBsize - head_block;
1394 error = xlog_write_log_records(log, (head_cycle - 1),
1395 head_block, distance, tail_cycle,
1396 tail_block);
1397
1398 if (error)
1399 return error;
1400
1401 /*
1402 * Now write the blocks at the start of the physical log.
1403 * This writes the remainder of the blocks we want to clear.
1404 * It uses the current cycle number since we're now on the
1405 * same cycle as the head so that we get:
1406 * n ... n ... | n - 1 ...
1407 * ^^^^^ blocks we're writing
1408 */
1409 distance = max_distance - (log->l_logBBsize - head_block);
1410 error = xlog_write_log_records(log, head_cycle, 0, distance,
1411 tail_cycle, tail_block);
1412 if (error)
1413 return error;
1414 }
1415
1416 return 0;
1417}
1418
1419/******************************************************************************
1420 *
1421 * Log recover routines
1422 *
1423 ******************************************************************************
1424 */
1425
1426STATIC xlog_recover_t *
1427xlog_recover_find_tid(
Dave Chinnerf0a76952010-01-11 11:49:57 +00001428 struct hlist_head *head,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 xlog_tid_t tid)
1430{
Dave Chinnerf0a76952010-01-11 11:49:57 +00001431 xlog_recover_t *trans;
1432 struct hlist_node *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433
Dave Chinnerf0a76952010-01-11 11:49:57 +00001434 hlist_for_each_entry(trans, n, head, r_list) {
1435 if (trans->r_log_tid == tid)
1436 return trans;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 }
Dave Chinnerf0a76952010-01-11 11:49:57 +00001438 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439}
1440
1441STATIC void
Dave Chinnerf0a76952010-01-11 11:49:57 +00001442xlog_recover_new_tid(
1443 struct hlist_head *head,
1444 xlog_tid_t tid,
1445 xfs_lsn_t lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446{
Dave Chinnerf0a76952010-01-11 11:49:57 +00001447 xlog_recover_t *trans;
1448
1449 trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
1450 trans->r_log_tid = tid;
1451 trans->r_lsn = lsn;
1452 INIT_LIST_HEAD(&trans->r_itemq);
1453
1454 INIT_HLIST_NODE(&trans->r_list);
1455 hlist_add_head(&trans->r_list, head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456}
1457
1458STATIC void
1459xlog_recover_add_item(
Dave Chinnerf0a76952010-01-11 11:49:57 +00001460 struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461{
1462 xlog_recover_item_t *item;
1463
1464 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
Dave Chinnerf0a76952010-01-11 11:49:57 +00001465 INIT_LIST_HEAD(&item->ri_list);
1466 list_add_tail(&item->ri_list, head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467}
1468
1469STATIC int
1470xlog_recover_add_to_cont_trans(
Dave Chinner9abbc532010-04-13 15:06:46 +10001471 struct log *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 xlog_recover_t *trans,
1473 xfs_caddr_t dp,
1474 int len)
1475{
1476 xlog_recover_item_t *item;
1477 xfs_caddr_t ptr, old_ptr;
1478 int old_len;
1479
Dave Chinnerf0a76952010-01-11 11:49:57 +00001480 if (list_empty(&trans->r_itemq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 /* finish copying rest of trans header */
1482 xlog_recover_add_item(&trans->r_itemq);
1483 ptr = (xfs_caddr_t) &trans->r_theader +
1484 sizeof(xfs_trans_header_t) - len;
1485 memcpy(ptr, dp, len); /* d, s, l */
1486 return 0;
1487 }
Dave Chinnerf0a76952010-01-11 11:49:57 +00001488 /* take the tail entry */
1489 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490
1491 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
1492 old_len = item->ri_buf[item->ri_cnt-1].i_len;
1493
Christoph Hellwig760dea62005-09-02 16:56:02 +10001494 ptr = kmem_realloc(old_ptr, len+old_len, old_len, 0u);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 memcpy(&ptr[old_len], dp, len); /* d, s, l */
1496 item->ri_buf[item->ri_cnt-1].i_len += len;
1497 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
Dave Chinner9abbc532010-04-13 15:06:46 +10001498 trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 return 0;
1500}
1501
1502/*
1503 * The next region to add is the start of a new region. It could be
1504 * a whole region or it could be the first part of a new region. Because
1505 * of this, the assumption here is that the type and size fields of all
1506 * format structures fit into the first 32 bits of the structure.
1507 *
1508 * This works because all regions must be 32 bit aligned. Therefore, we
1509 * either have both fields or we have neither field. In the case we have
1510 * neither field, the data part of the region is zero length. We only have
1511 * a log_op_header and can throw away the header since a new one will appear
1512 * later. If we have at least 4 bytes, then we can determine how many regions
1513 * will appear in the current log item.
1514 */
1515STATIC int
1516xlog_recover_add_to_trans(
Dave Chinner9abbc532010-04-13 15:06:46 +10001517 struct log *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 xlog_recover_t *trans,
1519 xfs_caddr_t dp,
1520 int len)
1521{
1522 xfs_inode_log_format_t *in_f; /* any will do */
1523 xlog_recover_item_t *item;
1524 xfs_caddr_t ptr;
1525
1526 if (!len)
1527 return 0;
Dave Chinnerf0a76952010-01-11 11:49:57 +00001528 if (list_empty(&trans->r_itemq)) {
David Chinner5a792c42008-10-30 17:40:09 +11001529 /* we need to catch log corruptions here */
1530 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001531 xfs_warn(log->l_mp, "%s: bad header magic number",
1532 __func__);
David Chinner5a792c42008-10-30 17:40:09 +11001533 ASSERT(0);
1534 return XFS_ERROR(EIO);
1535 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 if (len == sizeof(xfs_trans_header_t))
1537 xlog_recover_add_item(&trans->r_itemq);
1538 memcpy(&trans->r_theader, dp, len); /* d, s, l */
1539 return 0;
1540 }
1541
1542 ptr = kmem_alloc(len, KM_SLEEP);
1543 memcpy(ptr, dp, len);
1544 in_f = (xfs_inode_log_format_t *)ptr;
1545
Dave Chinnerf0a76952010-01-11 11:49:57 +00001546 /* take the tail entry */
1547 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
1548 if (item->ri_total != 0 &&
1549 item->ri_total == item->ri_cnt) {
1550 /* tail item is in use, get a new one */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 xlog_recover_add_item(&trans->r_itemq);
Dave Chinnerf0a76952010-01-11 11:49:57 +00001552 item = list_entry(trans->r_itemq.prev,
1553 xlog_recover_item_t, ri_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555
1556 if (item->ri_total == 0) { /* first region to be added */
Christoph Hellwige8fa6b42009-03-03 14:48:36 -05001557 if (in_f->ilf_size == 0 ||
1558 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001559 xfs_warn(log->l_mp,
1560 "bad number of regions (%d) in inode log format",
Christoph Hellwige8fa6b42009-03-03 14:48:36 -05001561 in_f->ilf_size);
1562 ASSERT(0);
1563 return XFS_ERROR(EIO);
1564 }
1565
1566 item->ri_total = in_f->ilf_size;
1567 item->ri_buf =
1568 kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
1569 KM_SLEEP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 }
1571 ASSERT(item->ri_total > item->ri_cnt);
1572 /* Description region is ri_buf[0] */
1573 item->ri_buf[item->ri_cnt].i_addr = ptr;
1574 item->ri_buf[item->ri_cnt].i_len = len;
1575 item->ri_cnt++;
Dave Chinner9abbc532010-04-13 15:06:46 +10001576 trace_xfs_log_recover_item_add(log, trans, item, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 return 0;
1578}
1579
Dave Chinnerf0a76952010-01-11 11:49:57 +00001580/*
1581 * Sort the log items in the transaction. Cancelled buffers need
1582 * to be put first so they are processed before any items that might
1583 * modify the buffers. If they are cancelled, then the modifications
1584 * don't need to be replayed.
1585 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586STATIC int
1587xlog_recover_reorder_trans(
Dave Chinner9abbc532010-04-13 15:06:46 +10001588 struct log *log,
1589 xlog_recover_t *trans,
1590 int pass)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591{
Dave Chinnerf0a76952010-01-11 11:49:57 +00001592 xlog_recover_item_t *item, *n;
1593 LIST_HEAD(sort_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594
Dave Chinnerf0a76952010-01-11 11:49:57 +00001595 list_splice_init(&trans->r_itemq, &sort_list);
1596 list_for_each_entry_safe(item, n, &sort_list, ri_list) {
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10001597 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
Dave Chinnerf0a76952010-01-11 11:49:57 +00001598
1599 switch (ITEM_TYPE(item)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 case XFS_LI_BUF:
Dave Chinnerc1155412010-05-07 11:05:19 +10001601 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
Dave Chinner9abbc532010-04-13 15:06:46 +10001602 trace_xfs_log_recover_item_reorder_head(log,
1603 trans, item, pass);
Dave Chinnerf0a76952010-01-11 11:49:57 +00001604 list_move(&item->ri_list, &trans->r_itemq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 break;
1606 }
1607 case XFS_LI_INODE:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 case XFS_LI_DQUOT:
1609 case XFS_LI_QUOTAOFF:
1610 case XFS_LI_EFD:
1611 case XFS_LI_EFI:
Dave Chinner9abbc532010-04-13 15:06:46 +10001612 trace_xfs_log_recover_item_reorder_tail(log,
1613 trans, item, pass);
Dave Chinnerf0a76952010-01-11 11:49:57 +00001614 list_move_tail(&item->ri_list, &trans->r_itemq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 break;
1616 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001617 xfs_warn(log->l_mp,
1618 "%s: unrecognized type of log operation",
1619 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 ASSERT(0);
1621 return XFS_ERROR(EIO);
1622 }
Dave Chinnerf0a76952010-01-11 11:49:57 +00001623 }
1624 ASSERT(list_empty(&sort_list));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 return 0;
1626}
1627
1628/*
1629 * Build up the table of buf cancel records so that we don't replay
1630 * cancelled data in the second pass. For buffer records that are
1631 * not cancel records, there is nothing to do here so we just return.
1632 *
1633 * If we get a cancel record which is already in the table, this indicates
1634 * that the buffer was cancelled multiple times. In order to ensure
1635 * that during pass 2 we keep the record in the table until we reach its
1636 * last occurrence in the log, we keep a reference count in the cancel
1637 * record in the table to tell us how many times we expect to see this
1638 * record during the second pass.
1639 */
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001640STATIC int
1641xlog_recover_buffer_pass1(
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001642 struct log *log,
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001643 xlog_recover_item_t *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001645 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001646 struct list_head *bucket;
1647 struct xfs_buf_cancel *bcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648
1649 /*
1650 * If this isn't a cancel buffer item, then just return.
1651 */
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001652 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
Dave Chinner9abbc532010-04-13 15:06:46 +10001653 trace_xfs_log_recover_buf_not_cancel(log, buf_f);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001654 return 0;
Dave Chinner9abbc532010-04-13 15:06:46 +10001655 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656
1657 /*
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001658 * Insert an xfs_buf_cancel record into the hash table of them.
1659 * If there is already an identical record, bump its reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 */
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001661 bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno);
1662 list_for_each_entry(bcp, bucket, bc_list) {
1663 if (bcp->bc_blkno == buf_f->blf_blkno &&
1664 bcp->bc_len == buf_f->blf_len) {
1665 bcp->bc_refcount++;
Dave Chinner9abbc532010-04-13 15:06:46 +10001666 trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001667 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669 }
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001670
1671 bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP);
1672 bcp->bc_blkno = buf_f->blf_blkno;
1673 bcp->bc_len = buf_f->blf_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 bcp->bc_refcount = 1;
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001675 list_add_tail(&bcp->bc_list, bucket);
1676
Dave Chinner9abbc532010-04-13 15:06:46 +10001677 trace_xfs_log_recover_buf_cancel_add(log, buf_f);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001678 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679}
1680
1681/*
1682 * Check to see whether the buffer being recovered has a corresponding
1683 * entry in the buffer cancel record table. If it does then return 1
1684 * so that it will be cancelled, otherwise return 0. If the buffer is
Dave Chinnerc1155412010-05-07 11:05:19 +10001685 * actually a buffer cancel item (XFS_BLF_CANCEL is set), then decrement
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 * the refcount on the entry in the table and remove it from the table
1687 * if this is the last reference.
1688 *
1689 * We remove the cancel record from the table when we encounter its
1690 * last occurrence in the log so that if the same buffer is re-used
1691 * again after its last cancellation we actually replay the changes
1692 * made at that point.
1693 */
1694STATIC int
1695xlog_check_buffer_cancelled(
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001696 struct log *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 xfs_daddr_t blkno,
1698 uint len,
1699 ushort flags)
1700{
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001701 struct list_head *bucket;
1702 struct xfs_buf_cancel *bcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703
1704 if (log->l_buf_cancel_table == NULL) {
1705 /*
1706 * There is nothing in the table built in pass one,
1707 * so this buffer must not be cancelled.
1708 */
Dave Chinnerc1155412010-05-07 11:05:19 +10001709 ASSERT(!(flags & XFS_BLF_CANCEL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 return 0;
1711 }
1712
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001713 /*
1714 * Search for an entry in the cancel table that matches our buffer.
1715 */
1716 bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno);
1717 list_for_each_entry(bcp, bucket, bc_list) {
1718 if (bcp->bc_blkno == blkno && bcp->bc_len == len)
1719 goto found;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 }
1721
1722 /*
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001723 * We didn't find a corresponding entry in the table, so return 0 so
1724 * that the buffer is NOT cancelled.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 */
Dave Chinnerc1155412010-05-07 11:05:19 +10001726 ASSERT(!(flags & XFS_BLF_CANCEL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 return 0;
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001728
1729found:
1730 /*
1731 * We've go a match, so return 1 so that the recovery of this buffer
1732 * is cancelled. If this buffer is actually a buffer cancel log
1733 * item, then decrement the refcount on the one in the table and
1734 * remove it if this is the last reference.
1735 */
1736 if (flags & XFS_BLF_CANCEL) {
1737 if (--bcp->bc_refcount == 0) {
1738 list_del(&bcp->bc_list);
1739 kmem_free(bcp);
1740 }
1741 }
1742 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743}
1744
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745/*
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001746 * Perform recovery for a buffer full of inodes. In these buffers, the only
1747 * data which should be recovered is that which corresponds to the
1748 * di_next_unlinked pointers in the on disk inode structures. The rest of the
1749 * data for the inodes is always logged through the inodes themselves rather
1750 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 *
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001752 * The only time when buffers full of inodes are fully recovered is when the
1753 * buffer is full of newly allocated inodes. In this case the buffer will
1754 * not be marked as an inode buffer and so will be sent to
1755 * xlog_recover_do_reg_buffer() below during recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 */
1757STATIC int
1758xlog_recover_do_inode_buffer(
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001759 struct xfs_mount *mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 xlog_recover_item_t *item,
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001761 struct xfs_buf *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 xfs_buf_log_format_t *buf_f)
1763{
1764 int i;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001765 int item_index = 0;
1766 int bit = 0;
1767 int nbits = 0;
1768 int reg_buf_offset = 0;
1769 int reg_buf_bytes = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 int next_unlinked_offset;
1771 int inodes_per_buf;
1772 xfs_agino_t *logged_nextp;
1773 xfs_agino_t *buffer_nextp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774
Dave Chinner9abbc532010-04-13 15:06:46 +10001775 trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f);
1776
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 inodes_per_buf = XFS_BUF_COUNT(bp) >> mp->m_sb.sb_inodelog;
1778 for (i = 0; i < inodes_per_buf; i++) {
1779 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
1780 offsetof(xfs_dinode_t, di_next_unlinked);
1781
1782 while (next_unlinked_offset >=
1783 (reg_buf_offset + reg_buf_bytes)) {
1784 /*
1785 * The next di_next_unlinked field is beyond
1786 * the current logged region. Find the next
1787 * logged region that contains or is beyond
1788 * the current di_next_unlinked field.
1789 */
1790 bit += nbits;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001791 bit = xfs_next_bit(buf_f->blf_data_map,
1792 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793
1794 /*
1795 * If there are no more logged regions in the
1796 * buffer, then we're done.
1797 */
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001798 if (bit == -1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001801 nbits = xfs_contig_bits(buf_f->blf_data_map,
1802 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 ASSERT(nbits > 0);
Dave Chinnerc1155412010-05-07 11:05:19 +10001804 reg_buf_offset = bit << XFS_BLF_SHIFT;
1805 reg_buf_bytes = nbits << XFS_BLF_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806 item_index++;
1807 }
1808
1809 /*
1810 * If the current logged region starts after the current
1811 * di_next_unlinked field, then move on to the next
1812 * di_next_unlinked field.
1813 */
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001814 if (next_unlinked_offset < reg_buf_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816
1817 ASSERT(item->ri_buf[item_index].i_addr != NULL);
Dave Chinnerc1155412010-05-07 11:05:19 +10001818 ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 ASSERT((reg_buf_offset + reg_buf_bytes) <= XFS_BUF_COUNT(bp));
1820
1821 /*
1822 * The current logged region contains a copy of the
1823 * current di_next_unlinked field. Extract its value
1824 * and copy it to the buffer copy.
1825 */
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10001826 logged_nextp = item->ri_buf[item_index].i_addr +
1827 next_unlinked_offset - reg_buf_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 if (unlikely(*logged_nextp == 0)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001829 xfs_alert(mp,
1830 "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
1831 "Trying to replay bad (0) inode di_next_unlinked field.",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 item, bp);
1833 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1834 XFS_ERRLEVEL_LOW, mp);
1835 return XFS_ERROR(EFSCORRUPTED);
1836 }
1837
1838 buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
1839 next_unlinked_offset);
Tim Shimmin87c199c2006-06-09 14:56:16 +10001840 *buffer_nextp = *logged_nextp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 }
1842
1843 return 0;
1844}
1845
1846/*
1847 * Perform a 'normal' buffer recovery. Each logged region of the
1848 * buffer should be copied over the corresponding region in the
1849 * given buffer. The bitmap in the buf log format structure indicates
1850 * where to place the logged data.
1851 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852STATIC void
1853xlog_recover_do_reg_buffer(
Dave Chinner9abbc532010-04-13 15:06:46 +10001854 struct xfs_mount *mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 xlog_recover_item_t *item,
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001856 struct xfs_buf *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857 xfs_buf_log_format_t *buf_f)
1858{
1859 int i;
1860 int bit;
1861 int nbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862 int error;
1863
Dave Chinner9abbc532010-04-13 15:06:46 +10001864 trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f);
1865
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 bit = 0;
1867 i = 1; /* 0 is the buf format structure */
1868 while (1) {
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001869 bit = xfs_next_bit(buf_f->blf_data_map,
1870 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 if (bit == -1)
1872 break;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001873 nbits = xfs_contig_bits(buf_f->blf_data_map,
1874 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 ASSERT(nbits > 0);
Christoph Hellwig4b809162007-08-16 15:37:36 +10001876 ASSERT(item->ri_buf[i].i_addr != NULL);
Dave Chinnerc1155412010-05-07 11:05:19 +10001877 ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 ASSERT(XFS_BUF_COUNT(bp) >=
Dave Chinnerc1155412010-05-07 11:05:19 +10001879 ((uint)bit << XFS_BLF_SHIFT)+(nbits<<XFS_BLF_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880
1881 /*
1882 * Do a sanity check if this is a dquot buffer. Just checking
1883 * the first dquot in the buffer should do. XXXThis is
1884 * probably a good thing to do for other buf types also.
1885 */
1886 error = 0;
Nathan Scottc8ad20f2005-06-21 15:38:48 +10001887 if (buf_f->blf_flags &
Dave Chinnerc1155412010-05-07 11:05:19 +10001888 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02001889 if (item->ri_buf[i].i_addr == NULL) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001890 xfs_alert(mp,
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02001891 "XFS: NULL dquot in %s.", __func__);
1892 goto next;
1893 }
Jan Rekorajski8ec6dba2009-11-16 11:57:02 +00001894 if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001895 xfs_alert(mp,
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02001896 "XFS: dquot too small (%d) in %s.",
1897 item->ri_buf[i].i_len, __func__);
1898 goto next;
1899 }
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001900 error = xfs_qm_dqcheck(mp, item->ri_buf[i].i_addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901 -1, 0, XFS_QMOPT_DOWARN,
1902 "dquot_buf_recover");
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02001903 if (error)
1904 goto next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905 }
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02001906
1907 memcpy(xfs_buf_offset(bp,
Dave Chinnerc1155412010-05-07 11:05:19 +10001908 (uint)bit << XFS_BLF_SHIFT), /* dest */
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02001909 item->ri_buf[i].i_addr, /* source */
Dave Chinnerc1155412010-05-07 11:05:19 +10001910 nbits<<XFS_BLF_SHIFT); /* length */
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02001911 next:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 i++;
1913 bit += nbits;
1914 }
1915
1916 /* Shouldn't be any more regions */
1917 ASSERT(i == item->ri_total);
1918}
1919
1920/*
1921 * Do some primitive error checking on ondisk dquot data structures.
1922 */
1923int
1924xfs_qm_dqcheck(
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001925 struct xfs_mount *mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 xfs_disk_dquot_t *ddq,
1927 xfs_dqid_t id,
1928 uint type, /* used only when IO_dorepair is true */
1929 uint flags,
1930 char *str)
1931{
1932 xfs_dqblk_t *d = (xfs_dqblk_t *)ddq;
1933 int errs = 0;
1934
1935 /*
1936 * We can encounter an uninitialized dquot buffer for 2 reasons:
1937 * 1. If we crash while deleting the quotainode(s), and those blks got
1938 * used for user data. This is because we take the path of regular
1939 * file deletion; however, the size field of quotainodes is never
1940 * updated, so all the tricks that we play in itruncate_finish
1941 * don't quite matter.
1942 *
1943 * 2. We don't play the quota buffers when there's a quotaoff logitem.
1944 * But the allocation will be replayed so we'll end up with an
1945 * uninitialized quota block.
1946 *
1947 * This is all fine; things are still consistent, and we haven't lost
1948 * any quota information. Just don't complain about bad dquot blks.
1949 */
Christoph Hellwig1149d962005-11-02 15:01:12 +11001950 if (be16_to_cpu(ddq->d_magic) != XFS_DQUOT_MAGIC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001952 xfs_alert(mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
Christoph Hellwig1149d962005-11-02 15:01:12 +11001954 str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 errs++;
1956 }
Christoph Hellwig1149d962005-11-02 15:01:12 +11001957 if (ddq->d_version != XFS_DQUOT_VERSION) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001959 xfs_alert(mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
Christoph Hellwig1149d962005-11-02 15:01:12 +11001961 str, id, ddq->d_version, XFS_DQUOT_VERSION);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 errs++;
1963 }
1964
Christoph Hellwig1149d962005-11-02 15:01:12 +11001965 if (ddq->d_flags != XFS_DQ_USER &&
1966 ddq->d_flags != XFS_DQ_PROJ &&
1967 ddq->d_flags != XFS_DQ_GROUP) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001969 xfs_alert(mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
Christoph Hellwig1149d962005-11-02 15:01:12 +11001971 str, id, ddq->d_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 errs++;
1973 }
1974
Christoph Hellwig1149d962005-11-02 15:01:12 +11001975 if (id != -1 && id != be32_to_cpu(ddq->d_id)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001977 xfs_alert(mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978 "%s : ondisk-dquot 0x%p, ID mismatch: "
1979 "0x%x expected, found id 0x%x",
Christoph Hellwig1149d962005-11-02 15:01:12 +11001980 str, ddq, id, be32_to_cpu(ddq->d_id));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 errs++;
1982 }
1983
1984 if (!errs && ddq->d_id) {
Christoph Hellwig1149d962005-11-02 15:01:12 +11001985 if (ddq->d_blk_softlimit &&
1986 be64_to_cpu(ddq->d_bcount) >=
1987 be64_to_cpu(ddq->d_blk_softlimit)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 if (!ddq->d_btimer) {
1989 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001990 xfs_alert(mp,
1991 "%s : Dquot ID 0x%x (0x%p) BLK TIMER NOT STARTED",
Christoph Hellwig1149d962005-11-02 15:01:12 +11001992 str, (int)be32_to_cpu(ddq->d_id), ddq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993 errs++;
1994 }
1995 }
Christoph Hellwig1149d962005-11-02 15:01:12 +11001996 if (ddq->d_ino_softlimit &&
1997 be64_to_cpu(ddq->d_icount) >=
1998 be64_to_cpu(ddq->d_ino_softlimit)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 if (!ddq->d_itimer) {
2000 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002001 xfs_alert(mp,
2002 "%s : Dquot ID 0x%x (0x%p) INODE TIMER NOT STARTED",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002003 str, (int)be32_to_cpu(ddq->d_id), ddq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 errs++;
2005 }
2006 }
Christoph Hellwig1149d962005-11-02 15:01:12 +11002007 if (ddq->d_rtb_softlimit &&
2008 be64_to_cpu(ddq->d_rtbcount) >=
2009 be64_to_cpu(ddq->d_rtb_softlimit)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010 if (!ddq->d_rtbtimer) {
2011 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002012 xfs_alert(mp,
2013 "%s : Dquot ID 0x%x (0x%p) RTBLK TIMER NOT STARTED",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002014 str, (int)be32_to_cpu(ddq->d_id), ddq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 errs++;
2016 }
2017 }
2018 }
2019
2020 if (!errs || !(flags & XFS_QMOPT_DQREPAIR))
2021 return errs;
2022
2023 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002024 xfs_notice(mp, "Re-initializing dquot ID 0x%x", id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025
2026 /*
2027 * Typically, a repair is only requested by quotacheck.
2028 */
2029 ASSERT(id != -1);
2030 ASSERT(flags & XFS_QMOPT_DQREPAIR);
2031 memset(d, 0, sizeof(xfs_dqblk_t));
Christoph Hellwig1149d962005-11-02 15:01:12 +11002032
2033 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
2034 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
2035 d->dd_diskdq.d_flags = type;
2036 d->dd_diskdq.d_id = cpu_to_be32(id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037
2038 return errs;
2039}
2040
2041/*
2042 * Perform a dquot buffer recovery.
2043 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2044 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2045 * Else, treat it as a regular buffer and do recovery.
2046 */
2047STATIC void
2048xlog_recover_do_dquot_buffer(
2049 xfs_mount_t *mp,
2050 xlog_t *log,
2051 xlog_recover_item_t *item,
2052 xfs_buf_t *bp,
2053 xfs_buf_log_format_t *buf_f)
2054{
2055 uint type;
2056
Dave Chinner9abbc532010-04-13 15:06:46 +10002057 trace_xfs_log_recover_buf_dquot_buf(log, buf_f);
2058
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059 /*
2060 * Filesystems are required to send in quota flags at mount time.
2061 */
2062 if (mp->m_qflags == 0) {
2063 return;
2064 }
2065
2066 type = 0;
Dave Chinnerc1155412010-05-07 11:05:19 +10002067 if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 type |= XFS_DQ_USER;
Dave Chinnerc1155412010-05-07 11:05:19 +10002069 if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF)
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002070 type |= XFS_DQ_PROJ;
Dave Chinnerc1155412010-05-07 11:05:19 +10002071 if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 type |= XFS_DQ_GROUP;
2073 /*
2074 * This type of quotas was turned off, so ignore this buffer
2075 */
2076 if (log->l_quotaoffs_flag & type)
2077 return;
2078
Dave Chinner9abbc532010-04-13 15:06:46 +10002079 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080}
2081
2082/*
2083 * This routine replays a modification made to a buffer at runtime.
2084 * There are actually two types of buffer, regular and inode, which
2085 * are handled differently. Inode buffers are handled differently
2086 * in that we only recover a specific set of data from them, namely
2087 * the inode di_next_unlinked fields. This is because all other inode
2088 * data is actually logged via inode records and any data we replay
2089 * here which overlaps that may be stale.
2090 *
2091 * When meta-data buffers are freed at run time we log a buffer item
Dave Chinnerc1155412010-05-07 11:05:19 +10002092 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093 * of the buffer in the log should not be replayed at recovery time.
2094 * This is so that if the blocks covered by the buffer are reused for
2095 * file data before we crash we don't end up replaying old, freed
2096 * meta-data into a user's file.
2097 *
2098 * To handle the cancellation of buffer log items, we make two passes
2099 * over the log during recovery. During the first we build a table of
2100 * those buffers which have been cancelled, and during the second we
2101 * only replay those buffers which do not have corresponding cancel
2102 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2103 * for more details on the implementation of the table of cancel records.
2104 */
2105STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002106xlog_recover_buffer_pass2(
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107 xlog_t *log,
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002108 xlog_recover_item_t *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109{
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002110 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002111 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 xfs_buf_t *bp;
2113 int error;
Christoph Hellwig6ad112b2009-11-24 18:02:23 +00002114 uint buf_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002116 /*
2117 * In this pass we only want to recover all the buffers which have
2118 * not been cancelled and are not cancellation buffers themselves.
2119 */
2120 if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno,
2121 buf_f->blf_len, buf_f->blf_flags)) {
2122 trace_xfs_log_recover_buf_cancel(log, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 }
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002125
Dave Chinner9abbc532010-04-13 15:06:46 +10002126 trace_xfs_log_recover_buf_recover(log, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127
Christoph Hellwig0cadda12010-01-19 09:56:44 +00002128 buf_flags = XBF_LOCK;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002129 if (!(buf_f->blf_flags & XFS_BLF_INODE_BUF))
Christoph Hellwig0cadda12010-01-19 09:56:44 +00002130 buf_flags |= XBF_MAPPED;
Christoph Hellwig6ad112b2009-11-24 18:02:23 +00002131
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002132 bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len,
2133 buf_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134 if (XFS_BUF_ISERROR(bp)) {
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002135 xfs_ioerror_alert("xlog_recover_do..(read#1)", mp,
2136 bp, buf_f->blf_blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 error = XFS_BUF_GETERROR(bp);
2138 xfs_buf_relse(bp);
2139 return error;
2140 }
2141
2142 error = 0;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002143 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002145 } else if (buf_f->blf_flags &
Dave Chinnerc1155412010-05-07 11:05:19 +10002146 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147 xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2148 } else {
Dave Chinner9abbc532010-04-13 15:06:46 +10002149 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150 }
2151 if (error)
2152 return XFS_ERROR(error);
2153
2154 /*
2155 * Perform delayed write on the buffer. Asynchronous writes will be
2156 * slower when taking into account all the buffers to be flushed.
2157 *
2158 * Also make sure that only inode buffers with good sizes stay in
2159 * the buffer cache. The kernel moves inodes in buffers of 1 block
2160 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2161 * buffers in the log can be a different size if the log was generated
2162 * by an older kernel using unclustered inode buffers or a newer kernel
2163 * running with a different inode cluster size. Regardless, if the
2164 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2165 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2166 * the buffer out of the buffer cache so that the buffer won't
2167 * overlap with future reads of those inodes.
2168 */
2169 if (XFS_DINODE_MAGIC ==
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002170 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 (XFS_BUF_COUNT(bp) != MAX(log->l_mp->m_sb.sb_blocksize,
2172 (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) {
2173 XFS_BUF_STALE(bp);
2174 error = xfs_bwrite(mp, bp);
2175 } else {
Dave Chinnerebad8612010-09-22 10:47:20 +10002176 ASSERT(bp->b_target->bt_mount == mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2178 xfs_bdwrite(mp, bp);
2179 }
2180
2181 return (error);
2182}
2183
2184STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002185xlog_recover_inode_pass2(
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186 xlog_t *log,
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002187 xlog_recover_item_t *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188{
2189 xfs_inode_log_format_t *in_f;
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002190 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191 xfs_buf_t *bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 xfs_dinode_t *dip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193 int len;
2194 xfs_caddr_t src;
2195 xfs_caddr_t dest;
2196 int error;
2197 int attr_index;
2198 uint fields;
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002199 xfs_icdinode_t *dicp;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002200 int need_free = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201
Tim Shimmin6d192a92006-06-09 14:55:38 +10002202 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002203 in_f = item->ri_buf[0].i_addr;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002204 } else {
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002205 in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002206 need_free = 1;
2207 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2208 if (error)
2209 goto error;
2210 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211
2212 /*
2213 * Inode buffers can be freed, look out for it,
2214 * and do not replay the inode.
2215 */
Christoph Hellwiga1941892008-11-28 14:23:40 +11002216 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
2217 in_f->ilf_len, 0)) {
Tim Shimmin6d192a92006-06-09 14:55:38 +10002218 error = 0;
Dave Chinner9abbc532010-04-13 15:06:46 +10002219 trace_xfs_log_recover_inode_cancel(log, in_f);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002220 goto error;
2221 }
Dave Chinner9abbc532010-04-13 15:06:46 +10002222 trace_xfs_log_recover_inode_recover(log, in_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223
Christoph Hellwig6ad112b2009-11-24 18:02:23 +00002224 bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len,
Christoph Hellwig0cadda12010-01-19 09:56:44 +00002225 XBF_LOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 if (XFS_BUF_ISERROR(bp)) {
2227 xfs_ioerror_alert("xlog_recover_do..(read#2)", mp,
Christoph Hellwiga1941892008-11-28 14:23:40 +11002228 bp, in_f->ilf_blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229 error = XFS_BUF_GETERROR(bp);
2230 xfs_buf_relse(bp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002231 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232 }
2233 error = 0;
2234 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
Christoph Hellwiga1941892008-11-28 14:23:40 +11002235 dip = (xfs_dinode_t *)xfs_buf_offset(bp, in_f->ilf_boffset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236
2237 /*
2238 * Make sure the place we're flushing out to really looks
2239 * like an inode!
2240 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002241 if (unlikely(be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 xfs_buf_relse(bp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002243 xfs_alert(mp,
2244 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
2245 __func__, dip, bp, in_f->ilf_ino);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002246 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 XFS_ERRLEVEL_LOW, mp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002248 error = EFSCORRUPTED;
2249 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 }
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002251 dicp = item->ri_buf[1].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
2253 xfs_buf_relse(bp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002254 xfs_alert(mp,
2255 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
2256 __func__, item, in_f->ilf_ino);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002257 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 XFS_ERRLEVEL_LOW, mp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002259 error = EFSCORRUPTED;
2260 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261 }
2262
2263 /* Skip replay when the on disk inode is newer than the log one */
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002264 if (dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265 /*
2266 * Deal with the wrap case, DI_MAX_FLUSH is less
2267 * than smaller numbers
2268 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002269 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002270 dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271 /* do nothing */
2272 } else {
2273 xfs_buf_relse(bp);
Dave Chinner9abbc532010-04-13 15:06:46 +10002274 trace_xfs_log_recover_inode_skip(log, in_f);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002275 error = 0;
2276 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277 }
2278 }
2279 /* Take the opportunity to reset the flush iteration count */
2280 dicp->di_flushiter = 0;
2281
2282 if (unlikely((dicp->di_mode & S_IFMT) == S_IFREG)) {
2283 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2284 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002285 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 XFS_ERRLEVEL_LOW, mp, dicp);
2287 xfs_buf_relse(bp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002288 xfs_alert(mp,
2289 "%s: Bad regular inode log record, rec ptr 0x%p, "
2290 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2291 __func__, item, dip, bp, in_f->ilf_ino);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002292 error = EFSCORRUPTED;
2293 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 }
2295 } else if (unlikely((dicp->di_mode & S_IFMT) == S_IFDIR)) {
2296 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2297 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2298 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002299 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 XFS_ERRLEVEL_LOW, mp, dicp);
2301 xfs_buf_relse(bp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002302 xfs_alert(mp,
2303 "%s: Bad dir inode log record, rec ptr 0x%p, "
2304 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2305 __func__, item, dip, bp, in_f->ilf_ino);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002306 error = EFSCORRUPTED;
2307 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 }
2309 }
2310 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002311 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312 XFS_ERRLEVEL_LOW, mp, dicp);
2313 xfs_buf_relse(bp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002314 xfs_alert(mp,
2315 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2316 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2317 __func__, item, dip, bp, in_f->ilf_ino,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318 dicp->di_nextents + dicp->di_anextents,
2319 dicp->di_nblocks);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002320 error = EFSCORRUPTED;
2321 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 }
2323 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002324 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325 XFS_ERRLEVEL_LOW, mp, dicp);
2326 xfs_buf_relse(bp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002327 xfs_alert(mp,
2328 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2329 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__,
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002330 item, dip, bp, in_f->ilf_ino, dicp->di_forkoff);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002331 error = EFSCORRUPTED;
2332 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 }
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002334 if (unlikely(item->ri_buf[1].i_len > sizeof(struct xfs_icdinode))) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002335 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 XFS_ERRLEVEL_LOW, mp, dicp);
2337 xfs_buf_relse(bp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002338 xfs_alert(mp,
2339 "%s: Bad inode log record length %d, rec ptr 0x%p",
2340 __func__, item->ri_buf[1].i_len, item);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002341 error = EFSCORRUPTED;
2342 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 }
2344
2345 /* The core is in in-core format */
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002346 xfs_dinode_to_disk(dip, item->ri_buf[1].i_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347
2348 /* the rest is in on-disk format */
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002349 if (item->ri_buf[1].i_len > sizeof(struct xfs_icdinode)) {
2350 memcpy((xfs_caddr_t) dip + sizeof(struct xfs_icdinode),
2351 item->ri_buf[1].i_addr + sizeof(struct xfs_icdinode),
2352 item->ri_buf[1].i_len - sizeof(struct xfs_icdinode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353 }
2354
2355 fields = in_f->ilf_fields;
2356 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2357 case XFS_ILOG_DEV:
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002358 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359 break;
2360 case XFS_ILOG_UUID:
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002361 memcpy(XFS_DFORK_DPTR(dip),
2362 &in_f->ilf_u.ilfu_uuid,
2363 sizeof(uuid_t));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364 break;
2365 }
2366
2367 if (in_f->ilf_size == 2)
2368 goto write_inode_buffer;
2369 len = item->ri_buf[2].i_len;
2370 src = item->ri_buf[2].i_addr;
2371 ASSERT(in_f->ilf_size <= 4);
2372 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2373 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2374 (len == in_f->ilf_dsize));
2375
2376 switch (fields & XFS_ILOG_DFORK) {
2377 case XFS_ILOG_DDATA:
2378 case XFS_ILOG_DEXT:
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002379 memcpy(XFS_DFORK_DPTR(dip), src, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380 break;
2381
2382 case XFS_ILOG_DBROOT:
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002383 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002384 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 XFS_DFORK_DSIZE(dip, mp));
2386 break;
2387
2388 default:
2389 /*
2390 * There are no data fork flags set.
2391 */
2392 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2393 break;
2394 }
2395
2396 /*
2397 * If we logged any attribute data, recover it. There may or
2398 * may not have been any other non-core data logged in this
2399 * transaction.
2400 */
2401 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2402 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2403 attr_index = 3;
2404 } else {
2405 attr_index = 2;
2406 }
2407 len = item->ri_buf[attr_index].i_len;
2408 src = item->ri_buf[attr_index].i_addr;
2409 ASSERT(len == in_f->ilf_asize);
2410
2411 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2412 case XFS_ILOG_ADATA:
2413 case XFS_ILOG_AEXT:
2414 dest = XFS_DFORK_APTR(dip);
2415 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2416 memcpy(dest, src, len);
2417 break;
2418
2419 case XFS_ILOG_ABROOT:
2420 dest = XFS_DFORK_APTR(dip);
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002421 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
2422 len, (xfs_bmdr_block_t*)dest,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 XFS_DFORK_ASIZE(dip, mp));
2424 break;
2425
2426 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002427 xfs_warn(log->l_mp, "%s: Invalid flag", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428 ASSERT(0);
2429 xfs_buf_relse(bp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002430 error = EIO;
2431 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002432 }
2433 }
2434
2435write_inode_buffer:
Dave Chinnerebad8612010-09-22 10:47:20 +10002436 ASSERT(bp->b_target->bt_mount == mp);
Christoph Hellwigdd0bbad2009-03-16 08:19:59 +01002437 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2438 xfs_bdwrite(mp, bp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002439error:
2440 if (need_free)
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002441 kmem_free(in_f);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002442 return XFS_ERROR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443}
2444
2445/*
2446 * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
2447 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2448 * of that type.
2449 */
2450STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002451xlog_recover_quotaoff_pass1(
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452 xlog_t *log,
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002453 xlog_recover_item_t *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002455 xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456 ASSERT(qoff_f);
2457
2458 /*
2459 * The logitem format's flag tells us if this was user quotaoff,
Nathan Scott77a7cce2006-01-11 15:35:57 +11002460 * group/project quotaoff or both.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461 */
2462 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2463 log->l_quotaoffs_flag |= XFS_DQ_USER;
Nathan Scott77a7cce2006-01-11 15:35:57 +11002464 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2465 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2467 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2468
2469 return (0);
2470}
2471
2472/*
2473 * Recover a dquot record
2474 */
2475STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002476xlog_recover_dquot_pass2(
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477 xlog_t *log,
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002478 xlog_recover_item_t *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002480 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481 xfs_buf_t *bp;
2482 struct xfs_disk_dquot *ddq, *recddq;
2483 int error;
2484 xfs_dq_logformat_t *dq_f;
2485 uint type;
2486
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487
2488 /*
2489 * Filesystems are required to send in quota flags at mount time.
2490 */
2491 if (mp->m_qflags == 0)
2492 return (0);
2493
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002494 recddq = item->ri_buf[1].i_addr;
2495 if (recddq == NULL) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002496 xfs_alert(log->l_mp, "NULL dquot in %s.", __func__);
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002497 return XFS_ERROR(EIO);
2498 }
Jan Rekorajski8ec6dba2009-11-16 11:57:02 +00002499 if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002500 xfs_alert(log->l_mp, "dquot too small (%d) in %s.",
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002501 item->ri_buf[1].i_len, __func__);
2502 return XFS_ERROR(EIO);
2503 }
2504
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505 /*
2506 * This type of quotas was turned off, so ignore this record.
2507 */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002508 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509 ASSERT(type);
2510 if (log->l_quotaoffs_flag & type)
2511 return (0);
2512
2513 /*
2514 * At this point we know that quota was _not_ turned off.
2515 * Since the mount flags are not indicating to us otherwise, this
2516 * must mean that quota is on, and the dquot needs to be replayed.
2517 * Remember that we may not have fully recovered the superblock yet,
2518 * so we can't do the usual trick of looking at the SB quota bits.
2519 *
2520 * The other possibility, of course, is that the quota subsystem was
2521 * removed since the last mount - ENOSYS.
2522 */
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002523 dq_f = item->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 ASSERT(dq_f);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002525 error = xfs_qm_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2526 "xlog_recover_dquot_pass2 (log copy)");
2527 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528 return XFS_ERROR(EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529 ASSERT(dq_f->qlf_len == 1);
2530
2531 error = xfs_read_buf(mp, mp->m_ddev_targp,
2532 dq_f->qlf_blkno,
2533 XFS_FSB_TO_BB(mp, dq_f->qlf_len),
2534 0, &bp);
2535 if (error) {
2536 xfs_ioerror_alert("xlog_recover_do..(read#3)", mp,
2537 bp, dq_f->qlf_blkno);
2538 return error;
2539 }
2540 ASSERT(bp);
2541 ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
2542
2543 /*
2544 * At least the magic num portion should be on disk because this
2545 * was among a chunk of dquots created earlier, and we did some
2546 * minimal initialization then.
2547 */
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002548 error = xfs_qm_dqcheck(mp, ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2549 "xlog_recover_dquot_pass2");
2550 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 xfs_buf_relse(bp);
2552 return XFS_ERROR(EIO);
2553 }
2554
2555 memcpy(ddq, recddq, item->ri_buf[1].i_len);
2556
2557 ASSERT(dq_f->qlf_size == 2);
Dave Chinnerebad8612010-09-22 10:47:20 +10002558 ASSERT(bp->b_target->bt_mount == mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2560 xfs_bdwrite(mp, bp);
2561
2562 return (0);
2563}
2564
2565/*
2566 * This routine is called to create an in-core extent free intent
2567 * item from the efi format structure which was logged on disk.
2568 * It allocates an in-core efi, copies the extents from the format
2569 * structure into it, and adds the efi to the AIL with the given
2570 * LSN.
2571 */
Tim Shimmin6d192a92006-06-09 14:55:38 +10002572STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002573xlog_recover_efi_pass2(
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574 xlog_t *log,
2575 xlog_recover_item_t *item,
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002576 xfs_lsn_t lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577{
Tim Shimmin6d192a92006-06-09 14:55:38 +10002578 int error;
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002579 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580 xfs_efi_log_item_t *efip;
2581 xfs_efi_log_format_t *efi_formatp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002583 efi_formatp = item->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002586 if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
2587 &(efip->efi_format)))) {
2588 xfs_efi_item_free(efip);
2589 return error;
2590 }
Dave Chinnerb199c8a2010-12-20 11:59:49 +11002591 atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592
David Chinnera9c21c12008-10-30 17:39:35 +11002593 spin_lock(&log->l_ailp->xa_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594 /*
David Chinner783a2f62008-10-30 17:39:58 +11002595 * xfs_trans_ail_update() drops the AIL lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596 */
Dave Chinnere6059942010-12-20 12:34:26 +11002597 xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002598 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599}
2600
2601
2602/*
2603 * This routine is called when an efd format structure is found in
2604 * a committed transaction in the log. It's purpose is to cancel
2605 * the corresponding efi if it was still in the log. To do this
2606 * it searches the AIL for the efi with an id equal to that in the
2607 * efd format structure. If we find it, we remove the efi from the
2608 * AIL and free it.
2609 */
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002610STATIC int
2611xlog_recover_efd_pass2(
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612 xlog_t *log,
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002613 xlog_recover_item_t *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002614{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615 xfs_efd_log_format_t *efd_formatp;
2616 xfs_efi_log_item_t *efip = NULL;
2617 xfs_log_item_t *lip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618 __uint64_t efi_id;
David Chinner27d8d5f2008-10-30 17:38:39 +11002619 struct xfs_ail_cursor cur;
David Chinner783a2f62008-10-30 17:39:58 +11002620 struct xfs_ail *ailp = log->l_ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002622 efd_formatp = item->ri_buf[0].i_addr;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002623 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
2624 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
2625 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
2626 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 efi_id = efd_formatp->efd_efi_id;
2628
2629 /*
2630 * Search for the efi with the id in the efd format structure
2631 * in the AIL.
2632 */
David Chinnera9c21c12008-10-30 17:39:35 +11002633 spin_lock(&ailp->xa_lock);
2634 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635 while (lip != NULL) {
2636 if (lip->li_type == XFS_LI_EFI) {
2637 efip = (xfs_efi_log_item_t *)lip;
2638 if (efip->efi_format.efi_id == efi_id) {
2639 /*
David Chinner783a2f62008-10-30 17:39:58 +11002640 * xfs_trans_ail_delete() drops the
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641 * AIL lock.
2642 */
David Chinner783a2f62008-10-30 17:39:58 +11002643 xfs_trans_ail_delete(ailp, lip);
David Chinner8ae2c0f2007-11-23 16:28:17 +11002644 xfs_efi_item_free(efip);
David Chinnera9c21c12008-10-30 17:39:35 +11002645 spin_lock(&ailp->xa_lock);
David Chinner27d8d5f2008-10-30 17:38:39 +11002646 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647 }
2648 }
David Chinnera9c21c12008-10-30 17:39:35 +11002649 lip = xfs_trans_ail_cursor_next(ailp, &cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650 }
David Chinnera9c21c12008-10-30 17:39:35 +11002651 xfs_trans_ail_cursor_done(ailp, &cur);
2652 spin_unlock(&ailp->xa_lock);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002653
2654 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655}
2656
2657/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658 * Free up any resources allocated by the transaction
2659 *
2660 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2661 */
2662STATIC void
2663xlog_recover_free_trans(
Christoph Hellwigd0450942010-12-01 22:06:23 +00002664 struct xlog_recover *trans)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665{
Dave Chinnerf0a76952010-01-11 11:49:57 +00002666 xlog_recover_item_t *item, *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667 int i;
2668
Dave Chinnerf0a76952010-01-11 11:49:57 +00002669 list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
2670 /* Free the regions in the item. */
2671 list_del(&item->ri_list);
2672 for (i = 0; i < item->ri_cnt; i++)
2673 kmem_free(item->ri_buf[i].i_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674 /* Free the item itself */
Dave Chinnerf0a76952010-01-11 11:49:57 +00002675 kmem_free(item->ri_buf);
2676 kmem_free(item);
2677 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678 /* Free the transaction recover structure */
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002679 kmem_free(trans);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680}
2681
2682STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002683xlog_recover_commit_pass1(
Christoph Hellwigd0450942010-12-01 22:06:23 +00002684 struct log *log,
2685 struct xlog_recover *trans,
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002686 xlog_recover_item_t *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002688 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1);
Christoph Hellwigd0450942010-12-01 22:06:23 +00002689
2690 switch (ITEM_TYPE(item)) {
2691 case XFS_LI_BUF:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002692 return xlog_recover_buffer_pass1(log, item);
Christoph Hellwigd0450942010-12-01 22:06:23 +00002693 case XFS_LI_QUOTAOFF:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002694 return xlog_recover_quotaoff_pass1(log, item);
2695 case XFS_LI_INODE:
2696 case XFS_LI_EFI:
2697 case XFS_LI_EFD:
2698 case XFS_LI_DQUOT:
2699 /* nothing to do in pass 1 */
2700 return 0;
Christoph Hellwigd0450942010-12-01 22:06:23 +00002701 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002702 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
2703 __func__, ITEM_TYPE(item));
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002704 ASSERT(0);
2705 return XFS_ERROR(EIO);
2706 }
2707}
2708
2709STATIC int
2710xlog_recover_commit_pass2(
2711 struct log *log,
2712 struct xlog_recover *trans,
2713 xlog_recover_item_t *item)
2714{
2715 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2);
2716
2717 switch (ITEM_TYPE(item)) {
2718 case XFS_LI_BUF:
2719 return xlog_recover_buffer_pass2(log, item);
2720 case XFS_LI_INODE:
2721 return xlog_recover_inode_pass2(log, item);
2722 case XFS_LI_EFI:
2723 return xlog_recover_efi_pass2(log, item, trans->r_lsn);
2724 case XFS_LI_EFD:
2725 return xlog_recover_efd_pass2(log, item);
2726 case XFS_LI_DQUOT:
2727 return xlog_recover_dquot_pass2(log, item);
2728 case XFS_LI_QUOTAOFF:
2729 /* nothing to do in pass2 */
2730 return 0;
2731 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002732 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
2733 __func__, ITEM_TYPE(item));
Christoph Hellwigd0450942010-12-01 22:06:23 +00002734 ASSERT(0);
2735 return XFS_ERROR(EIO);
2736 }
2737}
2738
2739/*
2740 * Perform the transaction.
2741 *
2742 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2743 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2744 */
2745STATIC int
2746xlog_recover_commit_trans(
2747 struct log *log,
2748 struct xlog_recover *trans,
2749 int pass)
2750{
2751 int error = 0;
2752 xlog_recover_item_t *item;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002753
Dave Chinnerf0a76952010-01-11 11:49:57 +00002754 hlist_del(&trans->r_list);
Christoph Hellwigd0450942010-12-01 22:06:23 +00002755
2756 error = xlog_recover_reorder_trans(log, trans, pass);
2757 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002758 return error;
Christoph Hellwigd0450942010-12-01 22:06:23 +00002759
2760 list_for_each_entry(item, &trans->r_itemq, ri_list) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002761 if (pass == XLOG_RECOVER_PASS1)
2762 error = xlog_recover_commit_pass1(log, trans, item);
2763 else
2764 error = xlog_recover_commit_pass2(log, trans, item);
Christoph Hellwigd0450942010-12-01 22:06:23 +00002765 if (error)
2766 return error;
2767 }
2768
2769 xlog_recover_free_trans(trans);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770 return 0;
2771}
2772
2773STATIC int
2774xlog_recover_unmount_trans(
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002775 struct log *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776 xlog_recover_t *trans)
2777{
2778 /* Do nothing now */
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002779 xfs_warn(log->l_mp, "%s: Unmount LR", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780 return 0;
2781}
2782
2783/*
2784 * There are two valid states of the r_state field. 0 indicates that the
2785 * transaction structure is in a normal state. We have either seen the
2786 * start of the transaction or the last operation we added was not a partial
2787 * operation. If the last operation we added to the transaction was a
2788 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2789 *
2790 * NOTE: skip LRs with 0 data length.
2791 */
2792STATIC int
2793xlog_recover_process_data(
2794 xlog_t *log,
Dave Chinnerf0a76952010-01-11 11:49:57 +00002795 struct hlist_head rhash[],
Linus Torvalds1da177e2005-04-16 15:20:36 -07002796 xlog_rec_header_t *rhead,
2797 xfs_caddr_t dp,
2798 int pass)
2799{
2800 xfs_caddr_t lp;
2801 int num_logops;
2802 xlog_op_header_t *ohead;
2803 xlog_recover_t *trans;
2804 xlog_tid_t tid;
2805 int error;
2806 unsigned long hash;
2807 uint flags;
2808
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002809 lp = dp + be32_to_cpu(rhead->h_len);
2810 num_logops = be32_to_cpu(rhead->h_num_logops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811
2812 /* check the log format matches our own - else we can't recover */
2813 if (xlog_header_check_recover(log->l_mp, rhead))
2814 return (XFS_ERROR(EIO));
2815
2816 while ((dp < lp) && num_logops) {
2817 ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
2818 ohead = (xlog_op_header_t *)dp;
2819 dp += sizeof(xlog_op_header_t);
2820 if (ohead->oh_clientid != XFS_TRANSACTION &&
2821 ohead->oh_clientid != XFS_LOG) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002822 xfs_warn(log->l_mp, "%s: bad clientid 0x%x",
2823 __func__, ohead->oh_clientid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002824 ASSERT(0);
2825 return (XFS_ERROR(EIO));
2826 }
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10002827 tid = be32_to_cpu(ohead->oh_tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828 hash = XLOG_RHASH(tid);
Dave Chinnerf0a76952010-01-11 11:49:57 +00002829 trans = xlog_recover_find_tid(&rhash[hash], tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830 if (trans == NULL) { /* not found; add new tid */
2831 if (ohead->oh_flags & XLOG_START_TRANS)
2832 xlog_recover_new_tid(&rhash[hash], tid,
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002833 be64_to_cpu(rhead->h_lsn));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834 } else {
Lachlan McIlroy9742bb92008-01-10 16:43:36 +11002835 if (dp + be32_to_cpu(ohead->oh_len) > lp) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002836 xfs_warn(log->l_mp, "%s: bad length 0x%x",
2837 __func__, be32_to_cpu(ohead->oh_len));
Lachlan McIlroy9742bb92008-01-10 16:43:36 +11002838 WARN_ON(1);
2839 return (XFS_ERROR(EIO));
2840 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841 flags = ohead->oh_flags & ~XLOG_END_TRANS;
2842 if (flags & XLOG_WAS_CONT_TRANS)
2843 flags &= ~XLOG_CONTINUE_TRANS;
2844 switch (flags) {
2845 case XLOG_COMMIT_TRANS:
2846 error = xlog_recover_commit_trans(log,
Dave Chinnerf0a76952010-01-11 11:49:57 +00002847 trans, pass);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002848 break;
2849 case XLOG_UNMOUNT_TRANS:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002850 error = xlog_recover_unmount_trans(log, trans);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851 break;
2852 case XLOG_WAS_CONT_TRANS:
Dave Chinner9abbc532010-04-13 15:06:46 +10002853 error = xlog_recover_add_to_cont_trans(log,
2854 trans, dp,
2855 be32_to_cpu(ohead->oh_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002856 break;
2857 case XLOG_START_TRANS:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002858 xfs_warn(log->l_mp, "%s: bad transaction",
2859 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002860 ASSERT(0);
2861 error = XFS_ERROR(EIO);
2862 break;
2863 case 0:
2864 case XLOG_CONTINUE_TRANS:
Dave Chinner9abbc532010-04-13 15:06:46 +10002865 error = xlog_recover_add_to_trans(log, trans,
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10002866 dp, be32_to_cpu(ohead->oh_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867 break;
2868 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002869 xfs_warn(log->l_mp, "%s: bad flag 0x%x",
2870 __func__, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002871 ASSERT(0);
2872 error = XFS_ERROR(EIO);
2873 break;
2874 }
2875 if (error)
2876 return error;
2877 }
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10002878 dp += be32_to_cpu(ohead->oh_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879 num_logops--;
2880 }
2881 return 0;
2882}
2883
2884/*
2885 * Process an extent free intent item that was recovered from
2886 * the log. We need to free the extents that it describes.
2887 */
David Chinner3c1e2bb2008-04-10 12:21:11 +10002888STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889xlog_recover_process_efi(
2890 xfs_mount_t *mp,
2891 xfs_efi_log_item_t *efip)
2892{
2893 xfs_efd_log_item_t *efdp;
2894 xfs_trans_t *tp;
2895 int i;
David Chinner3c1e2bb2008-04-10 12:21:11 +10002896 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002897 xfs_extent_t *extp;
2898 xfs_fsblock_t startblock_fsb;
2899
Dave Chinnerb199c8a2010-12-20 11:59:49 +11002900 ASSERT(!test_bit(XFS_EFI_RECOVERED, &efip->efi_flags));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002901
2902 /*
2903 * First check the validity of the extents described by the
2904 * EFI. If any are bad, then assume that all are bad and
2905 * just toss the EFI.
2906 */
2907 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
2908 extp = &(efip->efi_format.efi_extents[i]);
2909 startblock_fsb = XFS_BB_TO_FSB(mp,
2910 XFS_FSB_TO_DADDR(mp, extp->ext_start));
2911 if ((startblock_fsb == 0) ||
2912 (extp->ext_len == 0) ||
2913 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
2914 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
2915 /*
2916 * This will pull the EFI from the AIL and
2917 * free the memory associated with it.
2918 */
2919 xfs_efi_release(efip, efip->efi_format.efi_nextents);
David Chinner3c1e2bb2008-04-10 12:21:11 +10002920 return XFS_ERROR(EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002921 }
2922 }
2923
2924 tp = xfs_trans_alloc(mp, 0);
David Chinner3c1e2bb2008-04-10 12:21:11 +10002925 error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 0, 0);
David Chinnerfc6149d2008-04-10 12:21:53 +10002926 if (error)
2927 goto abort_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002928 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
2929
2930 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
2931 extp = &(efip->efi_format.efi_extents[i]);
David Chinnerfc6149d2008-04-10 12:21:53 +10002932 error = xfs_free_extent(tp, extp->ext_start, extp->ext_len);
2933 if (error)
2934 goto abort_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002935 xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
2936 extp->ext_len);
2937 }
2938
Dave Chinnerb199c8a2010-12-20 11:59:49 +11002939 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
David Chinnere5720ee2008-04-10 12:21:18 +10002940 error = xfs_trans_commit(tp, 0);
David Chinner3c1e2bb2008-04-10 12:21:11 +10002941 return error;
David Chinnerfc6149d2008-04-10 12:21:53 +10002942
2943abort_error:
2944 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
2945 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002946}
2947
2948/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949 * When this is called, all of the EFIs which did not have
2950 * corresponding EFDs should be in the AIL. What we do now
2951 * is free the extents associated with each one.
2952 *
2953 * Since we process the EFIs in normal transactions, they
2954 * will be removed at some point after the commit. This prevents
2955 * us from just walking down the list processing each one.
2956 * We'll use a flag in the EFI to skip those that we've already
2957 * processed and use the AIL iteration mechanism's generation
2958 * count to try to speed this up at least a bit.
2959 *
2960 * When we start, we know that the EFIs are the only things in
2961 * the AIL. As we process them, however, other items are added
2962 * to the AIL. Since everything added to the AIL must come after
2963 * everything already in the AIL, we stop processing as soon as
2964 * we see something other than an EFI in the AIL.
2965 */
David Chinner3c1e2bb2008-04-10 12:21:11 +10002966STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002967xlog_recover_process_efis(
2968 xlog_t *log)
2969{
2970 xfs_log_item_t *lip;
2971 xfs_efi_log_item_t *efip;
David Chinner3c1e2bb2008-04-10 12:21:11 +10002972 int error = 0;
David Chinner27d8d5f2008-10-30 17:38:39 +11002973 struct xfs_ail_cursor cur;
David Chinnera9c21c12008-10-30 17:39:35 +11002974 struct xfs_ail *ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975
David Chinnera9c21c12008-10-30 17:39:35 +11002976 ailp = log->l_ailp;
2977 spin_lock(&ailp->xa_lock);
2978 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002979 while (lip != NULL) {
2980 /*
2981 * We're done when we see something other than an EFI.
David Chinner27d8d5f2008-10-30 17:38:39 +11002982 * There should be no EFIs left in the AIL now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002983 */
2984 if (lip->li_type != XFS_LI_EFI) {
David Chinner27d8d5f2008-10-30 17:38:39 +11002985#ifdef DEBUG
David Chinnera9c21c12008-10-30 17:39:35 +11002986 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
David Chinner27d8d5f2008-10-30 17:38:39 +11002987 ASSERT(lip->li_type != XFS_LI_EFI);
2988#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989 break;
2990 }
2991
2992 /*
2993 * Skip EFIs that we've already processed.
2994 */
2995 efip = (xfs_efi_log_item_t *)lip;
Dave Chinnerb199c8a2010-12-20 11:59:49 +11002996 if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) {
David Chinnera9c21c12008-10-30 17:39:35 +11002997 lip = xfs_trans_ail_cursor_next(ailp, &cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998 continue;
2999 }
3000
David Chinnera9c21c12008-10-30 17:39:35 +11003001 spin_unlock(&ailp->xa_lock);
3002 error = xlog_recover_process_efi(log->l_mp, efip);
3003 spin_lock(&ailp->xa_lock);
David Chinner27d8d5f2008-10-30 17:38:39 +11003004 if (error)
3005 goto out;
David Chinnera9c21c12008-10-30 17:39:35 +11003006 lip = xfs_trans_ail_cursor_next(ailp, &cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003007 }
David Chinner27d8d5f2008-10-30 17:38:39 +11003008out:
David Chinnera9c21c12008-10-30 17:39:35 +11003009 xfs_trans_ail_cursor_done(ailp, &cur);
3010 spin_unlock(&ailp->xa_lock);
David Chinner3c1e2bb2008-04-10 12:21:11 +10003011 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003012}
3013
3014/*
3015 * This routine performs a transaction to null out a bad inode pointer
3016 * in an agi unlinked inode hash bucket.
3017 */
3018STATIC void
3019xlog_recover_clear_agi_bucket(
3020 xfs_mount_t *mp,
3021 xfs_agnumber_t agno,
3022 int bucket)
3023{
3024 xfs_trans_t *tp;
3025 xfs_agi_t *agi;
3026 xfs_buf_t *agibp;
3027 int offset;
3028 int error;
3029
3030 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003031 error = xfs_trans_reserve(tp, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp),
3032 0, 0, 0);
David Chinnere5720ee2008-04-10 12:21:18 +10003033 if (error)
3034 goto out_abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003035
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003036 error = xfs_read_agi(mp, tp, agno, &agibp);
3037 if (error)
David Chinnere5720ee2008-04-10 12:21:18 +10003038 goto out_abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003039
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003040 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11003041 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003042 offset = offsetof(xfs_agi_t, agi_unlinked) +
3043 (sizeof(xfs_agino_t) * bucket);
3044 xfs_trans_log_buf(tp, agibp, offset,
3045 (offset + sizeof(xfs_agino_t) - 1));
3046
David Chinnere5720ee2008-04-10 12:21:18 +10003047 error = xfs_trans_commit(tp, 0);
3048 if (error)
3049 goto out_error;
3050 return;
3051
3052out_abort:
3053 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3054out_error:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003055 xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno);
David Chinnere5720ee2008-04-10 12:21:18 +10003056 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003057}
3058
Christoph Hellwig23fac502008-11-28 14:23:40 +11003059STATIC xfs_agino_t
3060xlog_recover_process_one_iunlink(
3061 struct xfs_mount *mp,
3062 xfs_agnumber_t agno,
3063 xfs_agino_t agino,
3064 int bucket)
3065{
3066 struct xfs_buf *ibp;
3067 struct xfs_dinode *dip;
3068 struct xfs_inode *ip;
3069 xfs_ino_t ino;
3070 int error;
3071
3072 ino = XFS_AGINO_TO_INO(mp, agno, agino);
Dave Chinner7b6259e2010-06-24 11:35:17 +10003073 error = xfs_iget(mp, NULL, ino, 0, 0, &ip);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003074 if (error)
3075 goto fail;
3076
3077 /*
3078 * Get the on disk inode to find the next inode in the bucket.
3079 */
Christoph Hellwig0cadda12010-01-19 09:56:44 +00003080 error = xfs_itobp(mp, NULL, ip, &dip, &ibp, XBF_LOCK);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003081 if (error)
Christoph Hellwig0e446672008-11-28 14:23:42 +11003082 goto fail_iput;
Christoph Hellwig23fac502008-11-28 14:23:40 +11003083
Christoph Hellwig23fac502008-11-28 14:23:40 +11003084 ASSERT(ip->i_d.di_nlink == 0);
Christoph Hellwig0e446672008-11-28 14:23:42 +11003085 ASSERT(ip->i_d.di_mode != 0);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003086
3087 /* setup for the next pass */
3088 agino = be32_to_cpu(dip->di_next_unlinked);
3089 xfs_buf_relse(ibp);
3090
3091 /*
3092 * Prevent any DMAPI event from being sent when the reference on
3093 * the inode is dropped.
3094 */
3095 ip->i_d.di_dmevmask = 0;
3096
Christoph Hellwig0e446672008-11-28 14:23:42 +11003097 IRELE(ip);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003098 return agino;
3099
Christoph Hellwig0e446672008-11-28 14:23:42 +11003100 fail_iput:
3101 IRELE(ip);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003102 fail:
3103 /*
3104 * We can't read in the inode this bucket points to, or this inode
3105 * is messed up. Just ditch this bucket of inodes. We will lose
3106 * some inodes and space, but at least we won't hang.
3107 *
3108 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3109 * clear the inode pointer in the bucket.
3110 */
3111 xlog_recover_clear_agi_bucket(mp, agno, bucket);
3112 return NULLAGINO;
3113}
3114
Linus Torvalds1da177e2005-04-16 15:20:36 -07003115/*
3116 * xlog_iunlink_recover
3117 *
3118 * This is called during recovery to process any inodes which
3119 * we unlinked but not freed when the system crashed. These
3120 * inodes will be on the lists in the AGI blocks. What we do
3121 * here is scan all the AGIs and fully truncate and free any
3122 * inodes found on the lists. Each inode is removed from the
3123 * lists when it has been fully truncated and is freed. The
3124 * freeing of the inode and its removal from the list must be
3125 * atomic.
3126 */
Eric Sandeend96f8f82009-07-02 00:09:33 -05003127STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07003128xlog_recover_process_iunlinks(
3129 xlog_t *log)
3130{
3131 xfs_mount_t *mp;
3132 xfs_agnumber_t agno;
3133 xfs_agi_t *agi;
3134 xfs_buf_t *agibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003135 xfs_agino_t agino;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003136 int bucket;
3137 int error;
3138 uint mp_dmevmask;
3139
3140 mp = log->l_mp;
3141
3142 /*
3143 * Prevent any DMAPI event from being sent while in this function.
3144 */
3145 mp_dmevmask = mp->m_dmevmask;
3146 mp->m_dmevmask = 0;
3147
3148 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3149 /*
3150 * Find the agi for this ag.
3151 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003152 error = xfs_read_agi(mp, NULL, agno, &agibp);
3153 if (error) {
3154 /*
3155 * AGI is b0rked. Don't process it.
3156 *
3157 * We should probably mark the filesystem as corrupt
3158 * after we've recovered all the ag's we can....
3159 */
3160 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003161 }
3162 agi = XFS_BUF_TO_AGI(agibp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003163
3164 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
Christoph Hellwig16259e72005-11-02 15:11:25 +11003165 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003166 while (agino != NULLAGINO) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003167 /*
3168 * Release the agi buffer so that it can
3169 * be acquired in the normal course of the
3170 * transaction to truncate and free the inode.
3171 */
3172 xfs_buf_relse(agibp);
3173
Christoph Hellwig23fac502008-11-28 14:23:40 +11003174 agino = xlog_recover_process_one_iunlink(mp,
3175 agno, agino, bucket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003176
3177 /*
3178 * Reacquire the agibuffer and continue around
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003179 * the loop. This should never fail as we know
3180 * the buffer was good earlier on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003181 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003182 error = xfs_read_agi(mp, NULL, agno, &agibp);
3183 ASSERT(error == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003184 agi = XFS_BUF_TO_AGI(agibp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003185 }
3186 }
3187
3188 /*
3189 * Release the buffer for the current agi so we can
3190 * go on to the next one.
3191 */
3192 xfs_buf_relse(agibp);
3193 }
3194
3195 mp->m_dmevmask = mp_dmevmask;
3196}
3197
3198
3199#ifdef DEBUG
3200STATIC void
3201xlog_pack_data_checksum(
3202 xlog_t *log,
3203 xlog_in_core_t *iclog,
3204 int size)
3205{
3206 int i;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003207 __be32 *up;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003208 uint chksum = 0;
3209
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003210 up = (__be32 *)iclog->ic_datap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003211 /* divide length by 4 to get # words */
3212 for (i = 0; i < (size >> 2); i++) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003213 chksum ^= be32_to_cpu(*up);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003214 up++;
3215 }
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003216 iclog->ic_header.h_chksum = cpu_to_be32(chksum);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003217}
3218#else
3219#define xlog_pack_data_checksum(log, iclog, size)
3220#endif
3221
3222/*
3223 * Stamp cycle number in every block
3224 */
3225void
3226xlog_pack_data(
3227 xlog_t *log,
3228 xlog_in_core_t *iclog,
3229 int roundoff)
3230{
3231 int i, j, k;
3232 int size = iclog->ic_offset + roundoff;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003233 __be32 cycle_lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003234 xfs_caddr_t dp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003235
3236 xlog_pack_data_checksum(log, iclog, size);
3237
3238 cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn);
3239
3240 dp = iclog->ic_datap;
3241 for (i = 0; i < BTOBB(size) &&
3242 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003243 iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp;
3244 *(__be32 *)dp = cycle_lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003245 dp += BBSIZE;
3246 }
3247
Eric Sandeen62118702008-03-06 13:44:28 +11003248 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Christoph Hellwigb28708d2008-11-28 14:23:38 +11003249 xlog_in_core_2_t *xhdr = iclog->ic_data;
3250
Linus Torvalds1da177e2005-04-16 15:20:36 -07003251 for ( ; i < BTOBB(size); i++) {
3252 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3253 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003254 xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp;
3255 *(__be32 *)dp = cycle_lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003256 dp += BBSIZE;
3257 }
3258
3259 for (i = 1; i < log->l_iclog_heads; i++) {
3260 xhdr[i].hic_xheader.xh_cycle = cycle_lsn;
3261 }
3262 }
3263}
3264
Linus Torvalds1da177e2005-04-16 15:20:36 -07003265STATIC void
3266xlog_unpack_data(
3267 xlog_rec_header_t *rhead,
3268 xfs_caddr_t dp,
3269 xlog_t *log)
3270{
3271 int i, j, k;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003272
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003273 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07003274 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003275 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003276 dp += BBSIZE;
3277 }
3278
Eric Sandeen62118702008-03-06 13:44:28 +11003279 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Christoph Hellwigb28708d2008-11-28 14:23:38 +11003280 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003281 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003282 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3283 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003284 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003285 dp += BBSIZE;
3286 }
3287 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003288}
3289
3290STATIC int
3291xlog_valid_rec_header(
3292 xlog_t *log,
3293 xlog_rec_header_t *rhead,
3294 xfs_daddr_t blkno)
3295{
3296 int hlen;
3297
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003298 if (unlikely(be32_to_cpu(rhead->h_magicno) != XLOG_HEADER_MAGIC_NUM)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003299 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3300 XFS_ERRLEVEL_LOW, log->l_mp);
3301 return XFS_ERROR(EFSCORRUPTED);
3302 }
3303 if (unlikely(
3304 (!rhead->h_version ||
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003305 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003306 xfs_warn(log->l_mp, "%s: unrecognised log version (%d).",
Harvey Harrison34a622b2008-04-10 12:19:21 +10003307 __func__, be32_to_cpu(rhead->h_version));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003308 return XFS_ERROR(EIO);
3309 }
3310
3311 /* LR body must have data or it wouldn't have been written */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003312 hlen = be32_to_cpu(rhead->h_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003313 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
3314 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3315 XFS_ERRLEVEL_LOW, log->l_mp);
3316 return XFS_ERROR(EFSCORRUPTED);
3317 }
3318 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
3319 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3320 XFS_ERRLEVEL_LOW, log->l_mp);
3321 return XFS_ERROR(EFSCORRUPTED);
3322 }
3323 return 0;
3324}
3325
3326/*
3327 * Read the log from tail to head and process the log records found.
3328 * Handle the two cases where the tail and head are in the same cycle
3329 * and where the active portion of the log wraps around the end of
3330 * the physical log separately. The pass parameter is passed through
3331 * to the routines called to process the data and is not looked at
3332 * here.
3333 */
3334STATIC int
3335xlog_do_recovery_pass(
3336 xlog_t *log,
3337 xfs_daddr_t head_blk,
3338 xfs_daddr_t tail_blk,
3339 int pass)
3340{
3341 xlog_rec_header_t *rhead;
3342 xfs_daddr_t blk_no;
Andy Polingfc5bc4c2009-11-03 17:26:47 +00003343 xfs_caddr_t offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003344 xfs_buf_t *hbp, *dbp;
3345 int error = 0, h_size;
3346 int bblks, split_bblks;
3347 int hblks, split_hblks, wrapped_hblks;
Dave Chinnerf0a76952010-01-11 11:49:57 +00003348 struct hlist_head rhash[XLOG_RHASH_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003349
3350 ASSERT(head_blk != tail_blk);
3351
3352 /*
3353 * Read the header of the tail block and get the iclog buffer size from
3354 * h_size. Use this to tell how many sectors make up the log header.
3355 */
Eric Sandeen62118702008-03-06 13:44:28 +11003356 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003357 /*
3358 * When using variable length iclogs, read first sector of
3359 * iclog header and extract the header size from it. Get a
3360 * new hbp that is the correct size.
3361 */
3362 hbp = xlog_get_bp(log, 1);
3363 if (!hbp)
3364 return ENOMEM;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003365
3366 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
3367 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003368 goto bread_err1;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003369
Linus Torvalds1da177e2005-04-16 15:20:36 -07003370 rhead = (xlog_rec_header_t *)offset;
3371 error = xlog_valid_rec_header(log, rhead, tail_blk);
3372 if (error)
3373 goto bread_err1;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003374 h_size = be32_to_cpu(rhead->h_size);
3375 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07003376 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
3377 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
3378 if (h_size % XLOG_HEADER_CYCLE_SIZE)
3379 hblks++;
3380 xlog_put_bp(hbp);
3381 hbp = xlog_get_bp(log, hblks);
3382 } else {
3383 hblks = 1;
3384 }
3385 } else {
Alex Elder69ce58f2010-04-20 17:09:59 +10003386 ASSERT(log->l_sectBBsize == 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003387 hblks = 1;
3388 hbp = xlog_get_bp(log, 1);
3389 h_size = XLOG_BIG_RECORD_BSIZE;
3390 }
3391
3392 if (!hbp)
3393 return ENOMEM;
3394 dbp = xlog_get_bp(log, BTOBB(h_size));
3395 if (!dbp) {
3396 xlog_put_bp(hbp);
3397 return ENOMEM;
3398 }
3399
3400 memset(rhash, 0, sizeof(rhash));
3401 if (tail_blk <= head_blk) {
3402 for (blk_no = tail_blk; blk_no < head_blk; ) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003403 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
3404 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003405 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003406
Linus Torvalds1da177e2005-04-16 15:20:36 -07003407 rhead = (xlog_rec_header_t *)offset;
3408 error = xlog_valid_rec_header(log, rhead, blk_no);
3409 if (error)
3410 goto bread_err2;
3411
3412 /* blocks in data section */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003413 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003414 error = xlog_bread(log, blk_no + hblks, bblks, dbp,
3415 &offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003416 if (error)
3417 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003418
Linus Torvalds1da177e2005-04-16 15:20:36 -07003419 xlog_unpack_data(rhead, offset, log);
3420 if ((error = xlog_recover_process_data(log,
3421 rhash, rhead, offset, pass)))
3422 goto bread_err2;
3423 blk_no += bblks + hblks;
3424 }
3425 } else {
3426 /*
3427 * Perform recovery around the end of the physical log.
3428 * When the head is not on the same cycle number as the tail,
3429 * we can't do a sequential recovery as above.
3430 */
3431 blk_no = tail_blk;
3432 while (blk_no < log->l_logBBsize) {
3433 /*
3434 * Check for header wrapping around physical end-of-log
3435 */
Andy Polingfc5bc4c2009-11-03 17:26:47 +00003436 offset = XFS_BUF_PTR(hbp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003437 split_hblks = 0;
3438 wrapped_hblks = 0;
3439 if (blk_no + hblks <= log->l_logBBsize) {
3440 /* Read header in one read */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003441 error = xlog_bread(log, blk_no, hblks, hbp,
3442 &offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003443 if (error)
3444 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003445 } else {
3446 /* This LR is split across physical log end */
3447 if (blk_no != log->l_logBBsize) {
3448 /* some data before physical log end */
3449 ASSERT(blk_no <= INT_MAX);
3450 split_hblks = log->l_logBBsize - (int)blk_no;
3451 ASSERT(split_hblks > 0);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003452 error = xlog_bread(log, blk_no,
3453 split_hblks, hbp,
3454 &offset);
3455 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003456 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003457 }
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003458
Linus Torvalds1da177e2005-04-16 15:20:36 -07003459 /*
3460 * Note: this black magic still works with
3461 * large sector sizes (non-512) only because:
3462 * - we increased the buffer size originally
3463 * by 1 sector giving us enough extra space
3464 * for the second read;
3465 * - the log start is guaranteed to be sector
3466 * aligned;
3467 * - we read the log end (LR header start)
3468 * _first_, then the log start (LR header end)
3469 * - order is important.
3470 */
David Chinner234f56a2008-04-10 12:24:24 +10003471 wrapped_hblks = hblks - split_hblks;
Dave Chinner44396472011-04-21 09:34:27 +00003472 error = xlog_bread_offset(log, 0,
3473 wrapped_hblks, hbp,
3474 offset + BBTOB(split_hblks));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003475 if (error)
3476 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003477 }
3478 rhead = (xlog_rec_header_t *)offset;
3479 error = xlog_valid_rec_header(log, rhead,
3480 split_hblks ? blk_no : 0);
3481 if (error)
3482 goto bread_err2;
3483
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003484 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003485 blk_no += hblks;
3486
3487 /* Read in data for log record */
3488 if (blk_no + bblks <= log->l_logBBsize) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003489 error = xlog_bread(log, blk_no, bblks, dbp,
3490 &offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003491 if (error)
3492 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003493 } else {
3494 /* This log record is split across the
3495 * physical end of log */
Andy Polingfc5bc4c2009-11-03 17:26:47 +00003496 offset = XFS_BUF_PTR(dbp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003497 split_bblks = 0;
3498 if (blk_no != log->l_logBBsize) {
3499 /* some data is before the physical
3500 * end of log */
3501 ASSERT(!wrapped_hblks);
3502 ASSERT(blk_no <= INT_MAX);
3503 split_bblks =
3504 log->l_logBBsize - (int)blk_no;
3505 ASSERT(split_bblks > 0);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003506 error = xlog_bread(log, blk_no,
3507 split_bblks, dbp,
3508 &offset);
3509 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003510 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003511 }
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003512
Linus Torvalds1da177e2005-04-16 15:20:36 -07003513 /*
3514 * Note: this black magic still works with
3515 * large sector sizes (non-512) only because:
3516 * - we increased the buffer size originally
3517 * by 1 sector giving us enough extra space
3518 * for the second read;
3519 * - the log start is guaranteed to be sector
3520 * aligned;
3521 * - we read the log end (LR header start)
3522 * _first_, then the log start (LR header end)
3523 * - order is important.
3524 */
Dave Chinner44396472011-04-21 09:34:27 +00003525 error = xlog_bread_offset(log, 0,
3526 bblks - split_bblks, hbp,
3527 offset + BBTOB(split_bblks));
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003528 if (error)
3529 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003530 }
3531 xlog_unpack_data(rhead, offset, log);
3532 if ((error = xlog_recover_process_data(log, rhash,
3533 rhead, offset, pass)))
3534 goto bread_err2;
3535 blk_no += bblks;
3536 }
3537
3538 ASSERT(blk_no >= log->l_logBBsize);
3539 blk_no -= log->l_logBBsize;
3540
3541 /* read first part of physical log */
3542 while (blk_no < head_blk) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003543 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
3544 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003545 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003546
Linus Torvalds1da177e2005-04-16 15:20:36 -07003547 rhead = (xlog_rec_header_t *)offset;
3548 error = xlog_valid_rec_header(log, rhead, blk_no);
3549 if (error)
3550 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003551
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003552 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003553 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
3554 &offset);
3555 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003556 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003557
Linus Torvalds1da177e2005-04-16 15:20:36 -07003558 xlog_unpack_data(rhead, offset, log);
3559 if ((error = xlog_recover_process_data(log, rhash,
3560 rhead, offset, pass)))
3561 goto bread_err2;
3562 blk_no += bblks + hblks;
3563 }
3564 }
3565
3566 bread_err2:
3567 xlog_put_bp(dbp);
3568 bread_err1:
3569 xlog_put_bp(hbp);
3570 return error;
3571}
3572
3573/*
3574 * Do the recovery of the log. We actually do this in two phases.
3575 * The two passes are necessary in order to implement the function
3576 * of cancelling a record written into the log. The first pass
3577 * determines those things which have been cancelled, and the
3578 * second pass replays log items normally except for those which
3579 * have been cancelled. The handling of the replay and cancellations
3580 * takes place in the log item type specific routines.
3581 *
3582 * The table of items which have cancel records in the log is allocated
3583 * and freed at this level, since only here do we know when all of
3584 * the log recovery has been completed.
3585 */
3586STATIC int
3587xlog_do_log_recovery(
3588 xlog_t *log,
3589 xfs_daddr_t head_blk,
3590 xfs_daddr_t tail_blk)
3591{
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00003592 int error, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003593
3594 ASSERT(head_blk != tail_blk);
3595
3596 /*
3597 * First do a pass to find all of the cancelled buf log items.
3598 * Store them in the buf_cancel_table for use in the second pass.
3599 */
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00003600 log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE *
3601 sizeof(struct list_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003602 KM_SLEEP);
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00003603 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
3604 INIT_LIST_HEAD(&log->l_buf_cancel_table[i]);
3605
Linus Torvalds1da177e2005-04-16 15:20:36 -07003606 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3607 XLOG_RECOVER_PASS1);
3608 if (error != 0) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003609 kmem_free(log->l_buf_cancel_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003610 log->l_buf_cancel_table = NULL;
3611 return error;
3612 }
3613 /*
3614 * Then do a second pass to actually recover the items in the log.
3615 * When it is complete free the table of buf cancel items.
3616 */
3617 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3618 XLOG_RECOVER_PASS2);
3619#ifdef DEBUG
Tim Shimmin6d192a92006-06-09 14:55:38 +10003620 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003621 int i;
3622
3623 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00003624 ASSERT(list_empty(&log->l_buf_cancel_table[i]));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003625 }
3626#endif /* DEBUG */
3627
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003628 kmem_free(log->l_buf_cancel_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003629 log->l_buf_cancel_table = NULL;
3630
3631 return error;
3632}
3633
3634/*
3635 * Do the actual recovery
3636 */
3637STATIC int
3638xlog_do_recover(
3639 xlog_t *log,
3640 xfs_daddr_t head_blk,
3641 xfs_daddr_t tail_blk)
3642{
3643 int error;
3644 xfs_buf_t *bp;
3645 xfs_sb_t *sbp;
3646
3647 /*
3648 * First replay the images in the log.
3649 */
3650 error = xlog_do_log_recovery(log, head_blk, tail_blk);
3651 if (error) {
3652 return error;
3653 }
3654
3655 XFS_bflush(log->l_mp->m_ddev_targp);
3656
3657 /*
3658 * If IO errors happened during recovery, bail out.
3659 */
3660 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
3661 return (EIO);
3662 }
3663
3664 /*
3665 * We now update the tail_lsn since much of the recovery has completed
3666 * and there may be space available to use. If there were no extent
3667 * or iunlinks, we can free up the entire log and set the tail_lsn to
3668 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3669 * lsn of the last known good LR on disk. If there are extent frees
3670 * or iunlinks they will have some entries in the AIL; so we look at
3671 * the AIL to determine how to set the tail_lsn.
3672 */
3673 xlog_assign_tail_lsn(log->l_mp);
3674
3675 /*
3676 * Now that we've finished replaying all buffer and inode
3677 * updates, re-read in the superblock.
3678 */
3679 bp = xfs_getsb(log->l_mp, 0);
3680 XFS_BUF_UNDONE(bp);
Lachlan McIlroybebf9632007-10-15 13:18:02 +10003681 ASSERT(!(XFS_BUF_ISWRITE(bp)));
3682 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003683 XFS_BUF_READ(bp);
Lachlan McIlroybebf9632007-10-15 13:18:02 +10003684 XFS_BUF_UNASYNC(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003685 xfsbdstrat(log->l_mp, bp);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +00003686 error = xfs_buf_iowait(bp);
David Chinnerd64e31a2008-04-10 12:22:17 +10003687 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003688 xfs_ioerror_alert("xlog_do_recover",
3689 log->l_mp, bp, XFS_BUF_ADDR(bp));
3690 ASSERT(0);
3691 xfs_buf_relse(bp);
3692 return error;
3693 }
3694
3695 /* Convert superblock from on-disk format */
3696 sbp = &log->l_mp->m_sb;
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +10003697 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003698 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
Eric Sandeen62118702008-03-06 13:44:28 +11003699 ASSERT(xfs_sb_good_version(sbp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003700 xfs_buf_relse(bp);
3701
Lachlan McIlroy5478eea2007-02-10 18:36:29 +11003702 /* We've re-read the superblock so re-initialize per-cpu counters */
3703 xfs_icsb_reinit_counters(log->l_mp);
3704
Linus Torvalds1da177e2005-04-16 15:20:36 -07003705 xlog_recover_check_summary(log);
3706
3707 /* Normal transactions can now occur */
3708 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
3709 return 0;
3710}
3711
3712/*
3713 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3714 *
3715 * Return error or zero.
3716 */
3717int
3718xlog_recover(
Eric Sandeen65be6052006-01-11 15:34:19 +11003719 xlog_t *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003720{
3721 xfs_daddr_t head_blk, tail_blk;
3722 int error;
3723
3724 /* find the tail of the log */
Eric Sandeen65be6052006-01-11 15:34:19 +11003725 if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003726 return error;
3727
3728 if (tail_blk != head_blk) {
3729 /* There used to be a comment here:
3730 *
3731 * disallow recovery on read-only mounts. note -- mount
3732 * checks for ENOSPC and turns it into an intelligent
3733 * error message.
3734 * ...but this is no longer true. Now, unless you specify
3735 * NORECOVERY (in which case this function would never be
3736 * called), we just go ahead and recover. We do this all
3737 * under the vfs layer, so we can get away with it unless
3738 * the device itself is read-only, in which case we fail.
3739 */
Utako Kusaka3a02ee12007-05-08 13:50:06 +10003740 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003741 return error;
3742 }
3743
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003744 xfs_notice(log->l_mp, "Starting recovery (logdev: %s)",
3745 log->l_mp->m_logname ? log->l_mp->m_logname
3746 : "internal");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003747
3748 error = xlog_do_recover(log, head_blk, tail_blk);
3749 log->l_flags |= XLOG_RECOVERY_NEEDED;
3750 }
3751 return error;
3752}
3753
3754/*
3755 * In the first part of recovery we replay inodes and buffers and build
3756 * up the list of extent free items which need to be processed. Here
3757 * we process the extent free items and clean up the on disk unlinked
3758 * inode lists. This is separated from the first part of recovery so
3759 * that the root and real-time bitmap inodes can be read in from disk in
3760 * between the two stages. This is necessary so that we can free space
3761 * in the real-time portion of the file system.
3762 */
3763int
3764xlog_recover_finish(
Christoph Hellwig42490232008-08-13 16:49:32 +10003765 xlog_t *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003766{
3767 /*
3768 * Now we're ready to do the transactions needed for the
3769 * rest of recovery. Start with completing all the extent
3770 * free intent records and then process the unlinked inode
3771 * lists. At this point, we essentially run in normal mode
3772 * except that we're still performing recovery actions
3773 * rather than accepting new requests.
3774 */
3775 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
David Chinner3c1e2bb2008-04-10 12:21:11 +10003776 int error;
3777 error = xlog_recover_process_efis(log);
3778 if (error) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003779 xfs_alert(log->l_mp, "Failed to recover EFIs");
David Chinner3c1e2bb2008-04-10 12:21:11 +10003780 return error;
3781 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003782 /*
3783 * Sync the log to get all the EFIs out of the AIL.
3784 * This isn't absolutely necessary, but it helps in
3785 * case the unlink transactions would have problems
3786 * pushing the EFIs out of the way.
3787 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003788 xfs_log_force(log->l_mp, XFS_LOG_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003789
Christoph Hellwig42490232008-08-13 16:49:32 +10003790 xlog_recover_process_iunlinks(log);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003791
3792 xlog_recover_check_summary(log);
3793
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003794 xfs_notice(log->l_mp, "Ending recovery (logdev: %s)",
3795 log->l_mp->m_logname ? log->l_mp->m_logname
3796 : "internal");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003797 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
3798 } else {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003799 xfs_info(log->l_mp, "Ending clean mount");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003800 }
3801 return 0;
3802}
3803
3804
3805#if defined(DEBUG)
3806/*
3807 * Read all of the agf and agi counters and check that they
3808 * are consistent with the superblock counters.
3809 */
3810void
3811xlog_recover_check_summary(
3812 xlog_t *log)
3813{
3814 xfs_mount_t *mp;
3815 xfs_agf_t *agfp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003816 xfs_buf_t *agfbp;
3817 xfs_buf_t *agibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003818 xfs_agnumber_t agno;
3819 __uint64_t freeblks;
3820 __uint64_t itotal;
3821 __uint64_t ifree;
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003822 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003823
3824 mp = log->l_mp;
3825
3826 freeblks = 0LL;
3827 itotal = 0LL;
3828 ifree = 0LL;
3829 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
From: Christoph Hellwig48056212008-11-28 14:23:38 +11003830 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
3831 if (error) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003832 xfs_alert(mp, "%s agf read failed agno %d error %d",
3833 __func__, agno, error);
From: Christoph Hellwig48056212008-11-28 14:23:38 +11003834 } else {
3835 agfp = XFS_BUF_TO_AGF(agfbp);
3836 freeblks += be32_to_cpu(agfp->agf_freeblks) +
3837 be32_to_cpu(agfp->agf_flcount);
3838 xfs_buf_relse(agfbp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003839 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003840
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003841 error = xfs_read_agi(mp, NULL, agno, &agibp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003842 if (error) {
3843 xfs_alert(mp, "%s agi read failed agno %d error %d",
3844 __func__, agno, error);
3845 } else {
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003846 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003847
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003848 itotal += be32_to_cpu(agi->agi_count);
3849 ifree += be32_to_cpu(agi->agi_freecount);
3850 xfs_buf_relse(agibp);
3851 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003852 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003853}
3854#endif /* DEBUG */