| blob | a59eea09930ae3a729cae134f1bd26084e2fbc4c |
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>
52 /*
53 * Used in xfs_itruncate_extents(). This is the maximum number of extents
54 * freed from a file in a single transaction.
55 */
56 #define XFS_ITRUNC_MAX_EXTENTS 2
63 /*
64 * helper function to extract extent size hint from inode
65 */
66 xfs_extlen_t
69 {
75 }
77 #ifdef DEBUG
78 /*
79 * Make sure that the extents in the given memory buffer
80 * are valid.
81 */
82 STATIC void
86 xfs_exntfmt_t fmt)
87 {
88 xfs_bmbt_irec_t irec;
89 xfs_bmbt_rec_host_t rec;
99 }
100 }
102 #define xfs_validate_extents(ifp, nrecs, fmt)
105 /*
106 * Check that none of the inode's in the buffer have a next
107 * unlinked field of 0.
108 */
109 #if defined(DEBUG)
110 void
114 {
127 bp);
129 }
130 }
131 }
132 #endif
134 /*
135 * Find the buffer associated with the given inode map
136 * We do basic validation checks on the buffer once it has been
137 * retrieved from disk.
138 */
139 STATIC int
145 uint buf_flags,
146 uint iget_flags)
147 {
163 }
165 }
167 /*
168 * Validate the magic number and version of every inode in the buffer
169 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
170 */
171 #ifdef DEBUG
175 #endif
186 XFS_ERRTAG_ITOBP_INOTOBP,
187 XFS_RANDOM_ITOBP_INOTOBP))) {
191 }
194 #ifdef DEBUG
200 #endif
203 }
204 }
209 }
211 /*
212 * This routine is called to map an inode number within a file
213 * system to the buffer containing the on-disk version of the
214 * inode. It returns a pointer to the buffer containing the
215 * on-disk inode in the bpp parameter, and in the dip parameter
216 * it returns a pointer to the on-disk inode within that buffer.
217 *
218 * If a non-zero error is returned, then the contents of bpp and
219 * dipp are undefined.
220 *
221 * Use xfs_imap() to determine the size and location of the
222 * buffer to read from disk.
223 */
224 int
228 xfs_ino_t ino,
232 uint imap_flags)
233 {
251 }
254 /*
255 * This routine is called to map an inode to the buffer containing
256 * the on-disk version of the inode. It returns a pointer to the
257 * buffer containing the on-disk inode in the bpp parameter, and in
258 * the dip parameter it returns a pointer to the on-disk inode within
259 * that buffer.
260 *
261 * If a non-zero error is returned, then the contents of bpp and
262 * dipp are undefined.
263 *
264 * The inode is expected to already been mapped to its buffer and read
265 * in once, thus we can use the mapping information stored in the inode
266 * rather than calling xfs_imap(). This allows us to avoid the overhead
267 * of looking at the inode btree for small block file systems
268 * (see xfs_imap()).
269 */
270 int
277 uint buf_flags)
278 {
293 }
298 }
300 /*
301 * Move inode type and inode format specific information from the
302 * on-disk inode to the in-core inode. For fifos, devs, and sockets
303 * this means set if_rdev to the proper value. For files, directories,
304 * and symlinks this means to bring in the in-line data or extent
305 * pointers. For a file in B-tree format, only the root is immediately
306 * brought in-core. The rest will be in-lined in if_extents when it
307 * is first referenced (see xfs_iread_extents()).
308 */
309 STATIC int
313 {
317 xfs_fsize_t di_size;
332 }
341 }
351 }
362 }
372 /*
373 * no local regular files yet
374 */
380 XFS_ERRLEVEL_LOW,
383 }
392 XFS_ERRLEVEL_LOW,
395 }
410 }
416 }
419 }
437 XFS_ERRLEVEL_LOW,
440 }
453 }
458 }
460 }
462 /*
463 * The file is in-lined in the on-disk inode.
464 * If it fits into if_inline_data, then copy
465 * it there, otherwise allocate a buffer for it
466 * and copy the data there. Either way, set
467 * if_data to point at the data.
468 * If we allocate a buffer for the data, make
469 * sure that its size is a multiple of 4 and
470 * record the real size in i_real_bytes.
471 */
472 STATIC int
478 {
482 /*
483 * If the size is unreasonable, then something
484 * is wrong and we just bail out rather than crash in
485 * kmem_alloc() or memcpy() below.
486 */
495 }
505 }
513 }
515 /*
516 * The file consists of a set of extents all
517 * of which fit into the on-disk inode.
518 * If there are few enough extents to fit into
519 * the if_inline_ext, then copy them there.
520 * Otherwise allocate a buffer for them and copy
521 * them into it. Either way, set if_extents
522 * to point at the extents.
523 */
524 STATIC int
529 {
540 /*
541 * If the number of extents is unreasonable, then something
542 * is wrong and we just bail out rather than crash in
543 * kmem_alloc() or memcpy() below.
544 */
551 }
558 else
569 }
576 XFS_ERRLEVEL_LOW,
579 }
580 }
583 }
585 /*
586 * The file has too many extents to fit into
587 * the inode, so they are in B-tree format.
588 * Allocate a buffer for the root of the B-tree
589 * and copy the root into it. The i_extents
590 * field will remain NULL until all of the
591 * extents are read in (when they are needed).
592 */
593 STATIC int
598 {
601 /* REFERENCED */
610 /*
611 * blow out if -- fork has less extents than can fit in
612 * fork (fork shouldn't be a btree format), root btree
613 * block has more records than can fit into the fork,
614 * or the number of extents is greater than the number of
615 * blocks.
616 */
627 }
632 /*
633 * Copy and convert from the on-disk structure
634 * to the in-memory structure.
635 */
643 }
645 STATIC void
649 {
679 }
681 void
685 {
715 }
717 STATIC uint
719 __uint16_t di_flags)
720 {
752 }
755 }
757 uint
760 {
765 }
767 uint
770 {
773 }
775 /*
776 * Read the disk inode attributes into the in-core inode structure.
777 */
778 int
783 uint iget_flags)
784 {
789 /*
790 * Fill in the location information in the in-core inode.
791 */
796 /*
797 * Get pointers to the on-disk inode and the buffer containing it.
798 */
804 /*
805 * If we got something that isn't an inode it means someone
806 * (nfs or dmi) has a stale handle.
807 */
809 #ifdef DEBUG
816 }
818 /*
819 * If the on-disk inode is already linked to a directory
820 * entry, copy all of the inode into the in-core inode.
821 * xfs_iformat() handles copying in the inode format
822 * specific information.
823 * Otherwise, just get the truly permanent information.
824 */
829 #ifdef DEBUG
834 }
840 /*
841 * Make sure to pull in the mode here as well in
842 * case the inode is released without being used.
843 * This ensures that xfs_inactive() will see that
844 * the inode is already free and not try to mess
845 * with the uninitialized part of it.
846 */
848 }
850 /*
851 * The inode format changed when we moved the link count and
852 * made it 32 bits long. If this is an old format inode,
853 * convert it in memory to look like a new one. If it gets
854 * flushed to disk we will convert back before flushing or
855 * logging it. We zero out the new projid field and the old link
856 * count field. We'll handle clearing the pad field (the remains
857 * of the old uuid field) when we actually convert the inode to
858 * the new format. We don't change the version number so that we
859 * can distinguish this from a real new format inode.
860 */
865 }
869 /*
870 * Mark the buffer containing the inode as something to keep
871 * around for a while. This helps to keep recently accessed
872 * meta-data in-core longer.
873 */
876 /*
877 * Use xfs_trans_brelse() to release the buffer containing the
878 * on-disk inode, because it was acquired with xfs_trans_read_buf()
879 * in xfs_itobp() above. If tp is NULL, this is just a normal
880 * brelse(). If we're within a transaction, then xfs_trans_brelse()
881 * will only release the buffer if it is not dirty within the
882 * transaction. It will be OK to release the buffer in this case,
883 * because inodes on disk are never destroyed and we will be
884 * locking the new in-core inode before putting it in the hash
885 * table where other processes can find it. Thus we don't have
886 * to worry about the inode being changed just because we released
887 * the buffer.
888 */
889 out_brelse:
892 }
894 /*
895 * Read in extents from a btree-format inode.
896 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
897 */
898 int
903 {
906 xfs_extnum_t nextents;
912 }
916 /*
917 * We know that the size is valid (it's checked in iformat_btree)
918 */
927 }
930 }
932 /*
933 * Allocate an inode on disk and return a copy of its in-core version.
934 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
935 * appropriately within the inode. The uid and gid for the inode are
936 * set according to the contents of the given cred structure.
937 *
938 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
939 * has a free inode available, call xfs_iget()
940 * to obtain the in-core version of the allocated inode. Finally,
941 * fill in the inode and log its initial contents. In this case,
942 * ialloc_context would be set to NULL and call_again set to false.
943 *
944 * If xfs_dialloc() does not have an available inode,
945 * it will replenish its supply by doing an allocation. Since we can
946 * only do one allocation within a transaction without deadlocks, we
947 * must commit the current transaction before returning the inode itself.
948 * In this case, therefore, we will set call_again to true and return.
949 * The caller should then commit the current transaction, start a new
950 * transaction, and call xfs_ialloc() again to actually get the inode.
951 *
952 * To ensure that some other process does not grab the inode that
953 * was allocated during the first call to xfs_ialloc(), this routine
954 * also returns the [locked] bp pointing to the head of the freelist
955 * as ialloc_context. The caller should hold this buffer across
956 * the commit and pass it back into this routine on the second call.
957 *
958 * If we are allocating quota inodes, we do not have a parent inode
959 * to attach to or associate with (i.e. pip == NULL) because they
960 * are not linked into the directory structure - they are attached
961 * directly to the superblock - and so have no parent.
962 */
963 int
967 umode_t mode,
968 xfs_nlink_t nlink,
969 xfs_dev_t rdev,
970 prid_t prid,
975 {
976 xfs_ino_t ino;
978 uint flags;
980 timespec_t tv;
983 /*
984 * Call the space management code to pick
985 * the on-disk inode to be allocated.
986 */
994 }
997 /*
998 * Get the in-core inode with the lock held exclusively.
999 * This is because we're setting fields here we need
1000 * to prevent others from looking at until we're done.
1001 */
1017 /*
1018 * If the superblock version is up to where we support new format
1019 * inodes and this is currently an old format inode, then change
1020 * the inode version number now. This way we only do the conversion
1021 * here rather than here and in the flush/logging code.
1022 */
1026 /*
1027 * We've already zeroed the old link count, the projid field,
1028 * and the pad field.
1029 */
1030 }
1032 /*
1033 * Project ids won't be stored on disk if we are using a version 1 inode.
1034 */
1042 }
1043 }
1045 /*
1046 * If the group ID of the new file does not match the effective group
1047 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1048 * (and only if the irix_sgid_inherit compatibility variable is set).
1049 */
1054 }
1066 /*
1067 * di_gen will have been taken care of in xfs_iread.
1068 */
1085 /*
1086 * we can't set up filestreams until after the VFS inode
1087 * is set up properly.
1088 */
1091 /* fall through */
1102 }
1109 }
1110 }
1112 xfs_inherit_noatime)
1115 xfs_inherit_nodump)
1118 xfs_inherit_sync)
1121 xfs_inherit_nosymlinks)
1126 xfs_inherit_nodefrag)
1131 }
1132 /* FALLTHROUGH */
1141 }
1142 /*
1143 * Attribute fork settings for new inode.
1144 */
1148 /*
1149 * Log the new values stuffed into the inode.
1150 */
1154 /* now that we have an i_mode we can setup inode ops and unlock */
1157 /* now we have set up the vfs inode we can associate the filestream */
1164 }
1168 }
1170 /*
1171 * Free up the underlying blocks past new_size. The new size must be smaller
1172 * than the current size. This routine can be used both for the attribute and
1173 * data fork, and does not modify the inode size, which is left to the caller.
1174 *
1175 * The transaction passed to this routine must have made a permanent log
1176 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1177 * given transaction and start new ones, so make sure everything involved in
1178 * the transaction is tidy before calling here. Some transaction will be
1179 * returned to the caller to be committed. The incoming transaction must
1180 * already include the inode, and both inode locks must be held exclusively.
1181 * The inode must also be "held" within the transaction. On return the inode
1182 * will be "held" within the returned transaction. This routine does NOT
1183 * require any disk space to be reserved for it within the transaction.
1184 *
1185 * If we get an error, we must return with the inode locked and linked into the
1186 * current transaction. This keeps things simple for the higher level code,
1187 * because it always knows that the inode is locked and held in the transaction
1188 * that returns to it whether errors occur or not. We don't mark the inode
1189 * dirty on error so that transactions can be easily aborted if possible.
1190 */
1191 int
1196 xfs_fsize_t new_size)
1197 {
1201 xfs_bmap_free_t free_list;
1202 xfs_fsblock_t first_block;
1203 xfs_fileoff_t first_unmap_block;
1204 xfs_fileoff_t last_block;
1205 xfs_filblks_t unmap_len;
1219 /*
1220 * Since it is possible for space to become allocated beyond
1221 * the end of the file (in a crash where the space is allocated
1222 * but the inode size is not yet updated), simply remove any
1223 * blocks which show up between the new EOF and the maximum
1224 * possible file size. If the first block to be removed is
1225 * beyond the maximum file size (ie it is the same as last_block),
1226 * then there is nothing to do.
1227 */
1240 XFS_ITRUNC_MAX_EXTENTS,
1246 /*
1247 * Duplicate the transaction that has the permanent
1248 * reservation and commit the old transaction.
1249 */
1257 /*
1258 * Mark the inode dirty so it will be logged and
1259 * moved forward in the log as part of every commit.
1260 */
1262 }
1273 /*
1274 * Transaction commit worked ok so we can drop the extra ticket
1275 * reference that we gained in xfs_trans_dup()
1276 */
1280 XFS_TRANS_PERM_LOG_RES,
1281 XFS_ITRUNCATE_LOG_COUNT);
1284 }
1286 /*
1287 * Always re-log the inode so that our permanent transaction can keep
1288 * on rolling it forward in the log.
1289 */
1294 out:
1297 out_bmap_cancel:
1298 /*
1299 * If the bunmapi call encounters an error, return to the caller where
1300 * the transaction can be properly aborted. We just need to make sure
1301 * we're not holding any resources that we were not when we came in.
1302 */
1305 }
1307 /*
1308 * This is called when the inode's link count goes to 0.
1309 * We place the on-disk inode on a list in the AGI. It
1310 * will be pulled from this list when the inode is freed.
1311 */
1312 int
1316 {
1322 xfs_agino_t agino;
1332 /*
1333 * Get the agi buffer first. It ensures lock ordering
1334 * on the list.
1335 */
1341 /*
1342 * Get the index into the agi hash table for the
1343 * list this inode will go on.
1344 */
1352 /*
1353 * There is already another inode in the bucket we need
1354 * to add ourselves to. Add us at the front of the list.
1355 * Here we put the head pointer into our next pointer,
1356 * and then we fall through to point the head at us.
1357 */
1370 }
1372 /*
1373 * Point the bucket head pointer at the inode being inserted.
1374 */
1382 }
1384 /*
1385 * Pull the on-disk inode from the AGI unlinked list.
1386 */
1387 STATIC int
1391 {
1392 xfs_ino_t next_ino;
1398 xfs_agnumber_t agno;
1399 xfs_agino_t agino;
1400 xfs_agino_t next_agino;
1410 /*
1411 * Get the agi buffer first. It ensures lock ordering
1412 * on the list.
1413 */
1420 /*
1421 * Get the index into the agi hash table for the
1422 * list this inode will go on.
1423 */
1431 /*
1432 * We're at the head of the list. Get the inode's
1433 * on-disk buffer to see if there is anyone after us
1434 * on the list. Only modify our next pointer if it
1435 * is not already NULLAGINO. This saves us the overhead
1436 * of dealing with the buffer when there is no need to
1437 * change it.
1438 */
1444 }
1457 }
1458 /*
1459 * Point the bucket head pointer at the next inode.
1460 */
1469 /*
1470 * We need to search the list for the inode being freed.
1471 */
1475 /*
1476 * If the last inode wasn't the one pointing to
1477 * us, then release its buffer since we're not
1478 * going to do anything with it.
1479 */
1482 }
1491 }
1495 }
1496 /*
1497 * Now last_ibp points to the buffer previous to us on
1498 * the unlinked list. Pull us from the list.
1499 */
1505 }
1519 }
1520 /*
1521 * Point the previous inode on the list to the next inode.
1522 */
1530 }
1532 }
1534 /*
1535 * A big issue when freeing the inode cluster is is that we _cannot_ skip any
1536 * inodes that are in memory - they all must be marked stale and attached to
1537 * the cluster buffer.
1538 */
1539 STATIC int
1543 xfs_ino_t inum)
1544 {
1550 xfs_daddr_t blkno;
1567 }
1573 /*
1574 * We obtain and lock the backing buffer first in the process
1575 * here, as we have to ensure that any dirty inode that we
1576 * can't get the flush lock on is attached to the buffer.
1577 * If we scan the in-memory inodes first, then buffer IO can
1578 * complete before we get a lock on it, and hence we may fail
1579 * to mark all the active inodes on the buffer stale.
1580 */
1586 /*
1587 * Walk the inodes already attached to the buffer and mark them
1588 * stale. These will all have the flush locks held, so an
1589 * in-memory inode walk can't lock them. By marking them all
1590 * stale first, we will not attempt to lock them in the loop
1591 * below as the XFS_ISTALE flag will be set.
1592 */
1603 }
1605 }
1608 /*
1609 * For each inode in memory attempt to add it to the inode
1610 * buffer and set it up for being staled on buffer IO
1611 * completion. This is safe as we've locked out tail pushing
1612 * and flushing by locking the buffer.
1613 *
1614 * We have already marked every inode that was part of a
1615 * transaction stale above, which means there is no point in
1616 * even trying to lock them.
1617 */
1619 retry:
1624 /* Inode not in memory, nothing to do */
1628 }
1630 /*
1631 * because this is an RCU protected lookup, we could
1632 * find a recently freed or even reallocated inode
1633 * during the lookup. We need to check under the
1634 * i_flags_lock for a valid inode here. Skip it if it
1635 * is not valid, the wrong inode or stale.
1636 */
1643 }
1646 /*
1647 * Don't try to lock/unlock the current inode, but we
1648 * _cannot_ skip the other inodes that we did not find
1649 * in the list attached to the buffer and are not
1650 * already marked stale. If we can't lock it, back off
1651 * and retry.
1652 */
1658 }
1664 /*
1665 * we don't need to attach clean inodes or those only
1666 * with unlogged changes (which we throw away, anyway).
1667 */
1674 }
1687 }
1691 }
1695 }
1697 /*
1698 * This is called to return an inode to the inode free list.
1699 * The inode should already be truncated to 0 length and have
1700 * no pages associated with it. This routine also assumes that
1701 * the inode is already a part of the transaction.
1702 *
1703 * The on-disk copy of the inode will have been added to the list
1704 * of unlinked inodes in the AGI. We need to remove the inode from
1705 * that list atomically with respect to freeing it here.
1706 */
1707 int
1712 {
1715 xfs_ino_t first_ino;
1726 /*
1727 * Pull the on-disk inode from the AGI unlinked list.
1728 */
1732 }
1737 }
1744 /*
1745 * Bump the generation count so no one will be confused
1746 * by reincarnations of this inode.
1747 */
1756 /*
1757 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
1758 * from picking up this inode when it is reclaimed (its incore state
1759 * initialzed but not flushed to disk yet). The in-core di_mode is
1760 * already cleared and a corresponding transaction logged.
1761 * The hack here just synchronizes the in-core to on-disk
1762 * di_mode value in advance before the actual inode sync to disk.
1763 * This is OK because the inode is already unlinked and would never
1764 * change its di_mode again for this inode generation.
1765 * This is a temporary hack that would require a proper fix
1766 * in the future.
1767 */
1772 }
1775 }
1777 /*
1778 * Reallocate the space for if_broot based on the number of records
1779 * being added or deleted as indicated in rec_diff. Move the records
1780 * and pointers in if_broot to fit the new size. When shrinking this
1781 * will eliminate holes between the records and pointers created by
1782 * the caller. When growing this will create holes to be filled in
1783 * by the caller.
1784 *
1785 * The caller must not request to add more records than would fit in
1786 * the on-disk inode root. If the if_broot is currently NULL, then
1787 * if we adding records one will be allocated. The caller must also
1788 * not request that the number of records go below zero, although
1789 * it can go to zero.
1790 *
1791 * ip -- the inode whose if_broot area is changing
1792 * ext_diff -- the change in the number of records, positive or negative,
1793 * requested for the if_broot array.
1794 */
1795 void
1800 {
1810 /*
1811 * Handle the degenerate case quietly.
1812 */
1815 }
1819 /*
1820 * If there wasn't any memory allocated before, just
1821 * allocate it now and get out.
1822 */
1828 }
1830 /*
1831 * If there is already an existing if_broot, then we need
1832 * to realloc() it and shift the pointers to their new
1833 * location. The records don't change location because
1834 * they are kept butted up against the btree block header.
1835 */
1851 }
1853 /*
1854 * rec_diff is less than 0. In this case, we are shrinking the
1855 * if_broot buffer. It must already exist. If we go to zero
1856 * records, just get rid of the root and clear the status bit.
1857 */
1864 else
1868 /*
1869 * First copy over the btree block header.
1870 */
1875 }
1877 /*
1878 * Only copy the records and pointers if there are any.
1879 */
1881 /*
1882 * First copy the records.
1883 */
1888 /*
1889 * Then copy the pointers.
1890 */
1896 }
1903 }
1906 /*
1907 * This is called when the amount of space needed for if_data
1908 * is increased or decreased. The change in size is indicated by
1909 * the number of bytes that need to be added or deleted in the
1910 * byte_diff parameter.
1911 *
1912 * If the amount of space needed has decreased below the size of the
1913 * inline buffer, then switch to using the inline buffer. Otherwise,
1914 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
1915 * to what is needed.
1916 *
1917 * ip -- the inode whose if_data area is changing
1918 * byte_diff -- the change in the number of bytes, positive or negative,
1919 * requested for the if_data array.
1920 */
1921 void
1926 {
1933 }
1942 }
1946 /*
1947 * If the valid extents/data can fit in if_inline_ext/data,
1948 * copy them from the malloc'd vector and free it.
1949 */
1955 new_size);
1958 }
1961 /*
1962 * Stuck with malloc/realloc.
1963 * For inline data, the underlying buffer must be
1964 * a multiple of 4 bytes in size so that it can be
1965 * logged and stay on word boundaries. We enforce
1966 * that here.
1967 */
1974 /*
1975 * Only do the realloc if the underlying size
1976 * is really changing.
1977 */
1981 real_size,
1984 }
1991 }
1992 }
1996 }
1998 void
2002 {
2009 }
2011 /*
2012 * If the format is local, then we can't have an extents
2013 * array so just look for an inline data array. If we're
2014 * not local then we may or may not have an extents list,
2015 * so check and free it up if we do.
2016 */
2024 }
2031 }
2038 }
2039 }
2041 /*
2042 * This is called to unpin an inode. The caller must have the inode locked
2043 * in at least shared mode so that the buffer cannot be subsequently pinned
2044 * once someone is waiting for it to be unpinned.
2045 */
2046 static void
2049 {
2054 /* Give the log a push to start the unpinning I/O */
2057 }
2059 static void
2062 {
2074 }
2076 void
2079 {
2082 }
2084 /*
2085 * xfs_iextents_copy()
2086 *
2087 * This is called to copy the REAL extents (as opposed to the delayed
2088 * allocation extents) from the inode into the given buffer. It
2089 * returns the number of bytes copied into the buffer.
2090 *
2091 * If there are no delayed allocation extents, then we can just
2092 * memcpy() the extents into the buffer. Otherwise, we need to
2093 * examine each extent in turn and skip those which are delayed.
2094 */
2095 int
2100 {
2105 xfs_fsblock_t start_block;
2115 /*
2116 * There are some delayed allocation extents in the
2117 * inode, so copy the extents one at a time and skip
2118 * the delayed ones. There must be at least one
2119 * non-delayed extent.
2120 */
2126 /*
2127 * It's a delayed allocation extent, so skip it.
2128 */
2130 }
2132 /* Translate to on disk format */
2135 dp++;
2136 copied++;
2137 }
2142 }
2144 /*
2145 * Each of the following cases stores data into the same region
2146 * of the on-disk inode, so only one of them can be valid at
2147 * any given time. While it is possible to have conflicting formats
2148 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2149 * in EXTENTS format, this can only happen when the fork has
2150 * changed formats after being modified but before being flushed.
2151 * In these cases, the format always takes precedence, because the
2152 * format indicates the current state of the fork.
2153 */
2154 /*ARGSUSED*/
2155 STATIC void
2162 {
2166 #ifdef XFS_TRANS_DEBUG
2168 #endif
2179 /*
2180 * This can happen if we gave up in iformat in an error path,
2181 * for the attribute fork.
2182 */
2186 }
2196 }
2207 whichfork);
2208 }
2217 XFS_BROOT_SIZE_ADJ));
2221 }
2228 }
2237 }
2243 }
2244 }
2246 STATIC int
2250 {
2274 /* really need a gang lookup range call here */
2285 /*
2286 * because this is an RCU protected lookup, we could find a
2287 * recently freed or even reallocated inode during the lookup.
2288 * We need to check under the i_flags_lock for a valid inode
2289 * here. Skip it if it is not valid or the wrong inode.
2290 */
2296 }
2299 /*
2300 * Do an un-protected check to see if the inode is dirty and
2301 * is a candidate for flushing. These checks will be repeated
2302 * later after the appropriate locks are acquired.
2303 */
2307 /*
2308 * Try to get locks. If any are unavailable or it is pinned,
2309 * then this inode cannot be flushed and is skipped.
2310 */
2317 }
2322 }
2324 /*
2325 * arriving here means that this inode can be flushed. First
2326 * re-check that it's dirty before flushing.
2327 */
2334 }
2335 clcount++;
2338 }
2340 }
2345 }
2347 out_free:
2350 out_put:
2355 cluster_corrupt_out:
2356 /*
2357 * Corruption detected in the clustering loop. Invalidate the
2358 * inode buffer and shut down the filesystem.
2359 */
2361 /*
2362 * Clean up the buffer. If it was delwri, just release it --
2363 * brelse can handle it with no problems. If not, shut down the
2364 * filesystem before releasing the buffer.
2365 */
2373 /*
2374 * Just like incore_relse: if we have b_iodone functions,
2375 * mark the buffer as an error and call them. Otherwise
2376 * mark it as stale and brelse.
2377 */
2386 }
2387 }
2389 /*
2390 * Unlocks the flush lock
2391 */
2396 }
2398 /*
2399 * Flush dirty inode metadata into the backing buffer.
2400 *
2401 * The caller must have the inode lock and the inode flush lock held. The
2402 * inode lock will still be held upon return to the caller, and the inode
2403 * flush lock will be released after the inode has reached the disk.
2404 *
2405 * The caller must write out the buffer returned in *bpp and release it.
2406 */
2407 int
2411 {
2428 /*
2429 * For stale inodes we cannot rely on the backing buffer remaining
2430 * stale in cache for the remaining life of the stale inode and so
2431 * xfs_itobp() below may give us a buffer that no longer contains
2432 * inodes below. We have to check this after ensuring the inode is
2433 * unpinned so that it is safe to reclaim the stale inode after the
2434 * flush call.
2435 */
2439 }
2441 /*
2442 * This may have been unpinned because the filesystem is shutting
2443 * down forcibly. If that's the case we must not write this inode
2444 * to disk, because the log record didn't make it to disk.
2445 *
2446 * We also have to remove the log item from the AIL in this case,
2447 * as we wait for an empty AIL as part of the unmount process.
2448 */
2452 }
2454 /*
2455 * Get the buffer containing the on-disk inode.
2456 */
2461 }
2463 /*
2464 * First flush out the inode that xfs_iflush was called with.
2465 */
2470 /*
2471 * If the buffer is pinned then push on the log now so we won't
2472 * get stuck waiting in the write for too long.
2473 */
2477 /*
2478 * inode clustering:
2479 * see if other inodes can be gathered into this write
2480 */
2488 corrupt_out:
2491 cluster_corrupt_out:
2493 abort_out:
2494 /*
2495 * Unlocks the flush lock
2496 */
2499 }
2502 STATIC int
2506 {
2510 #ifdef XFS_TRANS_DEBUG
2512 #endif
2522 /* set *dip = inode's place in the buffer */
2531 }
2538 }
2548 }
2559 }
2560 }
2563 XFS_RANDOM_IFLUSH_5)) {
2565 "%s: detected corrupt incore inode %Lu, "
2571 }
2578 }
2579 /*
2580 * bump the flush iteration count, used to detect flushes which
2581 * postdate a log record during recovery.
2582 */
2586 /*
2587 * Copy the dirty parts of the inode into the on-disk
2588 * inode. We always copy out the core of the inode,
2589 * because if the inode is dirty at all the core must
2590 * be.
2591 */
2594 /* Wrap, we never let the log put out DI_MAX_FLUSH */
2598 /*
2599 * If this is really an old format inode and the superblock version
2600 * has not been updated to support only new format inodes, then
2601 * convert back to the old inode format. If the superblock version
2602 * has been updated, then make the conversion permanent.
2603 */
2607 /*
2608 * Convert it back.
2609 */
2613 /*
2614 * The superblock version has already been bumped,
2615 * so just make the conversion to the new inode
2616 * format permanent.
2617 */
2626 }
2627 }
2634 /*
2635 * We've recorded everything logged in the inode, so we'd like to clear
2636 * the ili_fields bits so we don't log and flush things unnecessarily.
2637 * However, we can't stop logging all this information until the data
2638 * we've copied into the disk buffer is written to disk. If we did we
2639 * might overwrite the copy of the inode in the log with all the data
2640 * after re-logging only part of it, and in the face of a crash we
2641 * wouldn't have all the data we need to recover.
2642 *
2643 * What we do is move the bits to the ili_last_fields field. When
2644 * logging the inode, these bits are moved back to the ili_fields field.
2645 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
2646 * know that the information those bits represent is permanently on
2647 * disk. As long as the flush completes before the inode is logged
2648 * again, then both ili_fields and ili_last_fields will be cleared.
2649 *
2650 * We can play with the ili_fields bits here, because the inode lock
2651 * must be held exclusively in order to set bits there and the flush
2652 * lock protects the ili_last_fields bits. Set ili_logged so the flush
2653 * done routine can tell whether or not to look in the AIL. Also, store
2654 * the current LSN of the inode so that we can tell whether the item has
2655 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
2656 * need the AIL lock, because it is a 64 bit value that cannot be read
2657 * atomically.
2658 */
2667 /*
2668 * Attach the function xfs_iflush_done to the inode's
2669 * buffer. This will remove the inode from the AIL
2670 * and unlock the inode's flush lock when the inode is
2671 * completely written to disk.
2672 */
2678 /*
2679 * We're flushing an inode which is not in the AIL and has
2680 * not been logged. For this case we can immediately drop
2681 * the inode flush lock because we can avoid the whole
2682 * AIL state thing. It's OK to drop the flush lock now,
2683 * because we've already locked the buffer and to do anything
2684 * you really need both.
2685 */
2690 }
2692 }
2696 corrupt_out:
2698 }
2700 /*
2701 * Return a pointer to the extent record at file index idx.
2702 */
2703 xfs_bmbt_rec_host_t *
2707 {
2724 }
2725 }
2727 /*
2728 * Insert new item(s) into the extent records for incore inode
2729 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
2730 */
2731 void
2738 {
2748 }
2750 /*
2751 * This is called when the amount of space required for incore file
2752 * extents needs to be increased. The ext_diff parameter stores the
2753 * number of new extents being added and the idx parameter contains
2754 * the extent index where the new extents will be added. If the new
2755 * extents are being appended, then we just need to (re)allocate and
2756 * initialize the space. Otherwise, if the new extents are being
2757 * inserted into the middle of the existing entries, a bit more work
2758 * is required to make room for the new extents to be inserted. The
2759 * caller is responsible for filling in the new extent entries upon
2760 * return.
2761 */
2762 void
2767 {
2776 /*
2777 * If the new number of extents (nextents + ext_diff)
2778 * fits inside the inode, then continue to use the inline
2779 * extent buffer.
2780 */
2787 }
2790 }
2791 /*
2792 * Otherwise use a linear (direct) extent list.
2793 * If the extents are currently inside the inode,
2794 * xfs_iext_realloc_direct will switch us from
2795 * inline to direct extent allocation mode.
2796 */
2804 }
2805 }
2806 /* Indirection array */
2819 }
2820 /* Extents fit in target extent page */
2828 }
2831 }
2832 /* Insert a new extent page */
2836 }
2837 /*
2838 * If extent(s) are being appended to the last page in
2839 * the indirection array and the new extent(s) don't fit
2840 * in the page, then erp is NULL and erp_idx is set to
2841 * the next index needed in the indirection array.
2842 */
2851 erp_idx++;
2852 }
2853 }
2854 }
2855 }
2857 }
2859 /*
2860 * This is called when incore extents are being added to the indirection
2861 * array and the new extents do not fit in the target extent list. The
2862 * erp_idx parameter contains the irec index for the target extent list
2863 * in the indirection array, and the idx parameter contains the extent
2864 * index within the list. The number of extents being added is stored
2865 * in the count parameter.
2866 *
2867 * |-------| |-------|
2868 * | | | | idx - number of extents before idx
2869 * | idx | | count |
2870 * | | | | count - number of extents being inserted at idx
2871 * |-------| |-------|
2872 * | count | | nex2 | nex2 - number of extents after idx + count
2873 * |-------| |-------|
2874 */
2875 void
2881 {
2895 /*
2896 * Save second part of target extent list
2897 * (all extents past */
2905 }
2907 /*
2908 * Add the new extents to the end of the target
2909 * list, then allocate new irec record(s) and
2910 * extent buffer(s) as needed to store the rest
2911 * of the new extents.
2912 */
2919 }
2921 erp_idx++;
2927 }
2929 /* Add nex2 extents back to indirection array */
2931 xfs_extnum_t ext_avail;
2937 /*
2938 * If nex2 extents fit in the current page, append
2939 * nex2_ep after the new extents.
2940 */
2943 }
2944 /*
2945 * Otherwise, check if space is available in the
2946 * next page.
2947 */
2951 erp_idx++;
2952 erp++;
2953 /* Create a hole for nex2 extents */
2956 }
2957 /*
2958 * Final choice, create a new extent page for
2959 * nex2 extents.
2960 */
2962 erp_idx++;
2964 }
2969 }
2970 }
2972 /*
2973 * This is called when the amount of space required for incore file
2974 * extents needs to be decreased. The ext_diff parameter stores the
2975 * number of extents to be removed and the idx parameter contains
2976 * the extent index where the extents will be removed from.
2977 *
2978 * If the amount of space needed has decreased below the linear
2979 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
2980 * extent array. Otherwise, use kmem_realloc() to adjust the
2981 * size to what is needed.
2982 */
2983 void
2989 {
3008 }
3010 }
3012 /*
3013 * This removes ext_diff extents from the inline buffer, beginning
3014 * at extent index idx.
3015 */
3016 void
3021 {
3040 }
3041 }
3043 /*
3044 * This removes ext_diff extents from a linear (direct) extent list,
3045 * beginning at extent index idx. If the extents are being removed
3046 * from the end of the list (ie. truncate) then we just need to re-
3047 * allocate the list to remove the extra space. Otherwise, if the
3048 * extents are being removed from the middle of the existing extent
3049 * entries, then we first need to move the extent records beginning
3050 * at idx + ext_diff up in the list to overwrite the records being
3051 * removed, then remove the extra space via kmem_realloc.
3052 */
3053 void
3058 {
3070 }
3071 /* Move extents up in the list (if needed) */
3077 }
3080 /*
3081 * Reallocate the direct extent list. If the extents
3082 * will fit inside the inode then xfs_iext_realloc_direct
3083 * will switch from direct to inline extent allocation
3084 * mode for us.
3085 */
3088 }
3090 /*
3091 * This is called when incore extents are being removed from the
3092 * indirection array and the extents being removed span multiple extent
3093 * buffers. The idx parameter contains the file extent index where we
3094 * want to begin removing extents, and the count parameter contains
3095 * how many extents need to be removed.
3096 *
3097 * |-------| |-------|
3098 * | nex1 | | | nex1 - number of extents before idx
3099 * |-------| | count |
3100 * | | | | count - number of extents being removed at idx
3101 * | count | |-------|
3102 * | | | nex2 | nex2 - number of extents after idx + count
3103 * |-------| |-------|
3104 */
3105 void
3110 {
3127 /*
3128 * Check for deletion of entire list;
3129 * xfs_iext_irec_remove() updates extent offsets.
3130 */
3137 XFS_IEXT_BUFSZ);
3143 }
3144 }
3145 /* Move extents up (if needed) */
3150 }
3151 /* Zero out rest of page */
3154 /* Update remaining counters */
3159 erp_idx++;
3160 erp++;
3161 }
3164 }
3166 /*
3167 * Create, destroy, or resize a linear (direct) block of extents.
3168 */
3169 void
3173 {
3182 /* Free extent records */
3185 }
3186 /* Resize direct extent list and zero any new bytes */
3188 /* Check if extents will fit inside the inode */
3194 }
3197 }
3201 rnew_size,
3203 }
3208 }
3209 }
3210 /*
3211 * Switch from the inline extent buffer to a direct
3212 * extent list. Be sure to include the inline extent
3213 * bytes in new_size.
3214 */
3219 }
3221 }
3224 }
3226 /*
3227 * Switch from linear (direct) extent records to inline buffer.
3228 */
3229 void
3233 {
3236 /*
3237 * The inline buffer was zeroed when we switched
3238 * from inline to direct extent allocation mode,
3239 * so we don't need to clear it here.
3240 */
3246 }
3248 /*
3249 * Switch from inline buffer to linear (direct) extent records.
3250 * new_size should already be rounded up to the next power of 2
3251 * by the caller (when appropriate), so use new_size as it is.
3252 * However, since new_size may be rounded up, we can't update
3253 * if_bytes here. It is the caller's responsibility to update
3254 * if_bytes upon return.
3255 */
3256 void
3260 {
3268 }
3270 }
3272 /*
3273 * Resize an extent indirection array to new_size bytes.
3274 */
3275 STATIC void
3279 {
3294 }
3295 }
3297 /*
3298 * Switch from indirection array to linear (direct) extent allocations.
3299 */
3300 STATIC void
3303 {
3323 }
3324 }
3326 /*
3327 * Free incore file extents.
3328 */
3329 void
3332 {
3340 }
3347 }
3351 }
3353 /*
3354 * Return a pointer to the extent record for file system block bno.
3355 */
3361 {
3376 }
3379 /* Find target extent list */
3387 }
3388 /* Binary search extent records */
3399 /* Convert back to file-based extent index */
3402 }
3405 }
3406 }
3407 /* Convert back to file-based extent index */
3410 }
3416 }
3417 }
3420 }
3422 /*
3423 * Return a pointer to the indirection array entry containing the
3424 * extent record for filesystem block bno. Store the index of the
3425 * target irec in *erp_idxp.
3426 */
3432 {
3456 }
3457 }
3460 }
3462 /*
3463 * Return a pointer to the indirection array entry containing the
3464 * extent record at file extent index *idxp. Store the index of the
3465 * target irec in *erp_idxp and store the page index of the target
3466 * extent record in *idxp.
3467 */
3468 xfs_ext_irec_t *
3474 {
3493 /* Binary search extent irec's */
3509 erp_idx++;
3515 }
3516 }
3520 }
3522 /*
3523 * Allocate and initialize an indirection array once the space needed
3524 * for incore extents increases above XFS_IEXT_BUFSZ.
3525 */
3526 void
3529 {
3545 }
3556 }
3558 /*
3559 * Allocate and initialize a new entry in the indirection array.
3560 */
3561 xfs_ext_irec_t *
3565 {
3573 /* Resize indirection array */
3576 /*
3577 * Move records down in the array so the
3578 * new page can use erp_idx.
3579 */
3583 }
3586 /* Initialize new extent record */
3595 }
3597 /*
3598 * Remove a record from the indirection array.
3599 */
3600 void
3604 {
3616 }
3617 /* Compact extent records */
3621 }
3622 /*
3623 * Manually free the last extent record from the indirection
3624 * array. A call to xfs_iext_realloc_indirect() with a size
3625 * of zero would result in a call to xfs_iext_destroy() which
3626 * would in turn call this function again, creating a nasty
3627 * infinite loop.
3628 */
3634 }
3636 }
3638 /*
3639 * This is called to clean up large amounts of unused memory allocated
3640 * by the indirection array. Before compacting anything though, verify
3641 * that the indirection array is still needed and switch back to the
3642 * linear extent list (or even the inline buffer) if possible. The
3643 * compaction policy is as follows:
3644 *
3645 * Full Compaction: Extents fit into a single page (or inline buffer)
3646 * Partial Compaction: Extents occupy less than 50% of allocated space
3647 * No Compaction: Extents occupy at least 50% of allocated space
3648 */
3649 void
3652 {
3669 }
3670 }
3672 /*
3673 * Combine extents from neighboring extent pages.
3674 */
3675 void
3678 {
3694 /*
3695 * Free page before removing extent record
3696 * so er_extoffs don't get modified in
3697 * xfs_iext_irec_remove.
3698 */
3704 erp_idx++;
3705 }
3706 }
3707 }
3709 /*
3710 * This is called to update the er_extoff field in the indirection
3711 * array when extents have been added or removed from one of the
3712 * extent lists. erp_idx contains the irec index to begin updating
3713 * at and ext_diff contains the number of extents that were added
3714 * or removed.
3715 */
3716 void
3721 {
3729 }
3730 }