| blob | 25cfc85dbaa7bd5e6f54d3fdcb050bac579d928d |
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
37 #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
39 STATIC int
42 xfs_daddr_t *);
43 STATIC int
46 xfs_lsn_t);
47 #if defined(DEBUG)
48 STATIC void
51 #else
52 #define xlog_recover_check_summary(log)
53 #endif
54 STATIC int
58 /*
59 * This structure is used during recovery to record the buf log items which
60 * have been canceled and should not be replayed.
61 */
63 xfs_daddr_t bc_blkno;
64 uint bc_len;
67 };
69 /*
70 * Sector aligned buffer routines for buffer create/read/write/access
71 */
73 /*
74 * Verify the log-relative block number and length in basic blocks are valid for
75 * an operation involving the given XFS log buffer. Returns true if the fields
76 * are valid, false otherwise.
77 */
81 xfs_daddr_t blk_no,
83 {
89 }
91 /*
92 * Allocate a buffer to hold log data. The buffer needs to be able to map to
93 * a range of nbblks basic blocks at any valid offset within the log.
94 */
99 {
102 /*
103 * Pass log block 0 since we don't have an addr yet, buffer will be
104 * verified on read.
105 */
108 nbblks);
110 }
112 /*
113 * We do log I/O in units of log sectors (a power-of-2 multiple of the
114 * basic block size), so we round up the requested size to accommodate
115 * the basic blocks required for complete log sectors.
116 *
117 * In addition, the buffer may be used for a non-sector-aligned block
118 * offset, in which case an I/O of the requested size could extend
119 * beyond the end of the buffer. If the requested size is only 1 basic
120 * block it will never straddle a sector boundary, so this won't be an
121 * issue. Nor will this be a problem if the log I/O is done in basic
122 * blocks (sector size 1). But otherwise we extend the buffer by one
123 * extra log sector to ensure there's space to accommodate this
124 * possibility.
125 */
130 }
132 /*
133 * Return the address of the start of the given block number's data
134 * in a log buffer. The buffer covers a log sector-aligned region.
135 */
139 xfs_daddr_t blk_no)
140 {
142 }
144 static int
147 xfs_daddr_t blk_no,
151 {
159 }
172 }
174 }
176 STATIC int
179 xfs_daddr_t blk_no,
182 {
184 }
186 STATIC int
189 xfs_daddr_t blk_no,
193 {
200 }
202 STATIC int
205 xfs_daddr_t blk_no,
208 {
210 }
212 #ifdef DEBUG
213 /*
214 * dump debug superblock and log record information
215 */
216 STATIC void
220 {
225 }
226 #else
227 #define xlog_header_check_dump(mp, head)
228 #endif
230 /*
231 * check log record header for recovery
232 */
233 STATIC int
237 {
240 /*
241 * IRIX doesn't write the h_fmt field and leaves it zeroed
242 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
243 * a dirty log created in IRIX.
244 */
250 }
257 }
259 }
261 /*
262 * read the head block of the log and check the header
263 */
264 STATIC int
268 {
272 /*
273 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
274 * h_fs_uuid is null, we assume this log was last mounted
275 * by IRIX and continue.
276 */
283 }
285 }
287 STATIC void
290 {
292 /*
293 * We're not going to bother about retrying
294 * this during recovery. One strike!
295 */
299 }
300 }
302 /*
303 * On v5 supers, a bli could be attached to update the metadata LSN.
304 * Clean it up.
305 */
312 }
314 /*
315 * This routine finds (to an approximation) the first block in the physical
316 * log which contains the given cycle. It uses a binary search algorithm.
317 * Note that the algorithm can not be perfect because the disk will not
318 * necessarily be perfect.
319 */
320 STATIC int
324 xfs_daddr_t first_blk,
326 uint cycle)
327 {
329 xfs_daddr_t mid_blk;
330 xfs_daddr_t end_blk;
331 uint mid_cycle;
343 else
346 }
353 }
355 /*
356 * Check that a range of blocks does not contain stop_on_cycle_no.
357 * Fill in *new_blk with the block offset where such a block is
358 * found, or with -1 (an invalid block number) if there is no such
359 * block in the range. The scan needs to occur from front to back
360 * and the pointer into the region must be updated since a later
361 * routine will need to perform another test.
362 */
363 STATIC int
366 xfs_daddr_t start_blk,
368 uint stop_on_cycle_no,
370 {
372 uint cycle;
374 xfs_daddr_t bufblks;
378 /*
379 * Greedily allocate a buffer big enough to handle the full
380 * range of basic blocks we'll be examining. If that fails,
381 * try a smaller size. We need to be able to read at least
382 * a log sector, or we're out of luck.
383 */
391 }
407 }
410 }
411 }
415 out:
418 }
420 /*
421 * Potentially backup over partial log record write.
422 *
423 * In the typical case, last_blk is the number of the block directly after
424 * a good log record. Therefore, we subtract one to get the block number
425 * of the last block in the given buffer. extra_bblks contains the number
426 * of blocks we would have read on a previous read. This happens when the
427 * last log record is split over the end of the physical log.
428 *
429 * extra_bblks is the number of blocks potentially verified on a previous
430 * call to this routine.
431 */
432 STATIC int
435 xfs_daddr_t start_blk,
438 {
439 xfs_daddr_t i;
461 }
465 /* valid log record not found */
471 }
477 }
486 }
488 /*
489 * We hit the beginning of the physical log & still no header. Return
490 * to caller. If caller can handle a return of -1, then this routine
491 * will be called again for the end of the physical log.
492 */
496 }
498 /*
499 * We have the final block of the good log (the first block
500 * of the log record _before_ the head. So we check the uuid.
501 */
505 /*
506 * We may have found a log record header before we expected one.
507 * last_blk will be the 1st block # with a given cycle #. We may end
508 * up reading an entire log record. In this case, we don't want to
509 * reset last_blk. Only when last_blk points in the middle of a log
510 * record do we update last_blk.
511 */
517 xhdrs++;
520 }
526 out:
529 }
531 /*
532 * Head is defined to be the point of the log where the next log write
533 * could go. This means that incomplete LR writes at the end are
534 * eliminated when calculating the head. We aren't guaranteed that previous
535 * LR have complete transactions. We only know that a cycle number of
536 * current cycle number -1 won't be present in the log if we start writing
537 * from our current block number.
538 *
539 * last_blk contains the block number of the first block with a given
540 * cycle number.
541 *
542 * Return: zero if normal, non-zero if error.
543 */
544 STATIC int
548 {
554 uint stop_on_cycle;
557 /* Is the end of the log device zeroed? */
562 }
566 /* Is the whole lot zeroed? */
568 /* Linux XFS shouldn't generate totally zeroed logs -
569 * mkfs etc write a dummy unmount record to a fresh
570 * log so we can store the uuid in there
571 */
573 }
576 }
597 /*
598 * If the 1st half cycle number is equal to the last half cycle number,
599 * then the entire log is stamped with the same cycle number. In this
600 * case, head_blk can't be set to zero (which makes sense). The below
601 * math doesn't work out properly with head_blk equal to zero. Instead,
602 * we set it to log_bbnum which is an invalid block number, but this
603 * value makes the math correct. If head_blk doesn't changed through
604 * all the tests below, *head_blk is set to zero at the very end rather
605 * than log_bbnum. In a sense, log_bbnum and zero are the same block
606 * in a circular file.
607 */
609 /*
610 * In this case we believe that the entire log should have
611 * cycle number last_half_cycle. We need to scan backwards
612 * from the end verifying that there are no holes still
613 * containing last_half_cycle - 1. If we find such a hole,
614 * then the start of that hole will be the new head. The
615 * simple case looks like
616 * x | x ... | x - 1 | x
617 * Another case that fits this picture would be
618 * x | x + 1 | x ... | x
619 * In this case the head really is somewhere at the end of the
620 * log, as one of the latest writes at the beginning was
621 * incomplete.
622 * One more case is
623 * x | x + 1 | x ... | x - 1 | x
624 * This is really the combination of the above two cases, and
625 * the head has to end up at the start of the x-1 hole at the
626 * end of the log.
627 *
628 * In the 256k log case, we will read from the beginning to the
629 * end of the log and search for cycle numbers equal to x-1.
630 * We don't worry about the x+1 blocks that we encounter,
631 * because we know that they cannot be the head since the log
632 * started with x.
633 */
637 /*
638 * In this case we want to find the first block with cycle
639 * number matching last_half_cycle. We expect the log to be
640 * some variation on
641 * x + 1 ... | x ... | x
642 * The first block with cycle number x (last_half_cycle) will
643 * be where the new head belongs. First we do a binary search
644 * for the first occurrence of last_half_cycle. The binary
645 * search may not be totally accurate, so then we scan back
646 * from there looking for occurrences of last_half_cycle before
647 * us. If that backwards scan wraps around the beginning of
648 * the log, then we look for occurrences of last_half_cycle - 1
649 * at the end of the log. The cases we're looking for look
650 * like
651 * v binary search stopped here
652 * x + 1 ... | x | x + 1 | x ... | x
653 * ^ but we want to locate this spot
654 * or
655 * <---------> less than scan distance
656 * x + 1 ... | x ... | x - 1 | x
657 * ^ we want to locate this spot
658 */
661 last_half_cycle);
664 }
666 /*
667 * Now validate the answer. Scan back some number of maximum possible
668 * blocks and make sure each one has the expected cycle number. The
669 * maximum is determined by the total possible amount of buffering
670 * in the in-core log. The following number can be made tighter if
671 * we actually look at the block size of the filesystem.
672 */
675 /*
676 * We are guaranteed that the entire check can be performed
677 * in one buffer.
678 */
687 /*
688 * We are going to scan backwards in the log in two parts.
689 * First we scan the physical end of the log. In this part
690 * of the log, we are looking for blocks with cycle number
691 * last_half_cycle - 1.
692 * If we find one, then we know that the log starts there, as
693 * we've found a hole that didn't get written in going around
694 * the end of the physical log. The simple case for this is
695 * x + 1 ... | x ... | x - 1 | x
696 * <---------> less than scan distance
697 * If all of the blocks at the end of the log have cycle number
698 * last_half_cycle, then we check the blocks at the start of
699 * the log looking for occurrences of last_half_cycle. If we
700 * find one, then our current estimate for the location of the
701 * first occurrence of last_half_cycle is wrong and we move
702 * back to the hole we've found. This case looks like
703 * x + 1 ... | x | x + 1 | x ...
704 * ^ binary search stopped here
705 * Another case we need to handle that only occurs in 256k
706 * logs is
707 * x + 1 ... | x ... | x+1 | x ...
708 * ^ binary search stops here
709 * In a 256k log, the scan at the end of the log will see the
710 * x + 1 blocks. We need to skip past those since that is
711 * certainly not the head of the log. By searching for
712 * last_half_cycle-1 we accomplish that.
713 */
724 }
726 /*
727 * Scan beginning of log now. The last part of the physical
728 * log is good. This scan needs to verify that it doesn't find
729 * the last_half_cycle.
730 */
739 }
741 validate_head:
742 /*
743 * Now we need to make sure head_blk is not pointing to a block in
744 * the middle of a log record.
745 */
750 /* start ptr at last block ptr before head_blk */
763 /* We hit the beginning of the log during our search */
779 }
784 else
786 /*
787 * When returning here, we have a good block number. Bad block
788 * means that during a previous crash, we didn't have a clean break
789 * from cycle number N to cycle number N-1. In this case, we need
790 * to find the first block with cycle number N-1.
791 */
794 out_free_buffer:
799 }
801 /*
802 * Seek backwards in the log for log record headers.
803 *
804 * Given a starting log block, walk backwards until we find the provided number
805 * of records or hit the provided tail block. The return value is the number of
806 * records encountered or a negative error code. The log block and buffer
807 * pointer of the last record seen are returned in rblk and rhead respectively.
808 */
809 STATIC int
812 xfs_daddr_t head_blk,
813 xfs_daddr_t tail_blk,
819 {
824 xfs_daddr_t end_blk;
828 /*
829 * Walk backwards from the head block until we hit the tail or the first
830 * block in the log.
831 */
843 }
844 }
846 /*
847 * If we haven't hit the tail block or the log record header count,
848 * start looking again from the end of the physical log. Note that
849 * callers can pass head == tail if the tail is not yet known.
850 */
864 }
865 }
866 }
870 out_error:
872 }
874 /*
875 * Seek forward in the log for log record headers.
876 *
877 * Given head and tail blocks, walk forward from the tail block until we find
878 * the provided number of records or hit the head block. The return value is the
879 * number of records encountered or a negative error code. The log block and
880 * buffer pointer of the last record seen are returned in rblk and rhead
881 * respectively.
882 */
883 STATIC int
886 xfs_daddr_t head_blk,
887 xfs_daddr_t tail_blk,
893 {
898 xfs_daddr_t end_blk;
902 /*
903 * Walk forward from the tail block until we hit the head or the last
904 * block in the log.
905 */
917 }
918 }
920 /*
921 * If we haven't hit the head block or the log record header count,
922 * start looking again from the start of the physical log.
923 */
937 }
938 }
939 }
943 out_error:
945 }
947 /*
948 * Calculate distance from head to tail (i.e., unused space in the log).
949 */
953 xfs_daddr_t head_blk,
954 xfs_daddr_t tail_blk)
955 {
960 }
962 /*
963 * Verify the log tail. This is particularly important when torn or incomplete
964 * writes have been detected near the front of the log and the head has been
965 * walked back accordingly.
966 *
967 * We also have to handle the case where the tail was pinned and the head
968 * blocked behind the tail right before a crash. If the tail had been pushed
969 * immediately prior to the crash and the subsequent checkpoint was only
970 * partially written, it's possible it overwrote the last referenced tail in the
971 * log with garbage. This is not a coherency problem because the tail must have
972 * been pushed before it can be overwritten, but appears as log corruption to
973 * recovery because we have no way to know the tail was updated if the
974 * subsequent checkpoint didn't write successfully.
975 *
976 * Therefore, CRC check the log from tail to head. If a failure occurs and the
977 * offending record is within max iclog bufs from the head, walk the tail
978 * forward and retry until a valid tail is found or corruption is detected out
979 * of the range of a possible overwrite.
980 */
981 STATIC int
984 xfs_daddr_t head_blk,
987 {
990 xfs_daddr_t first_bad;
993 xfs_daddr_t tmp_tail;
1000 /*
1001 * Make sure the tail points to a record (returns positive count on
1002 * success).
1003 */
1011 /*
1012 * Run a CRC check from the tail to the head. We can't just check
1013 * MAX_ICLOGS records past the tail because the tail may point to stale
1014 * blocks cleared during the search for the head/tail. These blocks are
1015 * overwritten with zero-length records and thus record count is not a
1016 * reliable indicator of the iclog state before a crash.
1017 */
1024 /*
1025 * Is corruption within range of the head? If so, retry from
1026 * the next record. Otherwise return an error.
1027 */
1032 /* skip to the next record; returns positive count on success */
1042 }
1048 out:
1051 }
1053 /*
1054 * Detect and trim torn writes from the head of the log.
1055 *
1056 * Storage without sector atomicity guarantees can result in torn writes in the
1057 * log in the event of a crash. Our only means to detect this scenario is via
1058 * CRC verification. While we can't always be certain that CRC verification
1059 * failure is due to a torn write vs. an unrelated corruption, we do know that
1060 * only a certain number (XLOG_MAX_ICLOGS) of log records can be written out at
1061 * one time. Therefore, CRC verify up to XLOG_MAX_ICLOGS records at the head of
1062 * the log and treat failures in this range as torn writes as a matter of
1063 * policy. In the event of CRC failure, the head is walked back to the last good
1064 * record in the log and the tail is updated from that record and verified.
1065 */
1066 STATIC int
1075 {
1078 xfs_daddr_t first_bad;
1079 xfs_daddr_t tmp_rhead_blk;
1084 /*
1085 * Check the head of the log for torn writes. Search backwards from the
1086 * head until we hit the tail or the maximum number of log record I/Os
1087 * that could have been in flight at one time. Use a temporary buffer so
1088 * we don't trash the rhead/buffer pointers from the caller.
1089 */
1100 /*
1101 * Now run a CRC verification pass over the records starting at the
1102 * block found above to the current head. If a CRC failure occurs, the
1103 * log block of the first bad record is saved in first_bad.
1104 */
1108 /*
1109 * We've hit a potential torn write. Reset the error and warn
1110 * about it.
1111 */
1117 /*
1118 * Get the header block and buffer pointer for the last good
1119 * record before the bad record.
1120 *
1121 * Note that xlog_find_tail() clears the blocks at the new head
1122 * (i.e., the records with invalid CRC) if the cycle number
1123 * matches the the current cycle.
1124 */
1132 /*
1133 * Reset the head block to the starting block of the first bad
1134 * log record and set the tail block based on the last good
1135 * record.
1136 *
1137 * Bail out if the updated head/tail match as this indicates
1138 * possible corruption outside of the acceptable
1139 * (XLOG_MAX_ICLOGS) range. This is a job for xfs_repair...
1140 */
1146 }
1147 }
1153 }
1155 /*
1156 * We need to make sure we handle log wrapping properly, so we can't use the
1157 * calculated logbno directly. Make sure it wraps to the correct bno inside the
1158 * log.
1159 *
1160 * The log is limited to 32 bit sizes, so we use the appropriate modulus
1161 * operation here and cast it back to a 64 bit daddr on return.
1162 */
1166 xfs_daddr_t bno)
1167 {
1172 }
1174 /*
1175 * Check whether the head of the log points to an unmount record. In other
1176 * words, determine whether the log is clean. If so, update the in-core state
1177 * appropriately.
1178 */
1179 static int
1185 xfs_daddr_t rhead_blk,
1188 {
1190 xfs_daddr_t umount_data_blk;
1191 xfs_daddr_t after_umount_blk;
1198 /*
1199 * Look for unmount record. If we find it, then we know there was a
1200 * clean unmount. Since 'i' could be the last block in the physical
1201 * log, we convert to a log block before comparing to the head_blk.
1202 *
1203 * Save the current tail lsn to use to pass to xlog_clear_stale_blocks()
1204 * below. We won't want to clear the unmount record if there is one, so
1205 * we pass the lsn of the unmount record rather than the block after it.
1206 */
1215 hblks++;
1218 }
1221 }
1235 /*
1236 * Set tail and last sync so that newly written log
1237 * records will point recovery to after the current
1238 * unmount record.
1239 */
1247 }
1248 }
1251 }
1253 static void
1256 xfs_daddr_t head_blk,
1258 xfs_daddr_t rhead_blk,
1260 {
1261 /*
1262 * Reset log values according to the state of the log when we
1263 * crashed. In the case where head_blk == 0, we bump curr_cycle
1264 * one because the next write starts a new cycle rather than
1265 * continuing the cycle of the last good log record. At this
1266 * point we have guaranteed that all partial log records have been
1267 * accounted for. Therefore, we know that the last good log record
1268 * written was complete and ended exactly on the end boundary
1269 * of the physical log.
1270 */
1282 }
1284 /*
1285 * Find the sync block number or the tail of the log.
1286 *
1287 * This will be the block number of the last record to have its
1288 * associated buffers synced to disk. Every log record header has
1289 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
1290 * to get a sync block number. The only concern is to figure out which
1291 * log record header to believe.
1292 *
1293 * The following algorithm uses the log record header with the largest
1294 * lsn. The entire log record does not need to be valid. We only care
1295 * that the header is valid.
1296 *
1297 * We could speed up search by using current head_blk buffer, but it is not
1298 * available.
1299 */
1300 STATIC int
1305 {
1310 xfs_daddr_t rhead_blk;
1311 xfs_lsn_t tail_lsn;
1315 /*
1316 * Find previous log record
1317 */
1332 /* leave all other log inited values alone */
1334 }
1335 }
1337 /*
1338 * Search backwards through the log looking for the log record header
1339 * block. This wraps all the way back around to the head so something is
1340 * seriously wrong if we can't find it.
1341 */
1350 }
1353 /*
1354 * Set the log state based on the current head record.
1355 */
1359 /*
1360 * Look for an unmount record at the head of the log. This sets the log
1361 * state to determine whether recovery is necessary.
1362 */
1368 /*
1369 * Verify the log head if the log is not clean (e.g., we have anything
1370 * but an unmount record at the head). This uses CRC verification to
1371 * detect and trim torn writes. If discovered, CRC failures are
1372 * considered torn writes and the log head is trimmed accordingly.
1373 *
1374 * Note that we can only run CRC verification when the log is dirty
1375 * because there's no guarantee that the log data behind an unmount
1376 * record is compatible with the current architecture.
1377 */
1386 /* update in-core state again if the head changed */
1389 wrapped);
1396 }
1397 }
1399 /*
1400 * Note that the unmount was clean. If the unmount was not clean, we
1401 * need to know this to rebuild the superblock counters from the perag
1402 * headers if we have a filesystem using non-persistent counters.
1403 */
1407 /*
1408 * Make sure that there are no blocks in front of the head
1409 * with the same cycle number as the head. This can happen
1410 * because we allow multiple outstanding log writes concurrently,
1411 * and the later writes might make it out before earlier ones.
1412 *
1413 * We use the lsn from before modifying it so that we'll never
1414 * overwrite the unmount record after a clean unmount.
1415 *
1416 * Do this only if we are going to recover the filesystem
1417 *
1418 * NOTE: This used to say "if (!readonly)"
1419 * However on Linux, we can & do recover a read-only filesystem.
1420 * We only skip recovery if NORECOVERY is specified on mount,
1421 * in which case we would not be here.
1422 *
1423 * But... if the -device- itself is readonly, just skip this.
1424 * We can't recover this device anyway, so it won't matter.
1425 */
1429 done:
1435 }
1437 /*
1438 * Is the log zeroed at all?
1439 *
1440 * The last binary search should be changed to perform an X block read
1441 * once X becomes small enough. You can then search linearly through
1442 * the X blocks. This will cut down on the number of reads we need to do.
1443 *
1444 * If the log is partially zeroed, this routine will pass back the blkno
1445 * of the first block with cycle number 0. It won't have a complete LR
1446 * preceding it.
1447 *
1448 * Return:
1449 * 0 => the log is completely written to
1450 * 1 => use *blk_no as the first block of the log
1451 * <0 => error has occurred
1452 */
1453 STATIC int
1457 {
1462 xfs_daddr_t num_scan_bblks;
1467 /* check totally zeroed log */
1480 }
1482 /* check partially zeroed log */
1491 }
1493 /* we have a partially zeroed log */
1499 /*
1500 * Validate the answer. Because there is no way to guarantee that
1501 * the entire log is made up of log records which are the same size,
1502 * we scan over the defined maximum blocks. At this point, the maximum
1503 * is not chosen to mean anything special. XXXmiken
1504 */
1512 /*
1513 * We search for any instances of cycle number 0 that occur before
1514 * our current estimate of the head. What we're trying to detect is
1515 * 1 ... | 0 | 1 | 0...
1516 * ^ binary search ends here
1517 */
1524 /*
1525 * Potentially backup over partial log record write. We don't need
1526 * to search the end of the log because we know it is zero.
1527 */
1535 out_free_buffer:
1540 }
1542 /*
1543 * These are simple subroutines used by xlog_clear_stale_blocks() below
1544 * to initialize a buffer full of empty log record headers and write
1545 * them into the log.
1546 */
1547 STATIC void
1555 {
1567 }
1569 STATIC int
1577 {
1587 /*
1588 * Greedily allocate a buffer big enough to handle the full
1589 * range of basic blocks to be written. If that fails, try
1590 * a smaller size. We need to be able to write at least a
1591 * log sector, or we're out of luck.
1592 */
1600 }
1602 /* We may need to do a read at the start to fill in part of
1603 * the buffer in the starting sector not covered by the first
1604 * write below.
1605 */
1613 }
1621 /* We may need to do a read at the end to fill in part of
1622 * the buffer in the final sector not covered by the write.
1623 * If this is the same sector as the above read, skip it.
1624 */
1632 }
1639 }
1645 }
1647 out_free_buffer:
1650 }
1652 /*
1653 * This routine is called to blow away any incomplete log writes out
1654 * in front of the log head. We do this so that we won't become confused
1655 * if we come up, write only a little bit more, and then crash again.
1656 * If we leave the partial log records out there, this situation could
1657 * cause us to think those partial writes are valid blocks since they
1658 * have the current cycle number. We get rid of them by overwriting them
1659 * with empty log records with the old cycle number rather than the
1660 * current one.
1661 *
1662 * The tail lsn is passed in rather than taken from
1663 * the log so that we will not write over the unmount record after a
1664 * clean unmount in a 512 block log. Doing so would leave the log without
1665 * any valid log records in it until a new one was written. If we crashed
1666 * during that time we would not be able to recover.
1667 */
1668 STATIC int
1671 xfs_lsn_t tail_lsn)
1672 {
1684 /*
1685 * Figure out the distance between the new head of the log
1686 * and the tail. We want to write over any blocks beyond the
1687 * head that we may have written just before the crash, but
1688 * we don't want to overwrite the tail of the log.
1689 */
1691 /*
1692 * The tail is behind the head in the physical log,
1693 * so the distance from the head to the tail is the
1694 * distance from the head to the end of the log plus
1695 * the distance from the beginning of the log to the
1696 * tail.
1697 */
1704 /*
1705 * The head is behind the tail in the physical log,
1706 * so the distance from the head to the tail is just
1707 * the tail block minus the head block.
1708 */
1714 }
1716 /*
1717 * If the head is right up against the tail, we can't clear
1718 * anything.
1719 */
1723 }
1726 /*
1727 * Take the smaller of the maximum amount of outstanding I/O
1728 * we could have and the distance to the tail to clear out.
1729 * We take the smaller so that we don't overwrite the tail and
1730 * we don't waste all day writing from the head to the tail
1731 * for no reason.
1732 */
1736 /*
1737 * We can stomp all the blocks we need to without
1738 * wrapping around the end of the log. Just do it
1739 * in a single write. Use the cycle number of the
1740 * current cycle minus one so that the log will look like:
1741 * n ... | n - 1 ...
1742 */
1745 tail_block);
1749 /*
1750 * We need to wrap around the end of the physical log in
1751 * order to clear all the blocks. Do it in two separate
1752 * I/Os. The first write should be from the head to the
1753 * end of the physical log, and it should use the current
1754 * cycle number minus one just like above.
1755 */
1759 tail_block);
1764 /*
1765 * Now write the blocks at the start of the physical log.
1766 * This writes the remainder of the blocks we want to clear.
1767 * It uses the current cycle number since we're now on the
1768 * same cycle as the head so that we get:
1769 * n ... n ... | n - 1 ...
1770 * ^^^^^ blocks we're writing
1771 */
1777 }
1780 }
1782 /******************************************************************************
1783 *
1784 * Log recover routines
1785 *
1786 ******************************************************************************
1787 */
1789 /*
1790 * Sort the log items in the transaction.
1791 *
1792 * The ordering constraints are defined by the inode allocation and unlink
1793 * behaviour. The rules are:
1794 *
1795 * 1. Every item is only logged once in a given transaction. Hence it
1796 * represents the last logged state of the item. Hence ordering is
1797 * dependent on the order in which operations need to be performed so
1798 * required initial conditions are always met.
1799 *
1800 * 2. Cancelled buffers are recorded in pass 1 in a separate table and
1801 * there's nothing to replay from them so we can simply cull them
1802 * from the transaction. However, we can't do that until after we've
1803 * replayed all the other items because they may be dependent on the
1804 * cancelled buffer and replaying the cancelled buffer can remove it
1805 * form the cancelled buffer table. Hence they have tobe done last.
1806 *
1807 * 3. Inode allocation buffers must be replayed before inode items that
1808 * read the buffer and replay changes into it. For filesystems using the
1809 * ICREATE transactions, this means XFS_LI_ICREATE objects need to get
1810 * treated the same as inode allocation buffers as they create and
1811 * initialise the buffers directly.
1812 *
1813 * 4. Inode unlink buffers must be replayed after inode items are replayed.
1814 * This ensures that inodes are completely flushed to the inode buffer
1815 * in a "free" state before we remove the unlinked inode list pointer.
1816 *
1817 * Hence the ordering needs to be inode allocation buffers first, inode items
1818 * second, inode unlink buffers third and cancelled buffers last.
1819 *
1820 * But there's a problem with that - we can't tell an inode allocation buffer
1821 * apart from a regular buffer, so we can't separate them. We can, however,
1822 * tell an inode unlink buffer from the others, and so we can separate them out
1823 * from all the other buffers and move them to last.
1824 *
1825 * Hence, 4 lists, in order from head to tail:
1826 * - buffer_list for all buffers except cancelled/inode unlink buffers
1827 * - item_list for all non-buffer items
1828 * - inode_buffer_list for inode unlink buffers
1829 * - cancel_list for the cancelled buffers
1830 *
1831 * Note that we add objects to the tail of the lists so that first-to-last
1832 * ordering is preserved within the lists. Adding objects to the head of the
1833 * list means when we traverse from the head we walk them in last-to-first
1834 * order. For cancelled buffers and inode unlink buffers this doesn't matter,
1835 * but for all other items there may be specific ordering that we need to
1836 * preserve.
1837 */
1838 STATIC int
1843 {
1866 }
1870 }
1891 __func__);
1893 /*
1894 * return the remaining items back to the transaction
1895 * item list so they can be freed in caller.
1896 */
1901 }
1902 }
1903 out:
1914 }
1916 /*
1917 * Build up the table of buf cancel records so that we don't replay
1918 * cancelled data in the second pass. For buffer records that are
1919 * not cancel records, there is nothing to do here so we just return.
1920 *
1921 * If we get a cancel record which is already in the table, this indicates
1922 * that the buffer was cancelled multiple times. In order to ensure
1923 * that during pass 2 we keep the record in the table until we reach its
1924 * last occurrence in the log, we keep a reference count in the cancel
1925 * record in the table to tell us how many times we expect to see this
1926 * record during the second pass.
1927 */
1928 STATIC int
1932 {
1941 }
1943 /*
1944 * If this isn't a cancel buffer item, then just return.
1945 */
1949 }
1951 /*
1952 * Insert an xfs_buf_cancel record into the hash table of them.
1953 * If there is already an identical record, bump its reference count.
1954 */
1962 }
1963 }
1973 }
1975 /*
1976 * Check to see whether the buffer being recovered has a corresponding
1977 * entry in the buffer cancel record table. If it is, return the cancel
1978 * buffer structure to the caller.
1979 */
1983 xfs_daddr_t blkno,
1984 uint len,
1986 {
1991 /* empty table means no cancelled buffers in the log */
1994 }
2000 }
2002 /*
2003 * We didn't find a corresponding entry in the table, so return 0 so
2004 * that the buffer is NOT cancelled.
2005 */
2008 }
2010 /*
2011 * If the buffer is being cancelled then return 1 so that it will be cancelled,
2012 * otherwise return 0. If the buffer is actually a buffer cancel item
2013 * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the
2014 * table and remove it from the table if this is the last reference.
2015 *
2016 * We remove the cancel record from the table when we encounter its last
2017 * occurrence in the log so that if the same buffer is re-used again after its
2018 * last cancellation we actually replay the changes made at that point.
2019 */
2020 STATIC int
2023 xfs_daddr_t blkno,
2024 uint len,
2026 {
2033 /*
2034 * We've go a match, so return 1 so that the recovery of this buffer
2035 * is cancelled. If this buffer is actually a buffer cancel log
2036 * item, then decrement the refcount on the one in the table and
2037 * remove it if this is the last reference.
2038 */
2043 }
2044 }
2046 }
2048 /*
2049 * Perform recovery for a buffer full of inodes. In these buffers, the only
2050 * data which should be recovered is that which corresponds to the
2051 * di_next_unlinked pointers in the on disk inode structures. The rest of the
2052 * data for the inodes is always logged through the inodes themselves rather
2053 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
2054 *
2055 * The only time when buffers full of inodes are fully recovered is when the
2056 * buffer is full of newly allocated inodes. In this case the buffer will
2057 * not be marked as an inode buffer and so will be sent to
2058 * xlog_recover_do_reg_buffer() below during recovery.
2059 */
2060 STATIC int
2066 {
2080 /*
2081 * Post recovery validation only works properly on CRC enabled
2082 * filesystems.
2083 */
2094 /*
2095 * The next di_next_unlinked field is beyond
2096 * the current logged region. Find the next
2097 * logged region that contains or is beyond
2098 * the current di_next_unlinked field.
2099 */
2104 /*
2105 * If there are no more logged regions in the
2106 * buffer, then we're done.
2107 */
2116 item_index++;
2117 }
2119 /*
2120 * If the current logged region starts after the current
2121 * di_next_unlinked field, then move on to the next
2122 * di_next_unlinked field.
2123 */
2131 /*
2132 * The current logged region contains a copy of the
2133 * current di_next_unlinked field. Extract its value
2134 * and copy it to the buffer copy.
2135 */
2144 }
2149 /*
2150 * If necessary, recalculate the CRC in the on-disk inode. We
2151 * have to leave the inode in a consistent state for whoever
2152 * reads it next....
2153 */
2157 }
2160 }
2162 /*
2163 * V5 filesystems know the age of the buffer on disk being recovered. We can
2164 * have newer objects on disk than we are replaying, and so for these cases we
2165 * don't want to replay the current change as that will make the buffer contents
2166 * temporarily invalid on disk.
2167 *
2168 * The magic number might not match the buffer type we are going to recover
2169 * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags. Hence
2170 * extract the LSN of the existing object in the buffer based on it's current
2171 * magic number. If we don't recognise the magic number in the buffer, then
2172 * return a LSN of -1 so that the caller knows it was an unrecognised block and
2173 * so can recover the buffer.
2174 *
2175 * Note: we cannot rely solely on magic number matches to determine that the
2176 * buffer has a valid LSN - we also need to verify that it belongs to this
2177 * filesystem, so we need to extract the object's LSN and compare it to that
2178 * which we read from the superblock. If the UUIDs don't match, then we've got a
2179 * stale metadata block from an old filesystem instance that we need to recover
2180 * over the top of.
2181 */
2182 static xfs_lsn_t
2186 {
2194 /* v4 filesystems always recover immediately */
2213 }
2221 }
2245 /*
2246 * Remote attr blocks are written synchronously, rather than
2247 * being logged. That means they do not contain a valid LSN
2248 * (i.e. transactionally ordered) in them, and hence any time we
2249 * see a buffer to replay over the top of a remote attribute
2250 * block we should simply do so.
2251 */
2254 /*
2255 * superblock uuids are magic. We may or may not have a
2256 * sb_meta_uuid on disk, but it will be set in the in-core
2257 * superblock. We set the uuid pointer for verification
2258 * according to the superblock feature mask to ensure we check
2259 * the relevant UUID in the superblock.
2260 */
2264 else
2269 }
2275 }
2287 }
2293 }
2295 /*
2296 * We do individual object checks on dquot and inode buffers as they
2297 * have their own individual LSN records. Also, we could have a stale
2298 * buffer here, so we have to at least recognise these buffer types.
2299 *
2300 * A notd complexity here is inode unlinked list processing - it logs
2301 * the inode directly in the buffer, but we don't know which inodes have
2302 * been modified, and there is no global buffer LSN. Hence we need to
2303 * recover all inode buffer types immediately. This problem will be
2304 * fixed by logical logging of the unlinked list modifications.
2305 */
2313 }
2315 /* unknown buffer contents, recover immediately */
2317 recover_immediately:
2320 }
2322 /*
2323 * Validate the recovered buffer is of the correct type and attach the
2324 * appropriate buffer operations to them for writeback. Magic numbers are in a
2325 * few places:
2326 * the first 16 bits of the buffer (inode buffer, dquot buffer),
2327 * the first 32 bits of the buffer (most blocks),
2328 * inside a struct xfs_da_blkinfo at the start of the buffer.
2329 */
2330 static void
2335 xfs_lsn_t current_lsn)
2336 {
2343 /*
2344 * We can only do post recovery validation on items on CRC enabled
2345 * fielsystems as we need to know when the buffer was written to be able
2346 * to determine if we should have replayed the item. If we replay old
2347 * metadata over a newer buffer, then it will enter a temporarily
2348 * inconsistent state resulting in verification failures. Hence for now
2349 * just avoid the verification stage for non-crc filesystems
2350 */
2389 }
2395 }
2402 }
2409 }
2415 #ifdef CONFIG_XFS_QUOTA
2419 }
2421 #else
2425 #endif
2431 }
2438 }
2446 }
2454 }
2462 }
2470 }
2478 }
2486 }
2494 }
2501 }
2508 }
2511 #ifdef CONFIG_XFS_RT
2514 /* no magic numbers for verification of RT buffers */
2522 }
2524 /*
2525 * Nothing else to do in the case of a NULL current LSN as this means
2526 * the buffer is more recent than the change in the log and will be
2527 * skipped.
2528 */
2535 }
2537 /*
2538 * We must update the metadata LSN of the buffer as it is written out to
2539 * ensure that older transactions never replay over this one and corrupt
2540 * the buffer. This can occur if log recovery is interrupted at some
2541 * point after the current transaction completes, at which point a
2542 * subsequent mount starts recovery from the beginning.
2543 *
2544 * Write verifiers update the metadata LSN from log items attached to
2545 * the buffer. Therefore, initialize a bli purely to carry the LSN to
2546 * the verifier. We'll clean it up in our ->iodone() callback.
2547 */
2556 }
2557 }
2559 /*
2560 * Perform a 'normal' buffer recovery. Each logged region of the
2561 * buffer should be copied over the corresponding region in the
2562 * given buffer. The bitmap in the buf log format structure indicates
2563 * where to place the logged data.
2564 */
2565 STATIC void
2571 xfs_lsn_t current_lsn)
2572 {
2576 xfs_failaddr_t fa;
2596 /*
2597 * The dirty regions logged in the buffer, even though
2598 * contiguous, may span multiple chunks. This is because the
2599 * dirty region may span a physical page boundary in a buffer
2600 * and hence be split into two separate vectors for writing into
2601 * the log. Hence we need to trim nbits back to the length of
2602 * the current region being copied out of the log.
2603 */
2607 /*
2608 * Do a sanity check if this is a dquot buffer. Just checking
2609 * the first dquot in the buffer should do. XXXThis is
2610 * probably a good thing to do for other buf types also.
2611 */
2619 }
2625 }
2633 }
2634 }
2640 next:
2641 i++;
2643 }
2645 /* Shouldn't be any more regions */
2649 }
2651 /*
2652 * Perform a dquot buffer recovery.
2653 * Simple algorithm: if we have found a QUOTAOFF log item of the same type
2654 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2655 * Else, treat it as a regular buffer and do recovery.
2656 *
2657 * Return false if the buffer was tossed and true if we recovered the buffer to
2658 * indicate to the caller if the buffer needs writing.
2659 */
2660 STATIC bool
2667 {
2668 uint type;
2672 /*
2673 * Filesystems are required to send in quota flags at mount time.
2674 */
2685 /*
2686 * This type of quotas was turned off, so ignore this buffer
2687 */
2693 }
2695 /*
2696 * This routine replays a modification made to a buffer at runtime.
2697 * There are actually two types of buffer, regular and inode, which
2698 * are handled differently. Inode buffers are handled differently
2699 * in that we only recover a specific set of data from them, namely
2700 * the inode di_next_unlinked fields. This is because all other inode
2701 * data is actually logged via inode records and any data we replay
2702 * here which overlaps that may be stale.
2703 *
2704 * When meta-data buffers are freed at run time we log a buffer item
2705 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2706 * of the buffer in the log should not be replayed at recovery time.
2707 * This is so that if the blocks covered by the buffer are reused for
2708 * file data before we crash we don't end up replaying old, freed
2709 * meta-data into a user's file.
2710 *
2711 * To handle the cancellation of buffer log items, we make two passes
2712 * over the log during recovery. During the first we build a table of
2713 * those buffers which have been cancelled, and during the second we
2714 * only replay those buffers which do not have corresponding cancel
2715 * records in the table. See xlog_recover_buffer_pass[1,2] above
2716 * for more details on the implementation of the table of cancel records.
2717 */
2718 STATIC int
2723 xfs_lsn_t current_lsn)
2724 {
2729 uint buf_flags;
2730 xfs_lsn_t lsn;
2732 /*
2733 * In this pass we only want to recover all the buffers which have
2734 * not been cancelled and are not cancellation buffers themselves.
2735 */
2740 }
2753 /*
2754 * Recover the buffer only if we get an LSN from it and it's less than
2755 * the lsn of the transaction we are replaying.
2756 *
2757 * Note that we have to be extremely careful of readahead here.
2758 * Readahead does not attach verfiers to the buffers so if we don't
2759 * actually do any replay after readahead because of the LSN we found
2760 * in the buffer if more recent than that current transaction then we
2761 * need to attach the verifier directly. Failure to do so can lead to
2762 * future recovery actions (e.g. EFI and unlinked list recovery) can
2763 * operate on the buffers and they won't get the verifier attached. This
2764 * can lead to blocks on disk having the correct content but a stale
2765 * CRC.
2766 *
2767 * It is safe to assume these clean buffers are currently up to date.
2768 * If the buffer is dirtied by a later transaction being replayed, then
2769 * the verifier will be reset to match whatever recover turns that
2770 * buffer into.
2771 */
2777 }
2792 }
2794 /*
2795 * Perform delayed write on the buffer. Asynchronous writes will be
2796 * slower when taking into account all the buffers to be flushed.
2797 *
2798 * Also make sure that only inode buffers with good sizes stay in
2799 * the buffer cache. The kernel moves inodes in buffers of 1 block
2800 * or inode_cluster_size bytes, whichever is bigger. The inode
2801 * buffers in the log can be a different size if the log was generated
2802 * by an older kernel using unclustered inode buffers or a newer kernel
2803 * running with a different inode cluster size. Regardless, if the
2804 * the inode buffer size isn't max(blocksize, inode_cluster_size)
2805 * for *our* value of inode_cluster_size, then we need to keep
2806 * the buffer out of the buffer cache so that the buffer won't
2807 * overlap with future reads of those inodes.
2808 */
2818 }
2820 out_release:
2823 }
2825 /*
2826 * Inode fork owner changes
2827 *
2828 * If we have been told that we have to reparent the inode fork, it's because an
2829 * extent swap operation on a CRC enabled filesystem has been done and we are
2830 * replaying it. We need to walk the BMBT of the appropriate fork and change the
2831 * owners of it.
2832 *
2833 * The complexity here is that we don't have an inode context to work with, so
2834 * after we've replayed the inode we need to instantiate one. This is where the
2835 * fun begins.
2836 *
2837 * We are in the middle of log recovery, so we can't run transactions. That
2838 * means we cannot use cache coherent inode instantiation via xfs_iget(), as
2839 * that will result in the corresponding iput() running the inode through
2840 * xfs_inactive(). If we've just replayed an inode core that changes the link
2841 * count to zero (i.e. it's been unlinked), then xfs_inactive() will run
2842 * transactions (bad!).
2843 *
2844 * So, to avoid this, we instantiate an inode directly from the inode core we've
2845 * just recovered. We have the buffer still locked, and all we really need to
2846 * instantiate is the inode core and the forks being modified. We can do this
2847 * manually, then run the inode btree owner change, and then tear down the
2848 * xfs_inode without having to run any transactions at all.
2849 *
2850 * Also, because we don't have a transaction context available here but need to
2851 * gather all the buffers we modify for writeback so we pass the buffer_list
2852 * instead for the operation to use.
2853 */
2855 STATIC int
2861 {
2871 /* instantiate the inode */
2882 }
2890 }
2898 }
2900 out_free_ip:
2903 }
2905 STATIC int
2910 xfs_lsn_t current_lsn)
2911 {
2921 uint fields;
2923 uint isize;
2934 }
2936 /*
2937 * Inode buffers can be freed, look out for it,
2938 * and do not replay the inode.
2939 */
2945 }
2955 /*
2956 * Make sure the place we're flushing out to really looks
2957 * like an inode!
2958 */
2965 }
2973 }
2975 /*
2976 * If the inode has an LSN in it, recover the inode only if it's less
2977 * than the lsn of the transaction we are replaying. Note: we still
2978 * need to replay an owner change even though the inode is more recent
2979 * than the transaction as there is no guarantee that all the btree
2980 * blocks are more recent than this transaction, too.
2981 */
2989 }
2990 }
2992 /*
2993 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
2994 * are transactional and if ordering is necessary we can determine that
2995 * more accurately by the LSN field in the V3 inode core. Don't trust
2996 * the inode versions we might be changing them here - use the
2997 * superblock flag to determine whether we need to look at di_flushiter
2998 * to skip replay when the on disk inode is newer than the log one
2999 */
3002 /*
3003 * Deal with the wrap case, DI_MAX_FLUSH is less
3004 * than smaller numbers
3005 */
3008 /* do nothing */
3013 }
3014 }
3016 /* Take the opportunity to reset the flush iteration count */
3031 }
3045 }
3046 }
3059 }
3070 }
3081 }
3083 /* recover the log dinode inode into the on disk inode */
3112 /*
3113 * There are no data fork flags set.
3114 */
3117 }
3119 /*
3120 * If we logged any attribute data, recover it. There may or
3121 * may not have been any other non-core data logged in this
3122 * transaction.
3123 */
3129 }
3154 }
3155 }
3157 out_owner_change:
3158 /* Recover the swapext owner change unless inode has been deleted */
3162 buffer_list);
3163 /* re-generate the checksum. */
3170 out_release:
3172 error:
3176 }
3178 /*
3179 * Recover QUOTAOFF records. We simply make a note of it in the xlog
3180 * structure, so that we know not to do any dquot item or dquot buffer recovery,
3181 * of that type.
3182 */
3183 STATIC int
3187 {
3191 /*
3192 * The logitem format's flag tells us if this was user quotaoff,
3193 * group/project quotaoff or both.
3194 */
3203 }
3205 /*
3206 * Recover a dquot record
3207 */
3208 STATIC int
3213 xfs_lsn_t current_lsn)
3214 {
3218 xfs_failaddr_t fa;
3221 uint type;
3224 /*
3225 * Filesystems are required to send in quota flags at mount time.
3226 */
3234 }
3239 }
3241 /*
3242 * This type of quotas was turned off, so ignore this record.
3243 */
3249 /*
3250 * At this point we know that quota was _not_ turned off.
3251 * Since the mount flags are not indicating to us otherwise, this
3252 * must mean that quota is on, and the dquot needs to be replayed.
3253 * Remember that we may not have fully recovered the superblock yet,
3254 * so we can't do the usual trick of looking at the SB quota bits.
3255 *
3256 * The other possibility, of course, is that the quota subsystem was
3257 * removed since the last mount - ENOSYS.
3258 */
3266 }
3269 /*
3270 * At this point we are assuming that the dquots have been allocated
3271 * and hence the buffer has valid dquots stamped in it. It should,
3272 * therefore, pass verifier validation. If the dquot is bad, then the
3273 * we'll return an error here, so we don't need to specifically check
3274 * the dquot in the buffer after the verifier has run.
3275 */
3285 /*
3286 * If the dquot has an LSN in it, recover the dquot only if it's less
3287 * than the lsn of the transaction we are replaying.
3288 */
3295 }
3296 }
3301 XFS_DQUOT_CRC_OFF);
3302 }
3309 out_release:
3312 }
3314 /*
3315 * This routine is called to create an in-core extent free intent
3316 * item from the efi format structure which was logged on disk.
3317 * It allocates an in-core efi, copies the extents from the format
3318 * structure into it, and adds the efi to the AIL with the given
3319 * LSN.
3320 */
3321 STATIC int
3325 xfs_lsn_t lsn)
3326 {
3339 }
3343 /*
3344 * The EFI has two references. One for the EFD and one for EFI to ensure
3345 * it makes it into the AIL. Insert the EFI into the AIL directly and
3346 * drop the EFI reference. Note that xfs_trans_ail_update() drops the
3347 * AIL lock.
3348 */
3352 }
3355 /*
3356 * This routine is called when an EFD format structure is found in a committed
3357 * transaction in the log. Its purpose is to cancel the corresponding EFI if it
3358 * was still in the log. To do this it searches the AIL for the EFI with an id
3359 * equal to that in the EFD format structure. If we find it we drop the EFD
3360 * reference, which removes the EFI from the AIL and frees it.
3361 */
3362 STATIC int
3366 {
3381 /*
3382 * Search for the EFI with the id in the EFD format structure in the
3383 * AIL.
3384 */
3391 /*
3392 * Drop the EFD reference to the EFI. This
3393 * removes the EFI from the AIL and frees it.
3394 */
3399 }
3400 }
3402 }
3408 }
3410 /*
3411 * This routine is called to create an in-core extent rmap update
3412 * item from the rui format structure which was logged on disk.
3413 * It allocates an in-core rui, copies the extents from the format
3414 * structure into it, and adds the rui to the AIL with the given
3415 * LSN.
3416 */
3417 STATIC int
3421 xfs_lsn_t lsn)
3422 {
3435 }
3439 /*
3440 * The RUI has two references. One for the RUD and one for RUI to ensure
3441 * it makes it into the AIL. Insert the RUI into the AIL directly and
3442 * drop the RUI reference. Note that xfs_trans_ail_update() drops the
3443 * AIL lock.
3444 */
3448 }
3451 /*
3452 * This routine is called when an RUD format structure is found in a committed
3453 * transaction in the log. Its purpose is to cancel the corresponding RUI if it
3454 * was still in the log. To do this it searches the AIL for the RUI with an id
3455 * equal to that in the RUD format structure. If we find it we drop the RUD
3456 * reference, which removes the RUI from the AIL and frees it.
3457 */
3458 STATIC int
3462 {
3474 /*
3475 * Search for the RUI with the id in the RUD format structure in the
3476 * AIL.
3477 */
3484 /*
3485 * Drop the RUD reference to the RUI. This
3486 * removes the RUI from the AIL and frees it.
3487 */
3492 }
3493 }
3495 }
3501 }
3503 /*
3504 * Copy an CUI format buffer from the given buf, and into the destination
3505 * CUI format structure. The CUI/CUD items were designed not to need any
3506 * special alignment handling.
3507 */
3508 static int
3512 {
3514 uint len;
3522 }
3525 }
3527 /*
3528 * This routine is called to create an in-core extent refcount update
3529 * item from the cui format structure which was logged on disk.
3530 * It allocates an in-core cui, copies the extents from the format
3531 * structure into it, and adds the cui to the AIL with the given
3532 * LSN.
3533 */
3534 STATIC int
3538 xfs_lsn_t lsn)
3539 {
3552 }
3556 /*
3557 * The CUI has two references. One for the CUD and one for CUI to ensure
3558 * it makes it into the AIL. Insert the CUI into the AIL directly and
3559 * drop the CUI reference. Note that xfs_trans_ail_update() drops the
3560 * AIL lock.
3561 */
3565 }
3568 /*
3569 * This routine is called when an CUD format structure is found in a committed
3570 * transaction in the log. Its purpose is to cancel the corresponding CUI if it
3571 * was still in the log. To do this it searches the AIL for the CUI with an id
3572 * equal to that in the CUD format structure. If we find it we drop the CUD
3573 * reference, which removes the CUI from the AIL and frees it.
3574 */
3575 STATIC int
3579 {
3591 }
3594 /*
3595 * Search for the CUI with the id in the CUD format structure in the
3596 * AIL.
3597 */
3604 /*
3605 * Drop the CUD reference to the CUI. This
3606 * removes the CUI from the AIL and frees it.
3607 */
3612 }
3613 }
3615 }
3621 }
3623 /*
3624 * Copy an BUI format buffer from the given buf, and into the destination
3625 * BUI format structure. The BUI/BUD items were designed not to need any
3626 * special alignment handling.
3627 */
3628 static int
3632 {
3634 uint len;
3642 }
3645 }
3647 /*
3648 * This routine is called to create an in-core extent bmap update
3649 * item from the bui format structure which was logged on disk.
3650 * It allocates an in-core bui, copies the extents from the format
3651 * structure into it, and adds the bui to the AIL with the given
3652 * LSN.
3653 */
3654 STATIC int
3658 xfs_lsn_t lsn)
3659 {
3670 }
3676 }
3680 /*
3681 * The RUI has two references. One for the RUD and one for RUI to ensure
3682 * it makes it into the AIL. Insert the RUI into the AIL directly and
3683 * drop the RUI reference. Note that xfs_trans_ail_update() drops the
3684 * AIL lock.
3685 */
3689 }
3692 /*
3693 * This routine is called when an BUD format structure is found in a committed
3694 * transaction in the log. Its purpose is to cancel the corresponding BUI if it
3695 * was still in the log. To do this it searches the AIL for the BUI with an id
3696 * equal to that in the BUD format structure. If we find it we drop the BUD
3697 * reference, which removes the BUI from the AIL and frees it.
3698 */
3699 STATIC int
3703 {
3715 }
3718 /*
3719 * Search for the BUI with the id in the BUD format structure in the
3720 * AIL.
3721 */
3728 /*
3729 * Drop the BUD reference to the BUI. This
3730 * removes the BUI from the AIL and frees it.
3731 */
3736 }
3737 }
3739 }
3745 }
3747 /*
3748 * This routine is called when an inode create format structure is found in a
3749 * committed transaction in the log. It's purpose is to initialise the inodes
3750 * being allocated on disk. This requires us to get inode cluster buffers that
3751 * match the range to be initialised, stamped with inode templates and written
3752 * by delayed write so that subsequent modifications will hit the cached buffer
3753 * and only need writing out at the end of recovery.
3754 */
3755 STATIC int
3760 {
3764 xfs_agnumber_t agno;
3765 xfs_agblock_t agbno;
3768 xfs_agblock_t length;
3778 }
3783 }
3789 }
3794 }
3799 }
3804 }
3809 }
3811 /*
3812 * The inode chunk is either full or sparse and we only support
3813 * m_ino_geo.ialloc_min_blks sized sparse allocations at this time.
3814 */
3820 }
3822 /* verify inode count is consistent with extent length */
3826 __FUNCTION__);
3828 }
3830 /*
3831 * The icreate transaction can cover multiple cluster buffers and these
3832 * buffers could have been freed and reused. Check the individual
3833 * buffers for cancellation so we don't overwrite anything written after
3834 * a cancellation.
3835 */
3839 xfs_daddr_t daddr;
3844 cancel_count++;
3845 }
3847 /*
3848 * We currently only use icreate for a single allocation at a time. This
3849 * means we should expect either all or none of the buffers to be
3850 * cancelled. Be conservative and skip replay if at least one buffer is
3851 * cancelled, but warn the user that something is awry if the buffers
3852 * are not consistent.
3853 *
3854 * XXX: This must be refined to only skip cancelled clusters once we use
3855 * icreate for multiple chunk allocations.
3856 */
3864 }
3869 }
3871 STATIC void
3875 {
3882 }
3886 }
3888 STATIC void
3892 {
3906 }
3913 }
3915 STATIC void
3919 {
3923 uint type;
3951 }
3953 STATIC void
3957 {
3979 }
3980 }
3982 STATIC int
3987 {
4006 /* nothing to do in pass 1 */
4013 }
4014 }
4016 STATIC int
4022 {
4054 /* nothing to do in pass2 */
4061 }
4062 }
4064 STATIC int
4070 {
4079 }
4082 }
4084 /*
4085 * Perform the transaction.
4086 *
4087 * If the transaction modifies a buffer or inode, do it now. Otherwise,
4088 * EFIs and EFDs get queued up by adding entries into the AIL for them.
4089 */
4090 STATIC int
4096 {
4104 #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100
4120 items_queued++;
4126 }
4131 }
4135 }
4137 out:
4143 }
4149 }
4151 STATIC void
4154 {
4160 }
4162 STATIC int
4168 {
4173 /*
4174 * If the transaction is empty, the header was split across this and the
4175 * previous record. Copy the rest of the header.
4176 */
4182 }
4189 }
4191 /* take the tail entry */
4203 }
4205 /*
4206 * The next region to add is the start of a new region. It could be
4207 * a whole region or it could be the first part of a new region. Because
4208 * of this, the assumption here is that the type and size fields of all
4209 * format structures fit into the first 32 bits of the structure.
4210 *
4211 * This works because all regions must be 32 bit aligned. Therefore, we
4212 * either have both fields or we have neither field. In the case we have
4213 * neither field, the data part of the region is zero length. We only have
4214 * a log_op_header and can throw away the header since a new one will appear
4215 * later. If we have at least 4 bytes, then we can determine how many regions
4216 * will appear in the current log item.
4217 */
4218 STATIC int
4224 {
4232 /* we need to catch log corruptions here */
4235 __func__);
4238 }
4244 }
4246 /*
4247 * The transaction header can be arbitrarily split across op
4248 * records. If we don't have the whole thing here, copy what we
4249 * do have and handle the rest in the next record.
4250 */
4255 }
4261 /* take the tail entry */
4265 /* tail item is in use, get a new one */
4269 }
4280 }
4286 }
4295 }
4297 /* Description region is ri_buf[0] */
4303 }
4305 /*
4306 * Free up any resources allocated by the transaction
4307 *
4308 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
4309 */
4310 STATIC void
4313 {
4320 /* Free the regions in the item. */
4324 /* Free the item itself */
4327 }
4328 /* Free the transaction recover structure */
4330 }
4332 /*
4333 * On error or completion, trans is freed.
4334 */
4335 STATIC int
4344 {
4348 /* mask off ophdr transaction container flags */
4353 /*
4354 * Callees must not free the trans structure. We'll decide if we need to
4355 * free it or not based on the operation being done and it's result.
4356 */
4358 /* expected flag values */
4368 buffer_list);
4369 /* success or fail, we are now done with this transaction. */
4373 /* unexpected flag values */
4375 /* just skip trans */
4385 }
4389 }
4391 /*
4392 * Lookup the transaction recovery structure associated with the ID in the
4393 * current ophdr. If the transaction doesn't exist and the start flag is set in
4394 * the ophdr, then allocate a new transaction for future ID matches to find.
4395 * Either way, return what we found during the lookup - an existing transaction
4396 * or nothing.
4397 */
4403 {
4405 xlog_tid_t tid;
4413 }
4415 /*
4416 * skip over non-start transaction headers - we could be
4417 * processing slack space before the next transaction starts
4418 */
4424 /*
4425 * This is a new transaction so allocate a new recovery container to
4426 * hold the recovery ops that will follow.
4427 */
4435 /*
4436 * Nothing more to do for this ophdr. Items to be added to this new
4437 * transaction will be in subsequent ophdr containers.
4438 */
4440 }
4442 STATIC int
4452 {
4457 /* Do we understand who wrote this op? */
4464 }
4466 /*
4467 * Check the ophdr contains all the data it is supposed to contain.
4468 */
4474 }
4478 /* nothing to do, so skip over this ophdr */
4480 }
4482 /*
4483 * The recovered buffer queue is drained only once we know that all
4484 * recovery items for the current LSN have been processed. This is
4485 * required because:
4486 *
4487 * - Buffer write submission updates the metadata LSN of the buffer.
4488 * - Log recovery skips items with a metadata LSN >= the current LSN of
4489 * the recovery item.
4490 * - Separate recovery items against the same metadata buffer can share
4491 * a current LSN. I.e., consider that the LSN of a recovery item is
4492 * defined as the starting LSN of the first record in which its
4493 * transaction appears, that a record can hold multiple transactions,
4494 * and/or that a transaction can span multiple records.
4495 *
4496 * In other words, we are allowed to submit a buffer from log recovery
4497 * once per current LSN. Otherwise, we may incorrectly skip recovery
4498 * items and cause corruption.
4499 *
4500 * We don't know up front whether buffers are updated multiple times per
4501 * LSN. Therefore, track the current LSN of each commit log record as it
4502 * is processed and drain the queue when it changes. Use commit records
4503 * because they are ordered correctly by the logging code.
4504 */
4511 }
4515 }
4517 /*
4518 * There are two valid states of the r_state field. 0 indicates that the
4519 * transaction structure is in a normal state. We have either seen the
4520 * start of the transaction or the last operation we added was not a partial
4521 * operation. If the last operation we added to the transaction was a
4522 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
4523 *
4524 * NOTE: skip LRs with 0 data length.
4525 */
4526 STATIC int
4534 {
4543 /* check the log format matches our own - else we can't recover */
4554 /* errors will abort recovery */
4561 num_logops--;
4562 }
4564 }
4566 /* Recover the EFI if necessary. */
4567 STATIC int
4572 {
4576 /*
4577 * Skip EFIs that we've already processed.
4578 */
4588 }
4590 /* Release the EFI since we're cancelling everything. */
4591 STATIC void
4596 {
4604 }
4606 /* Recover the RUI if necessary. */
4607 STATIC int
4612 {
4616 /*
4617 * Skip RUIs that we've already processed.
4618 */
4628 }
4630 /* Release the RUI since we're cancelling everything. */
4631 STATIC void
4636 {
4644 }
4646 /* Recover the CUI if necessary. */
4647 STATIC int
4652 {
4656 /*
4657 * Skip CUIs that we've already processed.
4658 */
4668 }
4670 /* Release the CUI since we're cancelling everything. */
4671 STATIC void
4676 {
4684 }
4686 /* Recover the BUI if necessary. */
4687 STATIC int
4692 {
4696 /*
4697 * Skip BUIs that we've already processed.
4698 */
4708 }
4710 /* Release the BUI since we're cancelling everything. */
4711 STATIC void
4716 {
4724 }
4726 /* Is this log item a deferred action intent? */
4728 {
4737 }
4738 }
4740 /* Take all the collected deferred ops and finish them in order. */
4741 static int
4744 {
4748 uint resblks;
4751 /*
4752 * We're finishing the defer_ops that accumulated as a result of
4753 * recovering unfinished intent items during log recovery. We
4754 * reserve an itruncate transaction because it is the largest
4755 * permanent transaction type. Since we're the only user of the fs
4756 * right now, take 93% (15/16) of the available free blocks. Use
4757 * weird math to avoid a 64-bit division.
4758 */
4768 /* transfer all collected dfops to this transaction */
4772 }
4774 /*
4775 * When this is called, all of the log intent items which did not have
4776 * corresponding log done items should be in the AIL. What we do now
4777 * is update the data structures associated with each one.
4778 *
4779 * Since we process the log intent items in normal transactions, they
4780 * will be removed at some point after the commit. This prevents us
4781 * from just walking down the list processing each one. We'll use a
4782 * flag in the intent item to skip those that we've already processed
4783 * and use the AIL iteration mechanism's generation count to try to
4784 * speed this up at least a bit.
4785 *
4786 * When we start, we know that the intents are the only things in the
4787 * AIL. As we process them, however, other items are added to the
4788 * AIL.
4789 */
4790 STATIC int
4793 {
4799 #if defined(DEBUG) || defined(XFS_WARN)
4800 xfs_lsn_t last_lsn;
4801 #endif
4803 /*
4804 * The intent recovery handlers commit transactions to complete recovery
4805 * for individual intents, but any new deferred operations that are
4806 * queued during that process are held off until the very end. The
4807 * purpose of this transaction is to serve as a container for deferred
4808 * operations. Each intent recovery handler must transfer dfops here
4809 * before its local transaction commits, and we'll finish the entire
4810 * list below.
4811 */
4819 #if defined(DEBUG) || defined(XFS_WARN)
4821 #endif
4823 /*
4824 * We're done when we see something other than an intent.
4825 * There should be no intents left in the AIL now.
4826 */
4828 #ifdef DEBUG
4831 #endif
4833 }
4835 /*
4836 * We should never see a redo item with a LSN higher than
4837 * the last transaction we found in the log at the start
4838 * of recovery.
4839 */
4842 /*
4843 * NOTE: If your intent processing routine can create more
4844 * deferred ops, you /must/ attach them to the dfops in this
4845 * routine or else those subsequent intents will get
4846 * replayed in the wrong order!
4847 */
4861 }
4865 }
4866 out:
4874 }
4876 /*
4877 * A cancel occurs when the mount has failed and we're bailing out.
4878 * Release all pending log intent items so they don't pin the AIL.
4879 */
4880 STATIC void
4883 {
4892 /*
4893 * We're done when we see something other than an intent.
4894 * There should be no intents left in the AIL now.
4895 */
4897 #ifdef DEBUG
4900 #endif
4902 }
4917 }
4920 }
4924 }
4926 /*
4927 * This routine performs a transaction to null out a bad inode pointer
4928 * in an agi unlinked inode hash bucket.
4929 */
4930 STATIC void
4933 xfs_agnumber_t agno,
4935 {
4962 out_abort:
4964 out_error:
4967 }
4969 STATIC xfs_agino_t
4972 xfs_agnumber_t agno,
4973 xfs_agino_t agino,
4975 {
4979 xfs_ino_t ino;
4987 /*
4988 * Get the on disk inode to find the next inode in the bucket.
4989 */
4998 /* setup for the next pass */
5002 /*
5003 * Prevent any DMAPI event from being sent when the reference on
5004 * the inode is dropped.
5005 */
5011 fail_iput:
5013 fail:
5014 /*
5015 * We can't read in the inode this bucket points to, or this inode
5016 * is messed up. Just ditch this bucket of inodes. We will lose
5017 * some inodes and space, but at least we won't hang.
5018 *
5019 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
5020 * clear the inode pointer in the bucket.
5021 */
5024 }
5026 /*
5027 * Recover AGI unlinked lists
5028 *
5029 * This is called during recovery to process any inodes which we unlinked but
5030 * not freed when the system crashed. These inodes will be on the lists in the
5031 * AGI blocks. What we do here is scan all the AGIs and fully truncate and free
5032 * any inodes found on the lists. Each inode is removed from the lists when it
5033 * has been fully truncated and is freed. The freeing of the inode and its
5034 * removal from the list must be atomic.
5035 *
5036 * If everything we touch in the agi processing loop is already in memory, this
5037 * loop can hold the cpu for a long time. It runs without lock contention,
5038 * memory allocation contention, the need wait for IO, etc, and so will run
5039 * until we either run out of inodes to process, run low on memory or we run out
5040 * of log space.
5041 *
5042 * This behaviour is bad for latency on single CPU and non-preemptible kernels,
5043 * and can prevent other filesytem work (such as CIL pushes) from running. This
5044 * can lead to deadlocks if the recovery process runs out of log reservation
5045 * space. Hence we need to yield the CPU when there is other kernel work
5046 * scheduled on this CPU to ensure other scheduled work can run without undue
5047 * latency.
5048 */
5049 STATIC void
5052 {
5054 xfs_agnumber_t agno;
5057 xfs_agino_t agino;
5064 /*
5065 * Find the agi for this ag.
5066 */
5069 /*
5070 * AGI is b0rked. Don't process it.
5071 *
5072 * We should probably mark the filesystem as corrupt
5073 * after we've recovered all the ag's we can....
5074 */
5076 }
5077 /*
5078 * Unlock the buffer so that it can be acquired in the normal
5079 * course of the transaction to truncate and free each inode.
5080 * Because we are not racing with anyone else here for the AGI
5081 * buffer, we don't even need to hold it locked to read the
5082 * initial unlinked bucket entries out of the buffer. We keep
5083 * buffer reference though, so that it stays pinned in memory
5084 * while we need the buffer.
5085 */
5095 }
5096 }
5098 }
5099 }
5101 STATIC void
5106 {
5113 }
5122 }
5123 }
5124 }
5126 /*
5127 * CRC check, unpack and process a log record.
5128 */
5129 STATIC int
5137 {
5139 __le32 crc;
5143 /*
5144 * Nothing else to do if this is a CRC verification pass. Just return
5145 * if this a record with a non-zero crc. Unfortunately, mkfs always
5146 * sets old_crc to 0 so we must consider this valid even on v5 supers.
5147 * Otherwise, return EFSBADCRC on failure so the callers up the stack
5148 * know precisely what failed.
5149 */
5154 }
5156 /*
5157 * We're in the normal recovery path. Issue a warning if and only if the
5158 * CRC in the header is non-zero. This is an advisory warning and the
5159 * zero CRC check prevents warnings from being emitted when upgrading
5160 * the kernel from one that does not add CRCs by default.
5161 */
5169 }
5171 /*
5172 * If the filesystem is CRC enabled, this mismatch becomes a
5173 * fatal log corruption failure.
5174 */
5178 }
5179 }
5184 buffer_list);
5185 }
5187 STATIC int
5191 xfs_daddr_t blkno)
5192 {
5205 }
5207 /* LR body must have data or it wouldn't have been written */
5215 }
5217 /*
5218 * Read the log from tail to head and process the log records found.
5219 * Handle the two cases where the tail and head are in the same cycle
5220 * and where the active portion of the log wraps around the end of
5221 * the physical log separately. The pass parameter is passed through
5222 * to the routines called to process the data and is not looked at
5223 * here.
5224 */
5225 STATIC int
5228 xfs_daddr_t head_blk,
5229 xfs_daddr_t tail_blk,
5232 {
5235 xfs_daddr_t rhead_blk;
5252 /*
5253 * Read the header of the tail block and get the iclog buffer size from
5254 * h_size. Use this to tell how many sectors make up the log header.
5255 */
5257 /*
5258 * When using variable length iclogs, read first sector of
5259 * iclog header and extract the header size from it. Get a
5260 * new hbp that is the correct size.
5261 */
5275 /*
5276 * xfsprogs has a bug where record length is based on lsunit but
5277 * h_size (iclog size) is hardcoded to 32k. Now that we
5278 * unconditionally CRC verify the unmount record, this means the
5279 * log buffer can be too small for the record and cause an
5280 * overrun.
5281 *
5282 * Detect this condition here. Use lsunit for the buffer size as
5283 * long as this looks like the mkfs case. Otherwise, return an
5284 * error to avoid a buffer overrun.
5285 */
5300 }
5301 }
5307 hblks++;
5312 }
5318 }
5326 }
5330 /*
5331 * Perform recovery around the end of the physical log.
5332 * When the head is not on the same cycle number as the tail,
5333 * we can't do a sequential recovery.
5334 */
5336 /*
5337 * Check for header wrapping around physical end-of-log
5338 */
5343 /* Read header in one read */
5349 /* This LR is split across physical log end */
5351 /* some data before physical log end */
5360 }
5362 /*
5363 * Note: this black magic still works with
5364 * large sector sizes (non-512) only because:
5365 * - we increased the buffer size originally
5366 * by 1 sector giving us enough extra space
5367 * for the second read;
5368 * - the log start is guaranteed to be sector
5369 * aligned;
5370 * - we read the log end (LR header start)
5371 * _first_, then the log start (LR header end)
5372 * - order is important.
5373 */
5376 wrapped_hblks,
5380 }
5390 /*
5391 * Read the log record data in multiple reads if it
5392 * wraps around the end of the log. Note that if the
5393 * header already wrapped, blk_no could point past the
5394 * end of the log. The record data is contiguous in
5395 * that case.
5396 */
5405 /* This log record is split across the
5406 * physical end of log */
5410 /* some data is before the physical
5411 * end of log */
5414 split_bblks =
5422 }
5424 /*
5425 * Note: this black magic still works with
5426 * large sector sizes (non-512) only because:
5427 * - we increased the buffer size originally
5428 * by 1 sector giving us enough extra space
5429 * for the second read;
5430 * - the log start is guaranteed to be sector
5431 * aligned;
5432 * - we read the log end (LR header start)
5433 * _first_, then the log start (LR header end)
5434 * - order is important.
5435 */
5441 }
5450 }
5455 }
5457 /* read first part of physical log */
5468 /* blocks in data section */
5482 }
5484 bread_err2:
5486 bread_err1:
5489 /*
5490 * Submit buffers that have been added from the last record processed,
5491 * regardless of error status.
5492 */
5499 /*
5500 * Transactions are freed at commit time but transactions without commit
5501 * records on disk are never committed. Free any that may be left in the
5502 * hash table.
5503 */
5510 }
5513 }
5515 /*
5516 * Do the recovery of the log. We actually do this in two phases.
5517 * The two passes are necessary in order to implement the function
5518 * of cancelling a record written into the log. The first pass
5519 * determines those things which have been cancelled, and the
5520 * second pass replays log items normally except for those which
5521 * have been cancelled. The handling of the replay and cancellations
5522 * takes place in the log item type specific routines.
5523 *
5524 * The table of items which have cancel records in the log is allocated
5525 * and freed at this level, since only here do we know when all of
5526 * the log recovery has been completed.
5527 */
5528 STATIC int
5531 xfs_daddr_t head_blk,
5532 xfs_daddr_t tail_blk)
5533 {
5538 /*
5539 * First do a pass to find all of the cancelled buf log items.
5540 * Store them in the buf_cancel_table for use in the second pass.
5541 */
5554 }
5555 /*
5556 * Then do a second pass to actually recover the items in the log.
5557 * When it is complete free the table of buf cancel items.
5558 */
5561 #ifdef DEBUG
5567 }
5574 }
5576 /*
5577 * Do the actual recovery
5578 */
5579 STATIC int
5582 xfs_daddr_t head_blk,
5583 xfs_daddr_t tail_blk)
5584 {
5592 /*
5593 * First replay the images in the log.
5594 */
5599 /*
5600 * If IO errors happened during recovery, bail out.
5601 */
5604 }
5606 /*
5607 * We now update the tail_lsn since much of the recovery has completed
5608 * and there may be space available to use. If there were no extent
5609 * or iunlinks, we can free up the entire log and set the tail_lsn to
5610 * be the last_sync_lsn. This was set in xlog_find_tail to be the
5611 * lsn of the last known good LR on disk. If there are extent frees
5612 * or iunlinks they will have some entries in the AIL; so we look at
5613 * the AIL to determine how to set the tail_lsn.
5614 */
5617 /*
5618 * Now that we've finished replaying all buffer and inode
5619 * updates, re-read in the superblock and reverify it.
5620 */
5632 }
5635 }
5637 /* Convert superblock from on-disk format */
5642 /* re-initialise in-core superblock and geometry structures */
5648 }
5653 /* Normal transactions can now occur */
5656 }
5658 /*
5659 * Perform recovery and re-initialize some log variables in xlog_find_tail.
5660 *
5661 * Return error or zero.
5662 */
5663 int
5666 {
5670 /* find the tail of the log */
5675 /*
5676 * The superblock was read before the log was available and thus the LSN
5677 * could not be verified. Check the superblock LSN against the current
5678 * LSN now that it's known.
5679 */
5685 /* There used to be a comment here:
5686 *
5687 * disallow recovery on read-only mounts. note -- mount
5688 * checks for ENOSPC and turns it into an intelligent
5689 * error message.
5690 * ...but this is no longer true. Now, unless you specify
5691 * NORECOVERY (in which case this function would never be
5692 * called), we just go ahead and recover. We do this all
5693 * under the vfs layer, so we can get away with it unless
5694 * the device itself is read-only, in which case we fail.
5695 */
5698 }
5700 /*
5701 * Version 5 superblock log feature mask validation. We know the
5702 * log is dirty so check if there are any unknown log features
5703 * in what we need to recover. If there are unknown features
5704 * (e.g. unsupported transactions, then simply reject the
5705 * attempt at recovery before touching anything.
5706 */
5709 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
5713 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
5719 }
5721 /*
5722 * Delay log recovery if the debug hook is set. This is debug
5723 * instrumention to coordinate simulation of I/O failures with
5724 * log recovery.
5725 */
5731 }
5739 }
5741 }
5743 /*
5744 * In the first part of recovery we replay inodes and buffers and build
5745 * up the list of extent free items which need to be processed. Here
5746 * we process the extent free items and clean up the on disk unlinked
5747 * inode lists. This is separated from the first part of recovery so
5748 * that the root and real-time bitmap inodes can be read in from disk in
5749 * between the two stages. This is necessary so that we can free space
5750 * in the real-time portion of the file system.
5751 */
5752 int
5755 {
5756 /*
5757 * Now we're ready to do the transactions needed for the
5758 * rest of recovery. Start with completing all the extent
5759 * free intent records and then process the unlinked inode
5760 * lists. At this point, we essentially run in normal mode
5761 * except that we're still performing recovery actions
5762 * rather than accepting new requests.
5763 */
5770 }
5772 /*
5773 * Sync the log to get all the intents out of the AIL.
5774 * This isn't absolutely necessary, but it helps in
5775 * case the unlink transactions would have problems
5776 * pushing the intents out of the way.
5777 */
5790 }
5792 }
5794 void
5797 {
5800 }
5802 #if defined(DEBUG)
5803 /*
5804 * Read all of the agf and agi counters and check that they
5805 * are consistent with the superblock counters.
5806 */
5807 STATIC void
5810 {
5815 xfs_agnumber_t agno;
5836 }
5848 }
5849 }
5850 }