| blob | 34c6872b5f7b8c972bc12549722da40c9d6c6eb9 |
1 /*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18 #include <linux/log2.h>
54 /*
55 * Used in xfs_itruncate_extents(). This is the maximum number of extents
56 * freed from a file in a single transaction.
57 */
58 #define XFS_ITRUNC_MAX_EXTENTS 2
65 #ifdef DEBUG
66 /*
67 * Make sure that the extents in the given memory buffer
68 * are valid.
69 */
70 STATIC void
74 xfs_exntfmt_t fmt)
75 {
76 xfs_bmbt_irec_t irec;
77 xfs_bmbt_rec_host_t rec;
87 }
88 }
90 #define xfs_validate_extents(ifp, nrecs, fmt)
93 /*
94 * Check that none of the inode's in the buffer have a next
95 * unlinked field of 0.
96 */
97 #if defined(DEBUG)
98 void
102 {
115 bp);
117 }
118 }
119 }
120 #endif
122 /*
123 * Find the buffer associated with the given inode map
124 * We do basic validation checks on the buffer once it has been
125 * retrieved from disk.
126 */
127 STATIC int
133 uint buf_flags,
134 uint iget_flags)
135 {
150 }
152 }
154 /*
155 * Validate the magic number and version of every inode in the buffer
156 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
157 */
158 #ifdef DEBUG
162 #endif
173 XFS_ERRTAG_ITOBP_INOTOBP,
174 XFS_RANDOM_ITOBP_INOTOBP))) {
178 }
181 #ifdef DEBUG
187 #endif
190 }
191 }
195 /*
196 * Mark the buffer as an inode buffer now that it looks good
197 */
202 }
204 /*
205 * This routine is called to map an inode number within a file
206 * system to the buffer containing the on-disk version of the
207 * inode. It returns a pointer to the buffer containing the
208 * on-disk inode in the bpp parameter, and in the dip parameter
209 * it returns a pointer to the on-disk inode within that buffer.
210 *
211 * If a non-zero error is returned, then the contents of bpp and
212 * dipp are undefined.
213 *
214 * Use xfs_imap() to determine the size and location of the
215 * buffer to read from disk.
216 */
217 int
221 xfs_ino_t ino,
225 uint imap_flags)
226 {
244 }
247 /*
248 * This routine is called to map an inode to the buffer containing
249 * the on-disk version of the inode. It returns a pointer to the
250 * buffer containing the on-disk inode in the bpp parameter, and in
251 * the dip parameter it returns a pointer to the on-disk inode within
252 * that buffer.
253 *
254 * If a non-zero error is returned, then the contents of bpp and
255 * dipp are undefined.
256 *
257 * The inode is expected to already been mapped to its buffer and read
258 * in once, thus we can use the mapping information stored in the inode
259 * rather than calling xfs_imap(). This allows us to avoid the overhead
260 * of looking at the inode btree for small block file systems
261 * (see xfs_imap()).
262 */
263 int
270 uint buf_flags)
271 {
286 }
291 }
293 /*
294 * Move inode type and inode format specific information from the
295 * on-disk inode to the in-core inode. For fifos, devs, and sockets
296 * this means set if_rdev to the proper value. For files, directories,
297 * and symlinks this means to bring in the in-line data or extent
298 * pointers. For a file in B-tree format, only the root is immediately
299 * brought in-core. The rest will be in-lined in if_extents when it
300 * is first referenced (see xfs_iread_extents()).
301 */
302 STATIC int
306 {
310 xfs_fsize_t di_size;
328 }
337 }
347 }
358 }
369 /*
370 * no local regular files yet
371 */
377 XFS_ERRLEVEL_LOW,
380 }
389 XFS_ERRLEVEL_LOW,
392 }
407 }
413 }
416 }
434 XFS_ERRLEVEL_LOW,
437 }
450 }
455 }
457 }
459 /*
460 * The file is in-lined in the on-disk inode.
461 * If it fits into if_inline_data, then copy
462 * it there, otherwise allocate a buffer for it
463 * and copy the data there. Either way, set
464 * if_data to point at the data.
465 * If we allocate a buffer for the data, make
466 * sure that its size is a multiple of 4 and
467 * record the real size in i_real_bytes.
468 */
469 STATIC int
475 {
479 /*
480 * If the size is unreasonable, then something
481 * is wrong and we just bail out rather than crash in
482 * kmem_alloc() or memcpy() below.
483 */
492 }
502 }
510 }
512 /*
513 * The file consists of a set of extents all
514 * of which fit into the on-disk inode.
515 * If there are few enough extents to fit into
516 * the if_inline_ext, then copy them there.
517 * Otherwise allocate a buffer for them and copy
518 * them into it. Either way, set if_extents
519 * to point at the extents.
520 */
521 STATIC int
526 {
537 /*
538 * If the number of extents is unreasonable, then something
539 * is wrong and we just bail out rather than crash in
540 * kmem_alloc() or memcpy() below.
541 */
548 }
555 else
566 }
573 XFS_ERRLEVEL_LOW,
576 }
577 }
580 }
582 /*
583 * The file has too many extents to fit into
584 * the inode, so they are in B-tree format.
585 * Allocate a buffer for the root of the B-tree
586 * and copy the root into it. The i_extents
587 * field will remain NULL until all of the
588 * extents are read in (when they are needed).
589 */
590 STATIC int
595 {
598 /* REFERENCED */
607 /*
608 * blow out if -- fork has less extents than can fit in
609 * fork (fork shouldn't be a btree format), root btree
610 * block has more records than can fit into the fork,
611 * or the number of extents is greater than the number of
612 * blocks.
613 */
623 }
628 /*
629 * Copy and convert from the on-disk structure
630 * to the in-memory structure.
631 */
639 }
641 STATIC void
645 {
675 }
677 void
681 {
711 }
713 STATIC uint
715 __uint16_t di_flags)
716 {
748 }
751 }
753 uint
756 {
761 }
763 uint
766 {
769 }
771 /*
772 * Read the disk inode attributes into the in-core inode structure.
773 */
774 int
779 uint iget_flags)
780 {
785 /*
786 * Fill in the location information in the in-core inode.
787 */
792 /*
793 * Get pointers to the on-disk inode and the buffer containing it.
794 */
801 /*
802 * If we got something that isn't an inode it means someone
803 * (nfs or dmi) has a stale handle.
804 */
806 #ifdef DEBUG
813 }
815 /*
816 * If the on-disk inode is already linked to a directory
817 * entry, copy all of the inode into the in-core inode.
818 * xfs_iformat() handles copying in the inode format
819 * specific information.
820 * Otherwise, just get the truly permanent information.
821 */
826 #ifdef DEBUG
831 }
837 /*
838 * Make sure to pull in the mode here as well in
839 * case the inode is released without being used.
840 * This ensures that xfs_inactive() will see that
841 * the inode is already free and not try to mess
842 * with the uninitialized part of it.
843 */
845 /*
846 * Initialize the per-fork minima and maxima for a new
847 * inode here. xfs_iformat will do it for old inodes.
848 */
851 }
853 /*
854 * The inode format changed when we moved the link count and
855 * made it 32 bits long. If this is an old format inode,
856 * convert it in memory to look like a new one. If it gets
857 * flushed to disk we will convert back before flushing or
858 * logging it. We zero out the new projid field and the old link
859 * count field. We'll handle clearing the pad field (the remains
860 * of the old uuid field) when we actually convert the inode to
861 * the new format. We don't change the version number so that we
862 * can distinguish this from a real new format inode.
863 */
868 }
873 /*
874 * Mark the buffer containing the inode as something to keep
875 * around for a while. This helps to keep recently accessed
876 * meta-data in-core longer.
877 */
880 /*
881 * Use xfs_trans_brelse() to release the buffer containing the
882 * on-disk inode, because it was acquired with xfs_trans_read_buf()
883 * in xfs_itobp() above. If tp is NULL, this is just a normal
884 * brelse(). If we're within a transaction, then xfs_trans_brelse()
885 * will only release the buffer if it is not dirty within the
886 * transaction. It will be OK to release the buffer in this case,
887 * because inodes on disk are never destroyed and we will be
888 * locking the new in-core inode before putting it in the hash
889 * table where other processes can find it. Thus we don't have
890 * to worry about the inode being changed just because we released
891 * the buffer.
892 */
893 out_brelse:
896 }
898 /*
899 * Read in extents from a btree-format inode.
900 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
901 */
902 int
907 {
910 xfs_extnum_t nextents;
916 }
920 /*
921 * We know that the size is valid (it's checked in iformat_btree)
922 */
931 }
934 }
936 /*
937 * Allocate an inode on disk and return a copy of its in-core version.
938 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
939 * appropriately within the inode. The uid and gid for the inode are
940 * set according to the contents of the given cred structure.
941 *
942 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
943 * has a free inode available, call xfs_iget()
944 * to obtain the in-core version of the allocated inode. Finally,
945 * fill in the inode and log its initial contents. In this case,
946 * ialloc_context would be set to NULL and call_again set to false.
947 *
948 * If xfs_dialloc() does not have an available inode,
949 * it will replenish its supply by doing an allocation. Since we can
950 * only do one allocation within a transaction without deadlocks, we
951 * must commit the current transaction before returning the inode itself.
952 * In this case, therefore, we will set call_again to true and return.
953 * The caller should then commit the current transaction, start a new
954 * transaction, and call xfs_ialloc() again to actually get the inode.
955 *
956 * To ensure that some other process does not grab the inode that
957 * was allocated during the first call to xfs_ialloc(), this routine
958 * also returns the [locked] bp pointing to the head of the freelist
959 * as ialloc_context. The caller should hold this buffer across
960 * the commit and pass it back into this routine on the second call.
961 *
962 * If we are allocating quota inodes, we do not have a parent inode
963 * to attach to or associate with (i.e. pip == NULL) because they
964 * are not linked into the directory structure - they are attached
965 * directly to the superblock - and so have no parent.
966 */
967 int
971 mode_t mode,
972 xfs_nlink_t nlink,
973 xfs_dev_t rdev,
974 prid_t prid,
979 {
980 xfs_ino_t ino;
982 uint flags;
984 timespec_t tv;
987 /*
988 * Call the space management code to pick
989 * the on-disk inode to be allocated.
990 */
998 }
1001 /*
1002 * Get the in-core inode with the lock held exclusively.
1003 * This is because we're setting fields here we need
1004 * to prevent others from looking at until we're done.
1005 */
1021 /*
1022 * If the superblock version is up to where we support new format
1023 * inodes and this is currently an old format inode, then change
1024 * the inode version number now. This way we only do the conversion
1025 * here rather than here and in the flush/logging code.
1026 */
1030 /*
1031 * We've already zeroed the old link count, the projid field,
1032 * and the pad field.
1033 */
1034 }
1036 /*
1037 * Project ids won't be stored on disk if we are using a version 1 inode.
1038 */
1046 }
1047 }
1049 /*
1050 * If the group ID of the new file does not match the effective group
1051 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1052 * (and only if the irix_sgid_inherit compatibility variable is set).
1053 */
1058 }
1071 /*
1072 * di_gen will have been taken care of in xfs_iread.
1073 */
1090 /*
1091 * we can't set up filestreams until after the VFS inode
1092 * is set up properly.
1093 */
1096 /* fall through */
1107 }
1114 }
1115 }
1117 xfs_inherit_noatime)
1120 xfs_inherit_nodump)
1123 xfs_inherit_sync)
1126 xfs_inherit_nosymlinks)
1131 xfs_inherit_nodefrag)
1136 }
1137 /* FALLTHROUGH */
1146 }
1147 /*
1148 * Attribute fork settings for new inode.
1149 */
1153 /*
1154 * Log the new values stuffed into the inode.
1155 */
1159 /* now that we have an i_mode we can setup inode ops and unlock */
1162 /* now we have set up the vfs inode we can associate the filestream */
1169 }
1173 }
1175 /*
1176 * Check to make sure that there are no blocks allocated to the
1177 * file beyond the size of the file. We don't check this for
1178 * files with fixed size extents or real time extents, but we
1179 * at least do it for regular files.
1180 */
1181 #ifdef DEBUG
1182 STATIC void
1185 xfs_fsize_t isize)
1186 {
1188 xfs_fileoff_t map_first;
1203 /*
1204 * The filesystem could be shutting down, so bmapi may return
1205 * an error.
1206 */
1210 map_first),
1212 NULL))
1216 }
1218 #define xfs_isize_check(ip, isize)
1221 /*
1222 * Free up the underlying blocks past new_size. The new size must be smaller
1223 * than the current size. This routine can be used both for the attribute and
1224 * data fork, and does not modify the inode size, which is left to the caller.
1225 *
1226 * The transaction passed to this routine must have made a permanent log
1227 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1228 * given transaction and start new ones, so make sure everything involved in
1229 * the transaction is tidy before calling here. Some transaction will be
1230 * returned to the caller to be committed. The incoming transaction must
1231 * already include the inode, and both inode locks must be held exclusively.
1232 * The inode must also be "held" within the transaction. On return the inode
1233 * will be "held" within the returned transaction. This routine does NOT
1234 * require any disk space to be reserved for it within the transaction.
1235 *
1236 * If we get an error, we must return with the inode locked and linked into the
1237 * current transaction. This keeps things simple for the higher level code,
1238 * because it always knows that the inode is locked and held in the transaction
1239 * that returns to it whether errors occur or not. We don't mark the inode
1240 * dirty on error so that transactions can be easily aborted if possible.
1241 */
1242 int
1247 xfs_fsize_t new_size)
1248 {
1252 xfs_bmap_free_t free_list;
1253 xfs_fsblock_t first_block;
1254 xfs_fileoff_t first_unmap_block;
1255 xfs_fileoff_t last_block;
1256 xfs_filblks_t unmap_len;
1268 /*
1269 * Since it is possible for space to become allocated beyond
1270 * the end of the file (in a crash where the space is allocated
1271 * but the inode size is not yet updated), simply remove any
1272 * blocks which show up between the new EOF and the maximum
1273 * possible file size. If the first block to be removed is
1274 * beyond the maximum file size (ie it is the same as last_block),
1275 * then there is nothing to do.
1276 */
1289 XFS_ITRUNC_MAX_EXTENTS,
1295 /*
1296 * Duplicate the transaction that has the permanent
1297 * reservation and commit the old transaction.
1298 */
1306 /*
1307 * Mark the inode dirty so it will be logged and
1308 * moved forward in the log as part of every commit.
1309 */
1311 }
1322 /*
1323 * Transaction commit worked ok so we can drop the extra ticket
1324 * reference that we gained in xfs_trans_dup()
1325 */
1329 XFS_TRANS_PERM_LOG_RES,
1330 XFS_ITRUNCATE_LOG_COUNT);
1333 }
1335 out:
1338 out_bmap_cancel:
1339 /*
1340 * If the bunmapi call encounters an error, return to the caller where
1341 * the transaction can be properly aborted. We just need to make sure
1342 * we're not holding any resources that we were not when we came in.
1343 */
1346 }
1348 int
1352 xfs_fsize_t new_size)
1353 {
1358 /*
1359 * The first thing we do is set the size to new_size permanently on
1360 * disk. This way we don't have to worry about anyone ever being able
1361 * to look at the data being freed even in the face of a crash.
1362 * What we're getting around here is the case where we free a block, it
1363 * is allocated to another file, it is written to, and then we crash.
1364 * If the new data gets written to the file but the log buffers
1365 * containing the free and reallocation don't, then we'd end up with
1366 * garbage in the blocks being freed. As long as we make the new_size
1367 * permanent before actually freeing any blocks it doesn't matter if
1368 * they get written to.
1369 */
1371 /*
1372 * If we are not changing the file size then do not update
1373 * the on-disk file size - we may be called from
1374 * xfs_inactive_free_eofblocks(). If we update the on-disk
1375 * file size and then the system crashes before the contents
1376 * of the file are flushed to disk then the files may be
1377 * full of holes (ie NULL files bug).
1378 */
1383 }
1384 }
1390 /*
1391 * If we are not changing the file size then do not update the on-disk
1392 * file size - we may be called from xfs_inactive_free_eofblocks().
1393 * If we update the on-disk file size and then the system crashes
1394 * before the contents of the file are flushed to disk then the files
1395 * may be full of holes (ie NULL files bug).
1396 */
1401 }
1406 /*
1407 * Always re-log the inode so that our permanent transaction can keep
1408 * on rolling it forward in the log.
1409 */
1414 }
1416 /*
1417 * This is called when the inode's link count goes to 0.
1418 * We place the on-disk inode on a list in the AGI. It
1419 * will be pulled from this list when the inode is freed.
1420 */
1421 int
1425 {
1431 xfs_agino_t agino;
1441 /*
1442 * Get the agi buffer first. It ensures lock ordering
1443 * on the list.
1444 */
1450 /*
1451 * Get the index into the agi hash table for the
1452 * list this inode will go on.
1453 */
1461 /*
1462 * There is already another inode in the bucket we need
1463 * to add ourselves to. Add us at the front of the list.
1464 * Here we put the head pointer into our next pointer,
1465 * and then we fall through to point the head at us.
1466 */
1479 }
1481 /*
1482 * Point the bucket head pointer at the inode being inserted.
1483 */
1491 }
1493 /*
1494 * Pull the on-disk inode from the AGI unlinked list.
1495 */
1496 STATIC int
1500 {
1501 xfs_ino_t next_ino;
1507 xfs_agnumber_t agno;
1508 xfs_agino_t agino;
1509 xfs_agino_t next_agino;
1519 /*
1520 * Get the agi buffer first. It ensures lock ordering
1521 * on the list.
1522 */
1529 /*
1530 * Get the index into the agi hash table for the
1531 * list this inode will go on.
1532 */
1540 /*
1541 * We're at the head of the list. Get the inode's
1542 * on-disk buffer to see if there is anyone after us
1543 * on the list. Only modify our next pointer if it
1544 * is not already NULLAGINO. This saves us the overhead
1545 * of dealing with the buffer when there is no need to
1546 * change it.
1547 */
1553 }
1566 }
1567 /*
1568 * Point the bucket head pointer at the next inode.
1569 */
1578 /*
1579 * We need to search the list for the inode being freed.
1580 */
1584 /*
1585 * If the last inode wasn't the one pointing to
1586 * us, then release its buffer since we're not
1587 * going to do anything with it.
1588 */
1591 }
1600 }
1604 }
1605 /*
1606 * Now last_ibp points to the buffer previous to us on
1607 * the unlinked list. Pull us from the list.
1608 */
1614 }
1628 }
1629 /*
1630 * Point the previous inode on the list to the next inode.
1631 */
1639 }
1641 }
1643 /*
1644 * A big issue when freeing the inode cluster is is that we _cannot_ skip any
1645 * inodes that are in memory - they all must be marked stale and attached to
1646 * the cluster buffer.
1647 */
1648 STATIC void
1652 xfs_ino_t inum)
1653 {
1659 xfs_daddr_t blkno;
1676 }
1682 /*
1683 * We obtain and lock the backing buffer first in the process
1684 * here, as we have to ensure that any dirty inode that we
1685 * can't get the flush lock on is attached to the buffer.
1686 * If we scan the in-memory inodes first, then buffer IO can
1687 * complete before we get a lock on it, and hence we may fail
1688 * to mark all the active inodes on the buffer stale.
1689 */
1692 XBF_LOCK);
1694 /*
1695 * Walk the inodes already attached to the buffer and mark them
1696 * stale. These will all have the flush locks held, so an
1697 * in-memory inode walk can't lock them. By marking them all
1698 * stale first, we will not attempt to lock them in the loop
1699 * below as the XFS_ISTALE flag will be set.
1700 */
1711 }
1713 }
1716 /*
1717 * For each inode in memory attempt to add it to the inode
1718 * buffer and set it up for being staled on buffer IO
1719 * completion. This is safe as we've locked out tail pushing
1720 * and flushing by locking the buffer.
1721 *
1722 * We have already marked every inode that was part of a
1723 * transaction stale above, which means there is no point in
1724 * even trying to lock them.
1725 */
1727 retry:
1732 /* Inode not in memory, nothing to do */
1736 }
1738 /*
1739 * because this is an RCU protected lookup, we could
1740 * find a recently freed or even reallocated inode
1741 * during the lookup. We need to check under the
1742 * i_flags_lock for a valid inode here. Skip it if it
1743 * is not valid, the wrong inode or stale.
1744 */
1751 }
1754 /*
1755 * Don't try to lock/unlock the current inode, but we
1756 * _cannot_ skip the other inodes that we did not find
1757 * in the list attached to the buffer and are not
1758 * already marked stale. If we can't lock it, back off
1759 * and retry.
1760 */
1766 }
1772 /*
1773 * we don't need to attach clean inodes or those only
1774 * with unlogged changes (which we throw away, anyway).
1775 */
1783 }
1796 }
1800 }
1803 }
1805 /*
1806 * This is called to return an inode to the inode free list.
1807 * The inode should already be truncated to 0 length and have
1808 * no pages associated with it. This routine also assumes that
1809 * the inode is already a part of the transaction.
1810 *
1811 * The on-disk copy of the inode will have been added to the list
1812 * of unlinked inodes in the AGI. We need to remove the inode from
1813 * that list atomically with respect to freeing it here.
1814 */
1815 int
1820 {
1823 xfs_ino_t first_ino;
1835 /*
1836 * Pull the on-disk inode from the AGI unlinked list.
1837 */
1841 }
1846 }
1855 /*
1856 * Bump the generation count so no one will be confused
1857 * by reincarnations of this inode.
1858 */
1867 /*
1868 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
1869 * from picking up this inode when it is reclaimed (its incore state
1870 * initialzed but not flushed to disk yet). The in-core di_mode is
1871 * already cleared and a corresponding transaction logged.
1872 * The hack here just synchronizes the in-core to on-disk
1873 * di_mode value in advance before the actual inode sync to disk.
1874 * This is OK because the inode is already unlinked and would never
1875 * change its di_mode again for this inode generation.
1876 * This is a temporary hack that would require a proper fix
1877 * in the future.
1878 */
1883 }
1886 }
1888 /*
1889 * Reallocate the space for if_broot based on the number of records
1890 * being added or deleted as indicated in rec_diff. Move the records
1891 * and pointers in if_broot to fit the new size. When shrinking this
1892 * will eliminate holes between the records and pointers created by
1893 * the caller. When growing this will create holes to be filled in
1894 * by the caller.
1895 *
1896 * The caller must not request to add more records than would fit in
1897 * the on-disk inode root. If the if_broot is currently NULL, then
1898 * if we adding records one will be allocated. The caller must also
1899 * not request that the number of records go below zero, although
1900 * it can go to zero.
1901 *
1902 * ip -- the inode whose if_broot area is changing
1903 * ext_diff -- the change in the number of records, positive or negative,
1904 * requested for the if_broot array.
1905 */
1906 void
1911 {
1921 /*
1922 * Handle the degenerate case quietly.
1923 */
1926 }
1930 /*
1931 * If there wasn't any memory allocated before, just
1932 * allocate it now and get out.
1933 */
1939 }
1941 /*
1942 * If there is already an existing if_broot, then we need
1943 * to realloc() it and shift the pointers to their new
1944 * location. The records don't change location because
1945 * they are kept butted up against the btree block header.
1946 */
1962 }
1964 /*
1965 * rec_diff is less than 0. In this case, we are shrinking the
1966 * if_broot buffer. It must already exist. If we go to zero
1967 * records, just get rid of the root and clear the status bit.
1968 */
1975 else
1979 /*
1980 * First copy over the btree block header.
1981 */
1986 }
1988 /*
1989 * Only copy the records and pointers if there are any.
1990 */
1992 /*
1993 * First copy the records.
1994 */
1999 /*
2000 * Then copy the pointers.
2001 */
2007 }
2014 }
2017 /*
2018 * This is called when the amount of space needed for if_data
2019 * is increased or decreased. The change in size is indicated by
2020 * the number of bytes that need to be added or deleted in the
2021 * byte_diff parameter.
2022 *
2023 * If the amount of space needed has decreased below the size of the
2024 * inline buffer, then switch to using the inline buffer. Otherwise,
2025 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2026 * to what is needed.
2027 *
2028 * ip -- the inode whose if_data area is changing
2029 * byte_diff -- the change in the number of bytes, positive or negative,
2030 * requested for the if_data array.
2031 */
2032 void
2037 {
2044 }
2053 }
2057 /*
2058 * If the valid extents/data can fit in if_inline_ext/data,
2059 * copy them from the malloc'd vector and free it.
2060 */
2066 new_size);
2069 }
2072 /*
2073 * Stuck with malloc/realloc.
2074 * For inline data, the underlying buffer must be
2075 * a multiple of 4 bytes in size so that it can be
2076 * logged and stay on word boundaries. We enforce
2077 * that here.
2078 */
2085 /*
2086 * Only do the realloc if the underlying size
2087 * is really changing.
2088 */
2092 real_size,
2095 }
2102 }
2103 }
2107 }
2109 void
2113 {
2120 }
2122 /*
2123 * If the format is local, then we can't have an extents
2124 * array so just look for an inline data array. If we're
2125 * not local then we may or may not have an extents list,
2126 * so check and free it up if we do.
2127 */
2135 }
2142 }
2149 }
2150 }
2152 /*
2153 * This is called to unpin an inode. The caller must have the inode locked
2154 * in at least shared mode so that the buffer cannot be subsequently pinned
2155 * once someone is waiting for it to be unpinned.
2156 */
2157 static void
2160 {
2165 /* Give the log a push to start the unpinning I/O */
2168 }
2170 void
2173 {
2177 }
2178 }
2180 /*
2181 * xfs_iextents_copy()
2182 *
2183 * This is called to copy the REAL extents (as opposed to the delayed
2184 * allocation extents) from the inode into the given buffer. It
2185 * returns the number of bytes copied into the buffer.
2186 *
2187 * If there are no delayed allocation extents, then we can just
2188 * memcpy() the extents into the buffer. Otherwise, we need to
2189 * examine each extent in turn and skip those which are delayed.
2190 */
2191 int
2196 {
2201 xfs_fsblock_t start_block;
2211 /*
2212 * There are some delayed allocation extents in the
2213 * inode, so copy the extents one at a time and skip
2214 * the delayed ones. There must be at least one
2215 * non-delayed extent.
2216 */
2222 /*
2223 * It's a delayed allocation extent, so skip it.
2224 */
2226 }
2228 /* Translate to on disk format */
2231 dp++;
2232 copied++;
2233 }
2238 }
2240 /*
2241 * Each of the following cases stores data into the same region
2242 * of the on-disk inode, so only one of them can be valid at
2243 * any given time. While it is possible to have conflicting formats
2244 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2245 * in EXTENTS format, this can only happen when the fork has
2246 * changed formats after being modified but before being flushed.
2247 * In these cases, the format always takes precedence, because the
2248 * format indicates the current state of the fork.
2249 */
2250 /*ARGSUSED*/
2251 STATIC void
2258 {
2262 #ifdef XFS_TRANS_DEBUG
2264 #endif
2275 /*
2276 * This can happen if we gave up in iformat in an error path,
2277 * for the attribute fork.
2278 */
2282 }
2292 }
2303 whichfork);
2304 }
2313 XFS_BROOT_SIZE_ADJ));
2317 }
2324 }
2333 }
2339 }
2340 }
2342 STATIC int
2346 {
2370 /* really need a gang lookup range call here */
2381 /*
2382 * because this is an RCU protected lookup, we could find a
2383 * recently freed or even reallocated inode during the lookup.
2384 * We need to check under the i_flags_lock for a valid inode
2385 * here. Skip it if it is not valid or the wrong inode.
2386 */
2392 }
2395 /*
2396 * Do an un-protected check to see if the inode is dirty and
2397 * is a candidate for flushing. These checks will be repeated
2398 * later after the appropriate locks are acquired.
2399 */
2403 /*
2404 * Try to get locks. If any are unavailable or it is pinned,
2405 * then this inode cannot be flushed and is skipped.
2406 */
2413 }
2418 }
2420 /*
2421 * arriving here means that this inode can be flushed. First
2422 * re-check that it's dirty before flushing.
2423 */
2430 }
2431 clcount++;
2434 }
2436 }
2441 }
2443 out_free:
2446 out_put:
2451 cluster_corrupt_out:
2452 /*
2453 * Corruption detected in the clustering loop. Invalidate the
2454 * inode buffer and shut down the filesystem.
2455 */
2457 /*
2458 * Clean up the buffer. If it was B_DELWRI, just release it --
2459 * brelse can handle it with no problems. If not, shut down the
2460 * filesystem before releasing the buffer.
2461 */
2469 /*
2470 * Just like incore_relse: if we have b_iodone functions,
2471 * mark the buffer as an error and call them. Otherwise
2472 * mark it as stale and brelse.
2473 */
2482 }
2483 }
2485 /*
2486 * Unlocks the flush lock
2487 */
2492 }
2494 /*
2495 * xfs_iflush() will write a modified inode's changes out to the
2496 * inode's on disk home. The caller must have the inode lock held
2497 * in at least shared mode and the inode flush completion must be
2498 * active as well. The inode lock will still be held upon return from
2499 * the call and the caller is free to unlock it.
2500 * The inode flush will be completed when the inode reaches the disk.
2501 * The flags indicate how the inode's buffer should be written out.
2502 */
2503 int
2506 uint flags)
2507 {
2524 /*
2525 * We can't flush the inode until it is unpinned, so wait for it if we
2526 * are allowed to block. We know no one new can pin it, because we are
2527 * holding the inode lock shared and you need to hold it exclusively to
2528 * pin the inode.
2529 *
2530 * If we are not allowed to block, force the log out asynchronously so
2531 * that when we come back the inode will be unpinned. If other inodes
2532 * in the same cluster are dirty, they will probably write the inode
2533 * out for us if they occur after the log force completes.
2534 */
2539 }
2542 /*
2543 * For stale inodes we cannot rely on the backing buffer remaining
2544 * stale in cache for the remaining life of the stale inode and so
2545 * xfs_itobp() below may give us a buffer that no longer contains
2546 * inodes below. We have to check this after ensuring the inode is
2547 * unpinned so that it is safe to reclaim the stale inode after the
2548 * flush call.
2549 */
2553 }
2555 /*
2556 * This may have been unpinned because the filesystem is shutting
2557 * down forcibly. If that's the case we must not write this inode
2558 * to disk, because the log record didn't make it to disk!
2559 */
2566 }
2568 /*
2569 * Get the buffer containing the on-disk inode.
2570 */
2576 }
2578 /*
2579 * First flush out the inode that xfs_iflush was called with.
2580 */
2585 /*
2586 * If the buffer is pinned then push on the log now so we won't
2587 * get stuck waiting in the write for too long.
2588 */
2592 /*
2593 * inode clustering:
2594 * see if other inodes can be gathered into this write
2595 */
2602 else
2606 corrupt_out:
2609 cluster_corrupt_out:
2610 /*
2611 * Unlocks the flush lock
2612 */
2615 }
2618 STATIC int
2622 {
2626 #ifdef XFS_TRANS_DEBUG
2628 #endif
2638 /* set *dip = inode's place in the buffer */
2641 /*
2642 * Clear i_update_core before copying out the data.
2643 * This is for coordination with our timestamp updates
2644 * that don't hold the inode lock. They will always
2645 * update the timestamps BEFORE setting i_update_core,
2646 * so if we clear i_update_core after they set it we
2647 * are guaranteed to see their updates to the timestamps.
2648 * I believe that this depends on strongly ordered memory
2649 * semantics, but we have that. We use the SYNCHRONIZE
2650 * macro to make sure that the compiler does not reorder
2651 * the i_update_core access below the data copy below.
2652 */
2656 /*
2657 * Make sure to get the latest timestamps from the Linux inode.
2658 */
2667 }
2674 }
2684 }
2695 }
2696 }
2699 XFS_RANDOM_IFLUSH_5)) {
2701 "%s: detected corrupt incore inode %Lu, "
2707 }
2714 }
2715 /*
2716 * bump the flush iteration count, used to detect flushes which
2717 * postdate a log record during recovery.
2718 */
2722 /*
2723 * Copy the dirty parts of the inode into the on-disk
2724 * inode. We always copy out the core of the inode,
2725 * because if the inode is dirty at all the core must
2726 * be.
2727 */
2730 /* Wrap, we never let the log put out DI_MAX_FLUSH */
2734 /*
2735 * If this is really an old format inode and the superblock version
2736 * has not been updated to support only new format inodes, then
2737 * convert back to the old inode format. If the superblock version
2738 * has been updated, then make the conversion permanent.
2739 */
2743 /*
2744 * Convert it back.
2745 */
2749 /*
2750 * The superblock version has already been bumped,
2751 * so just make the conversion to the new inode
2752 * format permanent.
2753 */
2762 }
2763 }
2770 /*
2771 * We've recorded everything logged in the inode, so we'd
2772 * like to clear the ilf_fields bits so we don't log and
2773 * flush things unnecessarily. However, we can't stop
2774 * logging all this information until the data we've copied
2775 * into the disk buffer is written to disk. If we did we might
2776 * overwrite the copy of the inode in the log with all the
2777 * data after re-logging only part of it, and in the face of
2778 * a crash we wouldn't have all the data we need to recover.
2779 *
2780 * What we do is move the bits to the ili_last_fields field.
2781 * When logging the inode, these bits are moved back to the
2782 * ilf_fields field. In the xfs_iflush_done() routine we
2783 * clear ili_last_fields, since we know that the information
2784 * those bits represent is permanently on disk. As long as
2785 * the flush completes before the inode is logged again, then
2786 * both ilf_fields and ili_last_fields will be cleared.
2787 *
2788 * We can play with the ilf_fields bits here, because the inode
2789 * lock must be held exclusively in order to set bits there
2790 * and the flush lock protects the ili_last_fields bits.
2791 * Set ili_logged so the flush done
2792 * routine can tell whether or not to look in the AIL.
2793 * Also, store the current LSN of the inode so that we can tell
2794 * whether the item has moved in the AIL from xfs_iflush_done().
2795 * In order to read the lsn we need the AIL lock, because
2796 * it is a 64 bit value that cannot be read atomically.
2797 */
2806 /*
2807 * Attach the function xfs_iflush_done to the inode's
2808 * buffer. This will remove the inode from the AIL
2809 * and unlock the inode's flush lock when the inode is
2810 * completely written to disk.
2811 */
2817 /*
2818 * We're flushing an inode which is not in the AIL and has
2819 * not been logged but has i_update_core set. For this
2820 * case we can use a B_DELWRI flush and immediately drop
2821 * the inode flush lock because we can avoid the whole
2822 * AIL state thing. It's OK to drop the flush lock now,
2823 * because we've already locked the buffer and to do anything
2824 * you really need both.
2825 */
2830 }
2832 }
2836 corrupt_out:
2838 }
2840 /*
2841 * Return a pointer to the extent record at file index idx.
2842 */
2843 xfs_bmbt_rec_host_t *
2847 {
2864 }
2865 }
2867 /*
2868 * Insert new item(s) into the extent records for incore inode
2869 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
2870 */
2871 void
2878 {
2888 }
2890 /*
2891 * This is called when the amount of space required for incore file
2892 * extents needs to be increased. The ext_diff parameter stores the
2893 * number of new extents being added and the idx parameter contains
2894 * the extent index where the new extents will be added. If the new
2895 * extents are being appended, then we just need to (re)allocate and
2896 * initialize the space. Otherwise, if the new extents are being
2897 * inserted into the middle of the existing entries, a bit more work
2898 * is required to make room for the new extents to be inserted. The
2899 * caller is responsible for filling in the new extent entries upon
2900 * return.
2901 */
2902 void
2907 {
2916 /*
2917 * If the new number of extents (nextents + ext_diff)
2918 * fits inside the inode, then continue to use the inline
2919 * extent buffer.
2920 */
2927 }
2930 }
2931 /*
2932 * Otherwise use a linear (direct) extent list.
2933 * If the extents are currently inside the inode,
2934 * xfs_iext_realloc_direct will switch us from
2935 * inline to direct extent allocation mode.
2936 */
2944 }
2945 }
2946 /* Indirection array */
2959 }
2960 /* Extents fit in target extent page */
2968 }
2971 }
2972 /* Insert a new extent page */
2976 }
2977 /*
2978 * If extent(s) are being appended to the last page in
2979 * the indirection array and the new extent(s) don't fit
2980 * in the page, then erp is NULL and erp_idx is set to
2981 * the next index needed in the indirection array.
2982 */
2991 erp_idx++;
2992 }
2993 }
2994 }
2995 }
2997 }
2999 /*
3000 * This is called when incore extents are being added to the indirection
3001 * array and the new extents do not fit in the target extent list. The
3002 * erp_idx parameter contains the irec index for the target extent list
3003 * in the indirection array, and the idx parameter contains the extent
3004 * index within the list. The number of extents being added is stored
3005 * in the count parameter.
3006 *
3007 * |-------| |-------|
3008 * | | | | idx - number of extents before idx
3009 * | idx | | count |
3010 * | | | | count - number of extents being inserted at idx
3011 * |-------| |-------|
3012 * | count | | nex2 | nex2 - number of extents after idx + count
3013 * |-------| |-------|
3014 */
3015 void
3021 {
3035 /*
3036 * Save second part of target extent list
3037 * (all extents past */
3045 }
3047 /*
3048 * Add the new extents to the end of the target
3049 * list, then allocate new irec record(s) and
3050 * extent buffer(s) as needed to store the rest
3051 * of the new extents.
3052 */
3059 }
3061 erp_idx++;
3067 }
3069 /* Add nex2 extents back to indirection array */
3071 xfs_extnum_t ext_avail;
3077 /*
3078 * If nex2 extents fit in the current page, append
3079 * nex2_ep after the new extents.
3080 */
3083 }
3084 /*
3085 * Otherwise, check if space is available in the
3086 * next page.
3087 */
3091 erp_idx++;
3092 erp++;
3093 /* Create a hole for nex2 extents */
3096 }
3097 /*
3098 * Final choice, create a new extent page for
3099 * nex2 extents.
3100 */
3102 erp_idx++;
3104 }
3109 }
3110 }
3112 /*
3113 * This is called when the amount of space required for incore file
3114 * extents needs to be decreased. The ext_diff parameter stores the
3115 * number of extents to be removed and the idx parameter contains
3116 * the extent index where the extents will be removed from.
3117 *
3118 * If the amount of space needed has decreased below the linear
3119 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3120 * extent array. Otherwise, use kmem_realloc() to adjust the
3121 * size to what is needed.
3122 */
3123 void
3129 {
3148 }
3150 }
3152 /*
3153 * This removes ext_diff extents from the inline buffer, beginning
3154 * at extent index idx.
3155 */
3156 void
3161 {
3180 }
3181 }
3183 /*
3184 * This removes ext_diff extents from a linear (direct) extent list,
3185 * beginning at extent index idx. If the extents are being removed
3186 * from the end of the list (ie. truncate) then we just need to re-
3187 * allocate the list to remove the extra space. Otherwise, if the
3188 * extents are being removed from the middle of the existing extent
3189 * entries, then we first need to move the extent records beginning
3190 * at idx + ext_diff up in the list to overwrite the records being
3191 * removed, then remove the extra space via kmem_realloc.
3192 */
3193 void
3198 {
3210 }
3211 /* Move extents up in the list (if needed) */
3217 }
3220 /*
3221 * Reallocate the direct extent list. If the extents
3222 * will fit inside the inode then xfs_iext_realloc_direct
3223 * will switch from direct to inline extent allocation
3224 * mode for us.
3225 */
3228 }
3230 /*
3231 * This is called when incore extents are being removed from the
3232 * indirection array and the extents being removed span multiple extent
3233 * buffers. The idx parameter contains the file extent index where we
3234 * want to begin removing extents, and the count parameter contains
3235 * how many extents need to be removed.
3236 *
3237 * |-------| |-------|
3238 * | nex1 | | | nex1 - number of extents before idx
3239 * |-------| | count |
3240 * | | | | count - number of extents being removed at idx
3241 * | count | |-------|
3242 * | | | nex2 | nex2 - number of extents after idx + count
3243 * |-------| |-------|
3244 */
3245 void
3250 {
3267 /*
3268 * Check for deletion of entire list;
3269 * xfs_iext_irec_remove() updates extent offsets.
3270 */
3277 XFS_IEXT_BUFSZ);
3283 }
3284 }
3285 /* Move extents up (if needed) */
3290 }
3291 /* Zero out rest of page */
3294 /* Update remaining counters */
3299 erp_idx++;
3300 erp++;
3301 }
3304 }
3306 /*
3307 * Create, destroy, or resize a linear (direct) block of extents.
3308 */
3309 void
3313 {
3322 /* Free extent records */
3325 }
3326 /* Resize direct extent list and zero any new bytes */
3328 /* Check if extents will fit inside the inode */
3334 }
3337 }
3341 rnew_size,
3343 }
3348 }
3349 }
3350 /*
3351 * Switch from the inline extent buffer to a direct
3352 * extent list. Be sure to include the inline extent
3353 * bytes in new_size.
3354 */
3359 }
3361 }
3364 }
3366 /*
3367 * Switch from linear (direct) extent records to inline buffer.
3368 */
3369 void
3373 {
3376 /*
3377 * The inline buffer was zeroed when we switched
3378 * from inline to direct extent allocation mode,
3379 * so we don't need to clear it here.
3380 */
3386 }
3388 /*
3389 * Switch from inline buffer to linear (direct) extent records.
3390 * new_size should already be rounded up to the next power of 2
3391 * by the caller (when appropriate), so use new_size as it is.
3392 * However, since new_size may be rounded up, we can't update
3393 * if_bytes here. It is the caller's responsibility to update
3394 * if_bytes upon return.
3395 */
3396 void
3400 {
3408 }
3410 }
3412 /*
3413 * Resize an extent indirection array to new_size bytes.
3414 */
3415 STATIC void
3419 {
3434 }
3435 }
3437 /*
3438 * Switch from indirection array to linear (direct) extent allocations.
3439 */
3440 STATIC void
3443 {
3463 }
3464 }
3466 /*
3467 * Free incore file extents.
3468 */
3469 void
3472 {
3480 }
3487 }
3491 }
3493 /*
3494 * Return a pointer to the extent record for file system block bno.
3495 */
3501 {
3516 }
3519 /* Find target extent list */
3527 }
3528 /* Binary search extent records */
3539 /* Convert back to file-based extent index */
3542 }
3545 }
3546 }
3547 /* Convert back to file-based extent index */
3550 }
3556 }
3557 }
3560 }
3562 /*
3563 * Return a pointer to the indirection array entry containing the
3564 * extent record for filesystem block bno. Store the index of the
3565 * target irec in *erp_idxp.
3566 */
3572 {
3596 }
3597 }
3600 }
3602 /*
3603 * Return a pointer to the indirection array entry containing the
3604 * extent record at file extent index *idxp. Store the index of the
3605 * target irec in *erp_idxp and store the page index of the target
3606 * extent record in *idxp.
3607 */
3608 xfs_ext_irec_t *
3614 {
3633 /* Binary search extent irec's */
3649 erp_idx++;
3655 }
3656 }
3660 }
3662 /*
3663 * Allocate and initialize an indirection array once the space needed
3664 * for incore extents increases above XFS_IEXT_BUFSZ.
3665 */
3666 void
3669 {
3685 }
3696 }
3698 /*
3699 * Allocate and initialize a new entry in the indirection array.
3700 */
3701 xfs_ext_irec_t *
3705 {
3713 /* Resize indirection array */
3716 /*
3717 * Move records down in the array so the
3718 * new page can use erp_idx.
3719 */
3723 }
3726 /* Initialize new extent record */
3735 }
3737 /*
3738 * Remove a record from the indirection array.
3739 */
3740 void
3744 {
3756 }
3757 /* Compact extent records */
3761 }
3762 /*
3763 * Manually free the last extent record from the indirection
3764 * array. A call to xfs_iext_realloc_indirect() with a size
3765 * of zero would result in a call to xfs_iext_destroy() which
3766 * would in turn call this function again, creating a nasty
3767 * infinite loop.
3768 */
3774 }
3776 }
3778 /*
3779 * This is called to clean up large amounts of unused memory allocated
3780 * by the indirection array. Before compacting anything though, verify
3781 * that the indirection array is still needed and switch back to the
3782 * linear extent list (or even the inline buffer) if possible. The
3783 * compaction policy is as follows:
3784 *
3785 * Full Compaction: Extents fit into a single page (or inline buffer)
3786 * Partial Compaction: Extents occupy less than 50% of allocated space
3787 * No Compaction: Extents occupy at least 50% of allocated space
3788 */
3789 void
3792 {
3809 }
3810 }
3812 /*
3813 * Combine extents from neighboring extent pages.
3814 */
3815 void
3818 {
3834 /*
3835 * Free page before removing extent record
3836 * so er_extoffs don't get modified in
3837 * xfs_iext_irec_remove.
3838 */
3844 erp_idx++;
3845 }
3846 }
3847 }
3849 /*
3850 * This is called to update the er_extoff field in the indirection
3851 * array when extents have been added or removed from one of the
3852 * extent lists. erp_idx contains the irec index to begin updating
3853 * at and ext_diff contains the number of extents that were added
3854 * or removed.
3855 */
3856 void
3861 {
3869 }
3870 }