| blob | b181b5f57a19edc2c0cfdfec9ce68a517d6008c9 |
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
40 #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
42 STATIC int
45 xfs_daddr_t *);
46 STATIC int
49 xfs_lsn_t);
50 #if defined(DEBUG)
51 STATIC void
54 #else
55 #define xlog_recover_check_summary(log)
56 #endif
57 STATIC int
61 /*
62 * This structure is used during recovery to record the buf log items which
63 * have been canceled and should not be replayed.
64 */
66 xfs_daddr_t bc_blkno;
67 uint bc_len;
70 };
72 /*
73 * Sector aligned buffer routines for buffer create/read/write/access
74 */
76 /*
77 * Verify the log-relative block number and length in basic blocks are valid for
78 * an operation involving the given XFS log buffer. Returns true if the fields
79 * are valid, false otherwise.
80 */
84 xfs_daddr_t blk_no,
86 {
92 }
94 /*
95 * Allocate a buffer to hold log data. The buffer needs to be able
96 * to map to a range of nbblks basic blocks at any valid (basic
97 * block) offset within the log.
98 */
99 STATIC xfs_buf_t *
103 {
106 /*
107 * Pass log block 0 since we don't have an addr yet, buffer will be
108 * verified on read.
109 */
112 nbblks);
115 }
117 /*
118 * We do log I/O in units of log sectors (a power-of-2
119 * multiple of the basic block size), so we round up the
120 * requested size to accommodate the basic blocks required
121 * for complete log sectors.
122 *
123 * In addition, the buffer may be used for a non-sector-
124 * aligned block offset, in which case an I/O of the
125 * requested size could extend beyond the end of the
126 * buffer. If the requested size is only 1 basic block it
127 * will never straddle a sector boundary, so this won't be
128 * an issue. Nor will this be a problem if the log I/O is
129 * done in basic blocks (sector size 1). But otherwise we
130 * extend the buffer by one extra log sector to ensure
131 * there's space to accommodate this possibility.
132 */
141 }
143 STATIC void
146 {
148 }
150 /*
151 * Return the address of the start of the given block number's data
152 * in a log buffer. The buffer covers a log sector-aligned region.
153 */
157 xfs_daddr_t blk_no,
160 {
165 }
168 /*
169 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
170 */
171 STATIC int
174 xfs_daddr_t blk_no,
177 {
186 }
203 }
205 STATIC int
208 xfs_daddr_t blk_no,
212 {
221 }
223 /*
224 * Read at an offset into the buffer. Returns with the buffer in it's original
225 * state regardless of the result of the read.
226 */
227 STATIC int
234 {
245 /* must reset buffer pointer even on error */
250 }
252 /*
253 * Write out the buffer at the given block for the given number of blocks.
254 * The buffer is kept locked across the write and is returned locked.
255 * This can only be used for synchronous log writes.
256 */
257 STATIC int
260 xfs_daddr_t blk_no,
263 {
272 }
291 }
293 #ifdef DEBUG
294 /*
295 * dump debug superblock and log record information
296 */
297 STATIC void
301 {
306 }
307 #else
308 #define xlog_header_check_dump(mp, head)
309 #endif
311 /*
312 * check log record header for recovery
313 */
314 STATIC int
318 {
321 /*
322 * IRIX doesn't write the h_fmt field and leaves it zeroed
323 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
324 * a dirty log created in IRIX.
325 */
340 }
342 }
344 /*
345 * read the head block of the log and check the header
346 */
347 STATIC int
351 {
355 /*
356 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
357 * h_fs_uuid is null, we assume this log was last mounted
358 * by IRIX and continue.
359 */
367 }
369 }
371 STATIC void
374 {
376 /*
377 * We're not going to bother about retrying
378 * this during recovery. One strike!
379 */
383 SHUTDOWN_META_IO_ERROR);
384 }
385 }
387 /*
388 * On v5 supers, a bli could be attached to update the metadata LSN.
389 * Clean it up.
390 */
397 }
399 /*
400 * This routine finds (to an approximation) the first block in the physical
401 * log which contains the given cycle. It uses a binary search algorithm.
402 * Note that the algorithm can not be perfect because the disk will not
403 * necessarily be perfect.
404 */
405 STATIC int
409 xfs_daddr_t first_blk,
411 uint cycle)
412 {
414 xfs_daddr_t mid_blk;
415 xfs_daddr_t end_blk;
416 uint mid_cycle;
428 else
431 }
438 }
440 /*
441 * Check that a range of blocks does not contain stop_on_cycle_no.
442 * Fill in *new_blk with the block offset where such a block is
443 * found, or with -1 (an invalid block number) if there is no such
444 * block in the range. The scan needs to occur from front to back
445 * and the pointer into the region must be updated since a later
446 * routine will need to perform another test.
447 */
448 STATIC int
451 xfs_daddr_t start_blk,
453 uint stop_on_cycle_no,
455 {
457 uint cycle;
459 xfs_daddr_t bufblks;
463 /*
464 * Greedily allocate a buffer big enough to handle the full
465 * range of basic blocks we'll be examining. If that fails,
466 * try a smaller size. We need to be able to read at least
467 * a log sector, or we're out of luck.
468 */
476 }
492 }
495 }
496 }
500 out:
503 }
505 /*
506 * Potentially backup over partial log record write.
507 *
508 * In the typical case, last_blk is the number of the block directly after
509 * a good log record. Therefore, we subtract one to get the block number
510 * of the last block in the given buffer. extra_bblks contains the number
511 * of blocks we would have read on a previous read. This happens when the
512 * last log record is split over the end of the physical log.
513 *
514 * extra_bblks is the number of blocks potentially verified on a previous
515 * call to this routine.
516 */
517 STATIC int
520 xfs_daddr_t start_blk,
523 {
524 xfs_daddr_t i;
544 }
548 /* valid log record not found */
554 }
560 }
569 }
571 /*
572 * We hit the beginning of the physical log & still no header. Return
573 * to caller. If caller can handle a return of -1, then this routine
574 * will be called again for the end of the physical log.
575 */
579 }
581 /*
582 * We have the final block of the good log (the first block
583 * of the log record _before_ the head. So we check the uuid.
584 */
588 /*
589 * We may have found a log record header before we expected one.
590 * last_blk will be the 1st block # with a given cycle #. We may end
591 * up reading an entire log record. In this case, we don't want to
592 * reset last_blk. Only when last_blk points in the middle of a log
593 * record do we update last_blk.
594 */
600 xhdrs++;
603 }
609 out:
612 }
614 /*
615 * Head is defined to be the point of the log where the next log write
616 * could go. This means that incomplete LR writes at the end are
617 * eliminated when calculating the head. We aren't guaranteed that previous
618 * LR have complete transactions. We only know that a cycle number of
619 * current cycle number -1 won't be present in the log if we start writing
620 * from our current block number.
621 *
622 * last_blk contains the block number of the first block with a given
623 * cycle number.
624 *
625 * Return: zero if normal, non-zero if error.
626 */
627 STATIC int
631 {
637 uint stop_on_cycle;
640 /* Is the end of the log device zeroed? */
645 }
649 /* Is the whole lot zeroed? */
651 /* Linux XFS shouldn't generate totally zeroed logs -
652 * mkfs etc write a dummy unmount record to a fresh
653 * log so we can store the uuid in there
654 */
656 }
659 }
680 /*
681 * If the 1st half cycle number is equal to the last half cycle number,
682 * then the entire log is stamped with the same cycle number. In this
683 * case, head_blk can't be set to zero (which makes sense). The below
684 * math doesn't work out properly with head_blk equal to zero. Instead,
685 * we set it to log_bbnum which is an invalid block number, but this
686 * value makes the math correct. If head_blk doesn't changed through
687 * all the tests below, *head_blk is set to zero at the very end rather
688 * than log_bbnum. In a sense, log_bbnum and zero are the same block
689 * in a circular file.
690 */
692 /*
693 * In this case we believe that the entire log should have
694 * cycle number last_half_cycle. We need to scan backwards
695 * from the end verifying that there are no holes still
696 * containing last_half_cycle - 1. If we find such a hole,
697 * then the start of that hole will be the new head. The
698 * simple case looks like
699 * x | x ... | x - 1 | x
700 * Another case that fits this picture would be
701 * x | x + 1 | x ... | x
702 * In this case the head really is somewhere at the end of the
703 * log, as one of the latest writes at the beginning was
704 * incomplete.
705 * One more case is
706 * x | x + 1 | x ... | x - 1 | x
707 * This is really the combination of the above two cases, and
708 * the head has to end up at the start of the x-1 hole at the
709 * end of the log.
710 *
711 * In the 256k log case, we will read from the beginning to the
712 * end of the log and search for cycle numbers equal to x-1.
713 * We don't worry about the x+1 blocks that we encounter,
714 * because we know that they cannot be the head since the log
715 * started with x.
716 */
720 /*
721 * In this case we want to find the first block with cycle
722 * number matching last_half_cycle. We expect the log to be
723 * some variation on
724 * x + 1 ... | x ... | x
725 * The first block with cycle number x (last_half_cycle) will
726 * be where the new head belongs. First we do a binary search
727 * for the first occurrence of last_half_cycle. The binary
728 * search may not be totally accurate, so then we scan back
729 * from there looking for occurrences of last_half_cycle before
730 * us. If that backwards scan wraps around the beginning of
731 * the log, then we look for occurrences of last_half_cycle - 1
732 * at the end of the log. The cases we're looking for look
733 * like
734 * v binary search stopped here
735 * x + 1 ... | x | x + 1 | x ... | x
736 * ^ but we want to locate this spot
737 * or
738 * <---------> less than scan distance
739 * x + 1 ... | x ... | x - 1 | x
740 * ^ we want to locate this spot
741 */
746 }
748 /*
749 * Now validate the answer. Scan back some number of maximum possible
750 * blocks and make sure each one has the expected cycle number. The
751 * maximum is determined by the total possible amount of buffering
752 * in the in-core log. The following number can be made tighter if
753 * we actually look at the block size of the filesystem.
754 */
757 /*
758 * We are guaranteed that the entire check can be performed
759 * in one buffer.
760 */
769 /*
770 * We are going to scan backwards in the log in two parts.
771 * First we scan the physical end of the log. In this part
772 * of the log, we are looking for blocks with cycle number
773 * last_half_cycle - 1.
774 * If we find one, then we know that the log starts there, as
775 * we've found a hole that didn't get written in going around
776 * the end of the physical log. The simple case for this is
777 * x + 1 ... | x ... | x - 1 | x
778 * <---------> less than scan distance
779 * If all of the blocks at the end of the log have cycle number
780 * last_half_cycle, then we check the blocks at the start of
781 * the log looking for occurrences of last_half_cycle. If we
782 * find one, then our current estimate for the location of the
783 * first occurrence of last_half_cycle is wrong and we move
784 * back to the hole we've found. This case looks like
785 * x + 1 ... | x | x + 1 | x ...
786 * ^ binary search stopped here
787 * Another case we need to handle that only occurs in 256k
788 * logs is
789 * x + 1 ... | x ... | x+1 | x ...
790 * ^ binary search stops here
791 * In a 256k log, the scan at the end of the log will see the
792 * x + 1 blocks. We need to skip past those since that is
793 * certainly not the head of the log. By searching for
794 * last_half_cycle-1 we accomplish that.
795 */
806 }
808 /*
809 * Scan beginning of log now. The last part of the physical
810 * log is good. This scan needs to verify that it doesn't find
811 * the last_half_cycle.
812 */
821 }
823 validate_head:
824 /*
825 * Now we need to make sure head_blk is not pointing to a block in
826 * the middle of a log record.
827 */
832 /* start ptr at last block ptr before head_blk */
845 /* We hit the beginning of the log during our search */
861 }
866 else
868 /*
869 * When returning here, we have a good block number. Bad block
870 * means that during a previous crash, we didn't have a clean break
871 * from cycle number N to cycle number N-1. In this case, we need
872 * to find the first block with cycle number N-1.
873 */
876 bp_err:
882 }
884 /*
885 * Seek backwards in the log for log record headers.
886 *
887 * Given a starting log block, walk backwards until we find the provided number
888 * of records or hit the provided tail block. The return value is the number of
889 * records encountered or a negative error code. The log block and buffer
890 * pointer of the last record seen are returned in rblk and rhead respectively.
891 */
892 STATIC int
895 xfs_daddr_t head_blk,
896 xfs_daddr_t tail_blk,
902 {
907 xfs_daddr_t end_blk;
911 /*
912 * Walk backwards from the head block until we hit the tail or the first
913 * block in the log.
914 */
926 }
927 }
929 /*
930 * If we haven't hit the tail block or the log record header count,
931 * start looking again from the end of the physical log. Note that
932 * callers can pass head == tail if the tail is not yet known.
933 */
947 }
948 }
949 }
953 out_error:
955 }
957 /*
958 * Seek forward in the log for log record headers.
959 *
960 * Given head and tail blocks, walk forward from the tail block until we find
961 * the provided number of records or hit the head block. The return value is the
962 * number of records encountered or a negative error code. The log block and
963 * buffer pointer of the last record seen are returned in rblk and rhead
964 * respectively.
965 */
966 STATIC int
969 xfs_daddr_t head_blk,
970 xfs_daddr_t tail_blk,
976 {
981 xfs_daddr_t end_blk;
985 /*
986 * Walk forward from the tail block until we hit the head or the last
987 * block in the log.
988 */
1000 }
1001 }
1003 /*
1004 * If we haven't hit the head block or the log record header count,
1005 * start looking again from the start of the physical log.
1006 */
1020 }
1021 }
1022 }
1026 out_error:
1028 }
1030 /*
1031 * Calculate distance from head to tail (i.e., unused space in the log).
1032 */
1036 xfs_daddr_t head_blk,
1037 xfs_daddr_t tail_blk)
1038 {
1043 }
1045 /*
1046 * Verify the log tail. This is particularly important when torn or incomplete
1047 * writes have been detected near the front of the log and the head has been
1048 * walked back accordingly.
1049 *
1050 * We also have to handle the case where the tail was pinned and the head
1051 * blocked behind the tail right before a crash. If the tail had been pushed
1052 * immediately prior to the crash and the subsequent checkpoint was only
1053 * partially written, it's possible it overwrote the last referenced tail in the
1054 * log with garbage. This is not a coherency problem because the tail must have
1055 * been pushed before it can be overwritten, but appears as log corruption to
1056 * recovery because we have no way to know the tail was updated if the
1057 * subsequent checkpoint didn't write successfully.
1058 *
1059 * Therefore, CRC check the log from tail to head. If a failure occurs and the
1060 * offending record is within max iclog bufs from the head, walk the tail
1061 * forward and retry until a valid tail is found or corruption is detected out
1062 * of the range of a possible overwrite.
1063 */
1064 STATIC int
1067 xfs_daddr_t head_blk,
1070 {
1073 xfs_daddr_t first_bad;
1076 xfs_daddr_t tmp_tail;
1083 /*
1084 * Make sure the tail points to a record (returns positive count on
1085 * success).
1086 */
1094 /*
1095 * Run a CRC check from the tail to the head. We can't just check
1096 * MAX_ICLOGS records past the tail because the tail may point to stale
1097 * blocks cleared during the search for the head/tail. These blocks are
1098 * overwritten with zero-length records and thus record count is not a
1099 * reliable indicator of the iclog state before a crash.
1100 */
1107 /*
1108 * Is corruption within range of the head? If so, retry from
1109 * the next record. Otherwise return an error.
1110 */
1115 /* skip to the next record; returns positive count on success */
1125 }
1131 out:
1134 }
1136 /*
1137 * Detect and trim torn writes from the head of the log.
1138 *
1139 * Storage without sector atomicity guarantees can result in torn writes in the
1140 * log in the event of a crash. Our only means to detect this scenario is via
1141 * CRC verification. While we can't always be certain that CRC verification
1142 * failure is due to a torn write vs. an unrelated corruption, we do know that
1143 * only a certain number (XLOG_MAX_ICLOGS) of log records can be written out at
1144 * one time. Therefore, CRC verify up to XLOG_MAX_ICLOGS records at the head of
1145 * the log and treat failures in this range as torn writes as a matter of
1146 * policy. In the event of CRC failure, the head is walked back to the last good
1147 * record in the log and the tail is updated from that record and verified.
1148 */
1149 STATIC int
1158 {
1161 xfs_daddr_t first_bad;
1162 xfs_daddr_t tmp_rhead_blk;
1167 /*
1168 * Check the head of the log for torn writes. Search backwards from the
1169 * head until we hit the tail or the maximum number of log record I/Os
1170 * that could have been in flight at one time. Use a temporary buffer so
1171 * we don't trash the rhead/bp pointers from the caller.
1172 */
1183 /*
1184 * Now run a CRC verification pass over the records starting at the
1185 * block found above to the current head. If a CRC failure occurs, the
1186 * log block of the first bad record is saved in first_bad.
1187 */
1191 /*
1192 * We've hit a potential torn write. Reset the error and warn
1193 * about it.
1194 */
1200 /*
1201 * Get the header block and buffer pointer for the last good
1202 * record before the bad record.
1203 *
1204 * Note that xlog_find_tail() clears the blocks at the new head
1205 * (i.e., the records with invalid CRC) if the cycle number
1206 * matches the the current cycle.
1207 */
1215 /*
1216 * Reset the head block to the starting block of the first bad
1217 * log record and set the tail block based on the last good
1218 * record.
1219 *
1220 * Bail out if the updated head/tail match as this indicates
1221 * possible corruption outside of the acceptable
1222 * (XLOG_MAX_ICLOGS) range. This is a job for xfs_repair...
1223 */
1229 }
1230 }
1236 }
1238 /*
1239 * We need to make sure we handle log wrapping properly, so we can't use the
1240 * calculated logbno directly. Make sure it wraps to the correct bno inside the
1241 * log.
1242 *
1243 * The log is limited to 32 bit sizes, so we use the appropriate modulus
1244 * operation here and cast it back to a 64 bit daddr on return.
1245 */
1249 xfs_daddr_t bno)
1250 {
1255 }
1257 /*
1258 * Check whether the head of the log points to an unmount record. In other
1259 * words, determine whether the log is clean. If so, update the in-core state
1260 * appropriately.
1261 */
1262 static int
1268 xfs_daddr_t rhead_blk,
1271 {
1273 xfs_daddr_t umount_data_blk;
1274 xfs_daddr_t after_umount_blk;
1281 /*
1282 * Look for unmount record. If we find it, then we know there was a
1283 * clean unmount. Since 'i' could be the last block in the physical
1284 * log, we convert to a log block before comparing to the head_blk.
1285 *
1286 * Save the current tail lsn to use to pass to xlog_clear_stale_blocks()
1287 * below. We won't want to clear the unmount record if there is one, so
1288 * we pass the lsn of the unmount record rather than the block after it.
1289 */
1298 hblks++;
1301 }
1304 }
1318 /*
1319 * Set tail and last sync so that newly written log
1320 * records will point recovery to after the current
1321 * unmount record.
1322 */
1330 }
1331 }
1334 }
1336 static void
1339 xfs_daddr_t head_blk,
1341 xfs_daddr_t rhead_blk,
1343 {
1344 /*
1345 * Reset log values according to the state of the log when we
1346 * crashed. In the case where head_blk == 0, we bump curr_cycle
1347 * one because the next write starts a new cycle rather than
1348 * continuing the cycle of the last good log record. At this
1349 * point we have guaranteed that all partial log records have been
1350 * accounted for. Therefore, we know that the last good log record
1351 * written was complete and ended exactly on the end boundary
1352 * of the physical log.
1353 */
1365 }
1367 /*
1368 * Find the sync block number or the tail of the log.
1369 *
1370 * This will be the block number of the last record to have its
1371 * associated buffers synced to disk. Every log record header has
1372 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
1373 * to get a sync block number. The only concern is to figure out which
1374 * log record header to believe.
1375 *
1376 * The following algorithm uses the log record header with the largest
1377 * lsn. The entire log record does not need to be valid. We only care
1378 * that the header is valid.
1379 *
1380 * We could speed up search by using current head_blk buffer, but it is not
1381 * available.
1382 */
1383 STATIC int
1388 {
1393 xfs_daddr_t rhead_blk;
1394 xfs_lsn_t tail_lsn;
1398 /*
1399 * Find previous log record
1400 */
1415 /* leave all other log inited values alone */
1417 }
1418 }
1420 /*
1421 * Search backwards through the log looking for the log record header
1422 * block. This wraps all the way back around to the head so something is
1423 * seriously wrong if we can't find it.
1424 */
1432 }
1435 /*
1436 * Set the log state based on the current head record.
1437 */
1441 /*
1442 * Look for an unmount record at the head of the log. This sets the log
1443 * state to determine whether recovery is necessary.
1444 */
1450 /*
1451 * Verify the log head if the log is not clean (e.g., we have anything
1452 * but an unmount record at the head). This uses CRC verification to
1453 * detect and trim torn writes. If discovered, CRC failures are
1454 * considered torn writes and the log head is trimmed accordingly.
1455 *
1456 * Note that we can only run CRC verification when the log is dirty
1457 * because there's no guarantee that the log data behind an unmount
1458 * record is compatible with the current architecture.
1459 */
1468 /* update in-core state again if the head changed */
1471 wrapped);
1478 }
1479 }
1481 /*
1482 * Note that the unmount was clean. If the unmount was not clean, we
1483 * need to know this to rebuild the superblock counters from the perag
1484 * headers if we have a filesystem using non-persistent counters.
1485 */
1489 /*
1490 * Make sure that there are no blocks in front of the head
1491 * with the same cycle number as the head. This can happen
1492 * because we allow multiple outstanding log writes concurrently,
1493 * and the later writes might make it out before earlier ones.
1494 *
1495 * We use the lsn from before modifying it so that we'll never
1496 * overwrite the unmount record after a clean unmount.
1497 *
1498 * Do this only if we are going to recover the filesystem
1499 *
1500 * NOTE: This used to say "if (!readonly)"
1501 * However on Linux, we can & do recover a read-only filesystem.
1502 * We only skip recovery if NORECOVERY is specified on mount,
1503 * in which case we would not be here.
1504 *
1505 * But... if the -device- itself is readonly, just skip this.
1506 * We can't recover this device anyway, so it won't matter.
1507 */
1511 done:
1517 }
1519 /*
1520 * Is the log zeroed at all?
1521 *
1522 * The last binary search should be changed to perform an X block read
1523 * once X becomes small enough. You can then search linearly through
1524 * the X blocks. This will cut down on the number of reads we need to do.
1525 *
1526 * If the log is partially zeroed, this routine will pass back the blkno
1527 * of the first block with cycle number 0. It won't have a complete LR
1528 * preceding it.
1529 *
1530 * Return:
1531 * 0 => the log is completely written to
1532 * 1 => use *blk_no as the first block of the log
1533 * <0 => error has occurred
1534 */
1535 STATIC int
1539 {
1544 xfs_daddr_t num_scan_bblks;
1549 /* check totally zeroed log */
1562 }
1564 /* check partially zeroed log */
1574 /*
1575 * If the cycle of the last block is zero, the cycle of
1576 * the first block must be 1. If it's not, maybe we're
1577 * not looking at a log... Bail out.
1578 */
1583 }
1585 /* we have a partially zeroed log */
1590 /*
1591 * Validate the answer. Because there is no way to guarantee that
1592 * the entire log is made up of log records which are the same size,
1593 * we scan over the defined maximum blocks. At this point, the maximum
1594 * is not chosen to mean anything special. XXXmiken
1595 */
1603 /*
1604 * We search for any instances of cycle number 0 that occur before
1605 * our current estimate of the head. What we're trying to detect is
1606 * 1 ... | 0 | 1 | 0...
1607 * ^ binary search ends here
1608 */
1615 /*
1616 * Potentially backup over partial log record write. We don't need
1617 * to search the end of the log because we know it is zero.
1618 */
1626 bp_err:
1631 }
1633 /*
1634 * These are simple subroutines used by xlog_clear_stale_blocks() below
1635 * to initialize a buffer full of empty log record headers and write
1636 * them into the log.
1637 */
1638 STATIC void
1646 {
1658 }
1660 STATIC int
1668 {
1678 /*
1679 * Greedily allocate a buffer big enough to handle the full
1680 * range of basic blocks to be written. If that fails, try
1681 * a smaller size. We need to be able to write at least a
1682 * log sector, or we're out of luck.
1683 */
1691 }
1693 /* We may need to do a read at the start to fill in part of
1694 * the buffer in the starting sector not covered by the first
1695 * write below.
1696 */
1704 }
1712 /* We may need to do a read at the end to fill in part of
1713 * the buffer in the final sector not covered by the write.
1714 * If this is the same sector as the above read, skip it.
1715 */
1724 }
1731 }
1737 }
1739 out_put_bp:
1742 }
1744 /*
1745 * This routine is called to blow away any incomplete log writes out
1746 * in front of the log head. We do this so that we won't become confused
1747 * if we come up, write only a little bit more, and then crash again.
1748 * If we leave the partial log records out there, this situation could
1749 * cause us to think those partial writes are valid blocks since they
1750 * have the current cycle number. We get rid of them by overwriting them
1751 * with empty log records with the old cycle number rather than the
1752 * current one.
1753 *
1754 * The tail lsn is passed in rather than taken from
1755 * the log so that we will not write over the unmount record after a
1756 * clean unmount in a 512 block log. Doing so would leave the log without
1757 * any valid log records in it until a new one was written. If we crashed
1758 * during that time we would not be able to recover.
1759 */
1760 STATIC int
1763 xfs_lsn_t tail_lsn)
1764 {
1776 /*
1777 * Figure out the distance between the new head of the log
1778 * and the tail. We want to write over any blocks beyond the
1779 * head that we may have written just before the crash, but
1780 * we don't want to overwrite the tail of the log.
1781 */
1783 /*
1784 * The tail is behind the head in the physical log,
1785 * so the distance from the head to the tail is the
1786 * distance from the head to the end of the log plus
1787 * the distance from the beginning of the log to the
1788 * tail.
1789 */
1794 }
1797 /*
1798 * The head is behind the tail in the physical log,
1799 * so the distance from the head to the tail is just
1800 * the tail block minus the head block.
1801 */
1806 }
1808 }
1810 /*
1811 * If the head is right up against the tail, we can't clear
1812 * anything.
1813 */
1817 }
1820 /*
1821 * Take the smaller of the maximum amount of outstanding I/O
1822 * we could have and the distance to the tail to clear out.
1823 * We take the smaller so that we don't overwrite the tail and
1824 * we don't waste all day writing from the head to the tail
1825 * for no reason.
1826 */
1830 /*
1831 * We can stomp all the blocks we need to without
1832 * wrapping around the end of the log. Just do it
1833 * in a single write. Use the cycle number of the
1834 * current cycle minus one so that the log will look like:
1835 * n ... | n - 1 ...
1836 */
1839 tail_block);
1843 /*
1844 * We need to wrap around the end of the physical log in
1845 * order to clear all the blocks. Do it in two separate
1846 * I/Os. The first write should be from the head to the
1847 * end of the physical log, and it should use the current
1848 * cycle number minus one just like above.
1849 */
1853 tail_block);
1858 /*
1859 * Now write the blocks at the start of the physical log.
1860 * This writes the remainder of the blocks we want to clear.
1861 * It uses the current cycle number since we're now on the
1862 * same cycle as the head so that we get:
1863 * n ... n ... | n - 1 ...
1864 * ^^^^^ blocks we're writing
1865 */
1871 }
1874 }
1876 /******************************************************************************
1877 *
1878 * Log recover routines
1879 *
1880 ******************************************************************************
1881 */
1883 /*
1884 * Sort the log items in the transaction.
1885 *
1886 * The ordering constraints are defined by the inode allocation and unlink
1887 * behaviour. The rules are:
1888 *
1889 * 1. Every item is only logged once in a given transaction. Hence it
1890 * represents the last logged state of the item. Hence ordering is
1891 * dependent on the order in which operations need to be performed so
1892 * required initial conditions are always met.
1893 *
1894 * 2. Cancelled buffers are recorded in pass 1 in a separate table and
1895 * there's nothing to replay from them so we can simply cull them
1896 * from the transaction. However, we can't do that until after we've
1897 * replayed all the other items because they may be dependent on the
1898 * cancelled buffer and replaying the cancelled buffer can remove it
1899 * form the cancelled buffer table. Hence they have tobe done last.
1900 *
1901 * 3. Inode allocation buffers must be replayed before inode items that
1902 * read the buffer and replay changes into it. For filesystems using the
1903 * ICREATE transactions, this means XFS_LI_ICREATE objects need to get
1904 * treated the same as inode allocation buffers as they create and
1905 * initialise the buffers directly.
1906 *
1907 * 4. Inode unlink buffers must be replayed after inode items are replayed.
1908 * This ensures that inodes are completely flushed to the inode buffer
1909 * in a "free" state before we remove the unlinked inode list pointer.
1910 *
1911 * Hence the ordering needs to be inode allocation buffers first, inode items
1912 * second, inode unlink buffers third and cancelled buffers last.
1913 *
1914 * But there's a problem with that - we can't tell an inode allocation buffer
1915 * apart from a regular buffer, so we can't separate them. We can, however,
1916 * tell an inode unlink buffer from the others, and so we can separate them out
1917 * from all the other buffers and move them to last.
1918 *
1919 * Hence, 4 lists, in order from head to tail:
1920 * - buffer_list for all buffers except cancelled/inode unlink buffers
1921 * - item_list for all non-buffer items
1922 * - inode_buffer_list for inode unlink buffers
1923 * - cancel_list for the cancelled buffers
1924 *
1925 * Note that we add objects to the tail of the lists so that first-to-last
1926 * ordering is preserved within the lists. Adding objects to the head of the
1927 * list means when we traverse from the head we walk them in last-to-first
1928 * order. For cancelled buffers and inode unlink buffers this doesn't matter,
1929 * but for all other items there may be specific ordering that we need to
1930 * preserve.
1931 */
1932 STATIC int
1937 {
1960 }
1964 }
1985 __func__);
1987 /*
1988 * return the remaining items back to the transaction
1989 * item list so they can be freed in caller.
1990 */
1995 }
1996 }
1997 out:
2008 }
2010 /*
2011 * Build up the table of buf cancel records so that we don't replay
2012 * cancelled data in the second pass. For buffer records that are
2013 * not cancel records, there is nothing to do here so we just return.
2014 *
2015 * If we get a cancel record which is already in the table, this indicates
2016 * that the buffer was cancelled multiple times. In order to ensure
2017 * that during pass 2 we keep the record in the table until we reach its
2018 * last occurrence in the log, we keep a reference count in the cancel
2019 * record in the table to tell us how many times we expect to see this
2020 * record during the second pass.
2021 */
2022 STATIC int
2026 {
2031 /*
2032 * If this isn't a cancel buffer item, then just return.
2033 */
2037 }
2039 /*
2040 * Insert an xfs_buf_cancel record into the hash table of them.
2041 * If there is already an identical record, bump its reference count.
2042 */
2050 }
2051 }
2061 }
2063 /*
2064 * Check to see whether the buffer being recovered has a corresponding
2065 * entry in the buffer cancel record table. If it is, return the cancel
2066 * buffer structure to the caller.
2067 */
2071 xfs_daddr_t blkno,
2072 uint len,
2074 {
2079 /* empty table means no cancelled buffers in the log */
2082 }
2088 }
2090 /*
2091 * We didn't find a corresponding entry in the table, so return 0 so
2092 * that the buffer is NOT cancelled.
2093 */
2096 }
2098 /*
2099 * If the buffer is being cancelled then return 1 so that it will be cancelled,
2100 * otherwise return 0. If the buffer is actually a buffer cancel item
2101 * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the
2102 * table and remove it from the table if this is the last reference.
2103 *
2104 * We remove the cancel record from the table when we encounter its last
2105 * occurrence in the log so that if the same buffer is re-used again after its
2106 * last cancellation we actually replay the changes made at that point.
2107 */
2108 STATIC int
2111 xfs_daddr_t blkno,
2112 uint len,
2114 {
2121 /*
2122 * We've go a match, so return 1 so that the recovery of this buffer
2123 * is cancelled. If this buffer is actually a buffer cancel log
2124 * item, then decrement the refcount on the one in the table and
2125 * remove it if this is the last reference.
2126 */
2131 }
2132 }
2134 }
2136 /*
2137 * Perform recovery for a buffer full of inodes. In these buffers, the only
2138 * data which should be recovered is that which corresponds to the
2139 * di_next_unlinked pointers in the on disk inode structures. The rest of the
2140 * data for the inodes is always logged through the inodes themselves rather
2141 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
2142 *
2143 * The only time when buffers full of inodes are fully recovered is when the
2144 * buffer is full of newly allocated inodes. In this case the buffer will
2145 * not be marked as an inode buffer and so will be sent to
2146 * xlog_recover_do_reg_buffer() below during recovery.
2147 */
2148 STATIC int
2154 {
2168 /*
2169 * Post recovery validation only works properly on CRC enabled
2170 * filesystems.
2171 */
2182 /*
2183 * The next di_next_unlinked field is beyond
2184 * the current logged region. Find the next
2185 * logged region that contains or is beyond
2186 * the current di_next_unlinked field.
2187 */
2192 /*
2193 * If there are no more logged regions in the
2194 * buffer, then we're done.
2195 */
2204 item_index++;
2205 }
2207 /*
2208 * If the current logged region starts after the current
2209 * di_next_unlinked field, then move on to the next
2210 * di_next_unlinked field.
2211 */
2220 /*
2221 * The current logged region contains a copy of the
2222 * current di_next_unlinked field. Extract its value
2223 * and copy it to the buffer copy.
2224 */
2235 }
2240 /*
2241 * If necessary, recalculate the CRC in the on-disk inode. We
2242 * have to leave the inode in a consistent state for whoever
2243 * reads it next....
2244 */
2248 }
2251 }
2253 /*
2254 * V5 filesystems know the age of the buffer on disk being recovered. We can
2255 * have newer objects on disk than we are replaying, and so for these cases we
2256 * don't want to replay the current change as that will make the buffer contents
2257 * temporarily invalid on disk.
2258 *
2259 * The magic number might not match the buffer type we are going to recover
2260 * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags. Hence
2261 * extract the LSN of the existing object in the buffer based on it's current
2262 * magic number. If we don't recognise the magic number in the buffer, then
2263 * return a LSN of -1 so that the caller knows it was an unrecognised block and
2264 * so can recover the buffer.
2265 *
2266 * Note: we cannot rely solely on magic number matches to determine that the
2267 * buffer has a valid LSN - we also need to verify that it belongs to this
2268 * filesystem, so we need to extract the object's LSN and compare it to that
2269 * which we read from the superblock. If the UUIDs don't match, then we've got a
2270 * stale metadata block from an old filesystem instance that we need to recover
2271 * over the top of.
2272 */
2273 static xfs_lsn_t
2277 {
2285 /* v4 filesystems always recover immediately */
2304 }
2312 }
2336 /*
2337 * Remote attr blocks are written synchronously, rather than
2338 * being logged. That means they do not contain a valid LSN
2339 * (i.e. transactionally ordered) in them, and hence any time we
2340 * see a buffer to replay over the top of a remote attribute
2341 * block we should simply do so.
2342 */
2345 /*
2346 * superblock uuids are magic. We may or may not have a
2347 * sb_meta_uuid on disk, but it will be set in the in-core
2348 * superblock. We set the uuid pointer for verification
2349 * according to the superblock feature mask to ensure we check
2350 * the relevant UUID in the superblock.
2351 */
2355 else
2360 }
2366 }
2378 }
2384 }
2386 /*
2387 * We do individual object checks on dquot and inode buffers as they
2388 * have their own individual LSN records. Also, we could have a stale
2389 * buffer here, so we have to at least recognise these buffer types.
2390 *
2391 * A notd complexity here is inode unlinked list processing - it logs
2392 * the inode directly in the buffer, but we don't know which inodes have
2393 * been modified, and there is no global buffer LSN. Hence we need to
2394 * recover all inode buffer types immediately. This problem will be
2395 * fixed by logical logging of the unlinked list modifications.
2396 */
2404 }
2406 /* unknown buffer contents, recover immediately */
2408 recover_immediately:
2411 }
2413 /*
2414 * Validate the recovered buffer is of the correct type and attach the
2415 * appropriate buffer operations to them for writeback. Magic numbers are in a
2416 * few places:
2417 * the first 16 bits of the buffer (inode buffer, dquot buffer),
2418 * the first 32 bits of the buffer (most blocks),
2419 * inside a struct xfs_da_blkinfo at the start of the buffer.
2420 */
2421 static void
2426 xfs_lsn_t current_lsn)
2427 {
2434 /*
2435 * We can only do post recovery validation on items on CRC enabled
2436 * fielsystems as we need to know when the buffer was written to be able
2437 * to determine if we should have replayed the item. If we replay old
2438 * metadata over a newer buffer, then it will enter a temporarily
2439 * inconsistent state resulting in verification failures. Hence for now
2440 * just avoid the verification stage for non-crc filesystems
2441 */
2476 }
2482 }
2489 }
2496 }
2502 #ifdef CONFIG_XFS_QUOTA
2506 }
2508 #else
2512 #endif
2518 }
2525 }
2533 }
2541 }
2549 }
2557 }
2565 }
2573 }
2581 }
2588 }
2595 }
2598 #ifdef CONFIG_XFS_RT
2601 /* no magic numbers for verification of RT buffers */
2609 }
2611 /*
2612 * Nothing else to do in the case of a NULL current LSN as this means
2613 * the buffer is more recent than the change in the log and will be
2614 * skipped.
2615 */
2622 }
2624 /*
2625 * We must update the metadata LSN of the buffer as it is written out to
2626 * ensure that older transactions never replay over this one and corrupt
2627 * the buffer. This can occur if log recovery is interrupted at some
2628 * point after the current transaction completes, at which point a
2629 * subsequent mount starts recovery from the beginning.
2630 *
2631 * Write verifiers update the metadata LSN from log items attached to
2632 * the buffer. Therefore, initialize a bli purely to carry the LSN to
2633 * the verifier. We'll clean it up in our ->iodone() callback.
2634 */
2643 }
2644 }
2646 /*
2647 * Perform a 'normal' buffer recovery. Each logged region of the
2648 * buffer should be copied over the corresponding region in the
2649 * given buffer. The bitmap in the buf log format structure indicates
2650 * where to place the logged data.
2651 */
2652 STATIC void
2658 xfs_lsn_t current_lsn)
2659 {
2663 xfs_failaddr_t fa;
2682 /*
2683 * The dirty regions logged in the buffer, even though
2684 * contiguous, may span multiple chunks. This is because the
2685 * dirty region may span a physical page boundary in a buffer
2686 * and hence be split into two separate vectors for writing into
2687 * the log. Hence we need to trim nbits back to the length of
2688 * the current region being copied out of the log.
2689 */
2693 /*
2694 * Do a sanity check if this is a dquot buffer. Just checking
2695 * the first dquot in the buffer should do. XXXThis is
2696 * probably a good thing to do for other buf types also.
2697 */
2705 }
2711 }
2719 }
2720 }
2726 next:
2727 i++;
2729 }
2731 /* Shouldn't be any more regions */
2735 }
2737 /*
2738 * Perform a dquot buffer recovery.
2739 * Simple algorithm: if we have found a QUOTAOFF log item of the same type
2740 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2741 * Else, treat it as a regular buffer and do recovery.
2742 *
2743 * Return false if the buffer was tossed and true if we recovered the buffer to
2744 * indicate to the caller if the buffer needs writing.
2745 */
2746 STATIC bool
2753 {
2754 uint type;
2758 /*
2759 * Filesystems are required to send in quota flags at mount time.
2760 */
2771 /*
2772 * This type of quotas was turned off, so ignore this buffer
2773 */
2779 }
2781 /*
2782 * This routine replays a modification made to a buffer at runtime.
2783 * There are actually two types of buffer, regular and inode, which
2784 * are handled differently. Inode buffers are handled differently
2785 * in that we only recover a specific set of data from them, namely
2786 * the inode di_next_unlinked fields. This is because all other inode
2787 * data is actually logged via inode records and any data we replay
2788 * here which overlaps that may be stale.
2789 *
2790 * When meta-data buffers are freed at run time we log a buffer item
2791 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2792 * of the buffer in the log should not be replayed at recovery time.
2793 * This is so that if the blocks covered by the buffer are reused for
2794 * file data before we crash we don't end up replaying old, freed
2795 * meta-data into a user's file.
2796 *
2797 * To handle the cancellation of buffer log items, we make two passes
2798 * over the log during recovery. During the first we build a table of
2799 * those buffers which have been cancelled, and during the second we
2800 * only replay those buffers which do not have corresponding cancel
2801 * records in the table. See xlog_recover_buffer_pass[1,2] above
2802 * for more details on the implementation of the table of cancel records.
2803 */
2804 STATIC int
2809 xfs_lsn_t current_lsn)
2810 {
2815 uint buf_flags;
2816 xfs_lsn_t lsn;
2818 /*
2819 * In this pass we only want to recover all the buffers which have
2820 * not been cancelled and are not cancellation buffers themselves.
2821 */
2826 }
2842 }
2844 /*
2845 * Recover the buffer only if we get an LSN from it and it's less than
2846 * the lsn of the transaction we are replaying.
2847 *
2848 * Note that we have to be extremely careful of readahead here.
2849 * Readahead does not attach verfiers to the buffers so if we don't
2850 * actually do any replay after readahead because of the LSN we found
2851 * in the buffer if more recent than that current transaction then we
2852 * need to attach the verifier directly. Failure to do so can lead to
2853 * future recovery actions (e.g. EFI and unlinked list recovery) can
2854 * operate on the buffers and they won't get the verifier attached. This
2855 * can lead to blocks on disk having the correct content but a stale
2856 * CRC.
2857 *
2858 * It is safe to assume these clean buffers are currently up to date.
2859 * If the buffer is dirtied by a later transaction being replayed, then
2860 * the verifier will be reset to match whatever recover turns that
2861 * buffer into.
2862 */
2868 }
2883 }
2885 /*
2886 * Perform delayed write on the buffer. Asynchronous writes will be
2887 * slower when taking into account all the buffers to be flushed.
2888 *
2889 * Also make sure that only inode buffers with good sizes stay in
2890 * the buffer cache. The kernel moves inodes in buffers of 1 block
2891 * or mp->m_inode_cluster_size bytes, whichever is bigger. The inode
2892 * buffers in the log can be a different size if the log was generated
2893 * by an older kernel using unclustered inode buffers or a newer kernel
2894 * running with a different inode cluster size. Regardless, if the
2895 * the inode buffer size isn't max(blocksize, mp->m_inode_cluster_size)
2896 * for *our* value of mp->m_inode_cluster_size, then we need to keep
2897 * the buffer out of the buffer cache so that the buffer won't
2898 * overlap with future reads of those inodes.
2899 */
2910 }
2912 out_release:
2915 }
2917 /*
2918 * Inode fork owner changes
2919 *
2920 * If we have been told that we have to reparent the inode fork, it's because an
2921 * extent swap operation on a CRC enabled filesystem has been done and we are
2922 * replaying it. We need to walk the BMBT of the appropriate fork and change the
2923 * owners of it.
2924 *
2925 * The complexity here is that we don't have an inode context to work with, so
2926 * after we've replayed the inode we need to instantiate one. This is where the
2927 * fun begins.
2928 *
2929 * We are in the middle of log recovery, so we can't run transactions. That
2930 * means we cannot use cache coherent inode instantiation via xfs_iget(), as
2931 * that will result in the corresponding iput() running the inode through
2932 * xfs_inactive(). If we've just replayed an inode core that changes the link
2933 * count to zero (i.e. it's been unlinked), then xfs_inactive() will run
2934 * transactions (bad!).
2935 *
2936 * So, to avoid this, we instantiate an inode directly from the inode core we've
2937 * just recovered. We have the buffer still locked, and all we really need to
2938 * instantiate is the inode core and the forks being modified. We can do this
2939 * manually, then run the inode btree owner change, and then tear down the
2940 * xfs_inode without having to run any transactions at all.
2941 *
2942 * Also, because we don't have a transaction context available here but need to
2943 * gather all the buffers we modify for writeback so we pass the buffer_list
2944 * instead for the operation to use.
2945 */
2947 STATIC int
2953 {
2963 /* instantiate the inode */
2974 }
2982 }
2990 }
2992 out_free_ip:
2995 }
2997 STATIC int
3002 xfs_lsn_t current_lsn)
3003 {
3013 uint fields;
3015 uint isize;
3026 }
3028 /*
3029 * Inode buffers can be freed, look out for it,
3030 * and do not replay the inode.
3031 */
3037 }
3045 }
3050 }
3054 /*
3055 * Make sure the place we're flushing out to really looks
3056 * like an inode!
3057 */
3066 }
3076 }
3078 /*
3079 * If the inode has an LSN in it, recover the inode only if it's less
3080 * than the lsn of the transaction we are replaying. Note: we still
3081 * need to replay an owner change even though the inode is more recent
3082 * than the transaction as there is no guarantee that all the btree
3083 * blocks are more recent than this transaction, too.
3084 */
3092 }
3093 }
3095 /*
3096 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
3097 * are transactional and if ordering is necessary we can determine that
3098 * more accurately by the LSN field in the V3 inode core. Don't trust
3099 * the inode versions we might be changing them here - use the
3100 * superblock flag to determine whether we need to look at di_flushiter
3101 * to skip replay when the on disk inode is newer than the log one
3102 */
3105 /*
3106 * Deal with the wrap case, DI_MAX_FLUSH is less
3107 * than smaller numbers
3108 */
3111 /* do nothing */
3116 }
3117 }
3119 /* Take the opportunity to reset the flush iteration count */
3134 }
3148 }
3149 }
3162 }
3173 }
3184 }
3186 /* recover the log dinode inode into the on disk inode */
3215 /*
3216 * There are no data fork flags set.
3217 */
3220 }
3222 /*
3223 * If we logged any attribute data, recover it. There may or
3224 * may not have been any other non-core data logged in this
3225 * transaction.
3226 */
3232 }
3257 }
3258 }
3260 out_owner_change:
3261 /* Recover the swapext owner change unless inode has been deleted */
3265 buffer_list);
3266 /* re-generate the checksum. */
3273 out_release:
3275 error:
3279 }
3281 /*
3282 * Recover QUOTAOFF records. We simply make a note of it in the xlog
3283 * structure, so that we know not to do any dquot item or dquot buffer recovery,
3284 * of that type.
3285 */
3286 STATIC int
3290 {
3294 /*
3295 * The logitem format's flag tells us if this was user quotaoff,
3296 * group/project quotaoff or both.
3297 */
3306 }
3308 /*
3309 * Recover a dquot record
3310 */
3311 STATIC int
3316 xfs_lsn_t current_lsn)
3317 {
3321 xfs_failaddr_t fa;
3324 uint type;
3327 /*
3328 * Filesystems are required to send in quota flags at mount time.
3329 */
3337 }
3342 }
3344 /*
3345 * This type of quotas was turned off, so ignore this record.
3346 */
3352 /*
3353 * At this point we know that quota was _not_ turned off.
3354 * Since the mount flags are not indicating to us otherwise, this
3355 * must mean that quota is on, and the dquot needs to be replayed.
3356 * Remember that we may not have fully recovered the superblock yet,
3357 * so we can't do the usual trick of looking at the SB quota bits.
3358 *
3359 * The other possibility, of course, is that the quota subsystem was
3360 * removed since the last mount - ENOSYS.
3361 */
3369 }
3372 /*
3373 * At this point we are assuming that the dquots have been allocated
3374 * and hence the buffer has valid dquots stamped in it. It should,
3375 * therefore, pass verifier validation. If the dquot is bad, then the
3376 * we'll return an error here, so we don't need to specifically check
3377 * the dquot in the buffer after the verifier has run.
3378 */
3388 /*
3389 * If the dquot has an LSN in it, recover the dquot only if it's less
3390 * than the lsn of the transaction we are replaying.
3391 */
3398 }
3399 }
3404 XFS_DQUOT_CRC_OFF);
3405 }
3412 out_release:
3415 }
3417 /*
3418 * This routine is called to create an in-core extent free intent
3419 * item from the efi format structure which was logged on disk.
3420 * It allocates an in-core efi, copies the extents from the format
3421 * structure into it, and adds the efi to the AIL with the given
3422 * LSN.
3423 */
3424 STATIC int
3428 xfs_lsn_t lsn)
3429 {
3442 }
3446 /*
3447 * The EFI has two references. One for the EFD and one for EFI to ensure
3448 * it makes it into the AIL. Insert the EFI into the AIL directly and
3449 * drop the EFI reference. Note that xfs_trans_ail_update() drops the
3450 * AIL lock.
3451 */
3455 }
3458 /*
3459 * This routine is called when an EFD format structure is found in a committed
3460 * transaction in the log. Its purpose is to cancel the corresponding EFI if it
3461 * was still in the log. To do this it searches the AIL for the EFI with an id
3462 * equal to that in the EFD format structure. If we find it we drop the EFD
3463 * reference, which removes the EFI from the AIL and frees it.
3464 */
3465 STATIC int
3469 {
3484 /*
3485 * Search for the EFI with the id in the EFD format structure in the
3486 * AIL.
3487 */
3494 /*
3495 * Drop the EFD reference to the EFI. This
3496 * removes the EFI from the AIL and frees it.
3497 */
3502 }
3503 }
3505 }
3511 }
3513 /*
3514 * This routine is called to create an in-core extent rmap update
3515 * item from the rui format structure which was logged on disk.
3516 * It allocates an in-core rui, copies the extents from the format
3517 * structure into it, and adds the rui to the AIL with the given
3518 * LSN.
3519 */
3520 STATIC int
3524 xfs_lsn_t lsn)
3525 {
3538 }
3542 /*
3543 * The RUI has two references. One for the RUD and one for RUI to ensure
3544 * it makes it into the AIL. Insert the RUI into the AIL directly and
3545 * drop the RUI reference. Note that xfs_trans_ail_update() drops the
3546 * AIL lock.
3547 */
3551 }
3554 /*
3555 * This routine is called when an RUD format structure is found in a committed
3556 * transaction in the log. Its purpose is to cancel the corresponding RUI if it
3557 * was still in the log. To do this it searches the AIL for the RUI with an id
3558 * equal to that in the RUD format structure. If we find it we drop the RUD
3559 * reference, which removes the RUI from the AIL and frees it.
3560 */
3561 STATIC int
3565 {
3577 /*
3578 * Search for the RUI with the id in the RUD format structure in the
3579 * AIL.
3580 */
3587 /*
3588 * Drop the RUD reference to the RUI. This
3589 * removes the RUI from the AIL and frees it.
3590 */
3595 }
3596 }
3598 }
3604 }
3606 /*
3607 * Copy an CUI format buffer from the given buf, and into the destination
3608 * CUI format structure. The CUI/CUD items were designed not to need any
3609 * special alignment handling.
3610 */
3611 static int
3615 {
3617 uint len;
3625 }
3627 }
3629 /*
3630 * This routine is called to create an in-core extent refcount update
3631 * item from the cui format structure which was logged on disk.
3632 * It allocates an in-core cui, copies the extents from the format
3633 * structure into it, and adds the cui to the AIL with the given
3634 * LSN.
3635 */
3636 STATIC int
3640 xfs_lsn_t lsn)
3641 {
3654 }
3658 /*
3659 * The CUI has two references. One for the CUD and one for CUI to ensure
3660 * it makes it into the AIL. Insert the CUI into the AIL directly and
3661 * drop the CUI reference. Note that xfs_trans_ail_update() drops the
3662 * AIL lock.
3663 */
3667 }
3670 /*
3671 * This routine is called when an CUD format structure is found in a committed
3672 * transaction in the log. Its purpose is to cancel the corresponding CUI if it
3673 * was still in the log. To do this it searches the AIL for the CUI with an id
3674 * equal to that in the CUD format structure. If we find it we drop the CUD
3675 * reference, which removes the CUI from the AIL and frees it.
3676 */
3677 STATIC int
3681 {
3694 /*
3695 * Search for the CUI with the id in the CUD format structure in the
3696 * AIL.
3697 */
3704 /*
3705 * Drop the CUD reference to the CUI. This
3706 * removes the CUI from the AIL and frees it.
3707 */
3712 }
3713 }
3715 }
3721 }
3723 /*
3724 * Copy an BUI format buffer from the given buf, and into the destination
3725 * BUI format structure. The BUI/BUD items were designed not to need any
3726 * special alignment handling.
3727 */
3728 static int
3732 {
3734 uint len;
3742 }
3744 }
3746 /*
3747 * This routine is called to create an in-core extent bmap update
3748 * item from the bui format structure which was logged on disk.
3749 * It allocates an in-core bui, copies the extents from the format
3750 * structure into it, and adds the bui to the AIL with the given
3751 * LSN.
3752 */
3753 STATIC int
3757 xfs_lsn_t lsn)
3758 {
3773 }
3777 /*
3778 * The RUI has two references. One for the RUD and one for RUI to ensure
3779 * it makes it into the AIL. Insert the RUI into the AIL directly and
3780 * drop the RUI reference. Note that xfs_trans_ail_update() drops the
3781 * AIL lock.
3782 */
3786 }
3789 /*
3790 * This routine is called when an BUD format structure is found in a committed
3791 * transaction in the log. Its purpose is to cancel the corresponding BUI if it
3792 * was still in the log. To do this it searches the AIL for the BUI with an id
3793 * equal to that in the BUD format structure. If we find it we drop the BUD
3794 * reference, which removes the BUI from the AIL and frees it.
3795 */
3796 STATIC int
3800 {
3813 /*
3814 * Search for the BUI with the id in the BUD format structure in the
3815 * AIL.
3816 */
3823 /*
3824 * Drop the BUD reference to the BUI. This
3825 * removes the BUI from the AIL and frees it.
3826 */
3831 }
3832 }
3834 }
3840 }
3842 /*
3843 * This routine is called when an inode create format structure is found in a
3844 * committed transaction in the log. It's purpose is to initialise the inodes
3845 * being allocated on disk. This requires us to get inode cluster buffers that
3846 * match the range to be initialised, stamped with inode templates and written
3847 * by delayed write so that subsequent modifications will hit the cached buffer
3848 * and only need writing out at the end of recovery.
3849 */
3850 STATIC int
3855 {
3858 xfs_agnumber_t agno;
3859 xfs_agblock_t agbno;
3862 xfs_agblock_t length;
3873 }
3878 }
3884 }
3889 }
3894 }
3899 }
3904 }
3906 /*
3907 * The inode chunk is either full or sparse and we only support
3908 * m_ialloc_min_blks sized sparse allocations at this time.
3909 */
3915 }
3917 /* verify inode count is consistent with extent length */
3921 __FUNCTION__);
3923 }
3925 /*
3926 * The icreate transaction can cover multiple cluster buffers and these
3927 * buffers could have been freed and reused. Check the individual
3928 * buffers for cancellation so we don't overwrite anything written after
3929 * a cancellation.
3930 */
3935 xfs_daddr_t daddr;
3940 cancel_count++;
3941 }
3943 /*
3944 * We currently only use icreate for a single allocation at a time. This
3945 * means we should expect either all or none of the buffers to be
3946 * cancelled. Be conservative and skip replay if at least one buffer is
3947 * cancelled, but warn the user that something is awry if the buffers
3948 * are not consistent.
3949 *
3950 * XXX: This must be refined to only skip cancelled clusters once we use
3951 * icreate for multiple chunk allocations.
3952 */
3960 }
3965 }
3967 STATIC void
3971 {
3978 }
3982 }
3984 STATIC void
3988 {
4002 }
4009 }
4011 STATIC void
4015 {
4019 uint type;
4047 }
4049 STATIC void
4053 {
4075 }
4076 }
4078 STATIC int
4083 {
4102 /* nothing to do in pass 1 */
4109 }
4110 }
4112 STATIC int
4118 {
4150 /* nothing to do in pass2 */
4157 }
4158 }
4160 STATIC int
4166 {
4175 }
4178 }
4180 /*
4181 * Perform the transaction.
4182 *
4183 * If the transaction modifies a buffer or inode, do it now. Otherwise,
4184 * EFIs and EFDs get queued up by adding entries into the AIL for them.
4185 */
4186 STATIC int
4192 {
4200 #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100
4216 items_queued++;
4222 }
4227 }
4231 }
4233 out:
4239 }
4245 }
4247 STATIC void
4250 {
4256 }
4258 STATIC int
4264 {
4269 /*
4270 * If the transaction is empty, the header was split across this and the
4271 * previous record. Copy the rest of the header.
4272 */
4278 }
4285 }
4287 /* take the tail entry */
4299 }
4301 /*
4302 * The next region to add is the start of a new region. It could be
4303 * a whole region or it could be the first part of a new region. Because
4304 * of this, the assumption here is that the type and size fields of all
4305 * format structures fit into the first 32 bits of the structure.
4306 *
4307 * This works because all regions must be 32 bit aligned. Therefore, we
4308 * either have both fields or we have neither field. In the case we have
4309 * neither field, the data part of the region is zero length. We only have
4310 * a log_op_header and can throw away the header since a new one will appear
4311 * later. If we have at least 4 bytes, then we can determine how many regions
4312 * will appear in the current log item.
4313 */
4314 STATIC int
4320 {
4328 /* we need to catch log corruptions here */
4331 __func__);
4334 }
4340 }
4342 /*
4343 * The transaction header can be arbitrarily split across op
4344 * records. If we don't have the whole thing here, copy what we
4345 * do have and handle the rest in the next record.
4346 */
4351 }
4357 /* take the tail entry */
4361 /* tail item is in use, get a new one */
4365 }
4376 }
4381 KM_SLEEP);
4382 }
4384 /* Description region is ri_buf[0] */
4390 }
4392 /*
4393 * Free up any resources allocated by the transaction
4394 *
4395 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
4396 */
4397 STATIC void
4400 {
4407 /* Free the regions in the item. */
4411 /* Free the item itself */
4414 }
4415 /* Free the transaction recover structure */
4417 }
4419 /*
4420 * On error or completion, trans is freed.
4421 */
4422 STATIC int
4431 {
4435 /* mask off ophdr transaction container flags */
4440 /*
4441 * Callees must not free the trans structure. We'll decide if we need to
4442 * free it or not based on the operation being done and it's result.
4443 */
4445 /* expected flag values */
4455 buffer_list);
4456 /* success or fail, we are now done with this transaction. */
4460 /* unexpected flag values */
4462 /* just skip trans */
4472 }
4476 }
4478 /*
4479 * Lookup the transaction recovery structure associated with the ID in the
4480 * current ophdr. If the transaction doesn't exist and the start flag is set in
4481 * the ophdr, then allocate a new transaction for future ID matches to find.
4482 * Either way, return what we found during the lookup - an existing transaction
4483 * or nothing.
4484 */
4490 {
4492 xlog_tid_t tid;
4500 }
4502 /*
4503 * skip over non-start transaction headers - we could be
4504 * processing slack space before the next transaction starts
4505 */
4511 /*
4512 * This is a new transaction so allocate a new recovery container to
4513 * hold the recovery ops that will follow.
4514 */
4522 /*
4523 * Nothing more to do for this ophdr. Items to be added to this new
4524 * transaction will be in subsequent ophdr containers.
4525 */
4527 }
4529 STATIC int
4539 {
4544 /* Do we understand who wrote this op? */
4551 }
4553 /*
4554 * Check the ophdr contains all the data it is supposed to contain.
4555 */
4561 }
4565 /* nothing to do, so skip over this ophdr */
4567 }
4569 /*
4570 * The recovered buffer queue is drained only once we know that all
4571 * recovery items for the current LSN have been processed. This is
4572 * required because:
4573 *
4574 * - Buffer write submission updates the metadata LSN of the buffer.
4575 * - Log recovery skips items with a metadata LSN >= the current LSN of
4576 * the recovery item.
4577 * - Separate recovery items against the same metadata buffer can share
4578 * a current LSN. I.e., consider that the LSN of a recovery item is
4579 * defined as the starting LSN of the first record in which its
4580 * transaction appears, that a record can hold multiple transactions,
4581 * and/or that a transaction can span multiple records.
4582 *
4583 * In other words, we are allowed to submit a buffer from log recovery
4584 * once per current LSN. Otherwise, we may incorrectly skip recovery
4585 * items and cause corruption.
4586 *
4587 * We don't know up front whether buffers are updated multiple times per
4588 * LSN. Therefore, track the current LSN of each commit log record as it
4589 * is processed and drain the queue when it changes. Use commit records
4590 * because they are ordered correctly by the logging code.
4591 */
4598 }
4602 }
4604 /*
4605 * There are two valid states of the r_state field. 0 indicates that the
4606 * transaction structure is in a normal state. We have either seen the
4607 * start of the transaction or the last operation we added was not a partial
4608 * operation. If the last operation we added to the transaction was a
4609 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
4610 *
4611 * NOTE: skip LRs with 0 data length.
4612 */
4613 STATIC int
4621 {
4630 /* check the log format matches our own - else we can't recover */
4641 /* errors will abort recovery */
4648 num_logops--;
4649 }
4651 }
4653 /* Recover the EFI if necessary. */
4654 STATIC int
4659 {
4663 /*
4664 * Skip EFIs that we've already processed.
4665 */
4675 }
4677 /* Release the EFI since we're cancelling everything. */
4678 STATIC void
4683 {
4691 }
4693 /* Recover the RUI if necessary. */
4694 STATIC int
4699 {
4703 /*
4704 * Skip RUIs that we've already processed.
4705 */
4715 }
4717 /* Release the RUI since we're cancelling everything. */
4718 STATIC void
4723 {
4731 }
4733 /* Recover the CUI if necessary. */
4734 STATIC int
4740 {
4744 /*
4745 * Skip CUIs that we've already processed.
4746 */
4756 }
4758 /* Release the CUI since we're cancelling everything. */
4759 STATIC void
4764 {
4772 }
4774 /* Recover the BUI if necessary. */
4775 STATIC int
4781 {
4785 /*
4786 * Skip BUIs that we've already processed.
4787 */
4797 }
4799 /* Release the BUI since we're cancelling everything. */
4800 STATIC void
4805 {
4813 }
4815 /* Is this log item a deferred action intent? */
4817 {
4826 }
4827 }
4829 /* Take all the collected deferred ops and finish them in order. */
4830 static int
4834 {
4837 uint resblks;
4840 /*
4841 * We're finishing the defer_ops that accumulated as a result of
4842 * recovering unfinished intent items during log recovery. We
4843 * reserve an itruncate transaction because it is the largest
4844 * permanent transaction type. Since we're the only user of the fs
4845 * right now, take 93% (15/16) of the available free blocks. Use
4846 * weird math to avoid a 64-bit division.
4847 */
4864 out_cancel:
4867 }
4869 /*
4870 * When this is called, all of the log intent items which did not have
4871 * corresponding log done items should be in the AIL. What we do now
4872 * is update the data structures associated with each one.
4873 *
4874 * Since we process the log intent items in normal transactions, they
4875 * will be removed at some point after the commit. This prevents us
4876 * from just walking down the list processing each one. We'll use a
4877 * flag in the intent item to skip those that we've already processed
4878 * and use the AIL iteration mechanism's generation count to try to
4879 * speed this up at least a bit.
4880 *
4881 * When we start, we know that the intents are the only things in the
4882 * AIL. As we process them, however, other items are added to the
4883 * AIL.
4884 */
4885 STATIC int
4888 {
4893 xfs_fsblock_t firstfsb;
4895 #if defined(DEBUG) || defined(XFS_WARN)
4896 xfs_lsn_t last_lsn;
4897 #endif
4902 #if defined(DEBUG) || defined(XFS_WARN)
4904 #endif
4907 /*
4908 * We're done when we see something other than an intent.
4909 * There should be no intents left in the AIL now.
4910 */
4912 #ifdef DEBUG
4915 #endif
4917 }
4919 /*
4920 * We should never see a redo item with a LSN higher than
4921 * the last transaction we found in the log at the start
4922 * of recovery.
4923 */
4926 /*
4927 * NOTE: If your intent processing routine can create more
4928 * deferred ops, you /must/ attach them to the dfops in this
4929 * routine or else those subsequent intents will get
4930 * replayed in the wrong order!
4931 */
4947 }
4951 }
4952 out:
4957 else
4961 }
4963 /*
4964 * A cancel occurs when the mount has failed and we're bailing out.
4965 * Release all pending log intent items so they don't pin the AIL.
4966 */
4967 STATIC int
4970 {
4980 /*
4981 * We're done when we see something other than an intent.
4982 * There should be no intents left in the AIL now.
4983 */
4985 #ifdef DEBUG
4988 #endif
4990 }
5005 }
5008 }
5013 }
5015 /*
5016 * This routine performs a transaction to null out a bad inode pointer
5017 * in an agi unlinked inode hash bucket.
5018 */
5019 STATIC void
5022 xfs_agnumber_t agno,
5024 {
5051 out_abort:
5053 out_error:
5056 }
5058 STATIC xfs_agino_t
5061 xfs_agnumber_t agno,
5062 xfs_agino_t agino,
5064 {
5068 xfs_ino_t ino;
5076 /*
5077 * Get the on disk inode to find the next inode in the bucket.
5078 */
5087 /* setup for the next pass */
5091 /*
5092 * Prevent any DMAPI event from being sent when the reference on
5093 * the inode is dropped.
5094 */
5100 fail_iput:
5102 fail:
5103 /*
5104 * We can't read in the inode this bucket points to, or this inode
5105 * is messed up. Just ditch this bucket of inodes. We will lose
5106 * some inodes and space, but at least we won't hang.
5107 *
5108 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
5109 * clear the inode pointer in the bucket.
5110 */
5113 }
5115 /*
5116 * xlog_iunlink_recover
5117 *
5118 * This is called during recovery to process any inodes which
5119 * we unlinked but not freed when the system crashed. These
5120 * inodes will be on the lists in the AGI blocks. What we do
5121 * here is scan all the AGIs and fully truncate and free any
5122 * inodes found on the lists. Each inode is removed from the
5123 * lists when it has been fully truncated and is freed. The
5124 * freeing of the inode and its removal from the list must be
5125 * atomic.
5126 */
5127 STATIC void
5130 {
5132 xfs_agnumber_t agno;
5135 xfs_agino_t agino;
5142 /*
5143 * Find the agi for this ag.
5144 */
5147 /*
5148 * AGI is b0rked. Don't process it.
5149 *
5150 * We should probably mark the filesystem as corrupt
5151 * after we've recovered all the ag's we can....
5152 */
5154 }
5155 /*
5156 * Unlock the buffer so that it can be acquired in the normal
5157 * course of the transaction to truncate and free each inode.
5158 * Because we are not racing with anyone else here for the AGI
5159 * buffer, we don't even need to hold it locked to read the
5160 * initial unlinked bucket entries out of the buffer. We keep
5161 * buffer reference though, so that it stays pinned in memory
5162 * while we need the buffer.
5163 */
5172 }
5173 }
5175 }
5176 }
5178 STATIC int
5183 {
5190 }
5199 }
5200 }
5203 }
5205 /*
5206 * CRC check, unpack and process a log record.
5207 */
5208 STATIC int
5216 {
5219 __le32 crc;
5224 /*
5225 * Nothing else to do if this is a CRC verification pass. Just return
5226 * if this a record with a non-zero crc. Unfortunately, mkfs always
5227 * sets old_crc to 0 so we must consider this valid even on v5 supers.
5228 * Otherwise, return EFSBADCRC on failure so the callers up the stack
5229 * know precisely what failed.
5230 */
5235 }
5237 /*
5238 * We're in the normal recovery path. Issue a warning if and only if the
5239 * CRC in the header is non-zero. This is an advisory warning and the
5240 * zero CRC check prevents warnings from being emitted when upgrading
5241 * the kernel from one that does not add CRCs by default.
5242 */
5250 }
5252 /*
5253 * If the filesystem is CRC enabled, this mismatch becomes a
5254 * fatal log corruption failure.
5255 */
5258 }
5265 buffer_list);
5266 }
5268 STATIC int
5272 xfs_daddr_t blkno)
5273 {
5280 }
5287 }
5289 /* LR body must have data or it wouldn't have been written */
5295 }
5300 }
5302 }
5304 /*
5305 * Read the log from tail to head and process the log records found.
5306 * Handle the two cases where the tail and head are in the same cycle
5307 * and where the active portion of the log wraps around the end of
5308 * the physical log separately. The pass parameter is passed through
5309 * to the routines called to process the data and is not looked at
5310 * here.
5311 */
5312 STATIC int
5315 xfs_daddr_t head_blk,
5316 xfs_daddr_t tail_blk,
5319 {
5322 xfs_daddr_t rhead_blk;
5339 /*
5340 * Read the header of the tail block and get the iclog buffer size from
5341 * h_size. Use this to tell how many sectors make up the log header.
5342 */
5344 /*
5345 * When using variable length iclogs, read first sector of
5346 * iclog header and extract the header size from it. Get a
5347 * new hbp that is the correct size.
5348 */
5362 /*
5363 * xfsprogs has a bug where record length is based on lsunit but
5364 * h_size (iclog size) is hardcoded to 32k. Now that we
5365 * unconditionally CRC verify the unmount record, this means the
5366 * log buffer can be too small for the record and cause an
5367 * overrun.
5368 *
5369 * Detect this condition here. Use lsunit for the buffer size as
5370 * long as this looks like the mkfs case. Otherwise, return an
5371 * error to avoid a buffer overrun.
5372 */
5384 }
5390 hblks++;
5395 }
5401 }
5409 }
5413 /*
5414 * Perform recovery around the end of the physical log.
5415 * When the head is not on the same cycle number as the tail,
5416 * we can't do a sequential recovery.
5417 */
5419 /*
5420 * Check for header wrapping around physical end-of-log
5421 */
5426 /* Read header in one read */
5432 /* This LR is split across physical log end */
5434 /* some data before physical log end */
5443 }
5445 /*
5446 * Note: this black magic still works with
5447 * large sector sizes (non-512) only because:
5448 * - we increased the buffer size originally
5449 * by 1 sector giving us enough extra space
5450 * for the second read;
5451 * - the log start is guaranteed to be sector
5452 * aligned;
5453 * - we read the log end (LR header start)
5454 * _first_, then the log start (LR header end)
5455 * - order is important.
5456 */
5463 }
5473 /*
5474 * Read the log record data in multiple reads if it
5475 * wraps around the end of the log. Note that if the
5476 * header already wrapped, blk_no could point past the
5477 * end of the log. The record data is contiguous in
5478 * that case.
5479 */
5488 /* This log record is split across the
5489 * physical end of log */
5493 /* some data is before the physical
5494 * end of log */
5497 split_bblks =
5505 }
5507 /*
5508 * Note: this black magic still works with
5509 * large sector sizes (non-512) only because:
5510 * - we increased the buffer size originally
5511 * by 1 sector giving us enough extra space
5512 * for the second read;
5513 * - the log start is guaranteed to be sector
5514 * aligned;
5515 * - we read the log end (LR header start)
5516 * _first_, then the log start (LR header end)
5517 * - order is important.
5518 */
5524 }
5533 }
5538 }
5540 /* read first part of physical log */
5551 /* blocks in data section */
5565 }
5567 bread_err2:
5569 bread_err1:
5572 /*
5573 * Submit buffers that have been added from the last record processed,
5574 * regardless of error status.
5575 */
5582 /*
5583 * Transactions are freed at commit time but transactions without commit
5584 * records on disk are never committed. Free any that may be left in the
5585 * hash table.
5586 */
5593 }
5596 }
5598 /*
5599 * Do the recovery of the log. We actually do this in two phases.
5600 * The two passes are necessary in order to implement the function
5601 * of cancelling a record written into the log. The first pass
5602 * determines those things which have been cancelled, and the
5603 * second pass replays log items normally except for those which
5604 * have been cancelled. The handling of the replay and cancellations
5605 * takes place in the log item type specific routines.
5606 *
5607 * The table of items which have cancel records in the log is allocated
5608 * and freed at this level, since only here do we know when all of
5609 * the log recovery has been completed.
5610 */
5611 STATIC int
5614 xfs_daddr_t head_blk,
5615 xfs_daddr_t tail_blk)
5616 {
5621 /*
5622 * First do a pass to find all of the cancelled buf log items.
5623 * Store them in the buf_cancel_table for use in the second pass.
5624 */
5627 KM_SLEEP);
5637 }
5638 /*
5639 * Then do a second pass to actually recover the items in the log.
5640 * When it is complete free the table of buf cancel items.
5641 */
5644 #ifdef DEBUG
5650 }
5657 }
5659 /*
5660 * Do the actual recovery
5661 */
5662 STATIC int
5665 xfs_daddr_t head_blk,
5666 xfs_daddr_t tail_blk)
5667 {
5675 /*
5676 * First replay the images in the log.
5677 */
5682 /*
5683 * If IO errors happened during recovery, bail out.
5684 */
5687 }
5689 /*
5690 * We now update the tail_lsn since much of the recovery has completed
5691 * and there may be space available to use. If there were no extent
5692 * or iunlinks, we can free up the entire log and set the tail_lsn to
5693 * be the last_sync_lsn. This was set in xlog_find_tail to be the
5694 * lsn of the last known good LR on disk. If there are extent frees
5695 * or iunlinks they will have some entries in the AIL; so we look at
5696 * the AIL to determine how to set the tail_lsn.
5697 */
5700 /*
5701 * Now that we've finished replaying all buffer and inode
5702 * updates, re-read in the superblock and reverify it.
5703 */
5715 }
5718 }
5720 /* Convert superblock from on-disk format */
5725 /* re-initialise in-core superblock and geometry structures */
5731 }
5736 /* Normal transactions can now occur */
5739 }
5741 /*
5742 * Perform recovery and re-initialize some log variables in xlog_find_tail.
5743 *
5744 * Return error or zero.
5745 */
5746 int
5749 {
5753 /* find the tail of the log */
5758 /*
5759 * The superblock was read before the log was available and thus the LSN
5760 * could not be verified. Check the superblock LSN against the current
5761 * LSN now that it's known.
5762 */
5768 /* There used to be a comment here:
5769 *
5770 * disallow recovery on read-only mounts. note -- mount
5771 * checks for ENOSPC and turns it into an intelligent
5772 * error message.
5773 * ...but this is no longer true. Now, unless you specify
5774 * NORECOVERY (in which case this function would never be
5775 * called), we just go ahead and recover. We do this all
5776 * under the vfs layer, so we can get away with it unless
5777 * the device itself is read-only, in which case we fail.
5778 */
5781 }
5783 /*
5784 * Version 5 superblock log feature mask validation. We know the
5785 * log is dirty so check if there are any unknown log features
5786 * in what we need to recover. If there are unknown features
5787 * (e.g. unsupported transactions, then simply reject the
5788 * attempt at recovery before touching anything.
5789 */
5792 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
5796 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
5802 }
5804 /*
5805 * Delay log recovery if the debug hook is set. This is debug
5806 * instrumention to coordinate simulation of I/O failures with
5807 * log recovery.
5808 */
5814 }
5822 }
5824 }
5826 /*
5827 * In the first part of recovery we replay inodes and buffers and build
5828 * up the list of extent free items which need to be processed. Here
5829 * we process the extent free items and clean up the on disk unlinked
5830 * inode lists. This is separated from the first part of recovery so
5831 * that the root and real-time bitmap inodes can be read in from disk in
5832 * between the two stages. This is necessary so that we can free space
5833 * in the real-time portion of the file system.
5834 */
5835 int
5838 {
5839 /*
5840 * Now we're ready to do the transactions needed for the
5841 * rest of recovery. Start with completing all the extent
5842 * free intent records and then process the unlinked inode
5843 * lists. At this point, we essentially run in normal mode
5844 * except that we're still performing recovery actions
5845 * rather than accepting new requests.
5846 */
5853 }
5855 /*
5856 * Sync the log to get all the intents out of the AIL.
5857 * This isn't absolutely necessary, but it helps in
5858 * case the unlink transactions would have problems
5859 * pushing the intents out of the way.
5860 */
5873 }
5875 }
5877 int
5880 {
5887 }
5889 #if defined(DEBUG)
5890 /*
5891 * Read all of the agf and agi counters and check that they
5892 * are consistent with the superblock counters.
5893 */
5894 STATIC void
5897 {
5902 xfs_agnumber_t agno;
5923 }
5935 }
5936 }
5937 }