| blob | 755ee8164880fe4122bac9de94119f1c7086a9b7 |
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>
53 /*
54 * Used in xfs_itruncate_extents(). This is the maximum number of extents
55 * freed from a file in a single transaction.
56 */
57 #define XFS_ITRUNC_MAX_EXTENTS 2
64 #ifdef DEBUG
65 /*
66 * Make sure that the extents in the given memory buffer
67 * are valid.
68 */
69 STATIC void
73 xfs_exntfmt_t fmt)
74 {
75 xfs_bmbt_irec_t irec;
76 xfs_bmbt_rec_host_t rec;
86 }
87 }
89 #define xfs_validate_extents(ifp, nrecs, fmt)
92 /*
93 * Check that none of the inode's in the buffer have a next
94 * unlinked field of 0.
95 */
96 #if defined(DEBUG)
97 void
101 {
114 bp);
116 }
117 }
118 }
119 #endif
121 /*
122 * Find the buffer associated with the given inode map
123 * We do basic validation checks on the buffer once it has been
124 * retrieved from disk.
125 */
126 STATIC int
132 uint buf_flags,
133 uint iget_flags)
134 {
149 }
151 }
153 /*
154 * Validate the magic number and version of every inode in the buffer
155 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
156 */
157 #ifdef DEBUG
161 #endif
172 XFS_ERRTAG_ITOBP_INOTOBP,
173 XFS_RANDOM_ITOBP_INOTOBP))) {
177 }
180 #ifdef DEBUG
186 #endif
189 }
190 }
195 }
197 /*
198 * This routine is called to map an inode number within a file
199 * system to the buffer containing the on-disk version of the
200 * inode. It returns a pointer to the buffer containing the
201 * on-disk inode in the bpp parameter, and in the dip parameter
202 * it returns a pointer to the on-disk inode within that buffer.
203 *
204 * If a non-zero error is returned, then the contents of bpp and
205 * dipp are undefined.
206 *
207 * Use xfs_imap() to determine the size and location of the
208 * buffer to read from disk.
209 */
210 int
214 xfs_ino_t ino,
218 uint imap_flags)
219 {
237 }
240 /*
241 * This routine is called to map an inode to the buffer containing
242 * the on-disk version of the inode. It returns a pointer to the
243 * buffer containing the on-disk inode in the bpp parameter, and in
244 * the dip parameter it returns a pointer to the on-disk inode within
245 * that buffer.
246 *
247 * If a non-zero error is returned, then the contents of bpp and
248 * dipp are undefined.
249 *
250 * The inode is expected to already been mapped to its buffer and read
251 * in once, thus we can use the mapping information stored in the inode
252 * rather than calling xfs_imap(). This allows us to avoid the overhead
253 * of looking at the inode btree for small block file systems
254 * (see xfs_imap()).
255 */
256 int
263 uint buf_flags)
264 {
279 }
284 }
286 /*
287 * Move inode type and inode format specific information from the
288 * on-disk inode to the in-core inode. For fifos, devs, and sockets
289 * this means set if_rdev to the proper value. For files, directories,
290 * and symlinks this means to bring in the in-line data or extent
291 * pointers. For a file in B-tree format, only the root is immediately
292 * brought in-core. The rest will be in-lined in if_extents when it
293 * is first referenced (see xfs_iread_extents()).
294 */
295 STATIC int
299 {
303 xfs_fsize_t di_size;
321 }
330 }
340 }
351 }
362 /*
363 * no local regular files yet
364 */
370 XFS_ERRLEVEL_LOW,
373 }
382 XFS_ERRLEVEL_LOW,
385 }
400 }
406 }
409 }
427 XFS_ERRLEVEL_LOW,
430 }
443 }
448 }
450 }
452 /*
453 * The file is in-lined in the on-disk inode.
454 * If it fits into if_inline_data, then copy
455 * it there, otherwise allocate a buffer for it
456 * and copy the data there. Either way, set
457 * if_data to point at the data.
458 * If we allocate a buffer for the data, make
459 * sure that its size is a multiple of 4 and
460 * record the real size in i_real_bytes.
461 */
462 STATIC int
468 {
472 /*
473 * If the size is unreasonable, then something
474 * is wrong and we just bail out rather than crash in
475 * kmem_alloc() or memcpy() below.
476 */
485 }
495 }
503 }
505 /*
506 * The file consists of a set of extents all
507 * of which fit into the on-disk inode.
508 * If there are few enough extents to fit into
509 * the if_inline_ext, then copy them there.
510 * Otherwise allocate a buffer for them and copy
511 * them into it. Either way, set if_extents
512 * to point at the extents.
513 */
514 STATIC int
519 {
530 /*
531 * If the number of extents is unreasonable, then something
532 * is wrong and we just bail out rather than crash in
533 * kmem_alloc() or memcpy() below.
534 */
541 }
548 else
559 }
566 XFS_ERRLEVEL_LOW,
569 }
570 }
573 }
575 /*
576 * The file has too many extents to fit into
577 * the inode, so they are in B-tree format.
578 * Allocate a buffer for the root of the B-tree
579 * and copy the root into it. The i_extents
580 * field will remain NULL until all of the
581 * extents are read in (when they are needed).
582 */
583 STATIC int
588 {
591 /* REFERENCED */
600 /*
601 * blow out if -- fork has less extents than can fit in
602 * fork (fork shouldn't be a btree format), root btree
603 * block has more records than can fit into the fork,
604 * or the number of extents is greater than the number of
605 * blocks.
606 */
616 }
621 /*
622 * Copy and convert from the on-disk structure
623 * to the in-memory structure.
624 */
632 }
634 STATIC void
638 {
668 }
670 void
674 {
704 }
706 STATIC uint
708 __uint16_t di_flags)
709 {
741 }
744 }
746 uint
749 {
754 }
756 uint
759 {
762 }
764 /*
765 * Read the disk inode attributes into the in-core inode structure.
766 */
767 int
772 uint iget_flags)
773 {
778 /*
779 * Fill in the location information in the in-core inode.
780 */
785 /*
786 * Get pointers to the on-disk inode and the buffer containing it.
787 */
794 /*
795 * If we got something that isn't an inode it means someone
796 * (nfs or dmi) has a stale handle.
797 */
799 #ifdef DEBUG
806 }
808 /*
809 * If the on-disk inode is already linked to a directory
810 * entry, copy all of the inode into the in-core inode.
811 * xfs_iformat() handles copying in the inode format
812 * specific information.
813 * Otherwise, just get the truly permanent information.
814 */
819 #ifdef DEBUG
824 }
830 /*
831 * Make sure to pull in the mode here as well in
832 * case the inode is released without being used.
833 * This ensures that xfs_inactive() will see that
834 * the inode is already free and not try to mess
835 * with the uninitialized part of it.
836 */
838 /*
839 * Initialize the per-fork minima and maxima for a new
840 * inode here. xfs_iformat will do it for old inodes.
841 */
844 }
846 /*
847 * The inode format changed when we moved the link count and
848 * made it 32 bits long. If this is an old format inode,
849 * convert it in memory to look like a new one. If it gets
850 * flushed to disk we will convert back before flushing or
851 * logging it. We zero out the new projid field and the old link
852 * count field. We'll handle clearing the pad field (the remains
853 * of the old uuid field) when we actually convert the inode to
854 * the new format. We don't change the version number so that we
855 * can distinguish this from a real new format inode.
856 */
861 }
866 /*
867 * Mark the buffer containing the inode as something to keep
868 * around for a while. This helps to keep recently accessed
869 * meta-data in-core longer.
870 */
873 /*
874 * Use xfs_trans_brelse() to release the buffer containing the
875 * on-disk inode, because it was acquired with xfs_trans_read_buf()
876 * in xfs_itobp() above. If tp is NULL, this is just a normal
877 * brelse(). If we're within a transaction, then xfs_trans_brelse()
878 * will only release the buffer if it is not dirty within the
879 * transaction. It will be OK to release the buffer in this case,
880 * because inodes on disk are never destroyed and we will be
881 * locking the new in-core inode before putting it in the hash
882 * table where other processes can find it. Thus we don't have
883 * to worry about the inode being changed just because we released
884 * the buffer.
885 */
886 out_brelse:
889 }
891 /*
892 * Read in extents from a btree-format inode.
893 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
894 */
895 int
900 {
903 xfs_extnum_t nextents;
909 }
913 /*
914 * We know that the size is valid (it's checked in iformat_btree)
915 */
924 }
927 }
929 /*
930 * Allocate an inode on disk and return a copy of its in-core version.
931 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
932 * appropriately within the inode. The uid and gid for the inode are
933 * set according to the contents of the given cred structure.
934 *
935 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
936 * has a free inode available, call xfs_iget()
937 * to obtain the in-core version of the allocated inode. Finally,
938 * fill in the inode and log its initial contents. In this case,
939 * ialloc_context would be set to NULL and call_again set to false.
940 *
941 * If xfs_dialloc() does not have an available inode,
942 * it will replenish its supply by doing an allocation. Since we can
943 * only do one allocation within a transaction without deadlocks, we
944 * must commit the current transaction before returning the inode itself.
945 * In this case, therefore, we will set call_again to true and return.
946 * The caller should then commit the current transaction, start a new
947 * transaction, and call xfs_ialloc() again to actually get the inode.
948 *
949 * To ensure that some other process does not grab the inode that
950 * was allocated during the first call to xfs_ialloc(), this routine
951 * also returns the [locked] bp pointing to the head of the freelist
952 * as ialloc_context. The caller should hold this buffer across
953 * the commit and pass it back into this routine on the second call.
954 *
955 * If we are allocating quota inodes, we do not have a parent inode
956 * to attach to or associate with (i.e. pip == NULL) because they
957 * are not linked into the directory structure - they are attached
958 * directly to the superblock - and so have no parent.
959 */
960 int
964 mode_t mode,
965 xfs_nlink_t nlink,
966 xfs_dev_t rdev,
967 prid_t prid,
972 {
973 xfs_ino_t ino;
975 uint flags;
977 timespec_t tv;
980 /*
981 * Call the space management code to pick
982 * the on-disk inode to be allocated.
983 */
991 }
994 /*
995 * Get the in-core inode with the lock held exclusively.
996 * This is because we're setting fields here we need
997 * to prevent others from looking at until we're done.
998 */
1014 /*
1015 * If the superblock version is up to where we support new format
1016 * inodes and this is currently an old format inode, then change
1017 * the inode version number now. This way we only do the conversion
1018 * here rather than here and in the flush/logging code.
1019 */
1023 /*
1024 * We've already zeroed the old link count, the projid field,
1025 * and the pad field.
1026 */
1027 }
1029 /*
1030 * Project ids won't be stored on disk if we are using a version 1 inode.
1031 */
1039 }
1040 }
1042 /*
1043 * If the group ID of the new file does not match the effective group
1044 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1045 * (and only if the irix_sgid_inherit compatibility variable is set).
1046 */
1051 }
1064 /*
1065 * di_gen will have been taken care of in xfs_iread.
1066 */
1083 /*
1084 * we can't set up filestreams until after the VFS inode
1085 * is set up properly.
1086 */
1089 /* fall through */
1100 }
1107 }
1108 }
1110 xfs_inherit_noatime)
1113 xfs_inherit_nodump)
1116 xfs_inherit_sync)
1119 xfs_inherit_nosymlinks)
1124 xfs_inherit_nodefrag)
1129 }
1130 /* FALLTHROUGH */
1139 }
1140 /*
1141 * Attribute fork settings for new inode.
1142 */
1146 /*
1147 * Log the new values stuffed into the inode.
1148 */
1152 /* now that we have an i_mode we can setup inode ops and unlock */
1155 /* now we have set up the vfs inode we can associate the filestream */
1162 }
1166 }
1168 /*
1169 * Check to make sure that there are no blocks allocated to the
1170 * file beyond the size of the file. We don't check this for
1171 * files with fixed size extents or real time extents, but we
1172 * at least do it for regular files.
1173 */
1174 #ifdef DEBUG
1175 STATIC void
1178 xfs_fsize_t isize)
1179 {
1181 xfs_fileoff_t map_first;
1197 /*
1198 * The filesystem could be shutting down, so bmapi may return
1199 * an error.
1200 */
1209 }
1211 #define xfs_isize_check(ip, isize)
1214 /*
1215 * Free up the underlying blocks past new_size. The new size must be smaller
1216 * than the current size. This routine can be used both for the attribute and
1217 * data fork, and does not modify the inode size, which is left to the caller.
1218 *
1219 * The transaction passed to this routine must have made a permanent log
1220 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1221 * given transaction and start new ones, so make sure everything involved in
1222 * the transaction is tidy before calling here. Some transaction will be
1223 * returned to the caller to be committed. The incoming transaction must
1224 * already include the inode, and both inode locks must be held exclusively.
1225 * The inode must also be "held" within the transaction. On return the inode
1226 * will be "held" within the returned transaction. This routine does NOT
1227 * require any disk space to be reserved for it within the transaction.
1228 *
1229 * If we get an error, we must return with the inode locked and linked into the
1230 * current transaction. This keeps things simple for the higher level code,
1231 * because it always knows that the inode is locked and held in the transaction
1232 * that returns to it whether errors occur or not. We don't mark the inode
1233 * dirty on error so that transactions can be easily aborted if possible.
1234 */
1235 int
1240 xfs_fsize_t new_size)
1241 {
1245 xfs_bmap_free_t free_list;
1246 xfs_fsblock_t first_block;
1247 xfs_fileoff_t first_unmap_block;
1248 xfs_fileoff_t last_block;
1249 xfs_filblks_t unmap_len;
1261 /*
1262 * Since it is possible for space to become allocated beyond
1263 * the end of the file (in a crash where the space is allocated
1264 * but the inode size is not yet updated), simply remove any
1265 * blocks which show up between the new EOF and the maximum
1266 * possible file size. If the first block to be removed is
1267 * beyond the maximum file size (ie it is the same as last_block),
1268 * then there is nothing to do.
1269 */
1282 XFS_ITRUNC_MAX_EXTENTS,
1288 /*
1289 * Duplicate the transaction that has the permanent
1290 * reservation and commit the old transaction.
1291 */
1299 /*
1300 * Mark the inode dirty so it will be logged and
1301 * moved forward in the log as part of every commit.
1302 */
1304 }
1315 /*
1316 * Transaction commit worked ok so we can drop the extra ticket
1317 * reference that we gained in xfs_trans_dup()
1318 */
1322 XFS_TRANS_PERM_LOG_RES,
1323 XFS_ITRUNCATE_LOG_COUNT);
1326 }
1328 out:
1331 out_bmap_cancel:
1332 /*
1333 * If the bunmapi call encounters an error, return to the caller where
1334 * the transaction can be properly aborted. We just need to make sure
1335 * we're not holding any resources that we were not when we came in.
1336 */
1339 }
1341 int
1345 xfs_fsize_t new_size)
1346 {
1351 /*
1352 * The first thing we do is set the size to new_size permanently on
1353 * disk. This way we don't have to worry about anyone ever being able
1354 * to look at the data being freed even in the face of a crash.
1355 * What we're getting around here is the case where we free a block, it
1356 * is allocated to another file, it is written to, and then we crash.
1357 * If the new data gets written to the file but the log buffers
1358 * containing the free and reallocation don't, then we'd end up with
1359 * garbage in the blocks being freed. As long as we make the new_size
1360 * permanent before actually freeing any blocks it doesn't matter if
1361 * they get written to.
1362 */
1364 /*
1365 * If we are not changing the file size then do not update
1366 * the on-disk file size - we may be called from
1367 * xfs_inactive_free_eofblocks(). If we update the on-disk
1368 * file size and then the system crashes before the contents
1369 * of the file are flushed to disk then the files may be
1370 * full of holes (ie NULL files bug).
1371 */
1376 }
1377 }
1383 /*
1384 * If we are not changing the file size then do not update the on-disk
1385 * file size - we may be called from xfs_inactive_free_eofblocks().
1386 * If we update the on-disk file size and then the system crashes
1387 * before the contents of the file are flushed to disk then the files
1388 * may be full of holes (ie NULL files bug).
1389 */
1394 }
1399 /*
1400 * Always re-log the inode so that our permanent transaction can keep
1401 * on rolling it forward in the log.
1402 */
1407 }
1409 /*
1410 * This is called when the inode's link count goes to 0.
1411 * We place the on-disk inode on a list in the AGI. It
1412 * will be pulled from this list when the inode is freed.
1413 */
1414 int
1418 {
1424 xfs_agino_t agino;
1434 /*
1435 * Get the agi buffer first. It ensures lock ordering
1436 * on the list.
1437 */
1443 /*
1444 * Get the index into the agi hash table for the
1445 * list this inode will go on.
1446 */
1454 /*
1455 * There is already another inode in the bucket we need
1456 * to add ourselves to. Add us at the front of the list.
1457 * Here we put the head pointer into our next pointer,
1458 * and then we fall through to point the head at us.
1459 */
1472 }
1474 /*
1475 * Point the bucket head pointer at the inode being inserted.
1476 */
1484 }
1486 /*
1487 * Pull the on-disk inode from the AGI unlinked list.
1488 */
1489 STATIC int
1493 {
1494 xfs_ino_t next_ino;
1500 xfs_agnumber_t agno;
1501 xfs_agino_t agino;
1502 xfs_agino_t next_agino;
1512 /*
1513 * Get the agi buffer first. It ensures lock ordering
1514 * on the list.
1515 */
1522 /*
1523 * Get the index into the agi hash table for the
1524 * list this inode will go on.
1525 */
1533 /*
1534 * We're at the head of the list. Get the inode's
1535 * on-disk buffer to see if there is anyone after us
1536 * on the list. Only modify our next pointer if it
1537 * is not already NULLAGINO. This saves us the overhead
1538 * of dealing with the buffer when there is no need to
1539 * change it.
1540 */
1546 }
1559 }
1560 /*
1561 * Point the bucket head pointer at the next inode.
1562 */
1571 /*
1572 * We need to search the list for the inode being freed.
1573 */
1577 /*
1578 * If the last inode wasn't the one pointing to
1579 * us, then release its buffer since we're not
1580 * going to do anything with it.
1581 */
1584 }
1593 }
1597 }
1598 /*
1599 * Now last_ibp points to the buffer previous to us on
1600 * the unlinked list. Pull us from the list.
1601 */
1607 }
1621 }
1622 /*
1623 * Point the previous inode on the list to the next inode.
1624 */
1632 }
1634 }
1636 /*
1637 * A big issue when freeing the inode cluster is is that we _cannot_ skip any
1638 * inodes that are in memory - they all must be marked stale and attached to
1639 * the cluster buffer.
1640 */
1641 STATIC int
1645 xfs_ino_t inum)
1646 {
1652 xfs_daddr_t blkno;
1669 }
1675 /*
1676 * We obtain and lock the backing buffer first in the process
1677 * here, as we have to ensure that any dirty inode that we
1678 * can't get the flush lock on is attached to the buffer.
1679 * If we scan the in-memory inodes first, then buffer IO can
1680 * complete before we get a lock on it, and hence we may fail
1681 * to mark all the active inodes on the buffer stale.
1682 */
1685 XBF_LOCK);
1689 /*
1690 * Walk the inodes already attached to the buffer and mark them
1691 * stale. These will all have the flush locks held, so an
1692 * in-memory inode walk can't lock them. By marking them all
1693 * stale first, we will not attempt to lock them in the loop
1694 * below as the XFS_ISTALE flag will be set.
1695 */
1706 }
1708 }
1711 /*
1712 * For each inode in memory attempt to add it to the inode
1713 * buffer and set it up for being staled on buffer IO
1714 * completion. This is safe as we've locked out tail pushing
1715 * and flushing by locking the buffer.
1716 *
1717 * We have already marked every inode that was part of a
1718 * transaction stale above, which means there is no point in
1719 * even trying to lock them.
1720 */
1722 retry:
1727 /* Inode not in memory, nothing to do */
1731 }
1733 /*
1734 * because this is an RCU protected lookup, we could
1735 * find a recently freed or even reallocated inode
1736 * during the lookup. We need to check under the
1737 * i_flags_lock for a valid inode here. Skip it if it
1738 * is not valid, the wrong inode or stale.
1739 */
1746 }
1749 /*
1750 * Don't try to lock/unlock the current inode, but we
1751 * _cannot_ skip the other inodes that we did not find
1752 * in the list attached to the buffer and are not
1753 * already marked stale. If we can't lock it, back off
1754 * and retry.
1755 */
1761 }
1767 /*
1768 * we don't need to attach clean inodes or those only
1769 * with unlogged changes (which we throw away, anyway).
1770 */
1778 }
1791 }
1795 }
1799 }
1801 /*
1802 * This is called to return an inode to the inode free list.
1803 * The inode should already be truncated to 0 length and have
1804 * no pages associated with it. This routine also assumes that
1805 * the inode is already a part of the transaction.
1806 *
1807 * The on-disk copy of the inode will have been added to the list
1808 * of unlinked inodes in the AGI. We need to remove the inode from
1809 * that list atomically with respect to freeing it here.
1810 */
1811 int
1816 {
1819 xfs_ino_t first_ino;
1831 /*
1832 * Pull the on-disk inode from the AGI unlinked list.
1833 */
1837 }
1842 }
1851 /*
1852 * Bump the generation count so no one will be confused
1853 * by reincarnations of this inode.
1854 */
1863 /*
1864 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
1865 * from picking up this inode when it is reclaimed (its incore state
1866 * initialzed but not flushed to disk yet). The in-core di_mode is
1867 * already cleared and a corresponding transaction logged.
1868 * The hack here just synchronizes the in-core to on-disk
1869 * di_mode value in advance before the actual inode sync to disk.
1870 * This is OK because the inode is already unlinked and would never
1871 * change its di_mode again for this inode generation.
1872 * This is a temporary hack that would require a proper fix
1873 * in the future.
1874 */
1879 }
1882 }
1884 /*
1885 * Reallocate the space for if_broot based on the number of records
1886 * being added or deleted as indicated in rec_diff. Move the records
1887 * and pointers in if_broot to fit the new size. When shrinking this
1888 * will eliminate holes between the records and pointers created by
1889 * the caller. When growing this will create holes to be filled in
1890 * by the caller.
1891 *
1892 * The caller must not request to add more records than would fit in
1893 * the on-disk inode root. If the if_broot is currently NULL, then
1894 * if we adding records one will be allocated. The caller must also
1895 * not request that the number of records go below zero, although
1896 * it can go to zero.
1897 *
1898 * ip -- the inode whose if_broot area is changing
1899 * ext_diff -- the change in the number of records, positive or negative,
1900 * requested for the if_broot array.
1901 */
1902 void
1907 {
1917 /*
1918 * Handle the degenerate case quietly.
1919 */
1922 }
1926 /*
1927 * If there wasn't any memory allocated before, just
1928 * allocate it now and get out.
1929 */
1935 }
1937 /*
1938 * If there is already an existing if_broot, then we need
1939 * to realloc() it and shift the pointers to their new
1940 * location. The records don't change location because
1941 * they are kept butted up against the btree block header.
1942 */
1958 }
1960 /*
1961 * rec_diff is less than 0. In this case, we are shrinking the
1962 * if_broot buffer. It must already exist. If we go to zero
1963 * records, just get rid of the root and clear the status bit.
1964 */
1971 else
1975 /*
1976 * First copy over the btree block header.
1977 */
1982 }
1984 /*
1985 * Only copy the records and pointers if there are any.
1986 */
1988 /*
1989 * First copy the records.
1990 */
1995 /*
1996 * Then copy the pointers.
1997 */
2003 }
2010 }
2013 /*
2014 * This is called when the amount of space needed for if_data
2015 * is increased or decreased. The change in size is indicated by
2016 * the number of bytes that need to be added or deleted in the
2017 * byte_diff parameter.
2018 *
2019 * If the amount of space needed has decreased below the size of the
2020 * inline buffer, then switch to using the inline buffer. Otherwise,
2021 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2022 * to what is needed.
2023 *
2024 * ip -- the inode whose if_data area is changing
2025 * byte_diff -- the change in the number of bytes, positive or negative,
2026 * requested for the if_data array.
2027 */
2028 void
2033 {
2040 }
2049 }
2053 /*
2054 * If the valid extents/data can fit in if_inline_ext/data,
2055 * copy them from the malloc'd vector and free it.
2056 */
2062 new_size);
2065 }
2068 /*
2069 * Stuck with malloc/realloc.
2070 * For inline data, the underlying buffer must be
2071 * a multiple of 4 bytes in size so that it can be
2072 * logged and stay on word boundaries. We enforce
2073 * that here.
2074 */
2081 /*
2082 * Only do the realloc if the underlying size
2083 * is really changing.
2084 */
2088 real_size,
2091 }
2098 }
2099 }
2103 }
2105 void
2109 {
2116 }
2118 /*
2119 * If the format is local, then we can't have an extents
2120 * array so just look for an inline data array. If we're
2121 * not local then we may or may not have an extents list,
2122 * so check and free it up if we do.
2123 */
2131 }
2138 }
2145 }
2146 }
2148 /*
2149 * This is called to unpin an inode. The caller must have the inode locked
2150 * in at least shared mode so that the buffer cannot be subsequently pinned
2151 * once someone is waiting for it to be unpinned.
2152 */
2153 static void
2156 {
2161 /* Give the log a push to start the unpinning I/O */
2164 }
2166 void
2169 {
2173 }
2174 }
2176 /*
2177 * xfs_iextents_copy()
2178 *
2179 * This is called to copy the REAL extents (as opposed to the delayed
2180 * allocation extents) from the inode into the given buffer. It
2181 * returns the number of bytes copied into the buffer.
2182 *
2183 * If there are no delayed allocation extents, then we can just
2184 * memcpy() the extents into the buffer. Otherwise, we need to
2185 * examine each extent in turn and skip those which are delayed.
2186 */
2187 int
2192 {
2197 xfs_fsblock_t start_block;
2207 /*
2208 * There are some delayed allocation extents in the
2209 * inode, so copy the extents one at a time and skip
2210 * the delayed ones. There must be at least one
2211 * non-delayed extent.
2212 */
2218 /*
2219 * It's a delayed allocation extent, so skip it.
2220 */
2222 }
2224 /* Translate to on disk format */
2227 dp++;
2228 copied++;
2229 }
2234 }
2236 /*
2237 * Each of the following cases stores data into the same region
2238 * of the on-disk inode, so only one of them can be valid at
2239 * any given time. While it is possible to have conflicting formats
2240 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2241 * in EXTENTS format, this can only happen when the fork has
2242 * changed formats after being modified but before being flushed.
2243 * In these cases, the format always takes precedence, because the
2244 * format indicates the current state of the fork.
2245 */
2246 /*ARGSUSED*/
2247 STATIC void
2254 {
2258 #ifdef XFS_TRANS_DEBUG
2260 #endif
2271 /*
2272 * This can happen if we gave up in iformat in an error path,
2273 * for the attribute fork.
2274 */
2278 }
2288 }
2299 whichfork);
2300 }
2309 XFS_BROOT_SIZE_ADJ));
2313 }
2320 }
2329 }
2335 }
2336 }
2338 STATIC int
2342 {
2366 /* really need a gang lookup range call here */
2377 /*
2378 * because this is an RCU protected lookup, we could find a
2379 * recently freed or even reallocated inode during the lookup.
2380 * We need to check under the i_flags_lock for a valid inode
2381 * here. Skip it if it is not valid or the wrong inode.
2382 */
2388 }
2391 /*
2392 * Do an un-protected check to see if the inode is dirty and
2393 * is a candidate for flushing. These checks will be repeated
2394 * later after the appropriate locks are acquired.
2395 */
2399 /*
2400 * Try to get locks. If any are unavailable or it is pinned,
2401 * then this inode cannot be flushed and is skipped.
2402 */
2409 }
2414 }
2416 /*
2417 * arriving here means that this inode can be flushed. First
2418 * re-check that it's dirty before flushing.
2419 */
2426 }
2427 clcount++;
2430 }
2432 }
2437 }
2439 out_free:
2442 out_put:
2447 cluster_corrupt_out:
2448 /*
2449 * Corruption detected in the clustering loop. Invalidate the
2450 * inode buffer and shut down the filesystem.
2451 */
2453 /*
2454 * Clean up the buffer. If it was B_DELWRI, just release it --
2455 * brelse can handle it with no problems. If not, shut down the
2456 * filesystem before releasing the buffer.
2457 */
2465 /*
2466 * Just like incore_relse: if we have b_iodone functions,
2467 * mark the buffer as an error and call them. Otherwise
2468 * mark it as stale and brelse.
2469 */
2478 }
2479 }
2481 /*
2482 * Unlocks the flush lock
2483 */
2488 }
2490 /*
2491 * xfs_iflush() will write a modified inode's changes out to the
2492 * inode's on disk home. The caller must have the inode lock held
2493 * in at least shared mode and the inode flush completion must be
2494 * active as well. The inode lock will still be held upon return from
2495 * the call and the caller is free to unlock it.
2496 * The inode flush will be completed when the inode reaches the disk.
2497 * The flags indicate how the inode's buffer should be written out.
2498 */
2499 int
2502 uint flags)
2503 {
2520 /*
2521 * We can't flush the inode until it is unpinned, so wait for it if we
2522 * are allowed to block. We know no one new can pin it, because we are
2523 * holding the inode lock shared and you need to hold it exclusively to
2524 * pin the inode.
2525 *
2526 * If we are not allowed to block, force the log out asynchronously so
2527 * that when we come back the inode will be unpinned. If other inodes
2528 * in the same cluster are dirty, they will probably write the inode
2529 * out for us if they occur after the log force completes.
2530 */
2535 }
2538 /*
2539 * For stale inodes we cannot rely on the backing buffer remaining
2540 * stale in cache for the remaining life of the stale inode and so
2541 * xfs_itobp() below may give us a buffer that no longer contains
2542 * inodes below. We have to check this after ensuring the inode is
2543 * unpinned so that it is safe to reclaim the stale inode after the
2544 * flush call.
2545 */
2549 }
2551 /*
2552 * This may have been unpinned because the filesystem is shutting
2553 * down forcibly. If that's the case we must not write this inode
2554 * to disk, because the log record didn't make it to disk!
2555 */
2562 }
2564 /*
2565 * Get the buffer containing the on-disk inode.
2566 */
2572 }
2574 /*
2575 * First flush out the inode that xfs_iflush was called with.
2576 */
2581 /*
2582 * If the buffer is pinned then push on the log now so we won't
2583 * get stuck waiting in the write for too long.
2584 */
2588 /*
2589 * inode clustering:
2590 * see if other inodes can be gathered into this write
2591 */
2598 else
2604 corrupt_out:
2607 cluster_corrupt_out:
2608 /*
2609 * Unlocks the flush lock
2610 */
2613 }
2616 STATIC int
2620 {
2624 #ifdef XFS_TRANS_DEBUG
2626 #endif
2636 /* set *dip = inode's place in the buffer */
2639 /*
2640 * Clear i_update_core before copying out the data.
2641 * This is for coordination with our timestamp updates
2642 * that don't hold the inode lock. They will always
2643 * update the timestamps BEFORE setting i_update_core,
2644 * so if we clear i_update_core after they set it we
2645 * are guaranteed to see their updates to the timestamps.
2646 * I believe that this depends on strongly ordered memory
2647 * semantics, but we have that. We use the SYNCHRONIZE
2648 * macro to make sure that the compiler does not reorder
2649 * the i_update_core access below the data copy below.
2650 */
2654 /*
2655 * Make sure to get the latest timestamps from the Linux inode.
2656 */
2665 }
2672 }
2682 }
2693 }
2694 }
2697 XFS_RANDOM_IFLUSH_5)) {
2699 "%s: detected corrupt incore inode %Lu, "
2705 }
2712 }
2713 /*
2714 * bump the flush iteration count, used to detect flushes which
2715 * postdate a log record during recovery.
2716 */
2720 /*
2721 * Copy the dirty parts of the inode into the on-disk
2722 * inode. We always copy out the core of the inode,
2723 * because if the inode is dirty at all the core must
2724 * be.
2725 */
2728 /* Wrap, we never let the log put out DI_MAX_FLUSH */
2732 /*
2733 * If this is really an old format inode and the superblock version
2734 * has not been updated to support only new format inodes, then
2735 * convert back to the old inode format. If the superblock version
2736 * has been updated, then make the conversion permanent.
2737 */
2741 /*
2742 * Convert it back.
2743 */
2747 /*
2748 * The superblock version has already been bumped,
2749 * so just make the conversion to the new inode
2750 * format permanent.
2751 */
2760 }
2761 }
2768 /*
2769 * We've recorded everything logged in the inode, so we'd
2770 * like to clear the ilf_fields bits so we don't log and
2771 * flush things unnecessarily. However, we can't stop
2772 * logging all this information until the data we've copied
2773 * into the disk buffer is written to disk. If we did we might
2774 * overwrite the copy of the inode in the log with all the
2775 * data after re-logging only part of it, and in the face of
2776 * a crash we wouldn't have all the data we need to recover.
2777 *
2778 * What we do is move the bits to the ili_last_fields field.
2779 * When logging the inode, these bits are moved back to the
2780 * ilf_fields field. In the xfs_iflush_done() routine we
2781 * clear ili_last_fields, since we know that the information
2782 * those bits represent is permanently on disk. As long as
2783 * the flush completes before the inode is logged again, then
2784 * both ilf_fields and ili_last_fields will be cleared.
2785 *
2786 * We can play with the ilf_fields bits here, because the inode
2787 * lock must be held exclusively in order to set bits there
2788 * and the flush lock protects the ili_last_fields bits.
2789 * Set ili_logged so the flush done
2790 * routine can tell whether or not to look in the AIL.
2791 * Also, store the current LSN of the inode so that we can tell
2792 * whether the item has moved in the AIL from xfs_iflush_done().
2793 * In order to read the lsn we need the AIL lock, because
2794 * it is a 64 bit value that cannot be read atomically.
2795 */
2804 /*
2805 * Attach the function xfs_iflush_done to the inode's
2806 * buffer. This will remove the inode from the AIL
2807 * and unlock the inode's flush lock when the inode is
2808 * completely written to disk.
2809 */
2815 /*
2816 * We're flushing an inode which is not in the AIL and has
2817 * not been logged but has i_update_core set. For this
2818 * case we can use a B_DELWRI flush and immediately drop
2819 * the inode flush lock because we can avoid the whole
2820 * AIL state thing. It's OK to drop the flush lock now,
2821 * because we've already locked the buffer and to do anything
2822 * you really need both.
2823 */
2828 }
2830 }
2834 corrupt_out:
2836 }
2838 void
2841 {
2854 }
2857 }
2859 /*
2860 * Return a pointer to the extent record at file index idx.
2861 */
2862 xfs_bmbt_rec_host_t *
2866 {
2883 }
2884 }
2886 /*
2887 * Insert new item(s) into the extent records for incore inode
2888 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
2889 */
2890 void
2897 {
2907 }
2909 /*
2910 * This is called when the amount of space required for incore file
2911 * extents needs to be increased. The ext_diff parameter stores the
2912 * number of new extents being added and the idx parameter contains
2913 * the extent index where the new extents will be added. If the new
2914 * extents are being appended, then we just need to (re)allocate and
2915 * initialize the space. Otherwise, if the new extents are being
2916 * inserted into the middle of the existing entries, a bit more work
2917 * is required to make room for the new extents to be inserted. The
2918 * caller is responsible for filling in the new extent entries upon
2919 * return.
2920 */
2921 void
2926 {
2935 /*
2936 * If the new number of extents (nextents + ext_diff)
2937 * fits inside the inode, then continue to use the inline
2938 * extent buffer.
2939 */
2946 }
2949 }
2950 /*
2951 * Otherwise use a linear (direct) extent list.
2952 * If the extents are currently inside the inode,
2953 * xfs_iext_realloc_direct will switch us from
2954 * inline to direct extent allocation mode.
2955 */
2963 }
2964 }
2965 /* Indirection array */
2978 }
2979 /* Extents fit in target extent page */
2987 }
2990 }
2991 /* Insert a new extent page */
2995 }
2996 /*
2997 * If extent(s) are being appended to the last page in
2998 * the indirection array and the new extent(s) don't fit
2999 * in the page, then erp is NULL and erp_idx is set to
3000 * the next index needed in the indirection array.
3001 */
3010 erp_idx++;
3011 }
3012 }
3013 }
3014 }
3016 }
3018 /*
3019 * This is called when incore extents are being added to the indirection
3020 * array and the new extents do not fit in the target extent list. The
3021 * erp_idx parameter contains the irec index for the target extent list
3022 * in the indirection array, and the idx parameter contains the extent
3023 * index within the list. The number of extents being added is stored
3024 * in the count parameter.
3025 *
3026 * |-------| |-------|
3027 * | | | | idx - number of extents before idx
3028 * | idx | | count |
3029 * | | | | count - number of extents being inserted at idx
3030 * |-------| |-------|
3031 * | count | | nex2 | nex2 - number of extents after idx + count
3032 * |-------| |-------|
3033 */
3034 void
3040 {
3054 /*
3055 * Save second part of target extent list
3056 * (all extents past */
3064 }
3066 /*
3067 * Add the new extents to the end of the target
3068 * list, then allocate new irec record(s) and
3069 * extent buffer(s) as needed to store the rest
3070 * of the new extents.
3071 */
3078 }
3080 erp_idx++;
3086 }
3088 /* Add nex2 extents back to indirection array */
3090 xfs_extnum_t ext_avail;
3096 /*
3097 * If nex2 extents fit in the current page, append
3098 * nex2_ep after the new extents.
3099 */
3102 }
3103 /*
3104 * Otherwise, check if space is available in the
3105 * next page.
3106 */
3110 erp_idx++;
3111 erp++;
3112 /* Create a hole for nex2 extents */
3115 }
3116 /*
3117 * Final choice, create a new extent page for
3118 * nex2 extents.
3119 */
3121 erp_idx++;
3123 }
3128 }
3129 }
3131 /*
3132 * This is called when the amount of space required for incore file
3133 * extents needs to be decreased. The ext_diff parameter stores the
3134 * number of extents to be removed and the idx parameter contains
3135 * the extent index where the extents will be removed from.
3136 *
3137 * If the amount of space needed has decreased below the linear
3138 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3139 * extent array. Otherwise, use kmem_realloc() to adjust the
3140 * size to what is needed.
3141 */
3142 void
3148 {
3167 }
3169 }
3171 /*
3172 * This removes ext_diff extents from the inline buffer, beginning
3173 * at extent index idx.
3174 */
3175 void
3180 {
3199 }
3200 }
3202 /*
3203 * This removes ext_diff extents from a linear (direct) extent list,
3204 * beginning at extent index idx. If the extents are being removed
3205 * from the end of the list (ie. truncate) then we just need to re-
3206 * allocate the list to remove the extra space. Otherwise, if the
3207 * extents are being removed from the middle of the existing extent
3208 * entries, then we first need to move the extent records beginning
3209 * at idx + ext_diff up in the list to overwrite the records being
3210 * removed, then remove the extra space via kmem_realloc.
3211 */
3212 void
3217 {
3229 }
3230 /* Move extents up in the list (if needed) */
3236 }
3239 /*
3240 * Reallocate the direct extent list. If the extents
3241 * will fit inside the inode then xfs_iext_realloc_direct
3242 * will switch from direct to inline extent allocation
3243 * mode for us.
3244 */
3247 }
3249 /*
3250 * This is called when incore extents are being removed from the
3251 * indirection array and the extents being removed span multiple extent
3252 * buffers. The idx parameter contains the file extent index where we
3253 * want to begin removing extents, and the count parameter contains
3254 * how many extents need to be removed.
3255 *
3256 * |-------| |-------|
3257 * | nex1 | | | nex1 - number of extents before idx
3258 * |-------| | count |
3259 * | | | | count - number of extents being removed at idx
3260 * | count | |-------|
3261 * | | | nex2 | nex2 - number of extents after idx + count
3262 * |-------| |-------|
3263 */
3264 void
3269 {
3286 /*
3287 * Check for deletion of entire list;
3288 * xfs_iext_irec_remove() updates extent offsets.
3289 */
3296 XFS_IEXT_BUFSZ);
3302 }
3303 }
3304 /* Move extents up (if needed) */
3309 }
3310 /* Zero out rest of page */
3313 /* Update remaining counters */
3318 erp_idx++;
3319 erp++;
3320 }
3323 }
3325 /*
3326 * Create, destroy, or resize a linear (direct) block of extents.
3327 */
3328 void
3332 {
3341 /* Free extent records */
3344 }
3345 /* Resize direct extent list and zero any new bytes */
3347 /* Check if extents will fit inside the inode */
3353 }
3356 }
3360 rnew_size,
3362 }
3367 }
3368 }
3369 /*
3370 * Switch from the inline extent buffer to a direct
3371 * extent list. Be sure to include the inline extent
3372 * bytes in new_size.
3373 */
3378 }
3380 }
3383 }
3385 /*
3386 * Switch from linear (direct) extent records to inline buffer.
3387 */
3388 void
3392 {
3395 /*
3396 * The inline buffer was zeroed when we switched
3397 * from inline to direct extent allocation mode,
3398 * so we don't need to clear it here.
3399 */
3405 }
3407 /*
3408 * Switch from inline buffer to linear (direct) extent records.
3409 * new_size should already be rounded up to the next power of 2
3410 * by the caller (when appropriate), so use new_size as it is.
3411 * However, since new_size may be rounded up, we can't update
3412 * if_bytes here. It is the caller's responsibility to update
3413 * if_bytes upon return.
3414 */
3415 void
3419 {
3427 }
3429 }
3431 /*
3432 * Resize an extent indirection array to new_size bytes.
3433 */
3434 STATIC void
3438 {
3453 }
3454 }
3456 /*
3457 * Switch from indirection array to linear (direct) extent allocations.
3458 */
3459 STATIC void
3462 {
3482 }
3483 }
3485 /*
3486 * Free incore file extents.
3487 */
3488 void
3491 {
3499 }
3506 }
3510 }
3512 /*
3513 * Return a pointer to the extent record for file system block bno.
3514 */
3520 {
3535 }
3538 /* Find target extent list */
3546 }
3547 /* Binary search extent records */
3558 /* Convert back to file-based extent index */
3561 }
3564 }
3565 }
3566 /* Convert back to file-based extent index */
3569 }
3575 }
3576 }
3579 }
3581 /*
3582 * Return a pointer to the indirection array entry containing the
3583 * extent record for filesystem block bno. Store the index of the
3584 * target irec in *erp_idxp.
3585 */
3591 {
3615 }
3616 }
3619 }
3621 /*
3622 * Return a pointer to the indirection array entry containing the
3623 * extent record at file extent index *idxp. Store the index of the
3624 * target irec in *erp_idxp and store the page index of the target
3625 * extent record in *idxp.
3626 */
3627 xfs_ext_irec_t *
3633 {
3652 /* Binary search extent irec's */
3668 erp_idx++;
3674 }
3675 }
3679 }
3681 /*
3682 * Allocate and initialize an indirection array once the space needed
3683 * for incore extents increases above XFS_IEXT_BUFSZ.
3684 */
3685 void
3688 {
3704 }
3715 }
3717 /*
3718 * Allocate and initialize a new entry in the indirection array.
3719 */
3720 xfs_ext_irec_t *
3724 {
3732 /* Resize indirection array */
3735 /*
3736 * Move records down in the array so the
3737 * new page can use erp_idx.
3738 */
3742 }
3745 /* Initialize new extent record */
3754 }
3756 /*
3757 * Remove a record from the indirection array.
3758 */
3759 void
3763 {
3775 }
3776 /* Compact extent records */
3780 }
3781 /*
3782 * Manually free the last extent record from the indirection
3783 * array. A call to xfs_iext_realloc_indirect() with a size
3784 * of zero would result in a call to xfs_iext_destroy() which
3785 * would in turn call this function again, creating a nasty
3786 * infinite loop.
3787 */
3793 }
3795 }
3797 /*
3798 * This is called to clean up large amounts of unused memory allocated
3799 * by the indirection array. Before compacting anything though, verify
3800 * that the indirection array is still needed and switch back to the
3801 * linear extent list (or even the inline buffer) if possible. The
3802 * compaction policy is as follows:
3803 *
3804 * Full Compaction: Extents fit into a single page (or inline buffer)
3805 * Partial Compaction: Extents occupy less than 50% of allocated space
3806 * No Compaction: Extents occupy at least 50% of allocated space
3807 */
3808 void
3811 {
3828 }
3829 }
3831 /*
3832 * Combine extents from neighboring extent pages.
3833 */
3834 void
3837 {
3853 /*
3854 * Free page before removing extent record
3855 * so er_extoffs don't get modified in
3856 * xfs_iext_irec_remove.
3857 */
3863 erp_idx++;
3864 }
3865 }
3866 }
3868 /*
3869 * This is called to update the er_extoff field in the indirection
3870 * array when extents have been added or removed from one of the
3871 * extent lists. erp_idx contains the irec index to begin updating
3872 * at and ext_diff contains the number of extents that were added
3873 * or removed.
3874 */
3875 void
3880 {
3888 }
3889 }