blob: ceeba45e0224b1ab385266f53a9c8cb5cae64785 [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_dir2.h"
28#include "xfs_dmapi.h"
29#include "xfs_mount.h"
30#include "xfs_error.h"
31#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110032#include "xfs_alloc_btree.h"
33#include "xfs_ialloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include "xfs_dir2_sf.h"
Nathan Scotta844f452005-11-02 14:38:42 +110035#include "xfs_attr_sf.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include "xfs_dinode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include "xfs_inode.h"
Nathan Scotta844f452005-11-02 14:38:42 +110038#include "xfs_inode_item.h"
Nathan Scotta844f452005-11-02 14:38:42 +110039#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include "xfs_ialloc.h"
41#include "xfs_log_priv.h"
42#include "xfs_buf_item.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include "xfs_log_recover.h"
44#include "xfs_extfree_item.h"
45#include "xfs_trans_priv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070046#include "xfs_quota.h"
47#include "xfs_rw.h"
Christoph Hellwig43355092008-03-27 18:01:08 +110048#include "xfs_utils.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
50STATIC int xlog_find_zeroed(xlog_t *, xfs_daddr_t *);
51STATIC int xlog_clear_stale_blocks(xlog_t *, xfs_lsn_t);
52STATIC void xlog_recover_insert_item_backq(xlog_recover_item_t **q,
53 xlog_recover_item_t *item);
54#if defined(DEBUG)
55STATIC void xlog_recover_check_summary(xlog_t *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070056#else
57#define xlog_recover_check_summary(log)
Linus Torvalds1da177e2005-04-16 15:20:36 -070058#endif
59
60
61/*
62 * Sector aligned buffer routines for buffer create/read/write/access
63 */
64
65#define XLOG_SECTOR_ROUNDUP_BBCOUNT(log, bbs) \
66 ( ((log)->l_sectbb_mask && (bbs & (log)->l_sectbb_mask)) ? \
67 ((bbs + (log)->l_sectbb_mask + 1) & ~(log)->l_sectbb_mask) : (bbs) )
68#define XLOG_SECTOR_ROUNDDOWN_BLKNO(log, bno) ((bno) & ~(log)->l_sectbb_mask)
69
70xfs_buf_t *
71xlog_get_bp(
72 xlog_t *log,
Dave Chinner32281492009-01-22 15:37:47 +110073 int nbblks)
Linus Torvalds1da177e2005-04-16 15:20:36 -070074{
Dave Chinner32281492009-01-22 15:37:47 +110075 if (nbblks <= 0 || nbblks > log->l_logBBsize) {
76 xlog_warn("XFS: Invalid block length (0x%x) given for buffer", nbblks);
77 XFS_ERROR_REPORT("xlog_get_bp(1)",
78 XFS_ERRLEVEL_HIGH, log->l_mp);
79 return NULL;
80 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070081
82 if (log->l_sectbb_log) {
Dave Chinner32281492009-01-22 15:37:47 +110083 if (nbblks > 1)
84 nbblks += XLOG_SECTOR_ROUNDUP_BBCOUNT(log, 1);
85 nbblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, nbblks);
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 }
Dave Chinner32281492009-01-22 15:37:47 +110087 return xfs_buf_get_noaddr(BBTOB(nbblks), log->l_mp->m_logdev_targp);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088}
89
90void
91xlog_put_bp(
92 xfs_buf_t *bp)
93{
94 xfs_buf_free(bp);
95}
96
97
98/*
99 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
100 */
101int
102xlog_bread(
103 xlog_t *log,
104 xfs_daddr_t blk_no,
105 int nbblks,
106 xfs_buf_t *bp)
107{
108 int error;
109
Dave Chinner32281492009-01-22 15:37:47 +1100110 if (nbblks <= 0 || nbblks > log->l_logBBsize) {
111 xlog_warn("XFS: Invalid block length (0x%x) given for buffer", nbblks);
112 XFS_ERROR_REPORT("xlog_bread(1)",
113 XFS_ERRLEVEL_HIGH, log->l_mp);
114 return EFSCORRUPTED;
115 }
116
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 if (log->l_sectbb_log) {
118 blk_no = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, blk_no);
119 nbblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, nbblks);
120 }
121
122 ASSERT(nbblks > 0);
123 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
124 ASSERT(bp);
125
126 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
127 XFS_BUF_READ(bp);
128 XFS_BUF_BUSY(bp);
129 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
130 XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
131
132 xfsbdstrat(log->l_mp, bp);
David Chinnerd64e31a2008-04-10 12:22:17 +1000133 error = xfs_iowait(bp);
134 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 xfs_ioerror_alert("xlog_bread", log->l_mp,
136 bp, XFS_BUF_ADDR(bp));
137 return error;
138}
139
140/*
141 * Write out the buffer at the given block for the given number of blocks.
142 * The buffer is kept locked across the write and is returned locked.
143 * This can only be used for synchronous log writes.
144 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000145STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146xlog_bwrite(
147 xlog_t *log,
148 xfs_daddr_t blk_no,
149 int nbblks,
150 xfs_buf_t *bp)
151{
152 int error;
153
Dave Chinner32281492009-01-22 15:37:47 +1100154 if (nbblks <= 0 || nbblks > log->l_logBBsize) {
155 xlog_warn("XFS: Invalid block length (0x%x) given for buffer", nbblks);
156 XFS_ERROR_REPORT("xlog_bwrite(1)",
157 XFS_ERRLEVEL_HIGH, log->l_mp);
158 return EFSCORRUPTED;
159 }
160
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 if (log->l_sectbb_log) {
162 blk_no = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, blk_no);
163 nbblks = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, nbblks);
164 }
165
166 ASSERT(nbblks > 0);
167 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
168
169 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
170 XFS_BUF_ZEROFLAGS(bp);
171 XFS_BUF_BUSY(bp);
172 XFS_BUF_HOLD(bp);
173 XFS_BUF_PSEMA(bp, PRIBIO);
174 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
175 XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
176
177 if ((error = xfs_bwrite(log->l_mp, bp)))
178 xfs_ioerror_alert("xlog_bwrite", log->l_mp,
179 bp, XFS_BUF_ADDR(bp));
180 return error;
181}
182
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000183STATIC xfs_caddr_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184xlog_align(
185 xlog_t *log,
186 xfs_daddr_t blk_no,
187 int nbblks,
188 xfs_buf_t *bp)
189{
190 xfs_caddr_t ptr;
191
192 if (!log->l_sectbb_log)
193 return XFS_BUF_PTR(bp);
194
195 ptr = XFS_BUF_PTR(bp) + BBTOB((int)blk_no & log->l_sectbb_mask);
196 ASSERT(XFS_BUF_SIZE(bp) >=
197 BBTOB(nbblks + (blk_no & log->l_sectbb_mask)));
198 return ptr;
199}
200
201#ifdef DEBUG
202/*
203 * dump debug superblock and log record information
204 */
205STATIC void
206xlog_header_check_dump(
207 xfs_mount_t *mp,
208 xlog_rec_header_t *head)
209{
210 int b;
211
Harvey Harrison34a622b2008-04-10 12:19:21 +1000212 cmn_err(CE_DEBUG, "%s: SB : uuid = ", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 for (b = 0; b < 16; b++)
Christoph Hellwiga5687782009-02-09 08:37:39 +0100214 cmn_err(CE_DEBUG, "%02x", ((__uint8_t *)&mp->m_sb.sb_uuid)[b]);
Nathan Scottb6574522006-06-09 15:29:40 +1000215 cmn_err(CE_DEBUG, ", fmt = %d\n", XLOG_FMT);
216 cmn_err(CE_DEBUG, " log : uuid = ");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 for (b = 0; b < 16; b++)
Christoph Hellwiga5687782009-02-09 08:37:39 +0100218 cmn_err(CE_DEBUG, "%02x", ((__uint8_t *)&head->h_fs_uuid)[b]);
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000219 cmn_err(CE_DEBUG, ", fmt = %d\n", be32_to_cpu(head->h_fmt));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220}
221#else
222#define xlog_header_check_dump(mp, head)
223#endif
224
225/*
226 * check log record header for recovery
227 */
228STATIC int
229xlog_header_check_recover(
230 xfs_mount_t *mp,
231 xlog_rec_header_t *head)
232{
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000233 ASSERT(be32_to_cpu(head->h_magicno) == XLOG_HEADER_MAGIC_NUM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234
235 /*
236 * IRIX doesn't write the h_fmt field and leaves it zeroed
237 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
238 * a dirty log created in IRIX.
239 */
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000240 if (unlikely(be32_to_cpu(head->h_fmt) != XLOG_FMT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 xlog_warn(
242 "XFS: dirty log written in incompatible format - can't recover");
243 xlog_header_check_dump(mp, head);
244 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
245 XFS_ERRLEVEL_HIGH, mp);
246 return XFS_ERROR(EFSCORRUPTED);
247 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
248 xlog_warn(
249 "XFS: dirty log entry has mismatched uuid - can't recover");
250 xlog_header_check_dump(mp, head);
251 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
252 XFS_ERRLEVEL_HIGH, mp);
253 return XFS_ERROR(EFSCORRUPTED);
254 }
255 return 0;
256}
257
258/*
259 * read the head block of the log and check the header
260 */
261STATIC int
262xlog_header_check_mount(
263 xfs_mount_t *mp,
264 xlog_rec_header_t *head)
265{
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000266 ASSERT(be32_to_cpu(head->h_magicno) == XLOG_HEADER_MAGIC_NUM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267
268 if (uuid_is_nil(&head->h_fs_uuid)) {
269 /*
270 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
271 * h_fs_uuid is nil, we assume this log was last mounted
272 * by IRIX and continue.
273 */
274 xlog_warn("XFS: nil uuid in log - IRIX style log");
275 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
276 xlog_warn("XFS: log has mismatched uuid - can't recover");
277 xlog_header_check_dump(mp, head);
278 XFS_ERROR_REPORT("xlog_header_check_mount",
279 XFS_ERRLEVEL_HIGH, mp);
280 return XFS_ERROR(EFSCORRUPTED);
281 }
282 return 0;
283}
284
285STATIC void
286xlog_recover_iodone(
287 struct xfs_buf *bp)
288{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 if (XFS_BUF_GETERROR(bp)) {
290 /*
291 * We're not going to bother about retrying
292 * this during recovery. One strike!
293 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 xfs_ioerror_alert("xlog_recover_iodone",
Christoph Hellwig15ac08a2008-12-09 04:47:30 -0500295 bp->b_mount, bp, XFS_BUF_ADDR(bp));
296 xfs_force_shutdown(bp->b_mount, SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 }
Christoph Hellwig15ac08a2008-12-09 04:47:30 -0500298 bp->b_mount = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 XFS_BUF_CLR_IODONE_FUNC(bp);
300 xfs_biodone(bp);
301}
302
303/*
304 * This routine finds (to an approximation) the first block in the physical
305 * log which contains the given cycle. It uses a binary search algorithm.
306 * Note that the algorithm can not be perfect because the disk will not
307 * necessarily be perfect.
308 */
David Chinnera8272ce2007-11-23 16:28:09 +1100309STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310xlog_find_cycle_start(
311 xlog_t *log,
312 xfs_buf_t *bp,
313 xfs_daddr_t first_blk,
314 xfs_daddr_t *last_blk,
315 uint cycle)
316{
317 xfs_caddr_t offset;
318 xfs_daddr_t mid_blk;
319 uint mid_cycle;
320 int error;
321
322 mid_blk = BLK_AVG(first_blk, *last_blk);
323 while (mid_blk != first_blk && mid_blk != *last_blk) {
324 if ((error = xlog_bread(log, mid_blk, 1, bp)))
325 return error;
326 offset = xlog_align(log, mid_blk, 1, bp);
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000327 mid_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 if (mid_cycle == cycle) {
329 *last_blk = mid_blk;
330 /* last_half_cycle == mid_cycle */
331 } else {
332 first_blk = mid_blk;
333 /* first_half_cycle == mid_cycle */
334 }
335 mid_blk = BLK_AVG(first_blk, *last_blk);
336 }
337 ASSERT((mid_blk == first_blk && mid_blk+1 == *last_blk) ||
338 (mid_blk == *last_blk && mid_blk-1 == first_blk));
339
340 return 0;
341}
342
343/*
344 * Check that the range of blocks does not contain the cycle number
345 * given. The scan needs to occur from front to back and the ptr into the
346 * region must be updated since a later routine will need to perform another
347 * test. If the region is completely good, we end up returning the same
348 * last block number.
349 *
350 * Set blkno to -1 if we encounter no errors. This is an invalid block number
351 * since we don't ever expect logs to get this large.
352 */
353STATIC int
354xlog_find_verify_cycle(
355 xlog_t *log,
356 xfs_daddr_t start_blk,
357 int nbblks,
358 uint stop_on_cycle_no,
359 xfs_daddr_t *new_blk)
360{
361 xfs_daddr_t i, j;
362 uint cycle;
363 xfs_buf_t *bp;
364 xfs_daddr_t bufblks;
365 xfs_caddr_t buf = NULL;
366 int error = 0;
367
368 bufblks = 1 << ffs(nbblks);
369
370 while (!(bp = xlog_get_bp(log, bufblks))) {
371 /* can't get enough memory to do everything in one big buffer */
372 bufblks >>= 1;
373 if (bufblks <= log->l_sectbb_log)
374 return ENOMEM;
375 }
376
377 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
378 int bcount;
379
380 bcount = min(bufblks, (start_blk + nbblks - i));
381
382 if ((error = xlog_bread(log, i, bcount, bp)))
383 goto out;
384
385 buf = xlog_align(log, i, bcount, bp);
386 for (j = 0; j < bcount; j++) {
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000387 cycle = xlog_get_cycle(buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 if (cycle == stop_on_cycle_no) {
389 *new_blk = i+j;
390 goto out;
391 }
392
393 buf += BBSIZE;
394 }
395 }
396
397 *new_blk = -1;
398
399out:
400 xlog_put_bp(bp);
401 return error;
402}
403
404/*
405 * Potentially backup over partial log record write.
406 *
407 * In the typical case, last_blk is the number of the block directly after
408 * a good log record. Therefore, we subtract one to get the block number
409 * of the last block in the given buffer. extra_bblks contains the number
410 * of blocks we would have read on a previous read. This happens when the
411 * last log record is split over the end of the physical log.
412 *
413 * extra_bblks is the number of blocks potentially verified on a previous
414 * call to this routine.
415 */
416STATIC int
417xlog_find_verify_log_record(
418 xlog_t *log,
419 xfs_daddr_t start_blk,
420 xfs_daddr_t *last_blk,
421 int extra_bblks)
422{
423 xfs_daddr_t i;
424 xfs_buf_t *bp;
425 xfs_caddr_t offset = NULL;
426 xlog_rec_header_t *head = NULL;
427 int error = 0;
428 int smallmem = 0;
429 int num_blks = *last_blk - start_blk;
430 int xhdrs;
431
432 ASSERT(start_blk != 0 || *last_blk != start_blk);
433
434 if (!(bp = xlog_get_bp(log, num_blks))) {
435 if (!(bp = xlog_get_bp(log, 1)))
436 return ENOMEM;
437 smallmem = 1;
438 } else {
439 if ((error = xlog_bread(log, start_blk, num_blks, bp)))
440 goto out;
441 offset = xlog_align(log, start_blk, num_blks, bp);
442 offset += ((num_blks - 1) << BBSHIFT);
443 }
444
445 for (i = (*last_blk) - 1; i >= 0; i--) {
446 if (i < start_blk) {
447 /* valid log record not found */
448 xlog_warn(
449 "XFS: Log inconsistent (didn't find previous header)");
450 ASSERT(0);
451 error = XFS_ERROR(EIO);
452 goto out;
453 }
454
455 if (smallmem) {
456 if ((error = xlog_bread(log, i, 1, bp)))
457 goto out;
458 offset = xlog_align(log, i, 1, bp);
459 }
460
461 head = (xlog_rec_header_t *)offset;
462
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000463 if (XLOG_HEADER_MAGIC_NUM == be32_to_cpu(head->h_magicno))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 break;
465
466 if (!smallmem)
467 offset -= BBSIZE;
468 }
469
470 /*
471 * We hit the beginning of the physical log & still no header. Return
472 * to caller. If caller can handle a return of -1, then this routine
473 * will be called again for the end of the physical log.
474 */
475 if (i == -1) {
476 error = -1;
477 goto out;
478 }
479
480 /*
481 * We have the final block of the good log (the first block
482 * of the log record _before_ the head. So we check the uuid.
483 */
484 if ((error = xlog_header_check_mount(log->l_mp, head)))
485 goto out;
486
487 /*
488 * We may have found a log record header before we expected one.
489 * last_blk will be the 1st block # with a given cycle #. We may end
490 * up reading an entire log record. In this case, we don't want to
491 * reset last_blk. Only when last_blk points in the middle of a log
492 * record do we update last_blk.
493 */
Eric Sandeen62118702008-03-06 13:44:28 +1100494 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000495 uint h_size = be32_to_cpu(head->h_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496
497 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
498 if (h_size % XLOG_HEADER_CYCLE_SIZE)
499 xhdrs++;
500 } else {
501 xhdrs = 1;
502 }
503
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000504 if (*last_blk - i + extra_bblks !=
505 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 *last_blk = i;
507
508out:
509 xlog_put_bp(bp);
510 return error;
511}
512
513/*
514 * Head is defined to be the point of the log where the next log write
515 * write could go. This means that incomplete LR writes at the end are
516 * eliminated when calculating the head. We aren't guaranteed that previous
517 * LR have complete transactions. We only know that a cycle number of
518 * current cycle number -1 won't be present in the log if we start writing
519 * from our current block number.
520 *
521 * last_blk contains the block number of the first block with a given
522 * cycle number.
523 *
524 * Return: zero if normal, non-zero if error.
525 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000526STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527xlog_find_head(
528 xlog_t *log,
529 xfs_daddr_t *return_head_blk)
530{
531 xfs_buf_t *bp;
532 xfs_caddr_t offset;
533 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
534 int num_scan_bblks;
535 uint first_half_cycle, last_half_cycle;
536 uint stop_on_cycle;
537 int error, log_bbnum = log->l_logBBsize;
538
539 /* Is the end of the log device zeroed? */
540 if ((error = xlog_find_zeroed(log, &first_blk)) == -1) {
541 *return_head_blk = first_blk;
542
543 /* Is the whole lot zeroed? */
544 if (!first_blk) {
545 /* Linux XFS shouldn't generate totally zeroed logs -
546 * mkfs etc write a dummy unmount record to a fresh
547 * log so we can store the uuid in there
548 */
549 xlog_warn("XFS: totally zeroed log");
550 }
551
552 return 0;
553 } else if (error) {
554 xlog_warn("XFS: empty log check failed");
555 return error;
556 }
557
558 first_blk = 0; /* get cycle # of 1st block */
559 bp = xlog_get_bp(log, 1);
560 if (!bp)
561 return ENOMEM;
562 if ((error = xlog_bread(log, 0, 1, bp)))
563 goto bp_err;
564 offset = xlog_align(log, 0, 1, bp);
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000565 first_half_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566
567 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
568 if ((error = xlog_bread(log, last_blk, 1, bp)))
569 goto bp_err;
570 offset = xlog_align(log, last_blk, 1, bp);
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000571 last_half_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 ASSERT(last_half_cycle != 0);
573
574 /*
575 * If the 1st half cycle number is equal to the last half cycle number,
576 * then the entire log is stamped with the same cycle number. In this
577 * case, head_blk can't be set to zero (which makes sense). The below
578 * math doesn't work out properly with head_blk equal to zero. Instead,
579 * we set it to log_bbnum which is an invalid block number, but this
580 * value makes the math correct. If head_blk doesn't changed through
581 * all the tests below, *head_blk is set to zero at the very end rather
582 * than log_bbnum. In a sense, log_bbnum and zero are the same block
583 * in a circular file.
584 */
585 if (first_half_cycle == last_half_cycle) {
586 /*
587 * In this case we believe that the entire log should have
588 * cycle number last_half_cycle. We need to scan backwards
589 * from the end verifying that there are no holes still
590 * containing last_half_cycle - 1. If we find such a hole,
591 * then the start of that hole will be the new head. The
592 * simple case looks like
593 * x | x ... | x - 1 | x
594 * Another case that fits this picture would be
595 * x | x + 1 | x ... | x
Nathan Scottc41564b2006-03-29 08:55:14 +1000596 * In this case the head really is somewhere at the end of the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 * log, as one of the latest writes at the beginning was
598 * incomplete.
599 * One more case is
600 * x | x + 1 | x ... | x - 1 | x
601 * This is really the combination of the above two cases, and
602 * the head has to end up at the start of the x-1 hole at the
603 * end of the log.
604 *
605 * In the 256k log case, we will read from the beginning to the
606 * end of the log and search for cycle numbers equal to x-1.
607 * We don't worry about the x+1 blocks that we encounter,
608 * because we know that they cannot be the head since the log
609 * started with x.
610 */
611 head_blk = log_bbnum;
612 stop_on_cycle = last_half_cycle - 1;
613 } else {
614 /*
615 * In this case we want to find the first block with cycle
616 * number matching last_half_cycle. We expect the log to be
617 * some variation on
618 * x + 1 ... | x ...
619 * The first block with cycle number x (last_half_cycle) will
620 * be where the new head belongs. First we do a binary search
621 * for the first occurrence of last_half_cycle. The binary
622 * search may not be totally accurate, so then we scan back
623 * from there looking for occurrences of last_half_cycle before
624 * us. If that backwards scan wraps around the beginning of
625 * the log, then we look for occurrences of last_half_cycle - 1
626 * at the end of the log. The cases we're looking for look
627 * like
628 * x + 1 ... | x | x + 1 | x ...
629 * ^ binary search stopped here
630 * or
631 * x + 1 ... | x ... | x - 1 | x
632 * <---------> less than scan distance
633 */
634 stop_on_cycle = last_half_cycle;
635 if ((error = xlog_find_cycle_start(log, bp, first_blk,
636 &head_blk, last_half_cycle)))
637 goto bp_err;
638 }
639
640 /*
641 * Now validate the answer. Scan back some number of maximum possible
642 * blocks and make sure each one has the expected cycle number. The
643 * maximum is determined by the total possible amount of buffering
644 * in the in-core log. The following number can be made tighter if
645 * we actually look at the block size of the filesystem.
646 */
647 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
648 if (head_blk >= num_scan_bblks) {
649 /*
650 * We are guaranteed that the entire check can be performed
651 * in one buffer.
652 */
653 start_blk = head_blk - num_scan_bblks;
654 if ((error = xlog_find_verify_cycle(log,
655 start_blk, num_scan_bblks,
656 stop_on_cycle, &new_blk)))
657 goto bp_err;
658 if (new_blk != -1)
659 head_blk = new_blk;
660 } else { /* need to read 2 parts of log */
661 /*
662 * We are going to scan backwards in the log in two parts.
663 * First we scan the physical end of the log. In this part
664 * of the log, we are looking for blocks with cycle number
665 * last_half_cycle - 1.
666 * If we find one, then we know that the log starts there, as
667 * we've found a hole that didn't get written in going around
668 * the end of the physical log. The simple case for this is
669 * x + 1 ... | x ... | x - 1 | x
670 * <---------> less than scan distance
671 * If all of the blocks at the end of the log have cycle number
672 * last_half_cycle, then we check the blocks at the start of
673 * the log looking for occurrences of last_half_cycle. If we
674 * find one, then our current estimate for the location of the
675 * first occurrence of last_half_cycle is wrong and we move
676 * back to the hole we've found. This case looks like
677 * x + 1 ... | x | x + 1 | x ...
678 * ^ binary search stopped here
679 * Another case we need to handle that only occurs in 256k
680 * logs is
681 * x + 1 ... | x ... | x+1 | x ...
682 * ^ binary search stops here
683 * In a 256k log, the scan at the end of the log will see the
684 * x + 1 blocks. We need to skip past those since that is
685 * certainly not the head of the log. By searching for
686 * last_half_cycle-1 we accomplish that.
687 */
688 start_blk = log_bbnum - num_scan_bblks + head_blk;
689 ASSERT(head_blk <= INT_MAX &&
690 (xfs_daddr_t) num_scan_bblks - head_blk >= 0);
691 if ((error = xlog_find_verify_cycle(log, start_blk,
692 num_scan_bblks - (int)head_blk,
693 (stop_on_cycle - 1), &new_blk)))
694 goto bp_err;
695 if (new_blk != -1) {
696 head_blk = new_blk;
697 goto bad_blk;
698 }
699
700 /*
701 * Scan beginning of log now. The last part of the physical
702 * log is good. This scan needs to verify that it doesn't find
703 * the last_half_cycle.
704 */
705 start_blk = 0;
706 ASSERT(head_blk <= INT_MAX);
707 if ((error = xlog_find_verify_cycle(log,
708 start_blk, (int)head_blk,
709 stop_on_cycle, &new_blk)))
710 goto bp_err;
711 if (new_blk != -1)
712 head_blk = new_blk;
713 }
714
715 bad_blk:
716 /*
717 * Now we need to make sure head_blk is not pointing to a block in
718 * the middle of a log record.
719 */
720 num_scan_bblks = XLOG_REC_SHIFT(log);
721 if (head_blk >= num_scan_bblks) {
722 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
723
724 /* start ptr at last block ptr before head_blk */
725 if ((error = xlog_find_verify_log_record(log, start_blk,
726 &head_blk, 0)) == -1) {
727 error = XFS_ERROR(EIO);
728 goto bp_err;
729 } else if (error)
730 goto bp_err;
731 } else {
732 start_blk = 0;
733 ASSERT(head_blk <= INT_MAX);
734 if ((error = xlog_find_verify_log_record(log, start_blk,
735 &head_blk, 0)) == -1) {
736 /* We hit the beginning of the log during our search */
737 start_blk = log_bbnum - num_scan_bblks + head_blk;
738 new_blk = log_bbnum;
739 ASSERT(start_blk <= INT_MAX &&
740 (xfs_daddr_t) log_bbnum-start_blk >= 0);
741 ASSERT(head_blk <= INT_MAX);
742 if ((error = xlog_find_verify_log_record(log,
743 start_blk, &new_blk,
744 (int)head_blk)) == -1) {
745 error = XFS_ERROR(EIO);
746 goto bp_err;
747 } else if (error)
748 goto bp_err;
749 if (new_blk != log_bbnum)
750 head_blk = new_blk;
751 } else if (error)
752 goto bp_err;
753 }
754
755 xlog_put_bp(bp);
756 if (head_blk == log_bbnum)
757 *return_head_blk = 0;
758 else
759 *return_head_blk = head_blk;
760 /*
761 * When returning here, we have a good block number. Bad block
762 * means that during a previous crash, we didn't have a clean break
763 * from cycle number N to cycle number N-1. In this case, we need
764 * to find the first block with cycle number N-1.
765 */
766 return 0;
767
768 bp_err:
769 xlog_put_bp(bp);
770
771 if (error)
772 xlog_warn("XFS: failed to find log head");
773 return error;
774}
775
776/*
777 * Find the sync block number or the tail of the log.
778 *
779 * This will be the block number of the last record to have its
780 * associated buffers synced to disk. Every log record header has
781 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
782 * to get a sync block number. The only concern is to figure out which
783 * log record header to believe.
784 *
785 * The following algorithm uses the log record header with the largest
786 * lsn. The entire log record does not need to be valid. We only care
787 * that the header is valid.
788 *
789 * We could speed up search by using current head_blk buffer, but it is not
790 * available.
791 */
792int
793xlog_find_tail(
794 xlog_t *log,
795 xfs_daddr_t *head_blk,
Eric Sandeen65be6052006-01-11 15:34:19 +1100796 xfs_daddr_t *tail_blk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797{
798 xlog_rec_header_t *rhead;
799 xlog_op_header_t *op_head;
800 xfs_caddr_t offset = NULL;
801 xfs_buf_t *bp;
802 int error, i, found;
803 xfs_daddr_t umount_data_blk;
804 xfs_daddr_t after_umount_blk;
805 xfs_lsn_t tail_lsn;
806 int hblks;
807
808 found = 0;
809
810 /*
811 * Find previous log record
812 */
813 if ((error = xlog_find_head(log, head_blk)))
814 return error;
815
816 bp = xlog_get_bp(log, 1);
817 if (!bp)
818 return ENOMEM;
819 if (*head_blk == 0) { /* special case */
820 if ((error = xlog_bread(log, 0, 1, bp)))
821 goto bread_err;
822 offset = xlog_align(log, 0, 1, bp);
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000823 if (xlog_get_cycle(offset) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 *tail_blk = 0;
825 /* leave all other log inited values alone */
826 goto exit;
827 }
828 }
829
830 /*
831 * Search backwards looking for log record header block
832 */
833 ASSERT(*head_blk < INT_MAX);
834 for (i = (int)(*head_blk) - 1; i >= 0; i--) {
835 if ((error = xlog_bread(log, i, 1, bp)))
836 goto bread_err;
837 offset = xlog_align(log, i, 1, bp);
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000838 if (XLOG_HEADER_MAGIC_NUM == be32_to_cpu(*(__be32 *)offset)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 found = 1;
840 break;
841 }
842 }
843 /*
844 * If we haven't found the log record header block, start looking
845 * again from the end of the physical log. XXXmiken: There should be
846 * a check here to make sure we didn't search more than N blocks in
847 * the previous code.
848 */
849 if (!found) {
850 for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) {
851 if ((error = xlog_bread(log, i, 1, bp)))
852 goto bread_err;
853 offset = xlog_align(log, i, 1, bp);
854 if (XLOG_HEADER_MAGIC_NUM ==
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000855 be32_to_cpu(*(__be32 *)offset)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 found = 2;
857 break;
858 }
859 }
860 }
861 if (!found) {
862 xlog_warn("XFS: xlog_find_tail: couldn't find sync record");
863 ASSERT(0);
864 return XFS_ERROR(EIO);
865 }
866
867 /* find blk_no of tail of log */
868 rhead = (xlog_rec_header_t *)offset;
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000869 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870
871 /*
872 * Reset log values according to the state of the log when we
873 * crashed. In the case where head_blk == 0, we bump curr_cycle
874 * one because the next write starts a new cycle rather than
875 * continuing the cycle of the last good log record. At this
876 * point we have guaranteed that all partial log records have been
877 * accounted for. Therefore, we know that the last good log record
878 * written was complete and ended exactly on the end boundary
879 * of the physical log.
880 */
881 log->l_prev_block = i;
882 log->l_curr_block = (int)*head_blk;
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000883 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 if (found == 2)
885 log->l_curr_cycle++;
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000886 log->l_tail_lsn = be64_to_cpu(rhead->h_tail_lsn);
887 log->l_last_sync_lsn = be64_to_cpu(rhead->h_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 log->l_grant_reserve_cycle = log->l_curr_cycle;
889 log->l_grant_reserve_bytes = BBTOB(log->l_curr_block);
890 log->l_grant_write_cycle = log->l_curr_cycle;
891 log->l_grant_write_bytes = BBTOB(log->l_curr_block);
892
893 /*
894 * Look for unmount record. If we find it, then we know there
895 * was a clean unmount. Since 'i' could be the last block in
896 * the physical log, we convert to a log block before comparing
897 * to the head_blk.
898 *
899 * Save the current tail lsn to use to pass to
900 * xlog_clear_stale_blocks() below. We won't want to clear the
901 * unmount record if there is one, so we pass the lsn of the
902 * unmount record rather than the block after it.
903 */
Eric Sandeen62118702008-03-06 13:44:28 +1100904 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000905 int h_size = be32_to_cpu(rhead->h_size);
906 int h_version = be32_to_cpu(rhead->h_version);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907
908 if ((h_version & XLOG_VERSION_2) &&
909 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
910 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
911 if (h_size % XLOG_HEADER_CYCLE_SIZE)
912 hblks++;
913 } else {
914 hblks = 1;
915 }
916 } else {
917 hblks = 1;
918 }
919 after_umount_blk = (i + hblks + (int)
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000920 BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 tail_lsn = log->l_tail_lsn;
922 if (*head_blk == after_umount_blk &&
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000923 be32_to_cpu(rhead->h_num_logops) == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 umount_data_blk = (i + hblks) % log->l_logBBsize;
925 if ((error = xlog_bread(log, umount_data_blk, 1, bp))) {
926 goto bread_err;
927 }
928 offset = xlog_align(log, umount_data_blk, 1, bp);
929 op_head = (xlog_op_header_t *)offset;
930 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
931 /*
932 * Set tail and last sync so that newly written
933 * log records will point recovery to after the
934 * current unmount record.
935 */
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000936 log->l_tail_lsn =
937 xlog_assign_lsn(log->l_curr_cycle,
938 after_umount_blk);
939 log->l_last_sync_lsn =
940 xlog_assign_lsn(log->l_curr_cycle,
941 after_umount_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 *tail_blk = after_umount_blk;
David Chinner92821e22007-05-24 15:26:31 +1000943
944 /*
945 * Note that the unmount was clean. If the unmount
946 * was not clean, we need to know this to rebuild the
947 * superblock counters from the perag headers if we
948 * have a filesystem using non-persistent counters.
949 */
950 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 }
952 }
953
954 /*
955 * Make sure that there are no blocks in front of the head
956 * with the same cycle number as the head. This can happen
957 * because we allow multiple outstanding log writes concurrently,
958 * and the later writes might make it out before earlier ones.
959 *
960 * We use the lsn from before modifying it so that we'll never
961 * overwrite the unmount record after a clean unmount.
962 *
963 * Do this only if we are going to recover the filesystem
964 *
965 * NOTE: This used to say "if (!readonly)"
966 * However on Linux, we can & do recover a read-only filesystem.
967 * We only skip recovery if NORECOVERY is specified on mount,
968 * in which case we would not be here.
969 *
970 * But... if the -device- itself is readonly, just skip this.
971 * We can't recover this device anyway, so it won't matter.
972 */
973 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp)) {
974 error = xlog_clear_stale_blocks(log, tail_lsn);
975 }
976
977bread_err:
978exit:
979 xlog_put_bp(bp);
980
981 if (error)
982 xlog_warn("XFS: failed to locate log tail");
983 return error;
984}
985
986/*
987 * Is the log zeroed at all?
988 *
989 * The last binary search should be changed to perform an X block read
990 * once X becomes small enough. You can then search linearly through
991 * the X blocks. This will cut down on the number of reads we need to do.
992 *
993 * If the log is partially zeroed, this routine will pass back the blkno
994 * of the first block with cycle number 0. It won't have a complete LR
995 * preceding it.
996 *
997 * Return:
998 * 0 => the log is completely written to
999 * -1 => use *blk_no as the first block of the log
1000 * >0 => error has occurred
1001 */
David Chinnera8272ce2007-11-23 16:28:09 +11001002STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003xlog_find_zeroed(
1004 xlog_t *log,
1005 xfs_daddr_t *blk_no)
1006{
1007 xfs_buf_t *bp;
1008 xfs_caddr_t offset;
1009 uint first_cycle, last_cycle;
1010 xfs_daddr_t new_blk, last_blk, start_blk;
1011 xfs_daddr_t num_scan_bblks;
1012 int error, log_bbnum = log->l_logBBsize;
1013
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10001014 *blk_no = 0;
1015
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 /* check totally zeroed log */
1017 bp = xlog_get_bp(log, 1);
1018 if (!bp)
1019 return ENOMEM;
1020 if ((error = xlog_bread(log, 0, 1, bp)))
1021 goto bp_err;
1022 offset = xlog_align(log, 0, 1, bp);
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10001023 first_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 if (first_cycle == 0) { /* completely zeroed log */
1025 *blk_no = 0;
1026 xlog_put_bp(bp);
1027 return -1;
1028 }
1029
1030 /* check partially zeroed log */
1031 if ((error = xlog_bread(log, log_bbnum-1, 1, bp)))
1032 goto bp_err;
1033 offset = xlog_align(log, log_bbnum-1, 1, bp);
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10001034 last_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 if (last_cycle != 0) { /* log completely written to */
1036 xlog_put_bp(bp);
1037 return 0;
1038 } else if (first_cycle != 1) {
1039 /*
1040 * If the cycle of the last block is zero, the cycle of
1041 * the first block must be 1. If it's not, maybe we're
1042 * not looking at a log... Bail out.
1043 */
1044 xlog_warn("XFS: Log inconsistent or not a log (last==0, first!=1)");
1045 return XFS_ERROR(EINVAL);
1046 }
1047
1048 /* we have a partially zeroed log */
1049 last_blk = log_bbnum-1;
1050 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1051 goto bp_err;
1052
1053 /*
1054 * Validate the answer. Because there is no way to guarantee that
1055 * the entire log is made up of log records which are the same size,
1056 * we scan over the defined maximum blocks. At this point, the maximum
1057 * is not chosen to mean anything special. XXXmiken
1058 */
1059 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1060 ASSERT(num_scan_bblks <= INT_MAX);
1061
1062 if (last_blk < num_scan_bblks)
1063 num_scan_bblks = last_blk;
1064 start_blk = last_blk - num_scan_bblks;
1065
1066 /*
1067 * We search for any instances of cycle number 0 that occur before
1068 * our current estimate of the head. What we're trying to detect is
1069 * 1 ... | 0 | 1 | 0...
1070 * ^ binary search ends here
1071 */
1072 if ((error = xlog_find_verify_cycle(log, start_blk,
1073 (int)num_scan_bblks, 0, &new_blk)))
1074 goto bp_err;
1075 if (new_blk != -1)
1076 last_blk = new_blk;
1077
1078 /*
1079 * Potentially backup over partial log record write. We don't need
1080 * to search the end of the log because we know it is zero.
1081 */
1082 if ((error = xlog_find_verify_log_record(log, start_blk,
1083 &last_blk, 0)) == -1) {
1084 error = XFS_ERROR(EIO);
1085 goto bp_err;
1086 } else if (error)
1087 goto bp_err;
1088
1089 *blk_no = last_blk;
1090bp_err:
1091 xlog_put_bp(bp);
1092 if (error)
1093 return error;
1094 return -1;
1095}
1096
1097/*
1098 * These are simple subroutines used by xlog_clear_stale_blocks() below
1099 * to initialize a buffer full of empty log record headers and write
1100 * them into the log.
1101 */
1102STATIC void
1103xlog_add_record(
1104 xlog_t *log,
1105 xfs_caddr_t buf,
1106 int cycle,
1107 int block,
1108 int tail_cycle,
1109 int tail_block)
1110{
1111 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1112
1113 memset(buf, 0, BBSIZE);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001114 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1115 recp->h_cycle = cpu_to_be32(cycle);
1116 recp->h_version = cpu_to_be32(
Eric Sandeen62118702008-03-06 13:44:28 +11001117 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001118 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1119 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1120 recp->h_fmt = cpu_to_be32(XLOG_FMT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1122}
1123
1124STATIC int
1125xlog_write_log_records(
1126 xlog_t *log,
1127 int cycle,
1128 int start_block,
1129 int blocks,
1130 int tail_cycle,
1131 int tail_block)
1132{
1133 xfs_caddr_t offset;
1134 xfs_buf_t *bp;
1135 int balign, ealign;
1136 int sectbb = XLOG_SECTOR_ROUNDUP_BBCOUNT(log, 1);
1137 int end_block = start_block + blocks;
1138 int bufblks;
1139 int error = 0;
1140 int i, j = 0;
1141
1142 bufblks = 1 << ffs(blocks);
1143 while (!(bp = xlog_get_bp(log, bufblks))) {
1144 bufblks >>= 1;
1145 if (bufblks <= log->l_sectbb_log)
1146 return ENOMEM;
1147 }
1148
1149 /* We may need to do a read at the start to fill in part of
1150 * the buffer in the starting sector not covered by the first
1151 * write below.
1152 */
1153 balign = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, start_block);
1154 if (balign != start_block) {
1155 if ((error = xlog_bread(log, start_block, 1, bp))) {
1156 xlog_put_bp(bp);
1157 return error;
1158 }
1159 j = start_block - balign;
1160 }
1161
1162 for (i = start_block; i < end_block; i += bufblks) {
1163 int bcount, endcount;
1164
1165 bcount = min(bufblks, end_block - start_block);
1166 endcount = bcount - j;
1167
1168 /* We may need to do a read at the end to fill in part of
1169 * the buffer in the final sector not covered by the write.
1170 * If this is the same sector as the above read, skip it.
1171 */
1172 ealign = XLOG_SECTOR_ROUNDDOWN_BLKNO(log, end_block);
1173 if (j == 0 && (start_block + endcount > ealign)) {
1174 offset = XFS_BUF_PTR(bp);
1175 balign = BBTOB(ealign - start_block);
David Chinner234f56a2008-04-10 12:24:24 +10001176 error = XFS_BUF_SET_PTR(bp, offset + balign,
1177 BBTOB(sectbb));
1178 if (!error)
1179 error = xlog_bread(log, ealign, sectbb, bp);
1180 if (!error)
1181 error = XFS_BUF_SET_PTR(bp, offset, bufblks);
1182 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 }
1185
1186 offset = xlog_align(log, start_block, endcount, bp);
1187 for (; j < endcount; j++) {
1188 xlog_add_record(log, offset, cycle, i+j,
1189 tail_cycle, tail_block);
1190 offset += BBSIZE;
1191 }
1192 error = xlog_bwrite(log, start_block, endcount, bp);
1193 if (error)
1194 break;
1195 start_block += endcount;
1196 j = 0;
1197 }
1198 xlog_put_bp(bp);
1199 return error;
1200}
1201
1202/*
1203 * This routine is called to blow away any incomplete log writes out
1204 * in front of the log head. We do this so that we won't become confused
1205 * if we come up, write only a little bit more, and then crash again.
1206 * If we leave the partial log records out there, this situation could
1207 * cause us to think those partial writes are valid blocks since they
1208 * have the current cycle number. We get rid of them by overwriting them
1209 * with empty log records with the old cycle number rather than the
1210 * current one.
1211 *
1212 * The tail lsn is passed in rather than taken from
1213 * the log so that we will not write over the unmount record after a
1214 * clean unmount in a 512 block log. Doing so would leave the log without
1215 * any valid log records in it until a new one was written. If we crashed
1216 * during that time we would not be able to recover.
1217 */
1218STATIC int
1219xlog_clear_stale_blocks(
1220 xlog_t *log,
1221 xfs_lsn_t tail_lsn)
1222{
1223 int tail_cycle, head_cycle;
1224 int tail_block, head_block;
1225 int tail_distance, max_distance;
1226 int distance;
1227 int error;
1228
1229 tail_cycle = CYCLE_LSN(tail_lsn);
1230 tail_block = BLOCK_LSN(tail_lsn);
1231 head_cycle = log->l_curr_cycle;
1232 head_block = log->l_curr_block;
1233
1234 /*
1235 * Figure out the distance between the new head of the log
1236 * and the tail. We want to write over any blocks beyond the
1237 * head that we may have written just before the crash, but
1238 * we don't want to overwrite the tail of the log.
1239 */
1240 if (head_cycle == tail_cycle) {
1241 /*
1242 * The tail is behind the head in the physical log,
1243 * so the distance from the head to the tail is the
1244 * distance from the head to the end of the log plus
1245 * the distance from the beginning of the log to the
1246 * tail.
1247 */
1248 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1249 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1250 XFS_ERRLEVEL_LOW, log->l_mp);
1251 return XFS_ERROR(EFSCORRUPTED);
1252 }
1253 tail_distance = tail_block + (log->l_logBBsize - head_block);
1254 } else {
1255 /*
1256 * The head is behind the tail in the physical log,
1257 * so the distance from the head to the tail is just
1258 * the tail block minus the head block.
1259 */
1260 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1261 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1262 XFS_ERRLEVEL_LOW, log->l_mp);
1263 return XFS_ERROR(EFSCORRUPTED);
1264 }
1265 tail_distance = tail_block - head_block;
1266 }
1267
1268 /*
1269 * If the head is right up against the tail, we can't clear
1270 * anything.
1271 */
1272 if (tail_distance <= 0) {
1273 ASSERT(tail_distance == 0);
1274 return 0;
1275 }
1276
1277 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1278 /*
1279 * Take the smaller of the maximum amount of outstanding I/O
1280 * we could have and the distance to the tail to clear out.
1281 * We take the smaller so that we don't overwrite the tail and
1282 * we don't waste all day writing from the head to the tail
1283 * for no reason.
1284 */
1285 max_distance = MIN(max_distance, tail_distance);
1286
1287 if ((head_block + max_distance) <= log->l_logBBsize) {
1288 /*
1289 * We can stomp all the blocks we need to without
1290 * wrapping around the end of the log. Just do it
1291 * in a single write. Use the cycle number of the
1292 * current cycle minus one so that the log will look like:
1293 * n ... | n - 1 ...
1294 */
1295 error = xlog_write_log_records(log, (head_cycle - 1),
1296 head_block, max_distance, tail_cycle,
1297 tail_block);
1298 if (error)
1299 return error;
1300 } else {
1301 /*
1302 * We need to wrap around the end of the physical log in
1303 * order to clear all the blocks. Do it in two separate
1304 * I/Os. The first write should be from the head to the
1305 * end of the physical log, and it should use the current
1306 * cycle number minus one just like above.
1307 */
1308 distance = log->l_logBBsize - head_block;
1309 error = xlog_write_log_records(log, (head_cycle - 1),
1310 head_block, distance, tail_cycle,
1311 tail_block);
1312
1313 if (error)
1314 return error;
1315
1316 /*
1317 * Now write the blocks at the start of the physical log.
1318 * This writes the remainder of the blocks we want to clear.
1319 * It uses the current cycle number since we're now on the
1320 * same cycle as the head so that we get:
1321 * n ... n ... | n - 1 ...
1322 * ^^^^^ blocks we're writing
1323 */
1324 distance = max_distance - (log->l_logBBsize - head_block);
1325 error = xlog_write_log_records(log, head_cycle, 0, distance,
1326 tail_cycle, tail_block);
1327 if (error)
1328 return error;
1329 }
1330
1331 return 0;
1332}
1333
1334/******************************************************************************
1335 *
1336 * Log recover routines
1337 *
1338 ******************************************************************************
1339 */
1340
1341STATIC xlog_recover_t *
1342xlog_recover_find_tid(
1343 xlog_recover_t *q,
1344 xlog_tid_t tid)
1345{
1346 xlog_recover_t *p = q;
1347
1348 while (p != NULL) {
1349 if (p->r_log_tid == tid)
1350 break;
1351 p = p->r_next;
1352 }
1353 return p;
1354}
1355
1356STATIC void
1357xlog_recover_put_hashq(
1358 xlog_recover_t **q,
1359 xlog_recover_t *trans)
1360{
1361 trans->r_next = *q;
1362 *q = trans;
1363}
1364
1365STATIC void
1366xlog_recover_add_item(
1367 xlog_recover_item_t **itemq)
1368{
1369 xlog_recover_item_t *item;
1370
1371 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
1372 xlog_recover_insert_item_backq(itemq, item);
1373}
1374
1375STATIC int
1376xlog_recover_add_to_cont_trans(
1377 xlog_recover_t *trans,
1378 xfs_caddr_t dp,
1379 int len)
1380{
1381 xlog_recover_item_t *item;
1382 xfs_caddr_t ptr, old_ptr;
1383 int old_len;
1384
1385 item = trans->r_itemq;
Christoph Hellwig4b809162007-08-16 15:37:36 +10001386 if (item == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 /* finish copying rest of trans header */
1388 xlog_recover_add_item(&trans->r_itemq);
1389 ptr = (xfs_caddr_t) &trans->r_theader +
1390 sizeof(xfs_trans_header_t) - len;
1391 memcpy(ptr, dp, len); /* d, s, l */
1392 return 0;
1393 }
1394 item = item->ri_prev;
1395
1396 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
1397 old_len = item->ri_buf[item->ri_cnt-1].i_len;
1398
Christoph Hellwig760dea62005-09-02 16:56:02 +10001399 ptr = kmem_realloc(old_ptr, len+old_len, old_len, 0u);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 memcpy(&ptr[old_len], dp, len); /* d, s, l */
1401 item->ri_buf[item->ri_cnt-1].i_len += len;
1402 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
1403 return 0;
1404}
1405
1406/*
1407 * The next region to add is the start of a new region. It could be
1408 * a whole region or it could be the first part of a new region. Because
1409 * of this, the assumption here is that the type and size fields of all
1410 * format structures fit into the first 32 bits of the structure.
1411 *
1412 * This works because all regions must be 32 bit aligned. Therefore, we
1413 * either have both fields or we have neither field. In the case we have
1414 * neither field, the data part of the region is zero length. We only have
1415 * a log_op_header and can throw away the header since a new one will appear
1416 * later. If we have at least 4 bytes, then we can determine how many regions
1417 * will appear in the current log item.
1418 */
1419STATIC int
1420xlog_recover_add_to_trans(
1421 xlog_recover_t *trans,
1422 xfs_caddr_t dp,
1423 int len)
1424{
1425 xfs_inode_log_format_t *in_f; /* any will do */
1426 xlog_recover_item_t *item;
1427 xfs_caddr_t ptr;
1428
1429 if (!len)
1430 return 0;
1431 item = trans->r_itemq;
Christoph Hellwig4b809162007-08-16 15:37:36 +10001432 if (item == NULL) {
David Chinner5a792c42008-10-30 17:40:09 +11001433 /* we need to catch log corruptions here */
1434 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
1435 xlog_warn("XFS: xlog_recover_add_to_trans: "
1436 "bad header magic number");
1437 ASSERT(0);
1438 return XFS_ERROR(EIO);
1439 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 if (len == sizeof(xfs_trans_header_t))
1441 xlog_recover_add_item(&trans->r_itemq);
1442 memcpy(&trans->r_theader, dp, len); /* d, s, l */
1443 return 0;
1444 }
1445
1446 ptr = kmem_alloc(len, KM_SLEEP);
1447 memcpy(ptr, dp, len);
1448 in_f = (xfs_inode_log_format_t *)ptr;
1449
1450 if (item->ri_prev->ri_total != 0 &&
1451 item->ri_prev->ri_total == item->ri_prev->ri_cnt) {
1452 xlog_recover_add_item(&trans->r_itemq);
1453 }
1454 item = trans->r_itemq;
1455 item = item->ri_prev;
1456
1457 if (item->ri_total == 0) { /* first region to be added */
Christoph Hellwige8fa6b42009-03-03 14:48:36 -05001458 if (in_f->ilf_size == 0 ||
1459 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
1460 xlog_warn(
1461 "XFS: bad number of regions (%d) in inode log format",
1462 in_f->ilf_size);
1463 ASSERT(0);
1464 return XFS_ERROR(EIO);
1465 }
1466
1467 item->ri_total = in_f->ilf_size;
1468 item->ri_buf =
1469 kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
1470 KM_SLEEP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 }
1472 ASSERT(item->ri_total > item->ri_cnt);
1473 /* Description region is ri_buf[0] */
1474 item->ri_buf[item->ri_cnt].i_addr = ptr;
1475 item->ri_buf[item->ri_cnt].i_len = len;
1476 item->ri_cnt++;
1477 return 0;
1478}
1479
1480STATIC void
1481xlog_recover_new_tid(
1482 xlog_recover_t **q,
1483 xlog_tid_t tid,
1484 xfs_lsn_t lsn)
1485{
1486 xlog_recover_t *trans;
1487
1488 trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
1489 trans->r_log_tid = tid;
1490 trans->r_lsn = lsn;
1491 xlog_recover_put_hashq(q, trans);
1492}
1493
1494STATIC int
1495xlog_recover_unlink_tid(
1496 xlog_recover_t **q,
1497 xlog_recover_t *trans)
1498{
1499 xlog_recover_t *tp;
1500 int found = 0;
1501
Christoph Hellwig4b809162007-08-16 15:37:36 +10001502 ASSERT(trans != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 if (trans == *q) {
1504 *q = (*q)->r_next;
1505 } else {
1506 tp = *q;
Christoph Hellwig4b809162007-08-16 15:37:36 +10001507 while (tp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 if (tp->r_next == trans) {
1509 found = 1;
1510 break;
1511 }
1512 tp = tp->r_next;
1513 }
1514 if (!found) {
1515 xlog_warn(
1516 "XFS: xlog_recover_unlink_tid: trans not found");
1517 ASSERT(0);
1518 return XFS_ERROR(EIO);
1519 }
1520 tp->r_next = tp->r_next->r_next;
1521 }
1522 return 0;
1523}
1524
1525STATIC void
1526xlog_recover_insert_item_backq(
1527 xlog_recover_item_t **q,
1528 xlog_recover_item_t *item)
1529{
Christoph Hellwig4b809162007-08-16 15:37:36 +10001530 if (*q == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 item->ri_prev = item->ri_next = item;
1532 *q = item;
1533 } else {
1534 item->ri_next = *q;
1535 item->ri_prev = (*q)->ri_prev;
1536 (*q)->ri_prev = item;
1537 item->ri_prev->ri_next = item;
1538 }
1539}
1540
1541STATIC void
1542xlog_recover_insert_item_frontq(
1543 xlog_recover_item_t **q,
1544 xlog_recover_item_t *item)
1545{
1546 xlog_recover_insert_item_backq(q, item);
1547 *q = item;
1548}
1549
1550STATIC int
1551xlog_recover_reorder_trans(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 xlog_recover_t *trans)
1553{
1554 xlog_recover_item_t *first_item, *itemq, *itemq_next;
1555 xfs_buf_log_format_t *buf_f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 ushort flags = 0;
1557
1558 first_item = itemq = trans->r_itemq;
1559 trans->r_itemq = NULL;
1560 do {
1561 itemq_next = itemq->ri_next;
1562 buf_f = (xfs_buf_log_format_t *)itemq->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563
1564 switch (ITEM_TYPE(itemq)) {
1565 case XFS_LI_BUF:
Eric Sandeen804195b2007-02-10 18:35:02 +11001566 flags = buf_f->blf_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 if (!(flags & XFS_BLI_CANCEL)) {
1568 xlog_recover_insert_item_frontq(&trans->r_itemq,
1569 itemq);
1570 break;
1571 }
1572 case XFS_LI_INODE:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 case XFS_LI_DQUOT:
1574 case XFS_LI_QUOTAOFF:
1575 case XFS_LI_EFD:
1576 case XFS_LI_EFI:
1577 xlog_recover_insert_item_backq(&trans->r_itemq, itemq);
1578 break;
1579 default:
1580 xlog_warn(
1581 "XFS: xlog_recover_reorder_trans: unrecognized type of log operation");
1582 ASSERT(0);
1583 return XFS_ERROR(EIO);
1584 }
1585 itemq = itemq_next;
1586 } while (first_item != itemq);
1587 return 0;
1588}
1589
1590/*
1591 * Build up the table of buf cancel records so that we don't replay
1592 * cancelled data in the second pass. For buffer records that are
1593 * not cancel records, there is nothing to do here so we just return.
1594 *
1595 * If we get a cancel record which is already in the table, this indicates
1596 * that the buffer was cancelled multiple times. In order to ensure
1597 * that during pass 2 we keep the record in the table until we reach its
1598 * last occurrence in the log, we keep a reference count in the cancel
1599 * record in the table to tell us how many times we expect to see this
1600 * record during the second pass.
1601 */
1602STATIC void
1603xlog_recover_do_buffer_pass1(
1604 xlog_t *log,
1605 xfs_buf_log_format_t *buf_f)
1606{
1607 xfs_buf_cancel_t *bcp;
1608 xfs_buf_cancel_t *nextp;
1609 xfs_buf_cancel_t *prevp;
1610 xfs_buf_cancel_t **bucket;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 xfs_daddr_t blkno = 0;
1612 uint len = 0;
1613 ushort flags = 0;
1614
1615 switch (buf_f->blf_type) {
1616 case XFS_LI_BUF:
1617 blkno = buf_f->blf_blkno;
1618 len = buf_f->blf_len;
1619 flags = buf_f->blf_flags;
1620 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 }
1622
1623 /*
1624 * If this isn't a cancel buffer item, then just return.
1625 */
1626 if (!(flags & XFS_BLI_CANCEL))
1627 return;
1628
1629 /*
1630 * Insert an xfs_buf_cancel record into the hash table of
1631 * them. If there is already an identical record, bump
1632 * its reference count.
1633 */
1634 bucket = &log->l_buf_cancel_table[(__uint64_t)blkno %
1635 XLOG_BC_TABLE_SIZE];
1636 /*
1637 * If the hash bucket is empty then just insert a new record into
1638 * the bucket.
1639 */
1640 if (*bucket == NULL) {
1641 bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t),
1642 KM_SLEEP);
1643 bcp->bc_blkno = blkno;
1644 bcp->bc_len = len;
1645 bcp->bc_refcount = 1;
1646 bcp->bc_next = NULL;
1647 *bucket = bcp;
1648 return;
1649 }
1650
1651 /*
1652 * The hash bucket is not empty, so search for duplicates of our
1653 * record. If we find one them just bump its refcount. If not
1654 * then add us at the end of the list.
1655 */
1656 prevp = NULL;
1657 nextp = *bucket;
1658 while (nextp != NULL) {
1659 if (nextp->bc_blkno == blkno && nextp->bc_len == len) {
1660 nextp->bc_refcount++;
1661 return;
1662 }
1663 prevp = nextp;
1664 nextp = nextp->bc_next;
1665 }
1666 ASSERT(prevp != NULL);
1667 bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t),
1668 KM_SLEEP);
1669 bcp->bc_blkno = blkno;
1670 bcp->bc_len = len;
1671 bcp->bc_refcount = 1;
1672 bcp->bc_next = NULL;
1673 prevp->bc_next = bcp;
1674}
1675
1676/*
1677 * Check to see whether the buffer being recovered has a corresponding
1678 * entry in the buffer cancel record table. If it does then return 1
1679 * so that it will be cancelled, otherwise return 0. If the buffer is
1680 * actually a buffer cancel item (XFS_BLI_CANCEL is set), then decrement
1681 * the refcount on the entry in the table and remove it from the table
1682 * if this is the last reference.
1683 *
1684 * We remove the cancel record from the table when we encounter its
1685 * last occurrence in the log so that if the same buffer is re-used
1686 * again after its last cancellation we actually replay the changes
1687 * made at that point.
1688 */
1689STATIC int
1690xlog_check_buffer_cancelled(
1691 xlog_t *log,
1692 xfs_daddr_t blkno,
1693 uint len,
1694 ushort flags)
1695{
1696 xfs_buf_cancel_t *bcp;
1697 xfs_buf_cancel_t *prevp;
1698 xfs_buf_cancel_t **bucket;
1699
1700 if (log->l_buf_cancel_table == NULL) {
1701 /*
1702 * There is nothing in the table built in pass one,
1703 * so this buffer must not be cancelled.
1704 */
1705 ASSERT(!(flags & XFS_BLI_CANCEL));
1706 return 0;
1707 }
1708
1709 bucket = &log->l_buf_cancel_table[(__uint64_t)blkno %
1710 XLOG_BC_TABLE_SIZE];
1711 bcp = *bucket;
1712 if (bcp == NULL) {
1713 /*
1714 * There is no corresponding entry in the table built
1715 * in pass one, so this buffer has not been cancelled.
1716 */
1717 ASSERT(!(flags & XFS_BLI_CANCEL));
1718 return 0;
1719 }
1720
1721 /*
1722 * Search for an entry in the buffer cancel table that
1723 * matches our buffer.
1724 */
1725 prevp = NULL;
1726 while (bcp != NULL) {
1727 if (bcp->bc_blkno == blkno && bcp->bc_len == len) {
1728 /*
1729 * We've go a match, so return 1 so that the
1730 * recovery of this buffer is cancelled.
1731 * If this buffer is actually a buffer cancel
1732 * log item, then decrement the refcount on the
1733 * one in the table and remove it if this is the
1734 * last reference.
1735 */
1736 if (flags & XFS_BLI_CANCEL) {
1737 bcp->bc_refcount--;
1738 if (bcp->bc_refcount == 0) {
1739 if (prevp == NULL) {
1740 *bucket = bcp->bc_next;
1741 } else {
1742 prevp->bc_next = bcp->bc_next;
1743 }
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10001744 kmem_free(bcp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745 }
1746 }
1747 return 1;
1748 }
1749 prevp = bcp;
1750 bcp = bcp->bc_next;
1751 }
1752 /*
1753 * We didn't find a corresponding entry in the table, so
1754 * return 0 so that the buffer is NOT cancelled.
1755 */
1756 ASSERT(!(flags & XFS_BLI_CANCEL));
1757 return 0;
1758}
1759
1760STATIC int
1761xlog_recover_do_buffer_pass2(
1762 xlog_t *log,
1763 xfs_buf_log_format_t *buf_f)
1764{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 xfs_daddr_t blkno = 0;
1766 ushort flags = 0;
1767 uint len = 0;
1768
1769 switch (buf_f->blf_type) {
1770 case XFS_LI_BUF:
1771 blkno = buf_f->blf_blkno;
1772 flags = buf_f->blf_flags;
1773 len = buf_f->blf_len;
1774 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 }
1776
1777 return xlog_check_buffer_cancelled(log, blkno, len, flags);
1778}
1779
1780/*
1781 * Perform recovery for a buffer full of inodes. In these buffers,
1782 * the only data which should be recovered is that which corresponds
1783 * to the di_next_unlinked pointers in the on disk inode structures.
1784 * The rest of the data for the inodes is always logged through the
1785 * inodes themselves rather than the inode buffer and is recovered
1786 * in xlog_recover_do_inode_trans().
1787 *
1788 * The only time when buffers full of inodes are fully recovered is
1789 * when the buffer is full of newly allocated inodes. In this case
1790 * the buffer will not be marked as an inode buffer and so will be
1791 * sent to xlog_recover_do_reg_buffer() below during recovery.
1792 */
1793STATIC int
1794xlog_recover_do_inode_buffer(
1795 xfs_mount_t *mp,
1796 xlog_recover_item_t *item,
1797 xfs_buf_t *bp,
1798 xfs_buf_log_format_t *buf_f)
1799{
1800 int i;
1801 int item_index;
1802 int bit;
1803 int nbits;
1804 int reg_buf_offset;
1805 int reg_buf_bytes;
1806 int next_unlinked_offset;
1807 int inodes_per_buf;
1808 xfs_agino_t *logged_nextp;
1809 xfs_agino_t *buffer_nextp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 unsigned int *data_map = NULL;
1811 unsigned int map_size = 0;
1812
1813 switch (buf_f->blf_type) {
1814 case XFS_LI_BUF:
1815 data_map = buf_f->blf_data_map;
1816 map_size = buf_f->blf_map_size;
1817 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 }
1819 /*
1820 * Set the variables corresponding to the current region to
1821 * 0 so that we'll initialize them on the first pass through
1822 * the loop.
1823 */
1824 reg_buf_offset = 0;
1825 reg_buf_bytes = 0;
1826 bit = 0;
1827 nbits = 0;
1828 item_index = 0;
1829 inodes_per_buf = XFS_BUF_COUNT(bp) >> mp->m_sb.sb_inodelog;
1830 for (i = 0; i < inodes_per_buf; i++) {
1831 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
1832 offsetof(xfs_dinode_t, di_next_unlinked);
1833
1834 while (next_unlinked_offset >=
1835 (reg_buf_offset + reg_buf_bytes)) {
1836 /*
1837 * The next di_next_unlinked field is beyond
1838 * the current logged region. Find the next
1839 * logged region that contains or is beyond
1840 * the current di_next_unlinked field.
1841 */
1842 bit += nbits;
1843 bit = xfs_next_bit(data_map, map_size, bit);
1844
1845 /*
1846 * If there are no more logged regions in the
1847 * buffer, then we're done.
1848 */
1849 if (bit == -1) {
1850 return 0;
1851 }
1852
1853 nbits = xfs_contig_bits(data_map, map_size,
1854 bit);
1855 ASSERT(nbits > 0);
1856 reg_buf_offset = bit << XFS_BLI_SHIFT;
1857 reg_buf_bytes = nbits << XFS_BLI_SHIFT;
1858 item_index++;
1859 }
1860
1861 /*
1862 * If the current logged region starts after the current
1863 * di_next_unlinked field, then move on to the next
1864 * di_next_unlinked field.
1865 */
1866 if (next_unlinked_offset < reg_buf_offset) {
1867 continue;
1868 }
1869
1870 ASSERT(item->ri_buf[item_index].i_addr != NULL);
1871 ASSERT((item->ri_buf[item_index].i_len % XFS_BLI_CHUNK) == 0);
1872 ASSERT((reg_buf_offset + reg_buf_bytes) <= XFS_BUF_COUNT(bp));
1873
1874 /*
1875 * The current logged region contains a copy of the
1876 * current di_next_unlinked field. Extract its value
1877 * and copy it to the buffer copy.
1878 */
1879 logged_nextp = (xfs_agino_t *)
1880 ((char *)(item->ri_buf[item_index].i_addr) +
1881 (next_unlinked_offset - reg_buf_offset));
1882 if (unlikely(*logged_nextp == 0)) {
1883 xfs_fs_cmn_err(CE_ALERT, mp,
1884 "bad inode buffer log record (ptr = 0x%p, bp = 0x%p). XFS trying to replay bad (0) inode di_next_unlinked field",
1885 item, bp);
1886 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1887 XFS_ERRLEVEL_LOW, mp);
1888 return XFS_ERROR(EFSCORRUPTED);
1889 }
1890
1891 buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
1892 next_unlinked_offset);
Tim Shimmin87c199c2006-06-09 14:56:16 +10001893 *buffer_nextp = *logged_nextp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 }
1895
1896 return 0;
1897}
1898
1899/*
1900 * Perform a 'normal' buffer recovery. Each logged region of the
1901 * buffer should be copied over the corresponding region in the
1902 * given buffer. The bitmap in the buf log format structure indicates
1903 * where to place the logged data.
1904 */
1905/*ARGSUSED*/
1906STATIC void
1907xlog_recover_do_reg_buffer(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908 xlog_recover_item_t *item,
1909 xfs_buf_t *bp,
1910 xfs_buf_log_format_t *buf_f)
1911{
1912 int i;
1913 int bit;
1914 int nbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915 unsigned int *data_map = NULL;
1916 unsigned int map_size = 0;
1917 int error;
1918
1919 switch (buf_f->blf_type) {
1920 case XFS_LI_BUF:
1921 data_map = buf_f->blf_data_map;
1922 map_size = buf_f->blf_map_size;
1923 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 }
1925 bit = 0;
1926 i = 1; /* 0 is the buf format structure */
1927 while (1) {
1928 bit = xfs_next_bit(data_map, map_size, bit);
1929 if (bit == -1)
1930 break;
1931 nbits = xfs_contig_bits(data_map, map_size, bit);
1932 ASSERT(nbits > 0);
Christoph Hellwig4b809162007-08-16 15:37:36 +10001933 ASSERT(item->ri_buf[i].i_addr != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 ASSERT(item->ri_buf[i].i_len % XFS_BLI_CHUNK == 0);
1935 ASSERT(XFS_BUF_COUNT(bp) >=
1936 ((uint)bit << XFS_BLI_SHIFT)+(nbits<<XFS_BLI_SHIFT));
1937
1938 /*
1939 * Do a sanity check if this is a dquot buffer. Just checking
1940 * the first dquot in the buffer should do. XXXThis is
1941 * probably a good thing to do for other buf types also.
1942 */
1943 error = 0;
Nathan Scottc8ad20f2005-06-21 15:38:48 +10001944 if (buf_f->blf_flags &
1945 (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 error = xfs_qm_dqcheck((xfs_disk_dquot_t *)
1947 item->ri_buf[i].i_addr,
1948 -1, 0, XFS_QMOPT_DOWARN,
1949 "dquot_buf_recover");
1950 }
Tim Shimmin053c59a2007-10-01 16:39:37 +10001951 if (!error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 memcpy(xfs_buf_offset(bp,
1953 (uint)bit << XFS_BLI_SHIFT), /* dest */
1954 item->ri_buf[i].i_addr, /* source */
1955 nbits<<XFS_BLI_SHIFT); /* length */
1956 i++;
1957 bit += nbits;
1958 }
1959
1960 /* Shouldn't be any more regions */
1961 ASSERT(i == item->ri_total);
1962}
1963
1964/*
1965 * Do some primitive error checking on ondisk dquot data structures.
1966 */
1967int
1968xfs_qm_dqcheck(
1969 xfs_disk_dquot_t *ddq,
1970 xfs_dqid_t id,
1971 uint type, /* used only when IO_dorepair is true */
1972 uint flags,
1973 char *str)
1974{
1975 xfs_dqblk_t *d = (xfs_dqblk_t *)ddq;
1976 int errs = 0;
1977
1978 /*
1979 * We can encounter an uninitialized dquot buffer for 2 reasons:
1980 * 1. If we crash while deleting the quotainode(s), and those blks got
1981 * used for user data. This is because we take the path of regular
1982 * file deletion; however, the size field of quotainodes is never
1983 * updated, so all the tricks that we play in itruncate_finish
1984 * don't quite matter.
1985 *
1986 * 2. We don't play the quota buffers when there's a quotaoff logitem.
1987 * But the allocation will be replayed so we'll end up with an
1988 * uninitialized quota block.
1989 *
1990 * This is all fine; things are still consistent, and we haven't lost
1991 * any quota information. Just don't complain about bad dquot blks.
1992 */
Christoph Hellwig1149d962005-11-02 15:01:12 +11001993 if (be16_to_cpu(ddq->d_magic) != XFS_DQUOT_MAGIC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 if (flags & XFS_QMOPT_DOWARN)
1995 cmn_err(CE_ALERT,
1996 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
Christoph Hellwig1149d962005-11-02 15:01:12 +11001997 str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 errs++;
1999 }
Christoph Hellwig1149d962005-11-02 15:01:12 +11002000 if (ddq->d_version != XFS_DQUOT_VERSION) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 if (flags & XFS_QMOPT_DOWARN)
2002 cmn_err(CE_ALERT,
2003 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002004 str, id, ddq->d_version, XFS_DQUOT_VERSION);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005 errs++;
2006 }
2007
Christoph Hellwig1149d962005-11-02 15:01:12 +11002008 if (ddq->d_flags != XFS_DQ_USER &&
2009 ddq->d_flags != XFS_DQ_PROJ &&
2010 ddq->d_flags != XFS_DQ_GROUP) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011 if (flags & XFS_QMOPT_DOWARN)
2012 cmn_err(CE_ALERT,
2013 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002014 str, id, ddq->d_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 errs++;
2016 }
2017
Christoph Hellwig1149d962005-11-02 15:01:12 +11002018 if (id != -1 && id != be32_to_cpu(ddq->d_id)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 if (flags & XFS_QMOPT_DOWARN)
2020 cmn_err(CE_ALERT,
2021 "%s : ondisk-dquot 0x%p, ID mismatch: "
2022 "0x%x expected, found id 0x%x",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002023 str, ddq, id, be32_to_cpu(ddq->d_id));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024 errs++;
2025 }
2026
2027 if (!errs && ddq->d_id) {
Christoph Hellwig1149d962005-11-02 15:01:12 +11002028 if (ddq->d_blk_softlimit &&
2029 be64_to_cpu(ddq->d_bcount) >=
2030 be64_to_cpu(ddq->d_blk_softlimit)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 if (!ddq->d_btimer) {
2032 if (flags & XFS_QMOPT_DOWARN)
2033 cmn_err(CE_ALERT,
2034 "%s : Dquot ID 0x%x (0x%p) "
2035 "BLK TIMER NOT STARTED",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002036 str, (int)be32_to_cpu(ddq->d_id), ddq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037 errs++;
2038 }
2039 }
Christoph Hellwig1149d962005-11-02 15:01:12 +11002040 if (ddq->d_ino_softlimit &&
2041 be64_to_cpu(ddq->d_icount) >=
2042 be64_to_cpu(ddq->d_ino_softlimit)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 if (!ddq->d_itimer) {
2044 if (flags & XFS_QMOPT_DOWARN)
2045 cmn_err(CE_ALERT,
2046 "%s : Dquot ID 0x%x (0x%p) "
2047 "INODE TIMER NOT STARTED",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002048 str, (int)be32_to_cpu(ddq->d_id), ddq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 errs++;
2050 }
2051 }
Christoph Hellwig1149d962005-11-02 15:01:12 +11002052 if (ddq->d_rtb_softlimit &&
2053 be64_to_cpu(ddq->d_rtbcount) >=
2054 be64_to_cpu(ddq->d_rtb_softlimit)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 if (!ddq->d_rtbtimer) {
2056 if (flags & XFS_QMOPT_DOWARN)
2057 cmn_err(CE_ALERT,
2058 "%s : Dquot ID 0x%x (0x%p) "
2059 "RTBLK TIMER NOT STARTED",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002060 str, (int)be32_to_cpu(ddq->d_id), ddq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 errs++;
2062 }
2063 }
2064 }
2065
2066 if (!errs || !(flags & XFS_QMOPT_DQREPAIR))
2067 return errs;
2068
2069 if (flags & XFS_QMOPT_DOWARN)
2070 cmn_err(CE_NOTE, "Re-initializing dquot ID 0x%x", id);
2071
2072 /*
2073 * Typically, a repair is only requested by quotacheck.
2074 */
2075 ASSERT(id != -1);
2076 ASSERT(flags & XFS_QMOPT_DQREPAIR);
2077 memset(d, 0, sizeof(xfs_dqblk_t));
Christoph Hellwig1149d962005-11-02 15:01:12 +11002078
2079 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
2080 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
2081 d->dd_diskdq.d_flags = type;
2082 d->dd_diskdq.d_id = cpu_to_be32(id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083
2084 return errs;
2085}
2086
2087/*
2088 * Perform a dquot buffer recovery.
2089 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2090 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2091 * Else, treat it as a regular buffer and do recovery.
2092 */
2093STATIC void
2094xlog_recover_do_dquot_buffer(
2095 xfs_mount_t *mp,
2096 xlog_t *log,
2097 xlog_recover_item_t *item,
2098 xfs_buf_t *bp,
2099 xfs_buf_log_format_t *buf_f)
2100{
2101 uint type;
2102
2103 /*
2104 * Filesystems are required to send in quota flags at mount time.
2105 */
2106 if (mp->m_qflags == 0) {
2107 return;
2108 }
2109
2110 type = 0;
2111 if (buf_f->blf_flags & XFS_BLI_UDQUOT_BUF)
2112 type |= XFS_DQ_USER;
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002113 if (buf_f->blf_flags & XFS_BLI_PDQUOT_BUF)
2114 type |= XFS_DQ_PROJ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 if (buf_f->blf_flags & XFS_BLI_GDQUOT_BUF)
2116 type |= XFS_DQ_GROUP;
2117 /*
2118 * This type of quotas was turned off, so ignore this buffer
2119 */
2120 if (log->l_quotaoffs_flag & type)
2121 return;
2122
Tim Shimmin053c59a2007-10-01 16:39:37 +10002123 xlog_recover_do_reg_buffer(item, bp, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124}
2125
2126/*
2127 * This routine replays a modification made to a buffer at runtime.
2128 * There are actually two types of buffer, regular and inode, which
2129 * are handled differently. Inode buffers are handled differently
2130 * in that we only recover a specific set of data from them, namely
2131 * the inode di_next_unlinked fields. This is because all other inode
2132 * data is actually logged via inode records and any data we replay
2133 * here which overlaps that may be stale.
2134 *
2135 * When meta-data buffers are freed at run time we log a buffer item
2136 * with the XFS_BLI_CANCEL bit set to indicate that previous copies
2137 * of the buffer in the log should not be replayed at recovery time.
2138 * This is so that if the blocks covered by the buffer are reused for
2139 * file data before we crash we don't end up replaying old, freed
2140 * meta-data into a user's file.
2141 *
2142 * To handle the cancellation of buffer log items, we make two passes
2143 * over the log during recovery. During the first we build a table of
2144 * those buffers which have been cancelled, and during the second we
2145 * only replay those buffers which do not have corresponding cancel
2146 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2147 * for more details on the implementation of the table of cancel records.
2148 */
2149STATIC int
2150xlog_recover_do_buffer_trans(
2151 xlog_t *log,
2152 xlog_recover_item_t *item,
2153 int pass)
2154{
2155 xfs_buf_log_format_t *buf_f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156 xfs_mount_t *mp;
2157 xfs_buf_t *bp;
2158 int error;
2159 int cancel;
2160 xfs_daddr_t blkno;
2161 int len;
2162 ushort flags;
2163
2164 buf_f = (xfs_buf_log_format_t *)item->ri_buf[0].i_addr;
2165
2166 if (pass == XLOG_RECOVER_PASS1) {
2167 /*
2168 * In this pass we're only looking for buf items
2169 * with the XFS_BLI_CANCEL bit set.
2170 */
2171 xlog_recover_do_buffer_pass1(log, buf_f);
2172 return 0;
2173 } else {
2174 /*
2175 * In this pass we want to recover all the buffers
2176 * which have not been cancelled and are not
2177 * cancellation buffers themselves. The routine
2178 * we call here will tell us whether or not to
2179 * continue with the replay of this buffer.
2180 */
2181 cancel = xlog_recover_do_buffer_pass2(log, buf_f);
2182 if (cancel) {
2183 return 0;
2184 }
2185 }
2186 switch (buf_f->blf_type) {
2187 case XFS_LI_BUF:
2188 blkno = buf_f->blf_blkno;
2189 len = buf_f->blf_len;
2190 flags = buf_f->blf_flags;
2191 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 default:
2193 xfs_fs_cmn_err(CE_ALERT, log->l_mp,
Nathan Scottfc1f8c12005-11-02 11:44:33 +11002194 "xfs_log_recover: unknown buffer type 0x%x, logdev %s",
2195 buf_f->blf_type, log->l_mp->m_logname ?
2196 log->l_mp->m_logname : "internal");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197 XFS_ERROR_REPORT("xlog_recover_do_buffer_trans",
2198 XFS_ERRLEVEL_LOW, log->l_mp);
2199 return XFS_ERROR(EFSCORRUPTED);
2200 }
2201
2202 mp = log->l_mp;
2203 if (flags & XFS_BLI_INODE_BUF) {
2204 bp = xfs_buf_read_flags(mp->m_ddev_targp, blkno, len,
2205 XFS_BUF_LOCK);
2206 } else {
2207 bp = xfs_buf_read(mp->m_ddev_targp, blkno, len, 0);
2208 }
2209 if (XFS_BUF_ISERROR(bp)) {
2210 xfs_ioerror_alert("xlog_recover_do..(read#1)", log->l_mp,
2211 bp, blkno);
2212 error = XFS_BUF_GETERROR(bp);
2213 xfs_buf_relse(bp);
2214 return error;
2215 }
2216
2217 error = 0;
2218 if (flags & XFS_BLI_INODE_BUF) {
2219 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002220 } else if (flags &
2221 (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2223 } else {
Tim Shimmin053c59a2007-10-01 16:39:37 +10002224 xlog_recover_do_reg_buffer(item, bp, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225 }
2226 if (error)
2227 return XFS_ERROR(error);
2228
2229 /*
2230 * Perform delayed write on the buffer. Asynchronous writes will be
2231 * slower when taking into account all the buffers to be flushed.
2232 *
2233 * Also make sure that only inode buffers with good sizes stay in
2234 * the buffer cache. The kernel moves inodes in buffers of 1 block
2235 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2236 * buffers in the log can be a different size if the log was generated
2237 * by an older kernel using unclustered inode buffers or a newer kernel
2238 * running with a different inode cluster size. Regardless, if the
2239 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2240 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2241 * the buffer out of the buffer cache so that the buffer won't
2242 * overlap with future reads of those inodes.
2243 */
2244 if (XFS_DINODE_MAGIC ==
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002245 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 (XFS_BUF_COUNT(bp) != MAX(log->l_mp->m_sb.sb_blocksize,
2247 (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) {
2248 XFS_BUF_STALE(bp);
2249 error = xfs_bwrite(mp, bp);
2250 } else {
Christoph Hellwig15ac08a2008-12-09 04:47:30 -05002251 ASSERT(bp->b_mount == NULL || bp->b_mount == mp);
2252 bp->b_mount = mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2254 xfs_bdwrite(mp, bp);
2255 }
2256
2257 return (error);
2258}
2259
2260STATIC int
2261xlog_recover_do_inode_trans(
2262 xlog_t *log,
2263 xlog_recover_item_t *item,
2264 int pass)
2265{
2266 xfs_inode_log_format_t *in_f;
2267 xfs_mount_t *mp;
2268 xfs_buf_t *bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269 xfs_dinode_t *dip;
2270 xfs_ino_t ino;
2271 int len;
2272 xfs_caddr_t src;
2273 xfs_caddr_t dest;
2274 int error;
2275 int attr_index;
2276 uint fields;
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002277 xfs_icdinode_t *dicp;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002278 int need_free = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279
2280 if (pass == XLOG_RECOVER_PASS1) {
2281 return 0;
2282 }
2283
Tim Shimmin6d192a92006-06-09 14:55:38 +10002284 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
2285 in_f = (xfs_inode_log_format_t *)item->ri_buf[0].i_addr;
2286 } else {
2287 in_f = (xfs_inode_log_format_t *)kmem_alloc(
2288 sizeof(xfs_inode_log_format_t), KM_SLEEP);
2289 need_free = 1;
2290 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2291 if (error)
2292 goto error;
2293 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 ino = in_f->ilf_ino;
2295 mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296
2297 /*
2298 * Inode buffers can be freed, look out for it,
2299 * and do not replay the inode.
2300 */
Christoph Hellwiga1941892008-11-28 14:23:40 +11002301 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
2302 in_f->ilf_len, 0)) {
Tim Shimmin6d192a92006-06-09 14:55:38 +10002303 error = 0;
2304 goto error;
2305 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306
Christoph Hellwiga1941892008-11-28 14:23:40 +11002307 bp = xfs_buf_read_flags(mp->m_ddev_targp, in_f->ilf_blkno,
2308 in_f->ilf_len, XFS_BUF_LOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309 if (XFS_BUF_ISERROR(bp)) {
2310 xfs_ioerror_alert("xlog_recover_do..(read#2)", mp,
Christoph Hellwiga1941892008-11-28 14:23:40 +11002311 bp, in_f->ilf_blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312 error = XFS_BUF_GETERROR(bp);
2313 xfs_buf_relse(bp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002314 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 }
2316 error = 0;
2317 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
Christoph Hellwiga1941892008-11-28 14:23:40 +11002318 dip = (xfs_dinode_t *)xfs_buf_offset(bp, in_f->ilf_boffset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319
2320 /*
2321 * Make sure the place we're flushing out to really looks
2322 * like an inode!
2323 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002324 if (unlikely(be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325 xfs_buf_relse(bp);
2326 xfs_fs_cmn_err(CE_ALERT, mp,
2327 "xfs_inode_recover: Bad inode magic number, dino ptr = 0x%p, dino bp = 0x%p, ino = %Ld",
2328 dip, bp, ino);
2329 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(1)",
2330 XFS_ERRLEVEL_LOW, mp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002331 error = EFSCORRUPTED;
2332 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 }
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002334 dicp = (xfs_icdinode_t *)(item->ri_buf[1].i_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
2336 xfs_buf_relse(bp);
2337 xfs_fs_cmn_err(CE_ALERT, mp,
2338 "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, ino %Ld",
2339 item, ino);
2340 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(2)",
2341 XFS_ERRLEVEL_LOW, mp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002342 error = EFSCORRUPTED;
2343 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344 }
2345
2346 /* Skip replay when the on disk inode is newer than the log one */
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002347 if (dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348 /*
2349 * Deal with the wrap case, DI_MAX_FLUSH is less
2350 * than smaller numbers
2351 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002352 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002353 dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354 /* do nothing */
2355 } else {
2356 xfs_buf_relse(bp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002357 error = 0;
2358 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359 }
2360 }
2361 /* Take the opportunity to reset the flush iteration count */
2362 dicp->di_flushiter = 0;
2363
2364 if (unlikely((dicp->di_mode & S_IFMT) == S_IFREG)) {
2365 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2366 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
2367 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(3)",
2368 XFS_ERRLEVEL_LOW, mp, dicp);
2369 xfs_buf_relse(bp);
2370 xfs_fs_cmn_err(CE_ALERT, mp,
2371 "xfs_inode_recover: Bad regular inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2372 item, dip, bp, ino);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002373 error = EFSCORRUPTED;
2374 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375 }
2376 } else if (unlikely((dicp->di_mode & S_IFMT) == S_IFDIR)) {
2377 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2378 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2379 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
2380 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(4)",
2381 XFS_ERRLEVEL_LOW, mp, dicp);
2382 xfs_buf_relse(bp);
2383 xfs_fs_cmn_err(CE_ALERT, mp,
2384 "xfs_inode_recover: Bad dir inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2385 item, dip, bp, ino);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002386 error = EFSCORRUPTED;
2387 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388 }
2389 }
2390 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
2391 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(5)",
2392 XFS_ERRLEVEL_LOW, mp, dicp);
2393 xfs_buf_relse(bp);
2394 xfs_fs_cmn_err(CE_ALERT, mp,
2395 "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2396 item, dip, bp, ino,
2397 dicp->di_nextents + dicp->di_anextents,
2398 dicp->di_nblocks);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002399 error = EFSCORRUPTED;
2400 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401 }
2402 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
2403 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(6)",
2404 XFS_ERRLEVEL_LOW, mp, dicp);
2405 xfs_buf_relse(bp);
2406 xfs_fs_cmn_err(CE_ALERT, mp,
2407 "xfs_inode_recover: Bad inode log rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, forkoff 0x%x",
2408 item, dip, bp, ino, dicp->di_forkoff);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002409 error = EFSCORRUPTED;
2410 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411 }
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002412 if (unlikely(item->ri_buf[1].i_len > sizeof(struct xfs_icdinode))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(7)",
2414 XFS_ERRLEVEL_LOW, mp, dicp);
2415 xfs_buf_relse(bp);
2416 xfs_fs_cmn_err(CE_ALERT, mp,
2417 "xfs_inode_recover: Bad inode log record length %d, rec ptr 0x%p",
2418 item->ri_buf[1].i_len, item);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002419 error = EFSCORRUPTED;
2420 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 }
2422
2423 /* The core is in in-core format */
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002424 xfs_dinode_to_disk(dip, (xfs_icdinode_t *)item->ri_buf[1].i_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425
2426 /* the rest is in on-disk format */
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002427 if (item->ri_buf[1].i_len > sizeof(struct xfs_icdinode)) {
2428 memcpy((xfs_caddr_t) dip + sizeof(struct xfs_icdinode),
2429 item->ri_buf[1].i_addr + sizeof(struct xfs_icdinode),
2430 item->ri_buf[1].i_len - sizeof(struct xfs_icdinode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431 }
2432
2433 fields = in_f->ilf_fields;
2434 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2435 case XFS_ILOG_DEV:
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002436 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437 break;
2438 case XFS_ILOG_UUID:
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002439 memcpy(XFS_DFORK_DPTR(dip),
2440 &in_f->ilf_u.ilfu_uuid,
2441 sizeof(uuid_t));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442 break;
2443 }
2444
2445 if (in_f->ilf_size == 2)
2446 goto write_inode_buffer;
2447 len = item->ri_buf[2].i_len;
2448 src = item->ri_buf[2].i_addr;
2449 ASSERT(in_f->ilf_size <= 4);
2450 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2451 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2452 (len == in_f->ilf_dsize));
2453
2454 switch (fields & XFS_ILOG_DFORK) {
2455 case XFS_ILOG_DDATA:
2456 case XFS_ILOG_DEXT:
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002457 memcpy(XFS_DFORK_DPTR(dip), src, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458 break;
2459
2460 case XFS_ILOG_DBROOT:
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002461 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002462 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463 XFS_DFORK_DSIZE(dip, mp));
2464 break;
2465
2466 default:
2467 /*
2468 * There are no data fork flags set.
2469 */
2470 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2471 break;
2472 }
2473
2474 /*
2475 * If we logged any attribute data, recover it. There may or
2476 * may not have been any other non-core data logged in this
2477 * transaction.
2478 */
2479 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2480 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2481 attr_index = 3;
2482 } else {
2483 attr_index = 2;
2484 }
2485 len = item->ri_buf[attr_index].i_len;
2486 src = item->ri_buf[attr_index].i_addr;
2487 ASSERT(len == in_f->ilf_asize);
2488
2489 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2490 case XFS_ILOG_ADATA:
2491 case XFS_ILOG_AEXT:
2492 dest = XFS_DFORK_APTR(dip);
2493 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2494 memcpy(dest, src, len);
2495 break;
2496
2497 case XFS_ILOG_ABROOT:
2498 dest = XFS_DFORK_APTR(dip);
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002499 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
2500 len, (xfs_bmdr_block_t*)dest,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002501 XFS_DFORK_ASIZE(dip, mp));
2502 break;
2503
2504 default:
2505 xlog_warn("XFS: xlog_recover_do_inode_trans: Invalid flag");
2506 ASSERT(0);
2507 xfs_buf_relse(bp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002508 error = EIO;
2509 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510 }
2511 }
2512
2513write_inode_buffer:
2514 if (ITEM_TYPE(item) == XFS_LI_INODE) {
Christoph Hellwig15ac08a2008-12-09 04:47:30 -05002515 ASSERT(bp->b_mount == NULL || bp->b_mount == mp);
2516 bp->b_mount = mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2518 xfs_bdwrite(mp, bp);
2519 } else {
2520 XFS_BUF_STALE(bp);
2521 error = xfs_bwrite(mp, bp);
2522 }
2523
Tim Shimmin6d192a92006-06-09 14:55:38 +10002524error:
2525 if (need_free)
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002526 kmem_free(in_f);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002527 return XFS_ERROR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528}
2529
2530/*
2531 * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
2532 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2533 * of that type.
2534 */
2535STATIC int
2536xlog_recover_do_quotaoff_trans(
2537 xlog_t *log,
2538 xlog_recover_item_t *item,
2539 int pass)
2540{
2541 xfs_qoff_logformat_t *qoff_f;
2542
2543 if (pass == XLOG_RECOVER_PASS2) {
2544 return (0);
2545 }
2546
2547 qoff_f = (xfs_qoff_logformat_t *)item->ri_buf[0].i_addr;
2548 ASSERT(qoff_f);
2549
2550 /*
2551 * The logitem format's flag tells us if this was user quotaoff,
Nathan Scott77a7cce2006-01-11 15:35:57 +11002552 * group/project quotaoff or both.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553 */
2554 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2555 log->l_quotaoffs_flag |= XFS_DQ_USER;
Nathan Scott77a7cce2006-01-11 15:35:57 +11002556 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2557 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2559 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2560
2561 return (0);
2562}
2563
2564/*
2565 * Recover a dquot record
2566 */
2567STATIC int
2568xlog_recover_do_dquot_trans(
2569 xlog_t *log,
2570 xlog_recover_item_t *item,
2571 int pass)
2572{
2573 xfs_mount_t *mp;
2574 xfs_buf_t *bp;
2575 struct xfs_disk_dquot *ddq, *recddq;
2576 int error;
2577 xfs_dq_logformat_t *dq_f;
2578 uint type;
2579
2580 if (pass == XLOG_RECOVER_PASS1) {
2581 return 0;
2582 }
2583 mp = log->l_mp;
2584
2585 /*
2586 * Filesystems are required to send in quota flags at mount time.
2587 */
2588 if (mp->m_qflags == 0)
2589 return (0);
2590
2591 recddq = (xfs_disk_dquot_t *)item->ri_buf[1].i_addr;
2592 ASSERT(recddq);
2593 /*
2594 * This type of quotas was turned off, so ignore this record.
2595 */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002596 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597 ASSERT(type);
2598 if (log->l_quotaoffs_flag & type)
2599 return (0);
2600
2601 /*
2602 * At this point we know that quota was _not_ turned off.
2603 * Since the mount flags are not indicating to us otherwise, this
2604 * must mean that quota is on, and the dquot needs to be replayed.
2605 * Remember that we may not have fully recovered the superblock yet,
2606 * so we can't do the usual trick of looking at the SB quota bits.
2607 *
2608 * The other possibility, of course, is that the quota subsystem was
2609 * removed since the last mount - ENOSYS.
2610 */
2611 dq_f = (xfs_dq_logformat_t *)item->ri_buf[0].i_addr;
2612 ASSERT(dq_f);
2613 if ((error = xfs_qm_dqcheck(recddq,
2614 dq_f->qlf_id,
2615 0, XFS_QMOPT_DOWARN,
2616 "xlog_recover_do_dquot_trans (log copy)"))) {
2617 return XFS_ERROR(EIO);
2618 }
2619 ASSERT(dq_f->qlf_len == 1);
2620
2621 error = xfs_read_buf(mp, mp->m_ddev_targp,
2622 dq_f->qlf_blkno,
2623 XFS_FSB_TO_BB(mp, dq_f->qlf_len),
2624 0, &bp);
2625 if (error) {
2626 xfs_ioerror_alert("xlog_recover_do..(read#3)", mp,
2627 bp, dq_f->qlf_blkno);
2628 return error;
2629 }
2630 ASSERT(bp);
2631 ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
2632
2633 /*
2634 * At least the magic num portion should be on disk because this
2635 * was among a chunk of dquots created earlier, and we did some
2636 * minimal initialization then.
2637 */
2638 if (xfs_qm_dqcheck(ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2639 "xlog_recover_do_dquot_trans")) {
2640 xfs_buf_relse(bp);
2641 return XFS_ERROR(EIO);
2642 }
2643
2644 memcpy(ddq, recddq, item->ri_buf[1].i_len);
2645
2646 ASSERT(dq_f->qlf_size == 2);
Christoph Hellwig15ac08a2008-12-09 04:47:30 -05002647 ASSERT(bp->b_mount == NULL || bp->b_mount == mp);
2648 bp->b_mount = mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649 XFS_BUF_SET_IODONE_FUNC(bp, xlog_recover_iodone);
2650 xfs_bdwrite(mp, bp);
2651
2652 return (0);
2653}
2654
2655/*
2656 * This routine is called to create an in-core extent free intent
2657 * item from the efi format structure which was logged on disk.
2658 * It allocates an in-core efi, copies the extents from the format
2659 * structure into it, and adds the efi to the AIL with the given
2660 * LSN.
2661 */
Tim Shimmin6d192a92006-06-09 14:55:38 +10002662STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663xlog_recover_do_efi_trans(
2664 xlog_t *log,
2665 xlog_recover_item_t *item,
2666 xfs_lsn_t lsn,
2667 int pass)
2668{
Tim Shimmin6d192a92006-06-09 14:55:38 +10002669 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670 xfs_mount_t *mp;
2671 xfs_efi_log_item_t *efip;
2672 xfs_efi_log_format_t *efi_formatp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673
2674 if (pass == XLOG_RECOVER_PASS1) {
Tim Shimmin6d192a92006-06-09 14:55:38 +10002675 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676 }
2677
2678 efi_formatp = (xfs_efi_log_format_t *)item->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679
2680 mp = log->l_mp;
2681 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002682 if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
2683 &(efip->efi_format)))) {
2684 xfs_efi_item_free(efip);
2685 return error;
2686 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687 efip->efi_next_extent = efi_formatp->efi_nextents;
2688 efip->efi_flags |= XFS_EFI_COMMITTED;
2689
David Chinnera9c21c12008-10-30 17:39:35 +11002690 spin_lock(&log->l_ailp->xa_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691 /*
David Chinner783a2f62008-10-30 17:39:58 +11002692 * xfs_trans_ail_update() drops the AIL lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693 */
David Chinner783a2f62008-10-30 17:39:58 +11002694 xfs_trans_ail_update(log->l_ailp, (xfs_log_item_t *)efip, lsn);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002695 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696}
2697
2698
2699/*
2700 * This routine is called when an efd format structure is found in
2701 * a committed transaction in the log. It's purpose is to cancel
2702 * the corresponding efi if it was still in the log. To do this
2703 * it searches the AIL for the efi with an id equal to that in the
2704 * efd format structure. If we find it, we remove the efi from the
2705 * AIL and free it.
2706 */
2707STATIC void
2708xlog_recover_do_efd_trans(
2709 xlog_t *log,
2710 xlog_recover_item_t *item,
2711 int pass)
2712{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713 xfs_efd_log_format_t *efd_formatp;
2714 xfs_efi_log_item_t *efip = NULL;
2715 xfs_log_item_t *lip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716 __uint64_t efi_id;
David Chinner27d8d5f2008-10-30 17:38:39 +11002717 struct xfs_ail_cursor cur;
David Chinner783a2f62008-10-30 17:39:58 +11002718 struct xfs_ail *ailp = log->l_ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002719
2720 if (pass == XLOG_RECOVER_PASS1) {
2721 return;
2722 }
2723
2724 efd_formatp = (xfs_efd_log_format_t *)item->ri_buf[0].i_addr;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002725 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
2726 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
2727 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
2728 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729 efi_id = efd_formatp->efd_efi_id;
2730
2731 /*
2732 * Search for the efi with the id in the efd format structure
2733 * in the AIL.
2734 */
David Chinnera9c21c12008-10-30 17:39:35 +11002735 spin_lock(&ailp->xa_lock);
2736 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737 while (lip != NULL) {
2738 if (lip->li_type == XFS_LI_EFI) {
2739 efip = (xfs_efi_log_item_t *)lip;
2740 if (efip->efi_format.efi_id == efi_id) {
2741 /*
David Chinner783a2f62008-10-30 17:39:58 +11002742 * xfs_trans_ail_delete() drops the
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743 * AIL lock.
2744 */
David Chinner783a2f62008-10-30 17:39:58 +11002745 xfs_trans_ail_delete(ailp, lip);
David Chinner8ae2c0f2007-11-23 16:28:17 +11002746 xfs_efi_item_free(efip);
David Chinnera9c21c12008-10-30 17:39:35 +11002747 spin_lock(&ailp->xa_lock);
David Chinner27d8d5f2008-10-30 17:38:39 +11002748 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002749 }
2750 }
David Chinnera9c21c12008-10-30 17:39:35 +11002751 lip = xfs_trans_ail_cursor_next(ailp, &cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752 }
David Chinnera9c21c12008-10-30 17:39:35 +11002753 xfs_trans_ail_cursor_done(ailp, &cur);
2754 spin_unlock(&ailp->xa_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755}
2756
2757/*
2758 * Perform the transaction
2759 *
2760 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2761 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2762 */
2763STATIC int
2764xlog_recover_do_trans(
2765 xlog_t *log,
2766 xlog_recover_t *trans,
2767 int pass)
2768{
2769 int error = 0;
2770 xlog_recover_item_t *item, *first_item;
2771
Eric Sandeene9ed9d22007-05-08 13:48:56 +10002772 if ((error = xlog_recover_reorder_trans(trans)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773 return error;
2774 first_item = item = trans->r_itemq;
2775 do {
2776 /*
2777 * we don't need to worry about the block number being
2778 * truncated in > 1 TB buffers because in user-land,
2779 * we're now n32 or 64-bit so xfs_daddr_t is 64-bits so
Nathan Scottc41564b2006-03-29 08:55:14 +10002780 * the blknos will get through the user-mode buffer
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781 * cache properly. The only bad case is o32 kernels
2782 * where xfs_daddr_t is 32-bits but mount will warn us
2783 * off a > 1 TB filesystem before we get here.
2784 */
Eric Sandeen804195b2007-02-10 18:35:02 +11002785 if ((ITEM_TYPE(item) == XFS_LI_BUF)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786 if ((error = xlog_recover_do_buffer_trans(log, item,
2787 pass)))
2788 break;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002789 } else if ((ITEM_TYPE(item) == XFS_LI_INODE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002790 if ((error = xlog_recover_do_inode_trans(log, item,
2791 pass)))
2792 break;
2793 } else if (ITEM_TYPE(item) == XFS_LI_EFI) {
Tim Shimmin6d192a92006-06-09 14:55:38 +10002794 if ((error = xlog_recover_do_efi_trans(log, item, trans->r_lsn,
2795 pass)))
2796 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797 } else if (ITEM_TYPE(item) == XFS_LI_EFD) {
2798 xlog_recover_do_efd_trans(log, item, pass);
2799 } else if (ITEM_TYPE(item) == XFS_LI_DQUOT) {
2800 if ((error = xlog_recover_do_dquot_trans(log, item,
2801 pass)))
2802 break;
2803 } else if ((ITEM_TYPE(item) == XFS_LI_QUOTAOFF)) {
2804 if ((error = xlog_recover_do_quotaoff_trans(log, item,
2805 pass)))
2806 break;
2807 } else {
2808 xlog_warn("XFS: xlog_recover_do_trans");
2809 ASSERT(0);
2810 error = XFS_ERROR(EIO);
2811 break;
2812 }
2813 item = item->ri_next;
2814 } while (first_item != item);
2815
2816 return error;
2817}
2818
2819/*
2820 * Free up any resources allocated by the transaction
2821 *
2822 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2823 */
2824STATIC void
2825xlog_recover_free_trans(
2826 xlog_recover_t *trans)
2827{
2828 xlog_recover_item_t *first_item, *item, *free_item;
2829 int i;
2830
2831 item = first_item = trans->r_itemq;
2832 do {
2833 free_item = item;
2834 item = item->ri_next;
2835 /* Free the regions in the item. */
2836 for (i = 0; i < free_item->ri_cnt; i++) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002837 kmem_free(free_item->ri_buf[i].i_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838 }
2839 /* Free the item itself */
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002840 kmem_free(free_item->ri_buf);
2841 kmem_free(free_item);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002842 } while (first_item != item);
2843 /* Free the transaction recover structure */
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002844 kmem_free(trans);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845}
2846
2847STATIC int
2848xlog_recover_commit_trans(
2849 xlog_t *log,
2850 xlog_recover_t **q,
2851 xlog_recover_t *trans,
2852 int pass)
2853{
2854 int error;
2855
2856 if ((error = xlog_recover_unlink_tid(q, trans)))
2857 return error;
2858 if ((error = xlog_recover_do_trans(log, trans, pass)))
2859 return error;
2860 xlog_recover_free_trans(trans); /* no error */
2861 return 0;
2862}
2863
2864STATIC int
2865xlog_recover_unmount_trans(
2866 xlog_recover_t *trans)
2867{
2868 /* Do nothing now */
2869 xlog_warn("XFS: xlog_recover_unmount_trans: Unmount LR");
2870 return 0;
2871}
2872
2873/*
2874 * There are two valid states of the r_state field. 0 indicates that the
2875 * transaction structure is in a normal state. We have either seen the
2876 * start of the transaction or the last operation we added was not a partial
2877 * operation. If the last operation we added to the transaction was a
2878 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2879 *
2880 * NOTE: skip LRs with 0 data length.
2881 */
2882STATIC int
2883xlog_recover_process_data(
2884 xlog_t *log,
2885 xlog_recover_t *rhash[],
2886 xlog_rec_header_t *rhead,
2887 xfs_caddr_t dp,
2888 int pass)
2889{
2890 xfs_caddr_t lp;
2891 int num_logops;
2892 xlog_op_header_t *ohead;
2893 xlog_recover_t *trans;
2894 xlog_tid_t tid;
2895 int error;
2896 unsigned long hash;
2897 uint flags;
2898
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002899 lp = dp + be32_to_cpu(rhead->h_len);
2900 num_logops = be32_to_cpu(rhead->h_num_logops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002901
2902 /* check the log format matches our own - else we can't recover */
2903 if (xlog_header_check_recover(log->l_mp, rhead))
2904 return (XFS_ERROR(EIO));
2905
2906 while ((dp < lp) && num_logops) {
2907 ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
2908 ohead = (xlog_op_header_t *)dp;
2909 dp += sizeof(xlog_op_header_t);
2910 if (ohead->oh_clientid != XFS_TRANSACTION &&
2911 ohead->oh_clientid != XFS_LOG) {
2912 xlog_warn(
2913 "XFS: xlog_recover_process_data: bad clientid");
2914 ASSERT(0);
2915 return (XFS_ERROR(EIO));
2916 }
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10002917 tid = be32_to_cpu(ohead->oh_tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918 hash = XLOG_RHASH(tid);
2919 trans = xlog_recover_find_tid(rhash[hash], tid);
2920 if (trans == NULL) { /* not found; add new tid */
2921 if (ohead->oh_flags & XLOG_START_TRANS)
2922 xlog_recover_new_tid(&rhash[hash], tid,
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002923 be64_to_cpu(rhead->h_lsn));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924 } else {
Lachlan McIlroy9742bb92008-01-10 16:43:36 +11002925 if (dp + be32_to_cpu(ohead->oh_len) > lp) {
2926 xlog_warn(
2927 "XFS: xlog_recover_process_data: bad length");
2928 WARN_ON(1);
2929 return (XFS_ERROR(EIO));
2930 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931 flags = ohead->oh_flags & ~XLOG_END_TRANS;
2932 if (flags & XLOG_WAS_CONT_TRANS)
2933 flags &= ~XLOG_CONTINUE_TRANS;
2934 switch (flags) {
2935 case XLOG_COMMIT_TRANS:
2936 error = xlog_recover_commit_trans(log,
2937 &rhash[hash], trans, pass);
2938 break;
2939 case XLOG_UNMOUNT_TRANS:
2940 error = xlog_recover_unmount_trans(trans);
2941 break;
2942 case XLOG_WAS_CONT_TRANS:
2943 error = xlog_recover_add_to_cont_trans(trans,
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10002944 dp, be32_to_cpu(ohead->oh_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002945 break;
2946 case XLOG_START_TRANS:
2947 xlog_warn(
2948 "XFS: xlog_recover_process_data: bad transaction");
2949 ASSERT(0);
2950 error = XFS_ERROR(EIO);
2951 break;
2952 case 0:
2953 case XLOG_CONTINUE_TRANS:
2954 error = xlog_recover_add_to_trans(trans,
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10002955 dp, be32_to_cpu(ohead->oh_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002956 break;
2957 default:
2958 xlog_warn(
2959 "XFS: xlog_recover_process_data: bad flag");
2960 ASSERT(0);
2961 error = XFS_ERROR(EIO);
2962 break;
2963 }
2964 if (error)
2965 return error;
2966 }
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10002967 dp += be32_to_cpu(ohead->oh_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002968 num_logops--;
2969 }
2970 return 0;
2971}
2972
2973/*
2974 * Process an extent free intent item that was recovered from
2975 * the log. We need to free the extents that it describes.
2976 */
David Chinner3c1e2bb2008-04-10 12:21:11 +10002977STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002978xlog_recover_process_efi(
2979 xfs_mount_t *mp,
2980 xfs_efi_log_item_t *efip)
2981{
2982 xfs_efd_log_item_t *efdp;
2983 xfs_trans_t *tp;
2984 int i;
David Chinner3c1e2bb2008-04-10 12:21:11 +10002985 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002986 xfs_extent_t *extp;
2987 xfs_fsblock_t startblock_fsb;
2988
2989 ASSERT(!(efip->efi_flags & XFS_EFI_RECOVERED));
2990
2991 /*
2992 * First check the validity of the extents described by the
2993 * EFI. If any are bad, then assume that all are bad and
2994 * just toss the EFI.
2995 */
2996 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
2997 extp = &(efip->efi_format.efi_extents[i]);
2998 startblock_fsb = XFS_BB_TO_FSB(mp,
2999 XFS_FSB_TO_DADDR(mp, extp->ext_start));
3000 if ((startblock_fsb == 0) ||
3001 (extp->ext_len == 0) ||
3002 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
3003 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
3004 /*
3005 * This will pull the EFI from the AIL and
3006 * free the memory associated with it.
3007 */
3008 xfs_efi_release(efip, efip->efi_format.efi_nextents);
David Chinner3c1e2bb2008-04-10 12:21:11 +10003009 return XFS_ERROR(EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003010 }
3011 }
3012
3013 tp = xfs_trans_alloc(mp, 0);
David Chinner3c1e2bb2008-04-10 12:21:11 +10003014 error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 0, 0);
David Chinnerfc6149d2008-04-10 12:21:53 +10003015 if (error)
3016 goto abort_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
3018
3019 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3020 extp = &(efip->efi_format.efi_extents[i]);
David Chinnerfc6149d2008-04-10 12:21:53 +10003021 error = xfs_free_extent(tp, extp->ext_start, extp->ext_len);
3022 if (error)
3023 goto abort_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024 xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
3025 extp->ext_len);
3026 }
3027
3028 efip->efi_flags |= XFS_EFI_RECOVERED;
David Chinnere5720ee2008-04-10 12:21:18 +10003029 error = xfs_trans_commit(tp, 0);
David Chinner3c1e2bb2008-04-10 12:21:11 +10003030 return error;
David Chinnerfc6149d2008-04-10 12:21:53 +10003031
3032abort_error:
3033 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3034 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003035}
3036
3037/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003038 * When this is called, all of the EFIs which did not have
3039 * corresponding EFDs should be in the AIL. What we do now
3040 * is free the extents associated with each one.
3041 *
3042 * Since we process the EFIs in normal transactions, they
3043 * will be removed at some point after the commit. This prevents
3044 * us from just walking down the list processing each one.
3045 * We'll use a flag in the EFI to skip those that we've already
3046 * processed and use the AIL iteration mechanism's generation
3047 * count to try to speed this up at least a bit.
3048 *
3049 * When we start, we know that the EFIs are the only things in
3050 * the AIL. As we process them, however, other items are added
3051 * to the AIL. Since everything added to the AIL must come after
3052 * everything already in the AIL, we stop processing as soon as
3053 * we see something other than an EFI in the AIL.
3054 */
David Chinner3c1e2bb2008-04-10 12:21:11 +10003055STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003056xlog_recover_process_efis(
3057 xlog_t *log)
3058{
3059 xfs_log_item_t *lip;
3060 xfs_efi_log_item_t *efip;
David Chinner3c1e2bb2008-04-10 12:21:11 +10003061 int error = 0;
David Chinner27d8d5f2008-10-30 17:38:39 +11003062 struct xfs_ail_cursor cur;
David Chinnera9c21c12008-10-30 17:39:35 +11003063 struct xfs_ail *ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003064
David Chinnera9c21c12008-10-30 17:39:35 +11003065 ailp = log->l_ailp;
3066 spin_lock(&ailp->xa_lock);
3067 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003068 while (lip != NULL) {
3069 /*
3070 * We're done when we see something other than an EFI.
David Chinner27d8d5f2008-10-30 17:38:39 +11003071 * There should be no EFIs left in the AIL now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003072 */
3073 if (lip->li_type != XFS_LI_EFI) {
David Chinner27d8d5f2008-10-30 17:38:39 +11003074#ifdef DEBUG
David Chinnera9c21c12008-10-30 17:39:35 +11003075 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
David Chinner27d8d5f2008-10-30 17:38:39 +11003076 ASSERT(lip->li_type != XFS_LI_EFI);
3077#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003078 break;
3079 }
3080
3081 /*
3082 * Skip EFIs that we've already processed.
3083 */
3084 efip = (xfs_efi_log_item_t *)lip;
3085 if (efip->efi_flags & XFS_EFI_RECOVERED) {
David Chinnera9c21c12008-10-30 17:39:35 +11003086 lip = xfs_trans_ail_cursor_next(ailp, &cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003087 continue;
3088 }
3089
David Chinnera9c21c12008-10-30 17:39:35 +11003090 spin_unlock(&ailp->xa_lock);
3091 error = xlog_recover_process_efi(log->l_mp, efip);
3092 spin_lock(&ailp->xa_lock);
David Chinner27d8d5f2008-10-30 17:38:39 +11003093 if (error)
3094 goto out;
David Chinnera9c21c12008-10-30 17:39:35 +11003095 lip = xfs_trans_ail_cursor_next(ailp, &cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003096 }
David Chinner27d8d5f2008-10-30 17:38:39 +11003097out:
David Chinnera9c21c12008-10-30 17:39:35 +11003098 xfs_trans_ail_cursor_done(ailp, &cur);
3099 spin_unlock(&ailp->xa_lock);
David Chinner3c1e2bb2008-04-10 12:21:11 +10003100 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003101}
3102
3103/*
3104 * This routine performs a transaction to null out a bad inode pointer
3105 * in an agi unlinked inode hash bucket.
3106 */
3107STATIC void
3108xlog_recover_clear_agi_bucket(
3109 xfs_mount_t *mp,
3110 xfs_agnumber_t agno,
3111 int bucket)
3112{
3113 xfs_trans_t *tp;
3114 xfs_agi_t *agi;
3115 xfs_buf_t *agibp;
3116 int offset;
3117 int error;
3118
3119 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003120 error = xfs_trans_reserve(tp, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp),
3121 0, 0, 0);
David Chinnere5720ee2008-04-10 12:21:18 +10003122 if (error)
3123 goto out_abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003124
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003125 error = xfs_read_agi(mp, tp, agno, &agibp);
3126 if (error)
David Chinnere5720ee2008-04-10 12:21:18 +10003127 goto out_abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003128
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003129 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11003130 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003131 offset = offsetof(xfs_agi_t, agi_unlinked) +
3132 (sizeof(xfs_agino_t) * bucket);
3133 xfs_trans_log_buf(tp, agibp, offset,
3134 (offset + sizeof(xfs_agino_t) - 1));
3135
David Chinnere5720ee2008-04-10 12:21:18 +10003136 error = xfs_trans_commit(tp, 0);
3137 if (error)
3138 goto out_error;
3139 return;
3140
3141out_abort:
3142 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3143out_error:
3144 xfs_fs_cmn_err(CE_WARN, mp, "xlog_recover_clear_agi_bucket: "
3145 "failed to clear agi %d. Continuing.", agno);
3146 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003147}
3148
Christoph Hellwig23fac502008-11-28 14:23:40 +11003149STATIC xfs_agino_t
3150xlog_recover_process_one_iunlink(
3151 struct xfs_mount *mp,
3152 xfs_agnumber_t agno,
3153 xfs_agino_t agino,
3154 int bucket)
3155{
3156 struct xfs_buf *ibp;
3157 struct xfs_dinode *dip;
3158 struct xfs_inode *ip;
3159 xfs_ino_t ino;
3160 int error;
3161
3162 ino = XFS_AGINO_TO_INO(mp, agno, agino);
3163 error = xfs_iget(mp, NULL, ino, 0, 0, &ip, 0);
3164 if (error)
3165 goto fail;
3166
3167 /*
3168 * Get the on disk inode to find the next inode in the bucket.
3169 */
Christoph Hellwig76d8b272008-11-28 14:23:40 +11003170 error = xfs_itobp(mp, NULL, ip, &dip, &ibp, XFS_BUF_LOCK);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003171 if (error)
Christoph Hellwig0e446672008-11-28 14:23:42 +11003172 goto fail_iput;
Christoph Hellwig23fac502008-11-28 14:23:40 +11003173
Christoph Hellwig23fac502008-11-28 14:23:40 +11003174 ASSERT(ip->i_d.di_nlink == 0);
Christoph Hellwig0e446672008-11-28 14:23:42 +11003175 ASSERT(ip->i_d.di_mode != 0);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003176
3177 /* setup for the next pass */
3178 agino = be32_to_cpu(dip->di_next_unlinked);
3179 xfs_buf_relse(ibp);
3180
3181 /*
3182 * Prevent any DMAPI event from being sent when the reference on
3183 * the inode is dropped.
3184 */
3185 ip->i_d.di_dmevmask = 0;
3186
Christoph Hellwig0e446672008-11-28 14:23:42 +11003187 IRELE(ip);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003188 return agino;
3189
Christoph Hellwig0e446672008-11-28 14:23:42 +11003190 fail_iput:
3191 IRELE(ip);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003192 fail:
3193 /*
3194 * We can't read in the inode this bucket points to, or this inode
3195 * is messed up. Just ditch this bucket of inodes. We will lose
3196 * some inodes and space, but at least we won't hang.
3197 *
3198 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3199 * clear the inode pointer in the bucket.
3200 */
3201 xlog_recover_clear_agi_bucket(mp, agno, bucket);
3202 return NULLAGINO;
3203}
3204
Linus Torvalds1da177e2005-04-16 15:20:36 -07003205/*
3206 * xlog_iunlink_recover
3207 *
3208 * This is called during recovery to process any inodes which
3209 * we unlinked but not freed when the system crashed. These
3210 * inodes will be on the lists in the AGI blocks. What we do
3211 * here is scan all the AGIs and fully truncate and free any
3212 * inodes found on the lists. Each inode is removed from the
3213 * lists when it has been fully truncated and is freed. The
3214 * freeing of the inode and its removal from the list must be
3215 * atomic.
3216 */
3217void
3218xlog_recover_process_iunlinks(
3219 xlog_t *log)
3220{
3221 xfs_mount_t *mp;
3222 xfs_agnumber_t agno;
3223 xfs_agi_t *agi;
3224 xfs_buf_t *agibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003225 xfs_agino_t agino;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003226 int bucket;
3227 int error;
3228 uint mp_dmevmask;
3229
3230 mp = log->l_mp;
3231
3232 /*
3233 * Prevent any DMAPI event from being sent while in this function.
3234 */
3235 mp_dmevmask = mp->m_dmevmask;
3236 mp->m_dmevmask = 0;
3237
3238 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3239 /*
3240 * Find the agi for this ag.
3241 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003242 error = xfs_read_agi(mp, NULL, agno, &agibp);
3243 if (error) {
3244 /*
3245 * AGI is b0rked. Don't process it.
3246 *
3247 * We should probably mark the filesystem as corrupt
3248 * after we've recovered all the ag's we can....
3249 */
3250 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003251 }
3252 agi = XFS_BUF_TO_AGI(agibp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003253
3254 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
Christoph Hellwig16259e72005-11-02 15:11:25 +11003255 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003256 while (agino != NULLAGINO) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003257 /*
3258 * Release the agi buffer so that it can
3259 * be acquired in the normal course of the
3260 * transaction to truncate and free the inode.
3261 */
3262 xfs_buf_relse(agibp);
3263
Christoph Hellwig23fac502008-11-28 14:23:40 +11003264 agino = xlog_recover_process_one_iunlink(mp,
3265 agno, agino, bucket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003266
3267 /*
3268 * Reacquire the agibuffer and continue around
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003269 * the loop. This should never fail as we know
3270 * the buffer was good earlier on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003271 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003272 error = xfs_read_agi(mp, NULL, agno, &agibp);
3273 ASSERT(error == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003274 agi = XFS_BUF_TO_AGI(agibp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003275 }
3276 }
3277
3278 /*
3279 * Release the buffer for the current agi so we can
3280 * go on to the next one.
3281 */
3282 xfs_buf_relse(agibp);
3283 }
3284
3285 mp->m_dmevmask = mp_dmevmask;
3286}
3287
3288
3289#ifdef DEBUG
3290STATIC void
3291xlog_pack_data_checksum(
3292 xlog_t *log,
3293 xlog_in_core_t *iclog,
3294 int size)
3295{
3296 int i;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003297 __be32 *up;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003298 uint chksum = 0;
3299
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003300 up = (__be32 *)iclog->ic_datap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003301 /* divide length by 4 to get # words */
3302 for (i = 0; i < (size >> 2); i++) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003303 chksum ^= be32_to_cpu(*up);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003304 up++;
3305 }
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003306 iclog->ic_header.h_chksum = cpu_to_be32(chksum);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003307}
3308#else
3309#define xlog_pack_data_checksum(log, iclog, size)
3310#endif
3311
3312/*
3313 * Stamp cycle number in every block
3314 */
3315void
3316xlog_pack_data(
3317 xlog_t *log,
3318 xlog_in_core_t *iclog,
3319 int roundoff)
3320{
3321 int i, j, k;
3322 int size = iclog->ic_offset + roundoff;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003323 __be32 cycle_lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003324 xfs_caddr_t dp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003325
3326 xlog_pack_data_checksum(log, iclog, size);
3327
3328 cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn);
3329
3330 dp = iclog->ic_datap;
3331 for (i = 0; i < BTOBB(size) &&
3332 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003333 iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp;
3334 *(__be32 *)dp = cycle_lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003335 dp += BBSIZE;
3336 }
3337
Eric Sandeen62118702008-03-06 13:44:28 +11003338 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Christoph Hellwigb28708d2008-11-28 14:23:38 +11003339 xlog_in_core_2_t *xhdr = iclog->ic_data;
3340
Linus Torvalds1da177e2005-04-16 15:20:36 -07003341 for ( ; i < BTOBB(size); i++) {
3342 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3343 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003344 xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp;
3345 *(__be32 *)dp = cycle_lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003346 dp += BBSIZE;
3347 }
3348
3349 for (i = 1; i < log->l_iclog_heads; i++) {
3350 xhdr[i].hic_xheader.xh_cycle = cycle_lsn;
3351 }
3352 }
3353}
3354
3355#if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
3356STATIC void
3357xlog_unpack_data_checksum(
3358 xlog_rec_header_t *rhead,
3359 xfs_caddr_t dp,
3360 xlog_t *log)
3361{
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003362 __be32 *up = (__be32 *)dp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003363 uint chksum = 0;
3364 int i;
3365
3366 /* divide length by 4 to get # words */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003367 for (i=0; i < be32_to_cpu(rhead->h_len) >> 2; i++) {
3368 chksum ^= be32_to_cpu(*up);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003369 up++;
3370 }
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003371 if (chksum != be32_to_cpu(rhead->h_chksum)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003372 if (rhead->h_chksum ||
3373 ((log->l_flags & XLOG_CHKSUM_MISMATCH) == 0)) {
3374 cmn_err(CE_DEBUG,
Nathan Scottb6574522006-06-09 15:29:40 +10003375 "XFS: LogR chksum mismatch: was (0x%x) is (0x%x)\n",
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003376 be32_to_cpu(rhead->h_chksum), chksum);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003377 cmn_err(CE_DEBUG,
3378"XFS: Disregard message if filesystem was created with non-DEBUG kernel");
Eric Sandeen62118702008-03-06 13:44:28 +11003379 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003380 cmn_err(CE_DEBUG,
Nathan Scottb6574522006-06-09 15:29:40 +10003381 "XFS: LogR this is a LogV2 filesystem\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003382 }
3383 log->l_flags |= XLOG_CHKSUM_MISMATCH;
3384 }
3385 }
3386}
3387#else
3388#define xlog_unpack_data_checksum(rhead, dp, log)
3389#endif
3390
3391STATIC void
3392xlog_unpack_data(
3393 xlog_rec_header_t *rhead,
3394 xfs_caddr_t dp,
3395 xlog_t *log)
3396{
3397 int i, j, k;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003398
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003399 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07003400 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003401 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003402 dp += BBSIZE;
3403 }
3404
Eric Sandeen62118702008-03-06 13:44:28 +11003405 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Christoph Hellwigb28708d2008-11-28 14:23:38 +11003406 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003407 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003408 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3409 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003410 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003411 dp += BBSIZE;
3412 }
3413 }
3414
3415 xlog_unpack_data_checksum(rhead, dp, log);
3416}
3417
3418STATIC int
3419xlog_valid_rec_header(
3420 xlog_t *log,
3421 xlog_rec_header_t *rhead,
3422 xfs_daddr_t blkno)
3423{
3424 int hlen;
3425
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003426 if (unlikely(be32_to_cpu(rhead->h_magicno) != XLOG_HEADER_MAGIC_NUM)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003427 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3428 XFS_ERRLEVEL_LOW, log->l_mp);
3429 return XFS_ERROR(EFSCORRUPTED);
3430 }
3431 if (unlikely(
3432 (!rhead->h_version ||
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003433 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003434 xlog_warn("XFS: %s: unrecognised log version (%d).",
Harvey Harrison34a622b2008-04-10 12:19:21 +10003435 __func__, be32_to_cpu(rhead->h_version));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003436 return XFS_ERROR(EIO);
3437 }
3438
3439 /* LR body must have data or it wouldn't have been written */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003440 hlen = be32_to_cpu(rhead->h_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003441 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
3442 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3443 XFS_ERRLEVEL_LOW, log->l_mp);
3444 return XFS_ERROR(EFSCORRUPTED);
3445 }
3446 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
3447 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3448 XFS_ERRLEVEL_LOW, log->l_mp);
3449 return XFS_ERROR(EFSCORRUPTED);
3450 }
3451 return 0;
3452}
3453
3454/*
3455 * Read the log from tail to head and process the log records found.
3456 * Handle the two cases where the tail and head are in the same cycle
3457 * and where the active portion of the log wraps around the end of
3458 * the physical log separately. The pass parameter is passed through
3459 * to the routines called to process the data and is not looked at
3460 * here.
3461 */
3462STATIC int
3463xlog_do_recovery_pass(
3464 xlog_t *log,
3465 xfs_daddr_t head_blk,
3466 xfs_daddr_t tail_blk,
3467 int pass)
3468{
3469 xlog_rec_header_t *rhead;
3470 xfs_daddr_t blk_no;
3471 xfs_caddr_t bufaddr, offset;
3472 xfs_buf_t *hbp, *dbp;
3473 int error = 0, h_size;
3474 int bblks, split_bblks;
3475 int hblks, split_hblks, wrapped_hblks;
3476 xlog_recover_t *rhash[XLOG_RHASH_SIZE];
3477
3478 ASSERT(head_blk != tail_blk);
3479
3480 /*
3481 * Read the header of the tail block and get the iclog buffer size from
3482 * h_size. Use this to tell how many sectors make up the log header.
3483 */
Eric Sandeen62118702008-03-06 13:44:28 +11003484 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003485 /*
3486 * When using variable length iclogs, read first sector of
3487 * iclog header and extract the header size from it. Get a
3488 * new hbp that is the correct size.
3489 */
3490 hbp = xlog_get_bp(log, 1);
3491 if (!hbp)
3492 return ENOMEM;
3493 if ((error = xlog_bread(log, tail_blk, 1, hbp)))
3494 goto bread_err1;
3495 offset = xlog_align(log, tail_blk, 1, hbp);
3496 rhead = (xlog_rec_header_t *)offset;
3497 error = xlog_valid_rec_header(log, rhead, tail_blk);
3498 if (error)
3499 goto bread_err1;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003500 h_size = be32_to_cpu(rhead->h_size);
3501 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07003502 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
3503 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
3504 if (h_size % XLOG_HEADER_CYCLE_SIZE)
3505 hblks++;
3506 xlog_put_bp(hbp);
3507 hbp = xlog_get_bp(log, hblks);
3508 } else {
3509 hblks = 1;
3510 }
3511 } else {
3512 ASSERT(log->l_sectbb_log == 0);
3513 hblks = 1;
3514 hbp = xlog_get_bp(log, 1);
3515 h_size = XLOG_BIG_RECORD_BSIZE;
3516 }
3517
3518 if (!hbp)
3519 return ENOMEM;
3520 dbp = xlog_get_bp(log, BTOBB(h_size));
3521 if (!dbp) {
3522 xlog_put_bp(hbp);
3523 return ENOMEM;
3524 }
3525
3526 memset(rhash, 0, sizeof(rhash));
3527 if (tail_blk <= head_blk) {
3528 for (blk_no = tail_blk; blk_no < head_blk; ) {
3529 if ((error = xlog_bread(log, blk_no, hblks, hbp)))
3530 goto bread_err2;
3531 offset = xlog_align(log, blk_no, hblks, hbp);
3532 rhead = (xlog_rec_header_t *)offset;
3533 error = xlog_valid_rec_header(log, rhead, blk_no);
3534 if (error)
3535 goto bread_err2;
3536
3537 /* blocks in data section */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003538 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003539 error = xlog_bread(log, blk_no + hblks, bblks, dbp);
3540 if (error)
3541 goto bread_err2;
3542 offset = xlog_align(log, blk_no + hblks, bblks, dbp);
3543 xlog_unpack_data(rhead, offset, log);
3544 if ((error = xlog_recover_process_data(log,
3545 rhash, rhead, offset, pass)))
3546 goto bread_err2;
3547 blk_no += bblks + hblks;
3548 }
3549 } else {
3550 /*
3551 * Perform recovery around the end of the physical log.
3552 * When the head is not on the same cycle number as the tail,
3553 * we can't do a sequential recovery as above.
3554 */
3555 blk_no = tail_blk;
3556 while (blk_no < log->l_logBBsize) {
3557 /*
3558 * Check for header wrapping around physical end-of-log
3559 */
3560 offset = NULL;
3561 split_hblks = 0;
3562 wrapped_hblks = 0;
3563 if (blk_no + hblks <= log->l_logBBsize) {
3564 /* Read header in one read */
3565 error = xlog_bread(log, blk_no, hblks, hbp);
3566 if (error)
3567 goto bread_err2;
3568 offset = xlog_align(log, blk_no, hblks, hbp);
3569 } else {
3570 /* This LR is split across physical log end */
3571 if (blk_no != log->l_logBBsize) {
3572 /* some data before physical log end */
3573 ASSERT(blk_no <= INT_MAX);
3574 split_hblks = log->l_logBBsize - (int)blk_no;
3575 ASSERT(split_hblks > 0);
3576 if ((error = xlog_bread(log, blk_no,
3577 split_hblks, hbp)))
3578 goto bread_err2;
3579 offset = xlog_align(log, blk_no,
3580 split_hblks, hbp);
3581 }
3582 /*
3583 * Note: this black magic still works with
3584 * large sector sizes (non-512) only because:
3585 * - we increased the buffer size originally
3586 * by 1 sector giving us enough extra space
3587 * for the second read;
3588 * - the log start is guaranteed to be sector
3589 * aligned;
3590 * - we read the log end (LR header start)
3591 * _first_, then the log start (LR header end)
3592 * - order is important.
3593 */
David Chinner234f56a2008-04-10 12:24:24 +10003594 wrapped_hblks = hblks - split_hblks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003595 bufaddr = XFS_BUF_PTR(hbp);
David Chinner234f56a2008-04-10 12:24:24 +10003596 error = XFS_BUF_SET_PTR(hbp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003597 bufaddr + BBTOB(split_hblks),
3598 BBTOB(hblks - split_hblks));
David Chinner234f56a2008-04-10 12:24:24 +10003599 if (!error)
3600 error = xlog_bread(log, 0,
3601 wrapped_hblks, hbp);
3602 if (!error)
3603 error = XFS_BUF_SET_PTR(hbp, bufaddr,
3604 BBTOB(hblks));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003605 if (error)
3606 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003607 if (!offset)
3608 offset = xlog_align(log, 0,
3609 wrapped_hblks, hbp);
3610 }
3611 rhead = (xlog_rec_header_t *)offset;
3612 error = xlog_valid_rec_header(log, rhead,
3613 split_hblks ? blk_no : 0);
3614 if (error)
3615 goto bread_err2;
3616
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003617 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003618 blk_no += hblks;
3619
3620 /* Read in data for log record */
3621 if (blk_no + bblks <= log->l_logBBsize) {
3622 error = xlog_bread(log, blk_no, bblks, dbp);
3623 if (error)
3624 goto bread_err2;
3625 offset = xlog_align(log, blk_no, bblks, dbp);
3626 } else {
3627 /* This log record is split across the
3628 * physical end of log */
3629 offset = NULL;
3630 split_bblks = 0;
3631 if (blk_no != log->l_logBBsize) {
3632 /* some data is before the physical
3633 * end of log */
3634 ASSERT(!wrapped_hblks);
3635 ASSERT(blk_no <= INT_MAX);
3636 split_bblks =
3637 log->l_logBBsize - (int)blk_no;
3638 ASSERT(split_bblks > 0);
3639 if ((error = xlog_bread(log, blk_no,
3640 split_bblks, dbp)))
3641 goto bread_err2;
3642 offset = xlog_align(log, blk_no,
3643 split_bblks, dbp);
3644 }
3645 /*
3646 * Note: this black magic still works with
3647 * large sector sizes (non-512) only because:
3648 * - we increased the buffer size originally
3649 * by 1 sector giving us enough extra space
3650 * for the second read;
3651 * - the log start is guaranteed to be sector
3652 * aligned;
3653 * - we read the log end (LR header start)
3654 * _first_, then the log start (LR header end)
3655 * - order is important.
3656 */
3657 bufaddr = XFS_BUF_PTR(dbp);
David Chinner234f56a2008-04-10 12:24:24 +10003658 error = XFS_BUF_SET_PTR(dbp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003659 bufaddr + BBTOB(split_bblks),
3660 BBTOB(bblks - split_bblks));
David Chinner234f56a2008-04-10 12:24:24 +10003661 if (!error)
3662 error = xlog_bread(log, wrapped_hblks,
3663 bblks - split_bblks,
3664 dbp);
3665 if (!error)
3666 error = XFS_BUF_SET_PTR(dbp, bufaddr,
3667 h_size);
3668 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003669 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003670 if (!offset)
3671 offset = xlog_align(log, wrapped_hblks,
3672 bblks - split_bblks, dbp);
3673 }
3674 xlog_unpack_data(rhead, offset, log);
3675 if ((error = xlog_recover_process_data(log, rhash,
3676 rhead, offset, pass)))
3677 goto bread_err2;
3678 blk_no += bblks;
3679 }
3680
3681 ASSERT(blk_no >= log->l_logBBsize);
3682 blk_no -= log->l_logBBsize;
3683
3684 /* read first part of physical log */
3685 while (blk_no < head_blk) {
3686 if ((error = xlog_bread(log, blk_no, hblks, hbp)))
3687 goto bread_err2;
3688 offset = xlog_align(log, blk_no, hblks, hbp);
3689 rhead = (xlog_rec_header_t *)offset;
3690 error = xlog_valid_rec_header(log, rhead, blk_no);
3691 if (error)
3692 goto bread_err2;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003693 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003694 if ((error = xlog_bread(log, blk_no+hblks, bblks, dbp)))
3695 goto bread_err2;
3696 offset = xlog_align(log, blk_no+hblks, bblks, dbp);
3697 xlog_unpack_data(rhead, offset, log);
3698 if ((error = xlog_recover_process_data(log, rhash,
3699 rhead, offset, pass)))
3700 goto bread_err2;
3701 blk_no += bblks + hblks;
3702 }
3703 }
3704
3705 bread_err2:
3706 xlog_put_bp(dbp);
3707 bread_err1:
3708 xlog_put_bp(hbp);
3709 return error;
3710}
3711
3712/*
3713 * Do the recovery of the log. We actually do this in two phases.
3714 * The two passes are necessary in order to implement the function
3715 * of cancelling a record written into the log. The first pass
3716 * determines those things which have been cancelled, and the
3717 * second pass replays log items normally except for those which
3718 * have been cancelled. The handling of the replay and cancellations
3719 * takes place in the log item type specific routines.
3720 *
3721 * The table of items which have cancel records in the log is allocated
3722 * and freed at this level, since only here do we know when all of
3723 * the log recovery has been completed.
3724 */
3725STATIC int
3726xlog_do_log_recovery(
3727 xlog_t *log,
3728 xfs_daddr_t head_blk,
3729 xfs_daddr_t tail_blk)
3730{
3731 int error;
3732
3733 ASSERT(head_blk != tail_blk);
3734
3735 /*
3736 * First do a pass to find all of the cancelled buf log items.
3737 * Store them in the buf_cancel_table for use in the second pass.
3738 */
3739 log->l_buf_cancel_table =
3740 (xfs_buf_cancel_t **)kmem_zalloc(XLOG_BC_TABLE_SIZE *
3741 sizeof(xfs_buf_cancel_t*),
3742 KM_SLEEP);
3743 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3744 XLOG_RECOVER_PASS1);
3745 if (error != 0) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003746 kmem_free(log->l_buf_cancel_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003747 log->l_buf_cancel_table = NULL;
3748 return error;
3749 }
3750 /*
3751 * Then do a second pass to actually recover the items in the log.
3752 * When it is complete free the table of buf cancel items.
3753 */
3754 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3755 XLOG_RECOVER_PASS2);
3756#ifdef DEBUG
Tim Shimmin6d192a92006-06-09 14:55:38 +10003757 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003758 int i;
3759
3760 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
3761 ASSERT(log->l_buf_cancel_table[i] == NULL);
3762 }
3763#endif /* DEBUG */
3764
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003765 kmem_free(log->l_buf_cancel_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003766 log->l_buf_cancel_table = NULL;
3767
3768 return error;
3769}
3770
3771/*
3772 * Do the actual recovery
3773 */
3774STATIC int
3775xlog_do_recover(
3776 xlog_t *log,
3777 xfs_daddr_t head_blk,
3778 xfs_daddr_t tail_blk)
3779{
3780 int error;
3781 xfs_buf_t *bp;
3782 xfs_sb_t *sbp;
3783
3784 /*
3785 * First replay the images in the log.
3786 */
3787 error = xlog_do_log_recovery(log, head_blk, tail_blk);
3788 if (error) {
3789 return error;
3790 }
3791
3792 XFS_bflush(log->l_mp->m_ddev_targp);
3793
3794 /*
3795 * If IO errors happened during recovery, bail out.
3796 */
3797 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
3798 return (EIO);
3799 }
3800
3801 /*
3802 * We now update the tail_lsn since much of the recovery has completed
3803 * and there may be space available to use. If there were no extent
3804 * or iunlinks, we can free up the entire log and set the tail_lsn to
3805 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3806 * lsn of the last known good LR on disk. If there are extent frees
3807 * or iunlinks they will have some entries in the AIL; so we look at
3808 * the AIL to determine how to set the tail_lsn.
3809 */
3810 xlog_assign_tail_lsn(log->l_mp);
3811
3812 /*
3813 * Now that we've finished replaying all buffer and inode
3814 * updates, re-read in the superblock.
3815 */
3816 bp = xfs_getsb(log->l_mp, 0);
3817 XFS_BUF_UNDONE(bp);
Lachlan McIlroybebf9632007-10-15 13:18:02 +10003818 ASSERT(!(XFS_BUF_ISWRITE(bp)));
3819 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003820 XFS_BUF_READ(bp);
Lachlan McIlroybebf9632007-10-15 13:18:02 +10003821 XFS_BUF_UNASYNC(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003822 xfsbdstrat(log->l_mp, bp);
David Chinnerd64e31a2008-04-10 12:22:17 +10003823 error = xfs_iowait(bp);
3824 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003825 xfs_ioerror_alert("xlog_do_recover",
3826 log->l_mp, bp, XFS_BUF_ADDR(bp));
3827 ASSERT(0);
3828 xfs_buf_relse(bp);
3829 return error;
3830 }
3831
3832 /* Convert superblock from on-disk format */
3833 sbp = &log->l_mp->m_sb;
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +10003834 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003835 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
Eric Sandeen62118702008-03-06 13:44:28 +11003836 ASSERT(xfs_sb_good_version(sbp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003837 xfs_buf_relse(bp);
3838
Lachlan McIlroy5478eea2007-02-10 18:36:29 +11003839 /* We've re-read the superblock so re-initialize per-cpu counters */
3840 xfs_icsb_reinit_counters(log->l_mp);
3841
Linus Torvalds1da177e2005-04-16 15:20:36 -07003842 xlog_recover_check_summary(log);
3843
3844 /* Normal transactions can now occur */
3845 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
3846 return 0;
3847}
3848
3849/*
3850 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3851 *
3852 * Return error or zero.
3853 */
3854int
3855xlog_recover(
Eric Sandeen65be6052006-01-11 15:34:19 +11003856 xlog_t *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003857{
3858 xfs_daddr_t head_blk, tail_blk;
3859 int error;
3860
3861 /* find the tail of the log */
Eric Sandeen65be6052006-01-11 15:34:19 +11003862 if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003863 return error;
3864
3865 if (tail_blk != head_blk) {
3866 /* There used to be a comment here:
3867 *
3868 * disallow recovery on read-only mounts. note -- mount
3869 * checks for ENOSPC and turns it into an intelligent
3870 * error message.
3871 * ...but this is no longer true. Now, unless you specify
3872 * NORECOVERY (in which case this function would never be
3873 * called), we just go ahead and recover. We do this all
3874 * under the vfs layer, so we can get away with it unless
3875 * the device itself is read-only, in which case we fail.
3876 */
Utako Kusaka3a02ee12007-05-08 13:50:06 +10003877 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003878 return error;
3879 }
3880
3881 cmn_err(CE_NOTE,
Nathan Scottfc1f8c12005-11-02 11:44:33 +11003882 "Starting XFS recovery on filesystem: %s (logdev: %s)",
3883 log->l_mp->m_fsname, log->l_mp->m_logname ?
3884 log->l_mp->m_logname : "internal");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003885
3886 error = xlog_do_recover(log, head_blk, tail_blk);
3887 log->l_flags |= XLOG_RECOVERY_NEEDED;
3888 }
3889 return error;
3890}
3891
3892/*
3893 * In the first part of recovery we replay inodes and buffers and build
3894 * up the list of extent free items which need to be processed. Here
3895 * we process the extent free items and clean up the on disk unlinked
3896 * inode lists. This is separated from the first part of recovery so
3897 * that the root and real-time bitmap inodes can be read in from disk in
3898 * between the two stages. This is necessary so that we can free space
3899 * in the real-time portion of the file system.
3900 */
3901int
3902xlog_recover_finish(
Christoph Hellwig42490232008-08-13 16:49:32 +10003903 xlog_t *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003904{
3905 /*
3906 * Now we're ready to do the transactions needed for the
3907 * rest of recovery. Start with completing all the extent
3908 * free intent records and then process the unlinked inode
3909 * lists. At this point, we essentially run in normal mode
3910 * except that we're still performing recovery actions
3911 * rather than accepting new requests.
3912 */
3913 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
David Chinner3c1e2bb2008-04-10 12:21:11 +10003914 int error;
3915 error = xlog_recover_process_efis(log);
3916 if (error) {
3917 cmn_err(CE_ALERT,
3918 "Failed to recover EFIs on filesystem: %s",
3919 log->l_mp->m_fsname);
3920 return error;
3921 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003922 /*
3923 * Sync the log to get all the EFIs out of the AIL.
3924 * This isn't absolutely necessary, but it helps in
3925 * case the unlink transactions would have problems
3926 * pushing the EFIs out of the way.
3927 */
3928 xfs_log_force(log->l_mp, (xfs_lsn_t)0,
3929 (XFS_LOG_FORCE | XFS_LOG_SYNC));
3930
Christoph Hellwig42490232008-08-13 16:49:32 +10003931 xlog_recover_process_iunlinks(log);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003932
3933 xlog_recover_check_summary(log);
3934
3935 cmn_err(CE_NOTE,
Nathan Scottfc1f8c12005-11-02 11:44:33 +11003936 "Ending XFS recovery on filesystem: %s (logdev: %s)",
3937 log->l_mp->m_fsname, log->l_mp->m_logname ?
3938 log->l_mp->m_logname : "internal");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003939 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
3940 } else {
3941 cmn_err(CE_DEBUG,
Nathan Scottb6574522006-06-09 15:29:40 +10003942 "!Ending clean XFS mount for filesystem: %s\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07003943 log->l_mp->m_fsname);
3944 }
3945 return 0;
3946}
3947
3948
3949#if defined(DEBUG)
3950/*
3951 * Read all of the agf and agi counters and check that they
3952 * are consistent with the superblock counters.
3953 */
3954void
3955xlog_recover_check_summary(
3956 xlog_t *log)
3957{
3958 xfs_mount_t *mp;
3959 xfs_agf_t *agfp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003960 xfs_buf_t *agfbp;
3961 xfs_buf_t *agibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003962 xfs_buf_t *sbbp;
3963#ifdef XFS_LOUD_RECOVERY
3964 xfs_sb_t *sbp;
3965#endif
3966 xfs_agnumber_t agno;
3967 __uint64_t freeblks;
3968 __uint64_t itotal;
3969 __uint64_t ifree;
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003970 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003971
3972 mp = log->l_mp;
3973
3974 freeblks = 0LL;
3975 itotal = 0LL;
3976 ifree = 0LL;
3977 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
From: Christoph Hellwig48056212008-11-28 14:23:38 +11003978 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
3979 if (error) {
3980 xfs_fs_cmn_err(CE_ALERT, mp,
3981 "xlog_recover_check_summary(agf)"
3982 "agf read failed agno %d error %d",
3983 agno, error);
3984 } else {
3985 agfp = XFS_BUF_TO_AGF(agfbp);
3986 freeblks += be32_to_cpu(agfp->agf_freeblks) +
3987 be32_to_cpu(agfp->agf_flcount);
3988 xfs_buf_relse(agfbp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003989 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003990
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003991 error = xfs_read_agi(mp, NULL, agno, &agibp);
3992 if (!error) {
3993 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003994
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003995 itotal += be32_to_cpu(agi->agi_count);
3996 ifree += be32_to_cpu(agi->agi_freecount);
3997 xfs_buf_relse(agibp);
3998 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003999 }
4000
4001 sbbp = xfs_getsb(mp, 0);
4002#ifdef XFS_LOUD_RECOVERY
4003 sbp = &mp->m_sb;
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +10004004 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(sbbp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004005 cmn_err(CE_NOTE,
4006 "xlog_recover_check_summary: sb_icount %Lu itotal %Lu",
4007 sbp->sb_icount, itotal);
4008 cmn_err(CE_NOTE,
4009 "xlog_recover_check_summary: sb_ifree %Lu itotal %Lu",
4010 sbp->sb_ifree, ifree);
4011 cmn_err(CE_NOTE,
4012 "xlog_recover_check_summary: sb_fdblocks %Lu freeblks %Lu",
4013 sbp->sb_fdblocks, freeblks);
4014#if 0
4015 /*
4016 * This is turned off until I account for the allocation
4017 * btree blocks which live in free space.
4018 */
4019 ASSERT(sbp->sb_icount == itotal);
4020 ASSERT(sbp->sb_ifree == ifree);
4021 ASSERT(sbp->sb_fdblocks == freeblks);
4022#endif
4023#endif
4024 xfs_buf_relse(sbbp);
4025}
4026#endif /* DEBUG */