| blob | 0af3d477197b2418170ad99c25dbc695633af9b3 |
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
32 #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
34 STATIC int
37 xfs_daddr_t *);
38 STATIC int
41 xfs_lsn_t);
42 STATIC int
46 /*
47 * Sector aligned buffer routines for buffer create/read/write/access
48 */
50 /*
51 * Verify the log-relative block number and length in basic blocks are valid for
52 * an operation involving the given XFS log buffer. Returns true if the fields
53 * are valid, false otherwise.
54 */
58 xfs_daddr_t blk_no,
60 {
66 }
68 /*
69 * Allocate a buffer to hold log data. The buffer needs to be able to map to
70 * a range of nbblks basic blocks at any valid offset within the log.
71 */
76 {
77 /*
78 * Pass log block 0 since we don't have an addr yet, buffer will be
79 * verified on read.
80 */
83 nbblks);
85 }
87 /*
88 * We do log I/O in units of log sectors (a power-of-2 multiple of the
89 * basic block size), so we round up the requested size to accommodate
90 * the basic blocks required for complete log sectors.
91 *
92 * In addition, the buffer may be used for a non-sector-aligned block
93 * offset, in which case an I/O of the requested size could extend
94 * beyond the end of the buffer. If the requested size is only 1 basic
95 * block it will never straddle a sector boundary, so this won't be an
96 * issue. Nor will this be a problem if the log I/O is done in basic
97 * blocks (sector size 1). But otherwise we extend the buffer by one
98 * extra log sector to ensure there's space to accommodate this
99 * possibility.
100 */
105 }
107 /*
108 * Return the address of the start of the given block number's data
109 * in a log buffer. The buffer covers a log sector-aligned region.
110 */
114 xfs_daddr_t blk_no)
115 {
117 }
119 static int
122 xfs_daddr_t blk_no,
126 {
134 }
147 }
149 }
151 STATIC int
154 xfs_daddr_t blk_no,
157 {
159 }
161 STATIC int
164 xfs_daddr_t blk_no,
168 {
175 }
177 STATIC int
180 xfs_daddr_t blk_no,
183 {
185 }
187 #ifdef DEBUG
188 /*
189 * dump debug superblock and log record information
190 */
191 STATIC void
195 {
200 }
201 #else
202 #define xlog_header_check_dump(mp, head)
203 #endif
205 /*
206 * check log record header for recovery
207 */
208 STATIC int
212 {
215 /*
216 * IRIX doesn't write the h_fmt field and leaves it zeroed
217 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
218 * a dirty log created in IRIX.
219 */
225 }
232 }
234 }
236 /*
237 * read the head block of the log and check the header
238 */
239 STATIC int
243 {
247 /*
248 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
249 * h_fs_uuid is null, we assume this log was last mounted
250 * by IRIX and continue.
251 */
258 }
260 }
262 /*
263 * This routine finds (to an approximation) the first block in the physical
264 * log which contains the given cycle. It uses a binary search algorithm.
265 * Note that the algorithm can not be perfect because the disk will not
266 * necessarily be perfect.
267 */
268 STATIC int
272 xfs_daddr_t first_blk,
274 uint cycle)
275 {
277 xfs_daddr_t mid_blk;
278 xfs_daddr_t end_blk;
279 uint mid_cycle;
291 else
294 }
301 }
303 /*
304 * Check that a range of blocks does not contain stop_on_cycle_no.
305 * Fill in *new_blk with the block offset where such a block is
306 * found, or with -1 (an invalid block number) if there is no such
307 * block in the range. The scan needs to occur from front to back
308 * and the pointer into the region must be updated since a later
309 * routine will need to perform another test.
310 */
311 STATIC int
314 xfs_daddr_t start_blk,
316 uint stop_on_cycle_no,
318 {
320 uint cycle;
322 xfs_daddr_t bufblks;
326 /*
327 * Greedily allocate a buffer big enough to handle the full
328 * range of basic blocks we'll be examining. If that fails,
329 * try a smaller size. We need to be able to read at least
330 * a log sector, or we're out of luck.
331 */
339 }
355 }
358 }
359 }
363 out:
366 }
370 {
377 }
379 }
381 /*
382 * Potentially backup over partial log record write.
383 *
384 * In the typical case, last_blk is the number of the block directly after
385 * a good log record. Therefore, we subtract one to get the block number
386 * of the last block in the given buffer. extra_bblks contains the number
387 * of blocks we would have read on a previous read. This happens when the
388 * last log record is split over the end of the physical log.
389 *
390 * extra_bblks is the number of blocks potentially verified on a previous
391 * call to this routine.
392 */
393 STATIC int
396 xfs_daddr_t start_blk,
399 {
400 xfs_daddr_t i;
422 }
426 /* valid log record not found */
432 }
438 }
447 }
449 /*
450 * We hit the beginning of the physical log & still no header. Return
451 * to caller. If caller can handle a return of -1, then this routine
452 * will be called again for the end of the physical log.
453 */
457 }
459 /*
460 * We have the final block of the good log (the first block
461 * of the log record _before_ the head. So we check the uuid.
462 */
466 /*
467 * We may have found a log record header before we expected one.
468 * last_blk will be the 1st block # with a given cycle #. We may end
469 * up reading an entire log record. In this case, we don't want to
470 * reset last_blk. Only when last_blk points in the middle of a log
471 * record do we update last_blk.
472 */
479 out:
482 }
484 /*
485 * Head is defined to be the point of the log where the next log write
486 * could go. This means that incomplete LR writes at the end are
487 * eliminated when calculating the head. We aren't guaranteed that previous
488 * LR have complete transactions. We only know that a cycle number of
489 * current cycle number -1 won't be present in the log if we start writing
490 * from our current block number.
491 *
492 * last_blk contains the block number of the first block with a given
493 * cycle number.
494 *
495 * Return: zero if normal, non-zero if error.
496 */
497 STATIC int
501 {
507 uint stop_on_cycle;
510 /* Is the end of the log device zeroed? */
515 }
519 /* Is the whole lot zeroed? */
521 /* Linux XFS shouldn't generate totally zeroed logs -
522 * mkfs etc write a dummy unmount record to a fresh
523 * log so we can store the uuid in there
524 */
526 }
529 }
550 /*
551 * If the 1st half cycle number is equal to the last half cycle number,
552 * then the entire log is stamped with the same cycle number. In this
553 * case, head_blk can't be set to zero (which makes sense). The below
554 * math doesn't work out properly with head_blk equal to zero. Instead,
555 * we set it to log_bbnum which is an invalid block number, but this
556 * value makes the math correct. If head_blk doesn't changed through
557 * all the tests below, *head_blk is set to zero at the very end rather
558 * than log_bbnum. In a sense, log_bbnum and zero are the same block
559 * in a circular file.
560 */
562 /*
563 * In this case we believe that the entire log should have
564 * cycle number last_half_cycle. We need to scan backwards
565 * from the end verifying that there are no holes still
566 * containing last_half_cycle - 1. If we find such a hole,
567 * then the start of that hole will be the new head. The
568 * simple case looks like
569 * x | x ... | x - 1 | x
570 * Another case that fits this picture would be
571 * x | x + 1 | x ... | x
572 * In this case the head really is somewhere at the end of the
573 * log, as one of the latest writes at the beginning was
574 * incomplete.
575 * One more case is
576 * x | x + 1 | x ... | x - 1 | x
577 * This is really the combination of the above two cases, and
578 * the head has to end up at the start of the x-1 hole at the
579 * end of the log.
580 *
581 * In the 256k log case, we will read from the beginning to the
582 * end of the log and search for cycle numbers equal to x-1.
583 * We don't worry about the x+1 blocks that we encounter,
584 * because we know that they cannot be the head since the log
585 * started with x.
586 */
590 /*
591 * In this case we want to find the first block with cycle
592 * number matching last_half_cycle. We expect the log to be
593 * some variation on
594 * x + 1 ... | x ... | x
595 * The first block with cycle number x (last_half_cycle) will
596 * be where the new head belongs. First we do a binary search
597 * for the first occurrence of last_half_cycle. The binary
598 * search may not be totally accurate, so then we scan back
599 * from there looking for occurrences of last_half_cycle before
600 * us. If that backwards scan wraps around the beginning of
601 * the log, then we look for occurrences of last_half_cycle - 1
602 * at the end of the log. The cases we're looking for look
603 * like
604 * v binary search stopped here
605 * x + 1 ... | x | x + 1 | x ... | x
606 * ^ but we want to locate this spot
607 * or
608 * <---------> less than scan distance
609 * x + 1 ... | x ... | x - 1 | x
610 * ^ we want to locate this spot
611 */
614 last_half_cycle);
617 }
619 /*
620 * Now validate the answer. Scan back some number of maximum possible
621 * blocks and make sure each one has the expected cycle number. The
622 * maximum is determined by the total possible amount of buffering
623 * in the in-core log. The following number can be made tighter if
624 * we actually look at the block size of the filesystem.
625 */
628 /*
629 * We are guaranteed that the entire check can be performed
630 * in one buffer.
631 */
640 /*
641 * We are going to scan backwards in the log in two parts.
642 * First we scan the physical end of the log. In this part
643 * of the log, we are looking for blocks with cycle number
644 * last_half_cycle - 1.
645 * If we find one, then we know that the log starts there, as
646 * we've found a hole that didn't get written in going around
647 * the end of the physical log. The simple case for this is
648 * x + 1 ... | x ... | x - 1 | x
649 * <---------> less than scan distance
650 * If all of the blocks at the end of the log have cycle number
651 * last_half_cycle, then we check the blocks at the start of
652 * the log looking for occurrences of last_half_cycle. If we
653 * find one, then our current estimate for the location of the
654 * first occurrence of last_half_cycle is wrong and we move
655 * back to the hole we've found. This case looks like
656 * x + 1 ... | x | x + 1 | x ...
657 * ^ binary search stopped here
658 * Another case we need to handle that only occurs in 256k
659 * logs is
660 * x + 1 ... | x ... | x+1 | x ...
661 * ^ binary search stops here
662 * In a 256k log, the scan at the end of the log will see the
663 * x + 1 blocks. We need to skip past those since that is
664 * certainly not the head of the log. By searching for
665 * last_half_cycle-1 we accomplish that.
666 */
677 }
679 /*
680 * Scan beginning of log now. The last part of the physical
681 * log is good. This scan needs to verify that it doesn't find
682 * the last_half_cycle.
683 */
692 }
694 validate_head:
695 /*
696 * Now we need to make sure head_blk is not pointing to a block in
697 * the middle of a log record.
698 */
703 /* start ptr at last block ptr before head_blk */
716 /* We hit the beginning of the log during our search */
732 }
737 else
739 /*
740 * When returning here, we have a good block number. Bad block
741 * means that during a previous crash, we didn't have a clean break
742 * from cycle number N to cycle number N-1. In this case, we need
743 * to find the first block with cycle number N-1.
744 */
747 out_free_buffer:
752 }
754 /*
755 * Seek backwards in the log for log record headers.
756 *
757 * Given a starting log block, walk backwards until we find the provided number
758 * of records or hit the provided tail block. The return value is the number of
759 * records encountered or a negative error code. The log block and buffer
760 * pointer of the last record seen are returned in rblk and rhead respectively.
761 */
762 STATIC int
765 xfs_daddr_t head_blk,
766 xfs_daddr_t tail_blk,
772 {
777 xfs_daddr_t end_blk;
781 /*
782 * Walk backwards from the head block until we hit the tail or the first
783 * block in the log.
784 */
796 }
797 }
799 /*
800 * If we haven't hit the tail block or the log record header count,
801 * start looking again from the end of the physical log. Note that
802 * callers can pass head == tail if the tail is not yet known.
803 */
817 }
818 }
819 }
823 out_error:
825 }
827 /*
828 * Seek forward in the log for log record headers.
829 *
830 * Given head and tail blocks, walk forward from the tail block until we find
831 * the provided number of records or hit the head block. The return value is the
832 * number of records encountered or a negative error code. The log block and
833 * buffer pointer of the last record seen are returned in rblk and rhead
834 * respectively.
835 */
836 STATIC int
839 xfs_daddr_t head_blk,
840 xfs_daddr_t tail_blk,
846 {
851 xfs_daddr_t end_blk;
855 /*
856 * Walk forward from the tail block until we hit the head or the last
857 * block in the log.
858 */
870 }
871 }
873 /*
874 * If we haven't hit the head block or the log record header count,
875 * start looking again from the start of the physical log.
876 */
890 }
891 }
892 }
896 out_error:
898 }
900 /*
901 * Calculate distance from head to tail (i.e., unused space in the log).
902 */
906 xfs_daddr_t head_blk,
907 xfs_daddr_t tail_blk)
908 {
913 }
915 /*
916 * Verify the log tail. This is particularly important when torn or incomplete
917 * writes have been detected near the front of the log and the head has been
918 * walked back accordingly.
919 *
920 * We also have to handle the case where the tail was pinned and the head
921 * blocked behind the tail right before a crash. If the tail had been pushed
922 * immediately prior to the crash and the subsequent checkpoint was only
923 * partially written, it's possible it overwrote the last referenced tail in the
924 * log with garbage. This is not a coherency problem because the tail must have
925 * been pushed before it can be overwritten, but appears as log corruption to
926 * recovery because we have no way to know the tail was updated if the
927 * subsequent checkpoint didn't write successfully.
928 *
929 * Therefore, CRC check the log from tail to head. If a failure occurs and the
930 * offending record is within max iclog bufs from the head, walk the tail
931 * forward and retry until a valid tail is found or corruption is detected out
932 * of the range of a possible overwrite.
933 */
934 STATIC int
937 xfs_daddr_t head_blk,
940 {
943 xfs_daddr_t first_bad;
946 xfs_daddr_t tmp_tail;
953 /*
954 * Make sure the tail points to a record (returns positive count on
955 * success).
956 */
964 /*
965 * Run a CRC check from the tail to the head. We can't just check
966 * MAX_ICLOGS records past the tail because the tail may point to stale
967 * blocks cleared during the search for the head/tail. These blocks are
968 * overwritten with zero-length records and thus record count is not a
969 * reliable indicator of the iclog state before a crash.
970 */
977 /*
978 * Is corruption within range of the head? If so, retry from
979 * the next record. Otherwise return an error.
980 */
985 /* skip to the next record; returns positive count on success */
995 }
1001 out:
1004 }
1006 /*
1007 * Detect and trim torn writes from the head of the log.
1008 *
1009 * Storage without sector atomicity guarantees can result in torn writes in the
1010 * log in the event of a crash. Our only means to detect this scenario is via
1011 * CRC verification. While we can't always be certain that CRC verification
1012 * failure is due to a torn write vs. an unrelated corruption, we do know that
1013 * only a certain number (XLOG_MAX_ICLOGS) of log records can be written out at
1014 * one time. Therefore, CRC verify up to XLOG_MAX_ICLOGS records at the head of
1015 * the log and treat failures in this range as torn writes as a matter of
1016 * policy. In the event of CRC failure, the head is walked back to the last good
1017 * record in the log and the tail is updated from that record and verified.
1018 */
1019 STATIC int
1028 {
1031 xfs_daddr_t first_bad;
1032 xfs_daddr_t tmp_rhead_blk;
1037 /*
1038 * Check the head of the log for torn writes. Search backwards from the
1039 * head until we hit the tail or the maximum number of log record I/Os
1040 * that could have been in flight at one time. Use a temporary buffer so
1041 * we don't trash the rhead/buffer pointers from the caller.
1042 */
1053 /*
1054 * Now run a CRC verification pass over the records starting at the
1055 * block found above to the current head. If a CRC failure occurs, the
1056 * log block of the first bad record is saved in first_bad.
1057 */
1061 /*
1062 * We've hit a potential torn write. Reset the error and warn
1063 * about it.
1064 */
1070 /*
1071 * Get the header block and buffer pointer for the last good
1072 * record before the bad record.
1073 *
1074 * Note that xlog_find_tail() clears the blocks at the new head
1075 * (i.e., the records with invalid CRC) if the cycle number
1076 * matches the current cycle.
1077 */
1085 /*
1086 * Reset the head block to the starting block of the first bad
1087 * log record and set the tail block based on the last good
1088 * record.
1089 *
1090 * Bail out if the updated head/tail match as this indicates
1091 * possible corruption outside of the acceptable
1092 * (XLOG_MAX_ICLOGS) range. This is a job for xfs_repair...
1093 */
1099 }
1100 }
1106 }
1108 /*
1109 * We need to make sure we handle log wrapping properly, so we can't use the
1110 * calculated logbno directly. Make sure it wraps to the correct bno inside the
1111 * log.
1112 *
1113 * The log is limited to 32 bit sizes, so we use the appropriate modulus
1114 * operation here and cast it back to a 64 bit daddr on return.
1115 */
1119 xfs_daddr_t bno)
1120 {
1125 }
1127 /*
1128 * Check whether the head of the log points to an unmount record. In other
1129 * words, determine whether the log is clean. If so, update the in-core state
1130 * appropriately.
1131 */
1132 static int
1138 xfs_daddr_t rhead_blk,
1141 {
1143 xfs_daddr_t umount_data_blk;
1144 xfs_daddr_t after_umount_blk;
1151 /*
1152 * Look for unmount record. If we find it, then we know there was a
1153 * clean unmount. Since 'i' could be the last block in the physical
1154 * log, we convert to a log block before comparing to the head_blk.
1155 *
1156 * Save the current tail lsn to use to pass to xlog_clear_stale_blocks()
1157 * below. We won't want to clear the unmount record if there is one, so
1158 * we pass the lsn of the unmount record rather than the block after it.
1159 */
1173 /*
1174 * Set tail and last sync so that newly written log
1175 * records will point recovery to after the current
1176 * unmount record.
1177 */
1185 }
1186 }
1189 }
1191 static void
1194 xfs_daddr_t head_blk,
1196 xfs_daddr_t rhead_blk,
1198 {
1199 /*
1200 * Reset log values according to the state of the log when we
1201 * crashed. In the case where head_blk == 0, we bump curr_cycle
1202 * one because the next write starts a new cycle rather than
1203 * continuing the cycle of the last good log record. At this
1204 * point we have guaranteed that all partial log records have been
1205 * accounted for. Therefore, we know that the last good log record
1206 * written was complete and ended exactly on the end boundary
1207 * of the physical log.
1208 */
1216 }
1218 /*
1219 * Find the sync block number or the tail of the log.
1220 *
1221 * This will be the block number of the last record to have its
1222 * associated buffers synced to disk. Every log record header has
1223 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
1224 * to get a sync block number. The only concern is to figure out which
1225 * log record header to believe.
1226 *
1227 * The following algorithm uses the log record header with the largest
1228 * lsn. The entire log record does not need to be valid. We only care
1229 * that the header is valid.
1230 *
1231 * We could speed up search by using current head_blk buffer, but it is not
1232 * available.
1233 */
1234 STATIC int
1239 {
1244 xfs_daddr_t rhead_blk;
1245 xfs_lsn_t tail_lsn;
1249 /*
1250 * Find previous log record
1251 */
1266 /* leave all other log inited values alone */
1268 }
1269 }
1271 /*
1272 * Search backwards through the log looking for the log record header
1273 * block. This wraps all the way back around to the head so something is
1274 * seriously wrong if we can't find it.
1275 */
1284 }
1287 /*
1288 * Set the log state based on the current head record.
1289 */
1293 /*
1294 * Look for an unmount record at the head of the log. This sets the log
1295 * state to determine whether recovery is necessary.
1296 */
1302 /*
1303 * Verify the log head if the log is not clean (e.g., we have anything
1304 * but an unmount record at the head). This uses CRC verification to
1305 * detect and trim torn writes. If discovered, CRC failures are
1306 * considered torn writes and the log head is trimmed accordingly.
1307 *
1308 * Note that we can only run CRC verification when the log is dirty
1309 * because there's no guarantee that the log data behind an unmount
1310 * record is compatible with the current architecture.
1311 */
1320 /* update in-core state again if the head changed */
1323 wrapped);
1330 }
1331 }
1333 /*
1334 * Note that the unmount was clean. If the unmount was not clean, we
1335 * need to know this to rebuild the superblock counters from the perag
1336 * headers if we have a filesystem using non-persistent counters.
1337 */
1341 /*
1342 * Make sure that there are no blocks in front of the head
1343 * with the same cycle number as the head. This can happen
1344 * because we allow multiple outstanding log writes concurrently,
1345 * and the later writes might make it out before earlier ones.
1346 *
1347 * We use the lsn from before modifying it so that we'll never
1348 * overwrite the unmount record after a clean unmount.
1349 *
1350 * Do this only if we are going to recover the filesystem
1351 *
1352 * NOTE: This used to say "if (!readonly)"
1353 * However on Linux, we can & do recover a read-only filesystem.
1354 * We only skip recovery if NORECOVERY is specified on mount,
1355 * in which case we would not be here.
1356 *
1357 * But... if the -device- itself is readonly, just skip this.
1358 * We can't recover this device anyway, so it won't matter.
1359 */
1363 done:
1369 }
1371 /*
1372 * Is the log zeroed at all?
1373 *
1374 * The last binary search should be changed to perform an X block read
1375 * once X becomes small enough. You can then search linearly through
1376 * the X blocks. This will cut down on the number of reads we need to do.
1377 *
1378 * If the log is partially zeroed, this routine will pass back the blkno
1379 * of the first block with cycle number 0. It won't have a complete LR
1380 * preceding it.
1381 *
1382 * Return:
1383 * 0 => the log is completely written to
1384 * 1 => use *blk_no as the first block of the log
1385 * <0 => error has occurred
1386 */
1387 STATIC int
1391 {
1396 xfs_daddr_t num_scan_bblks;
1402 /* check totally zeroed log */
1414 }
1416 /* check partially zeroed log */
1425 }
1427 /* we have a partially zeroed log */
1433 /*
1434 * Validate the answer. Because there is no way to guarantee that
1435 * the entire log is made up of log records which are the same size,
1436 * we scan over the defined maximum blocks. At this point, the maximum
1437 * is not chosen to mean anything special. XXXmiken
1438 */
1446 /*
1447 * We search for any instances of cycle number 0 that occur before
1448 * our current estimate of the head. What we're trying to detect is
1449 * 1 ... | 0 | 1 | 0...
1450 * ^ binary search ends here
1451 */
1458 /*
1459 * Potentially backup over partial log record write. We don't need
1460 * to search the end of the log because we know it is zero.
1461 */
1469 out_free_buffer:
1474 }
1476 /*
1477 * These are simple subroutines used by xlog_clear_stale_blocks() below
1478 * to initialize a buffer full of empty log record headers and write
1479 * them into the log.
1480 */
1481 STATIC void
1489 {
1501 }
1503 STATIC int
1511 {
1521 /*
1522 * Greedily allocate a buffer big enough to handle the full
1523 * range of basic blocks to be written. If that fails, try
1524 * a smaller size. We need to be able to write at least a
1525 * log sector, or we're out of luck.
1526 */
1534 }
1536 /* We may need to do a read at the start to fill in part of
1537 * the buffer in the starting sector not covered by the first
1538 * write below.
1539 */
1547 }
1555 /* We may need to do a read at the end to fill in part of
1556 * the buffer in the final sector not covered by the write.
1557 * If this is the same sector as the above read, skip it.
1558 */
1566 }
1573 }
1579 }
1581 out_free_buffer:
1584 }
1586 /*
1587 * This routine is called to blow away any incomplete log writes out
1588 * in front of the log head. We do this so that we won't become confused
1589 * if we come up, write only a little bit more, and then crash again.
1590 * If we leave the partial log records out there, this situation could
1591 * cause us to think those partial writes are valid blocks since they
1592 * have the current cycle number. We get rid of them by overwriting them
1593 * with empty log records with the old cycle number rather than the
1594 * current one.
1595 *
1596 * The tail lsn is passed in rather than taken from
1597 * the log so that we will not write over the unmount record after a
1598 * clean unmount in a 512 block log. Doing so would leave the log without
1599 * any valid log records in it until a new one was written. If we crashed
1600 * during that time we would not be able to recover.
1601 */
1602 STATIC int
1605 xfs_lsn_t tail_lsn)
1606 {
1618 /*
1619 * Figure out the distance between the new head of the log
1620 * and the tail. We want to write over any blocks beyond the
1621 * head that we may have written just before the crash, but
1622 * we don't want to overwrite the tail of the log.
1623 */
1625 /*
1626 * The tail is behind the head in the physical log,
1627 * so the distance from the head to the tail is the
1628 * distance from the head to the end of the log plus
1629 * the distance from the beginning of the log to the
1630 * tail.
1631 */
1638 /*
1639 * The head is behind the tail in the physical log,
1640 * so the distance from the head to the tail is just
1641 * the tail block minus the head block.
1642 */
1648 }
1650 /*
1651 * If the head is right up against the tail, we can't clear
1652 * anything.
1653 */
1657 }
1660 /*
1661 * Take the smaller of the maximum amount of outstanding I/O
1662 * we could have and the distance to the tail to clear out.
1663 * We take the smaller so that we don't overwrite the tail and
1664 * we don't waste all day writing from the head to the tail
1665 * for no reason.
1666 */
1670 /*
1671 * We can stomp all the blocks we need to without
1672 * wrapping around the end of the log. Just do it
1673 * in a single write. Use the cycle number of the
1674 * current cycle minus one so that the log will look like:
1675 * n ... | n - 1 ...
1676 */
1679 tail_block);
1683 /*
1684 * We need to wrap around the end of the physical log in
1685 * order to clear all the blocks. Do it in two separate
1686 * I/Os. The first write should be from the head to the
1687 * end of the physical log, and it should use the current
1688 * cycle number minus one just like above.
1689 */
1693 tail_block);
1698 /*
1699 * Now write the blocks at the start of the physical log.
1700 * This writes the remainder of the blocks we want to clear.
1701 * It uses the current cycle number since we're now on the
1702 * same cycle as the head so that we get:
1703 * n ... n ... | n - 1 ...
1704 * ^^^^^ blocks we're writing
1705 */
1711 }
1714 }
1716 /*
1717 * Release the recovered intent item in the AIL that matches the given intent
1718 * type and intent id.
1719 */
1720 void
1725 {
1739 }
1740 }
1742 int
1745 xfs_ino_t ino,
1747 {
1758 }
1764 }
1766 /*
1767 * Get an inode so that we can recover a log operation.
1768 *
1769 * Log intent items that target inodes effectively contain a file handle.
1770 * Check that the generation number matches the intent item like we do for
1771 * other file handles. Log intent items defined after this validation weakness
1772 * was identified must use this function.
1773 */
1774 int
1777 xfs_ino_t ino,
1780 {
1791 }
1795 }
1797 /******************************************************************************
1798 *
1799 * Log recover routines
1800 *
1801 ******************************************************************************
1802 */
1823 };
1828 {
1836 }
1838 /*
1839 * Sort the log items in the transaction.
1840 *
1841 * The ordering constraints are defined by the inode allocation and unlink
1842 * behaviour. The rules are:
1843 *
1844 * 1. Every item is only logged once in a given transaction. Hence it
1845 * represents the last logged state of the item. Hence ordering is
1846 * dependent on the order in which operations need to be performed so
1847 * required initial conditions are always met.
1848 *
1849 * 2. Cancelled buffers are recorded in pass 1 in a separate table and
1850 * there's nothing to replay from them so we can simply cull them
1851 * from the transaction. However, we can't do that until after we've
1852 * replayed all the other items because they may be dependent on the
1853 * cancelled buffer and replaying the cancelled buffer can remove it
1854 * form the cancelled buffer table. Hence they have to be done last.
1855 *
1856 * 3. Inode allocation buffers must be replayed before inode items that
1857 * read the buffer and replay changes into it. For filesystems using the
1858 * ICREATE transactions, this means XFS_LI_ICREATE objects need to get
1859 * treated the same as inode allocation buffers as they create and
1860 * initialise the buffers directly.
1861 *
1862 * 4. Inode unlink buffers must be replayed after inode items are replayed.
1863 * This ensures that inodes are completely flushed to the inode buffer
1864 * in a "free" state before we remove the unlinked inode list pointer.
1865 *
1866 * Hence the ordering needs to be inode allocation buffers first, inode items
1867 * second, inode unlink buffers third and cancelled buffers last.
1868 *
1869 * But there's a problem with that - we can't tell an inode allocation buffer
1870 * apart from a regular buffer, so we can't separate them. We can, however,
1871 * tell an inode unlink buffer from the others, and so we can separate them out
1872 * from all the other buffers and move them to last.
1873 *
1874 * Hence, 4 lists, in order from head to tail:
1875 * - buffer_list for all buffers except cancelled/inode unlink buffers
1876 * - item_list for all non-buffer items
1877 * - inode_buffer_list for inode unlink buffers
1878 * - cancel_list for the cancelled buffers
1879 *
1880 * Note that we add objects to the tail of the lists so that first-to-last
1881 * ordering is preserved within the lists. Adding objects to the head of the
1882 * list means when we traverse from the head we walk them in last-to-first
1883 * order. For cancelled buffers and inode unlink buffers this doesn't matter,
1884 * but for all other items there may be specific ordering that we need to
1885 * preserve.
1886 */
1887 STATIC int
1892 {
1911 /*
1912 * return the remaining items back to the transaction
1913 * item list so they can be freed in caller.
1914 */
1919 }
1941 }
1942 }
1954 }
1956 void
1959 xfs_daddr_t blkno,
1960 uint len,
1962 {
1965 }
1967 /*
1968 * Create a deferred work structure for resuming and tracking the progress of a
1969 * log intent item that was found during recovery.
1970 */
1971 void
1975 xfs_lsn_t lsn,
1977 {
1982 /*
1983 * Insert the intent into the AIL directly and drop one reference so
1984 * that finishing or canceling the work will drop the other.
1985 */
1988 }
1990 STATIC int
1996 {
2002 XLOG_RECOVER_PASS2);
2009 }
2012 }
2014 /*
2015 * Perform the transaction.
2016 *
2017 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2018 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2019 */
2020 STATIC int
2026 {
2034 #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100
2054 items_queued++;
2060 }
2065 }
2069 }
2071 out:
2077 }
2083 }
2085 STATIC void
2088 {
2095 }
2097 STATIC int
2103 {
2108 /*
2109 * If the transaction is empty, the header was split across this and the
2110 * previous record. Copy the rest of the header.
2111 */
2117 }
2124 }
2126 /* take the tail entry */
2128 ri_list);
2141 }
2143 /*
2144 * The next region to add is the start of a new region. It could be
2145 * a whole region or it could be the first part of a new region. Because
2146 * of this, the assumption here is that the type and size fields of all
2147 * format structures fit into the first 32 bits of the structure.
2148 *
2149 * This works because all regions must be 32 bit aligned. Therefore, we
2150 * either have both fields or we have neither field. In the case we have
2151 * neither field, the data part of the region is zero length. We only have
2152 * a log_op_header and can throw away the header since a new one will appear
2153 * later. If we have at least 4 bytes, then we can determine how many regions
2154 * will appear in the current log item.
2155 */
2156 STATIC int
2162 {
2170 /* we need to catch log corruptions here */
2173 __func__);
2176 }
2182 }
2184 /*
2185 * The transaction header can be arbitrarily split across op
2186 * records. If we don't have the whole thing here, copy what we
2187 * do have and handle the rest in the next record.
2188 */
2193 }
2199 /* take the tail entry */
2201 ri_list);
2204 /* tail item is in use, get a new one */
2208 }
2219 }
2224 }
2233 }
2235 /* Description region is ri_buf[0] */
2241 }
2243 /*
2244 * Free up any resources allocated by the transaction
2245 *
2246 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2247 */
2248 STATIC void
2251 {
2258 /* Free the regions in the item. */
2262 /* Free the item itself */
2265 }
2266 /* Free the transaction recover structure */
2268 }
2270 /*
2271 * On error or completion, trans is freed.
2272 */
2273 STATIC int
2282 {
2286 /* mask off ophdr transaction container flags */
2291 /*
2292 * Callees must not free the trans structure. We'll decide if we need to
2293 * free it or not based on the operation being done and it's result.
2294 */
2296 /* expected flag values */
2306 buffer_list);
2307 /* success or fail, we are now done with this transaction. */
2311 /* unexpected flag values */
2313 /* just skip trans */
2323 }
2327 }
2329 /*
2330 * Lookup the transaction recovery structure associated with the ID in the
2331 * current ophdr. If the transaction doesn't exist and the start flag is set in
2332 * the ophdr, then allocate a new transaction for future ID matches to find.
2333 * Either way, return what we found during the lookup - an existing transaction
2334 * or nothing.
2335 */
2341 {
2343 xlog_tid_t tid;
2351 }
2353 /*
2354 * skip over non-start transaction headers - we could be
2355 * processing slack space before the next transaction starts
2356 */
2362 /*
2363 * This is a new transaction so allocate a new recovery container to
2364 * hold the recovery ops that will follow.
2365 */
2373 /*
2374 * Nothing more to do for this ophdr. Items to be added to this new
2375 * transaction will be in subsequent ophdr containers.
2376 */
2378 }
2380 STATIC int
2390 {
2395 /* Do we understand who wrote this op? */
2402 }
2404 /*
2405 * Check the ophdr contains all the data it is supposed to contain.
2406 */
2412 }
2416 /* nothing to do, so skip over this ophdr */
2418 }
2420 /*
2421 * The recovered buffer queue is drained only once we know that all
2422 * recovery items for the current LSN have been processed. This is
2423 * required because:
2424 *
2425 * - Buffer write submission updates the metadata LSN of the buffer.
2426 * - Log recovery skips items with a metadata LSN >= the current LSN of
2427 * the recovery item.
2428 * - Separate recovery items against the same metadata buffer can share
2429 * a current LSN. I.e., consider that the LSN of a recovery item is
2430 * defined as the starting LSN of the first record in which its
2431 * transaction appears, that a record can hold multiple transactions,
2432 * and/or that a transaction can span multiple records.
2433 *
2434 * In other words, we are allowed to submit a buffer from log recovery
2435 * once per current LSN. Otherwise, we may incorrectly skip recovery
2436 * items and cause corruption.
2437 *
2438 * We don't know up front whether buffers are updated multiple times per
2439 * LSN. Therefore, track the current LSN of each commit log record as it
2440 * is processed and drain the queue when it changes. Use commit records
2441 * because they are ordered correctly by the logging code.
2442 */
2449 }
2453 }
2455 /*
2456 * There are two valid states of the r_state field. 0 indicates that the
2457 * transaction structure is in a normal state. We have either seen the
2458 * start of the transaction or the last operation we added was not a partial
2459 * operation. If the last operation we added to the transaction was a
2460 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2461 *
2462 * NOTE: skip LRs with 0 data length.
2463 */
2464 STATIC int
2472 {
2481 /* check the log format matches our own - else we can't recover */
2493 }
2495 /* errors will abort recovery */
2502 num_logops--;
2503 }
2505 }
2507 /* Take all the collected deferred ops and finish them in order. */
2508 static int
2512 {
2521 /*
2522 * Create a new transaction reservation from the captured
2523 * information. Set logcount to 1 to force the new transaction
2524 * to regrant every roll so that we can make forward progress
2525 * in recovery no matter how full the log might be.
2526 */
2536 }
2538 /*
2539 * Transfer to this new transaction all the dfops we captured
2540 * from recovering a single intent item.
2541 */
2548 }
2552 }
2554 /* Release all the captured defer ops and capture structures in this list. */
2555 static void
2559 {
2566 }
2567 }
2569 /*
2570 * When this is called, all of the log intent items which did not have
2571 * corresponding log done items should be in the AIL. What we do now is update
2572 * the data structures associated with each one.
2573 *
2574 * Since we process the log intent items in normal transactions, they will be
2575 * removed at some point after the commit. This prevents us from just walking
2576 * down the list processing each one. We'll use a flag in the intent item to
2577 * skip those that we've already processed and use the AIL iteration mechanism's
2578 * generation count to try to speed this up at least a bit.
2579 *
2580 * When we start, we know that the intents are the only things in the AIL. As we
2581 * process them, however, other items are added to the AIL. Hence we know we
2582 * have started recovery on all the pending intents when we find an non-intent
2583 * item in the AIL.
2584 */
2585 STATIC int
2588 {
2592 #if defined(DEBUG) || defined(XFS_WARN)
2593 xfs_lsn_t last_lsn;
2596 #endif
2601 /*
2602 * We should never see a redo item with a LSN higher than
2603 * the last transaction we found in the log at the start
2604 * of recovery.
2605 */
2608 /*
2609 * NOTE: If your intent processing routine can create more
2610 * deferred ops, you /must/ attach them to the capture list in
2611 * the recover routine or else those subsequent intents will be
2612 * replayed in the wrong order!
2613 *
2614 * The recovery function can free the log item, so we must not
2615 * access dfp->dfp_intent after it returns. It must dispose of
2616 * @dfp if it returns 0.
2617 */
2622 }
2631 err:
2634 }
2636 /*
2637 * A cancel occurs when the mount has failed and we're bailing out. Release all
2638 * pending log intent items that we haven't started recovery on so they don't
2639 * pin the AIL.
2640 */
2641 STATIC void
2644 {
2651 }
2652 }
2654 /*
2655 * Transfer ownership of the recovered pending work to the recovery transaction
2656 * and try to finish the work. If there is more work to be done, the dfp will
2657 * remain attached to the transaction. If not, the dfp is freed.
2658 */
2659 int
2663 {
2671 }
2673 /*
2674 * This routine performs a transaction to null out a bad inode pointer
2675 * in an agi unlinked inode hash bucket.
2676 */
2677 STATIC void
2681 {
2709 out_abort:
2711 out_error:
2715 }
2717 static int
2722 {
2745 /*
2746 * Ensure the inode is removed from the unlinked list
2747 * before we continue so that it won't race with
2748 * building the in-memory list here. This could be
2749 * serialised with the agibp lock, but that just
2750 * serialises via lockstepping and it's much simpler
2751 * just to flush the inodegc queue and wait for it to
2752 * complete.
2753 */
2757 }
2761 }
2772 }
2774 }
2776 /*
2777 * Recover AGI unlinked lists
2778 *
2779 * This is called during recovery to process any inodes which we unlinked but
2780 * not freed when the system crashed. These inodes will be on the lists in the
2781 * AGI blocks. What we do here is scan all the AGIs and fully truncate and free
2782 * any inodes found on the lists. Each inode is removed from the lists when it
2783 * has been fully truncated and is freed. The freeing of the inode and its
2784 * removal from the list must be atomic.
2785 *
2786 * If everything we touch in the agi processing loop is already in memory, this
2787 * loop can hold the cpu for a long time. It runs without lock contention,
2788 * memory allocation contention, the need wait for IO, etc, and so will run
2789 * until we either run out of inodes to process, run low on memory or we run out
2790 * of log space.
2791 *
2792 * This behaviour is bad for latency on single CPU and non-preemptible kernels,
2793 * and can prevent other filesystem work (such as CIL pushes) from running. This
2794 * can lead to deadlocks if the recovery process runs out of log reservation
2795 * space. Hence we need to yield the CPU when there is other kernel work
2796 * scheduled on this CPU to ensure other scheduled work can run without undue
2797 * latency.
2798 */
2799 static void
2802 {
2810 /*
2811 * AGI is b0rked. Don't process it.
2812 *
2813 * We should probably mark the filesystem as corrupt after we've
2814 * recovered all the ag's we can....
2815 */
2817 }
2819 /*
2820 * Unlock the buffer so that it can be acquired in the normal course of
2821 * the transaction to truncate and free each inode. Because we are not
2822 * racing with anyone else here for the AGI buffer, we don't even need
2823 * to hold it locked to read the initial unlinked bucket entries out of
2824 * the buffer. We keep buffer reference though, so that it stays pinned
2825 * in memory while we need the buffer.
2826 */
2833 /*
2834 * Bucket is unrecoverable, so only a repair scan can
2835 * free the remaining unlinked inodes. Just empty the
2836 * bucket and remaining inodes on it unreferenced and
2837 * unfreeable.
2838 */
2840 }
2841 }
2844 }
2846 static void
2849 {
2854 }
2856 STATIC void
2861 {
2868 }
2877 }
2878 }
2879 }
2881 /*
2882 * CRC check, unpack and process a log record.
2883 */
2884 STATIC int
2892 {
2894 __le32 crc;
2898 /*
2899 * Nothing else to do if this is a CRC verification pass. Just return
2900 * if this a record with a non-zero crc. Unfortunately, mkfs always
2901 * sets old_crc to 0 so we must consider this valid even on v5 supers.
2902 * Otherwise, return EFSBADCRC on failure so the callers up the stack
2903 * know precisely what failed.
2904 */
2909 }
2911 /*
2912 * We're in the normal recovery path. Issue a warning if and only if the
2913 * CRC in the header is non-zero. This is an advisory warning and the
2914 * zero CRC check prevents warnings from being emitted when upgrading
2915 * the kernel from one that does not add CRCs by default.
2916 */
2924 }
2926 /*
2927 * If the filesystem is CRC enabled, this mismatch becomes a
2928 * fatal log corruption failure.
2929 */
2933 }
2934 }
2939 buffer_list);
2940 }
2942 STATIC int
2946 xfs_daddr_t blkno,
2948 {
2961 }
2963 /*
2964 * LR body must have data (or it wouldn't have been written)
2965 * and h_len must not be greater than LR buffer size.
2966 */
2975 }
2977 /*
2978 * Read the log from tail to head and process the log records found.
2979 * Handle the two cases where the tail and head are in the same cycle
2980 * and where the active portion of the log wraps around the end of
2981 * the physical log separately. The pass parameter is passed through
2982 * to the routines called to process the data and is not looked at
2983 * here.
2984 */
2985 STATIC int
2988 xfs_daddr_t head_blk,
2989 xfs_daddr_t tail_blk,
2992 {
2995 xfs_daddr_t rhead_blk;
3016 /*
3017 * Read the header of the tail block and get the iclog buffer size from
3018 * h_size. Use this to tell how many sectors make up the log header.
3019 */
3021 /*
3022 * When using variable length iclogs, read first sector of
3023 * iclog header and extract the header size from it. Get a
3024 * new hbp that is the correct size.
3025 */
3032 /*
3033 * xfsprogs has a bug where record length is based on lsunit but
3034 * h_size (iclog size) is hardcoded to 32k. Now that we
3035 * unconditionally CRC verify the unmount record, this means the
3036 * log buffer can be too small for the record and cause an
3037 * overrun.
3038 *
3039 * Detect this condition here. Use lsunit for the buffer size as
3040 * long as this looks like the mkfs case. Otherwise, return an
3041 * error to avoid a buffer overrun.
3042 */
3051 }
3057 /*
3058 * This open codes xlog_logrec_hblks so that we can reuse the
3059 * fixed up h_size value calculated above. Without that we'd
3060 * still allocate the buffer based on the incorrect on-disk
3061 * size.
3062 */
3071 }
3072 }
3076 }
3082 }
3086 /*
3087 * Perform recovery around the end of the physical log.
3088 * When the head is not on the same cycle number as the tail,
3089 * we can't do a sequential recovery.
3090 */
3092 /*
3093 * Check for header wrapping around physical end-of-log
3094 */
3099 /* Read header in one read */
3105 /* This LR is split across physical log end */
3107 /* some data before physical log end */
3116 }
3118 /*
3119 * Note: this black magic still works with
3120 * large sector sizes (non-512) only because:
3121 * - we increased the buffer size originally
3122 * by 1 sector giving us enough extra space
3123 * for the second read;
3124 * - the log start is guaranteed to be sector
3125 * aligned;
3126 * - we read the log end (LR header start)
3127 * _first_, then the log start (LR header end)
3128 * - order is important.
3129 */
3132 wrapped_hblks,
3136 }
3146 /*
3147 * Read the log record data in multiple reads if it
3148 * wraps around the end of the log. Note that if the
3149 * header already wrapped, blk_no could point past the
3150 * end of the log. The record data is contiguous in
3151 * that case.
3152 */
3161 /* This log record is split across the
3162 * physical end of log */
3166 /* some data is before the physical
3167 * end of log */
3170 split_bblks =
3178 }
3180 /*
3181 * Note: this black magic still works with
3182 * large sector sizes (non-512) only because:
3183 * - we increased the buffer size originally
3184 * by 1 sector giving us enough extra space
3185 * for the second read;
3186 * - the log start is guaranteed to be sector
3187 * aligned;
3188 * - we read the log end (LR header start)
3189 * _first_, then the log start (LR header end)
3190 * - order is important.
3191 */
3197 }
3206 }
3211 }
3213 /* read first part of physical log */
3224 /* blocks in data section */
3238 }
3240 bread_err2:
3242 bread_err1:
3245 /*
3246 * Submit buffers that have been dirtied by the last record recovered.
3247 */
3250 /*
3251 * If there has been an item recovery error then we
3252 * cannot allow partial checkpoint writeback to
3253 * occur. We might have multiple checkpoints with the
3254 * same start LSN in this buffer list, and partial
3255 * writeback of a checkpoint in this situation can
3256 * prevent future recovery of all the changes in the
3257 * checkpoints at this start LSN.
3258 *
3259 * Note: Shutting down the filesystem will result in the
3260 * delwri submission marking all the buffers stale,
3261 * completing them and cleaning up _XBF_LOGRECOVERY
3262 * state without doing any IO.
3263 */
3265 }
3267 }
3272 /*
3273 * Transactions are freed at commit time but transactions without commit
3274 * records on disk are never committed. Free any that may be left in the
3275 * hash table.
3276 */
3283 }
3286 }
3288 /*
3289 * Do the recovery of the log. We actually do this in two phases.
3290 * The two passes are necessary in order to implement the function
3291 * of cancelling a record written into the log. The first pass
3292 * determines those things which have been cancelled, and the
3293 * second pass replays log items normally except for those which
3294 * have been cancelled. The handling of the replay and cancellations
3295 * takes place in the log item type specific routines.
3296 *
3297 * The table of items which have cancel records in the log is allocated
3298 * and freed at this level, since only here do we know when all of
3299 * the log recovery has been completed.
3300 */
3301 STATIC int
3304 xfs_daddr_t head_blk,
3305 xfs_daddr_t tail_blk)
3306 {
3311 /*
3312 * First do a pass to find all of the cancelled buf log items.
3313 * Store them in the buf_cancel_table for use in the second pass.
3314 */
3324 /*
3325 * Then do a second pass to actually recover the items in the log.
3326 * When it is complete free the table of buf cancel items.
3327 */
3332 out_cancel:
3335 }
3337 /*
3338 * Do the actual recovery
3339 */
3340 STATIC int
3343 xfs_daddr_t head_blk,
3344 xfs_daddr_t tail_blk)
3345 {
3353 /*
3354 * First replay the images in the log.
3355 */
3363 /*
3364 * We now update the tail_lsn since much of the recovery has completed
3365 * and there may be space available to use. If there were no extent or
3366 * iunlinks, we can free up the entire log. This was set in
3367 * xlog_find_tail to be the lsn of the last known good LR on disk. If
3368 * there are extent frees or iunlinks they will have some entries in the
3369 * AIL; so we look at the AIL to determine how to set the tail_lsn.
3370 */
3373 /*
3374 * Now that we've finished replaying all buffer and inode updates,
3375 * re-read the superblock and reverify it.
3376 */
3384 }
3387 }
3389 /* Convert superblock from on-disk format */
3393 /* re-initialise in-core superblock and geometry structures */
3397 /* Normal transactions can now occur */
3400 }
3402 /*
3403 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3404 *
3405 * Return error or zero.
3406 */
3407 int
3410 {
3414 /* find the tail of the log */
3419 /*
3420 * The superblock was read before the log was available and thus the LSN
3421 * could not be verified. Check the superblock LSN against the current
3422 * LSN now that it's known.
3423 */
3429 /* There used to be a comment here:
3430 *
3431 * disallow recovery on read-only mounts. note -- mount
3432 * checks for ENOSPC and turns it into an intelligent
3433 * error message.
3434 * ...but this is no longer true. Now, unless you specify
3435 * NORECOVERY (in which case this function would never be
3436 * called), we just go ahead and recover. We do this all
3437 * under the vfs layer, so we can get away with it unless
3438 * the device itself is read-only, in which case we fail.
3439 */
3442 }
3444 /*
3445 * Version 5 superblock log feature mask validation. We know the
3446 * log is dirty so check if there are any unknown log features
3447 * in what we need to recover. If there are unknown features
3448 * (e.g. unsupported transactions, then simply reject the
3449 * attempt at recovery before touching anything.
3450 */
3453 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
3457 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
3463 }
3465 /*
3466 * Delay log recovery if the debug hook is set. This is debug
3467 * instrumentation to coordinate simulation of I/O failures with
3468 * log recovery.
3469 */
3475 }
3483 }
3485 }
3487 /*
3488 * In the first part of recovery we replay inodes and buffers and build up the
3489 * list of intents which need to be processed. Here we process the intents and
3490 * clean up the on disk unlinked inode lists. This is separated from the first
3491 * part of recovery so that the root and real-time bitmap inodes can be read in
3492 * from disk in between the two stages. This is necessary so that we can free
3493 * space in the real-time portion of the file system.
3494 *
3495 * We run this whole process under GFP_NOFS allocation context. We do a
3496 * combination of non-transactional and transactional work, yet we really don't
3497 * want to recurse into the filesystem from direct reclaim during any of this
3498 * processing. This allows all the recovery code run here not to care about the
3499 * memory allocation context it is running in.
3500 */
3501 int
3504 {
3510 /*
3511 * Cancel all the unprocessed intent items now so that we don't
3512 * leave them pinned in the AIL. This can cause the AIL to
3513 * livelock on the pinned item if anyone tries to push the AIL
3514 * (inode reclaim does this) before we get around to
3515 * xfs_log_mount_cancel.
3516 */
3521 }
3523 /*
3524 * Sync the log to get all the intents out of the AIL. This isn't
3525 * absolutely necessary, but it helps in case the unlink transactions
3526 * would have problems pushing the intents out of the way.
3527 */
3532 /*
3533 * Recover any CoW staging blocks that are still referenced by the
3534 * ondisk refcount metadata. During mount there cannot be any live
3535 * staging extents as we have not permitted any user modifications.
3536 * Therefore, it is safe to free them all right now, even on a
3537 * read-only mount.
3538 */
3543 error);
3544 /*
3545 * If we get an error here, make sure the log is shut down
3546 * but return zero so that any log items committed since the
3547 * end of intents processing can be pushed through the CIL
3548 * and AIL.
3549 */
3553 }
3555 out_error:
3558 }
3560 void
3563 {
3566 }