| blob | b21022499c2e8f302699f80ca2af344301fee941 |
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;
318 }
327 }
337 }
348 }
358 /*
359 * no local regular files yet
360 */
366 XFS_ERRLEVEL_LOW,
369 }
378 XFS_ERRLEVEL_LOW,
381 }
396 }
402 }
405 }
423 XFS_ERRLEVEL_LOW,
426 }
439 }
444 }
446 }
448 /*
449 * The file is in-lined in the on-disk inode.
450 * If it fits into if_inline_data, then copy
451 * it there, otherwise allocate a buffer for it
452 * and copy the data there. Either way, set
453 * if_data to point at the data.
454 * If we allocate a buffer for the data, make
455 * sure that its size is a multiple of 4 and
456 * record the real size in i_real_bytes.
457 */
458 STATIC int
464 {
468 /*
469 * If the size is unreasonable, then something
470 * is wrong and we just bail out rather than crash in
471 * kmem_alloc() or memcpy() below.
472 */
481 }
491 }
499 }
501 /*
502 * The file consists of a set of extents all
503 * of which fit into the on-disk inode.
504 * If there are few enough extents to fit into
505 * the if_inline_ext, then copy them there.
506 * Otherwise allocate a buffer for them and copy
507 * them into it. Either way, set if_extents
508 * to point at the extents.
509 */
510 STATIC int
515 {
526 /*
527 * If the number of extents is unreasonable, then something
528 * is wrong and we just bail out rather than crash in
529 * kmem_alloc() or memcpy() below.
530 */
537 }
544 else
555 }
562 XFS_ERRLEVEL_LOW,
565 }
566 }
569 }
571 /*
572 * The file has too many extents to fit into
573 * the inode, so they are in B-tree format.
574 * Allocate a buffer for the root of the B-tree
575 * and copy the root into it. The i_extents
576 * field will remain NULL until all of the
577 * extents are read in (when they are needed).
578 */
579 STATIC int
584 {
587 /* REFERENCED */
596 /*
597 * blow out if -- fork has less extents than can fit in
598 * fork (fork shouldn't be a btree format), root btree
599 * block has more records than can fit into the fork,
600 * or the number of extents is greater than the number of
601 * blocks.
602 */
613 }
618 /*
619 * Copy and convert from the on-disk structure
620 * to the in-memory structure.
621 */
629 }
631 STATIC void
635 {
665 }
667 void
671 {
701 }
703 STATIC uint
705 __uint16_t di_flags)
706 {
738 }
741 }
743 uint
746 {
751 }
753 uint
756 {
759 }
761 /*
762 * Read the disk inode attributes into the in-core inode structure.
763 */
764 int
769 uint iget_flags)
770 {
775 /*
776 * Fill in the location information in the in-core inode.
777 */
782 /*
783 * Get pointers to the on-disk inode and the buffer containing it.
784 */
791 /*
792 * If we got something that isn't an inode it means someone
793 * (nfs or dmi) has a stale handle.
794 */
796 #ifdef DEBUG
803 }
805 /*
806 * If the on-disk inode is already linked to a directory
807 * entry, copy all of the inode into the in-core inode.
808 * xfs_iformat() handles copying in the inode format
809 * specific information.
810 * Otherwise, just get the truly permanent information.
811 */
816 #ifdef DEBUG
821 }
827 /*
828 * Make sure to pull in the mode here as well in
829 * case the inode is released without being used.
830 * This ensures that xfs_inactive() will see that
831 * the inode is already free and not try to mess
832 * with the uninitialized part of it.
833 */
835 }
837 /*
838 * The inode format changed when we moved the link count and
839 * made it 32 bits long. If this is an old format inode,
840 * convert it in memory to look like a new one. If it gets
841 * flushed to disk we will convert back before flushing or
842 * logging it. We zero out the new projid field and the old link
843 * count field. We'll handle clearing the pad field (the remains
844 * of the old uuid field) when we actually convert the inode to
845 * the new format. We don't change the version number so that we
846 * can distinguish this from a real new format inode.
847 */
852 }
856 /*
857 * Mark the buffer containing the inode as something to keep
858 * around for a while. This helps to keep recently accessed
859 * meta-data in-core longer.
860 */
863 /*
864 * Use xfs_trans_brelse() to release the buffer containing the
865 * on-disk inode, because it was acquired with xfs_trans_read_buf()
866 * in xfs_itobp() above. If tp is NULL, this is just a normal
867 * brelse(). If we're within a transaction, then xfs_trans_brelse()
868 * will only release the buffer if it is not dirty within the
869 * transaction. It will be OK to release the buffer in this case,
870 * because inodes on disk are never destroyed and we will be
871 * locking the new in-core inode before putting it in the hash
872 * table where other processes can find it. Thus we don't have
873 * to worry about the inode being changed just because we released
874 * the buffer.
875 */
876 out_brelse:
879 }
881 /*
882 * Read in extents from a btree-format inode.
883 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
884 */
885 int
890 {
893 xfs_extnum_t nextents;
899 }
903 /*
904 * We know that the size is valid (it's checked in iformat_btree)
905 */
914 }
917 }
919 /*
920 * Allocate an inode on disk and return a copy of its in-core version.
921 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
922 * appropriately within the inode. The uid and gid for the inode are
923 * set according to the contents of the given cred structure.
924 *
925 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
926 * has a free inode available, call xfs_iget()
927 * to obtain the in-core version of the allocated inode. Finally,
928 * fill in the inode and log its initial contents. In this case,
929 * ialloc_context would be set to NULL and call_again set to false.
930 *
931 * If xfs_dialloc() does not have an available inode,
932 * it will replenish its supply by doing an allocation. Since we can
933 * only do one allocation within a transaction without deadlocks, we
934 * must commit the current transaction before returning the inode itself.
935 * In this case, therefore, we will set call_again to true and return.
936 * The caller should then commit the current transaction, start a new
937 * transaction, and call xfs_ialloc() again to actually get the inode.
938 *
939 * To ensure that some other process does not grab the inode that
940 * was allocated during the first call to xfs_ialloc(), this routine
941 * also returns the [locked] bp pointing to the head of the freelist
942 * as ialloc_context. The caller should hold this buffer across
943 * the commit and pass it back into this routine on the second call.
944 *
945 * If we are allocating quota inodes, we do not have a parent inode
946 * to attach to or associate with (i.e. pip == NULL) because they
947 * are not linked into the directory structure - they are attached
948 * directly to the superblock - and so have no parent.
949 */
950 int
954 umode_t mode,
955 xfs_nlink_t nlink,
956 xfs_dev_t rdev,
957 prid_t prid,
962 {
963 xfs_ino_t ino;
965 uint flags;
967 timespec_t tv;
970 /*
971 * Call the space management code to pick
972 * the on-disk inode to be allocated.
973 */
981 }
984 /*
985 * Get the in-core inode with the lock held exclusively.
986 * This is because we're setting fields here we need
987 * to prevent others from looking at until we're done.
988 */
1004 /*
1005 * If the superblock version is up to where we support new format
1006 * inodes and this is currently an old format inode, then change
1007 * the inode version number now. This way we only do the conversion
1008 * here rather than here and in the flush/logging code.
1009 */
1013 /*
1014 * We've already zeroed the old link count, the projid field,
1015 * and the pad field.
1016 */
1017 }
1019 /*
1020 * Project ids won't be stored on disk if we are using a version 1 inode.
1021 */
1029 }
1030 }
1032 /*
1033 * If the group ID of the new file does not match the effective group
1034 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1035 * (and only if the irix_sgid_inherit compatibility variable is set).
1036 */
1041 }
1053 /*
1054 * di_gen will have been taken care of in xfs_iread.
1055 */
1072 /*
1073 * we can't set up filestreams until after the VFS inode
1074 * is set up properly.
1075 */
1078 /* fall through */
1089 }
1096 }
1097 }
1099 xfs_inherit_noatime)
1102 xfs_inherit_nodump)
1105 xfs_inherit_sync)
1108 xfs_inherit_nosymlinks)
1113 xfs_inherit_nodefrag)
1118 }
1119 /* FALLTHROUGH */
1128 }
1129 /*
1130 * Attribute fork settings for new inode.
1131 */
1135 /*
1136 * Log the new values stuffed into the inode.
1137 */
1141 /* now that we have an i_mode we can setup inode ops and unlock */
1144 /* now we have set up the vfs inode we can associate the filestream */
1151 }
1155 }
1157 /*
1158 * Free up the underlying blocks past new_size. The new size must be smaller
1159 * than the current size. This routine can be used both for the attribute and
1160 * data fork, and does not modify the inode size, which is left to the caller.
1161 *
1162 * The transaction passed to this routine must have made a permanent log
1163 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1164 * given transaction and start new ones, so make sure everything involved in
1165 * the transaction is tidy before calling here. Some transaction will be
1166 * returned to the caller to be committed. The incoming transaction must
1167 * already include the inode, and both inode locks must be held exclusively.
1168 * The inode must also be "held" within the transaction. On return the inode
1169 * will be "held" within the returned transaction. This routine does NOT
1170 * require any disk space to be reserved for it within the transaction.
1171 *
1172 * If we get an error, we must return with the inode locked and linked into the
1173 * current transaction. This keeps things simple for the higher level code,
1174 * because it always knows that the inode is locked and held in the transaction
1175 * that returns to it whether errors occur or not. We don't mark the inode
1176 * dirty on error so that transactions can be easily aborted if possible.
1177 */
1178 int
1183 xfs_fsize_t new_size)
1184 {
1188 xfs_bmap_free_t free_list;
1189 xfs_fsblock_t first_block;
1190 xfs_fileoff_t first_unmap_block;
1191 xfs_fileoff_t last_block;
1192 xfs_filblks_t unmap_len;
1206 /*
1207 * Since it is possible for space to become allocated beyond
1208 * the end of the file (in a crash where the space is allocated
1209 * but the inode size is not yet updated), simply remove any
1210 * blocks which show up between the new EOF and the maximum
1211 * possible file size. If the first block to be removed is
1212 * beyond the maximum file size (ie it is the same as last_block),
1213 * then there is nothing to do.
1214 */
1227 XFS_ITRUNC_MAX_EXTENTS,
1233 /*
1234 * Duplicate the transaction that has the permanent
1235 * reservation and commit the old transaction.
1236 */
1244 /*
1245 * Mark the inode dirty so it will be logged and
1246 * moved forward in the log as part of every commit.
1247 */
1249 }
1260 /*
1261 * Transaction commit worked ok so we can drop the extra ticket
1262 * reference that we gained in xfs_trans_dup()
1263 */
1267 XFS_TRANS_PERM_LOG_RES,
1268 XFS_ITRUNCATE_LOG_COUNT);
1271 }
1273 /*
1274 * Always re-log the inode so that our permanent transaction can keep
1275 * on rolling it forward in the log.
1276 */
1281 out:
1284 out_bmap_cancel:
1285 /*
1286 * If the bunmapi call encounters an error, return to the caller where
1287 * the transaction can be properly aborted. We just need to make sure
1288 * we're not holding any resources that we were not when we came in.
1289 */
1292 }
1294 /*
1295 * This is called when the inode's link count goes to 0.
1296 * We place the on-disk inode on a list in the AGI. It
1297 * will be pulled from this list when the inode is freed.
1298 */
1299 int
1303 {
1309 xfs_agino_t agino;
1319 /*
1320 * Get the agi buffer first. It ensures lock ordering
1321 * on the list.
1322 */
1328 /*
1329 * Get the index into the agi hash table for the
1330 * list this inode will go on.
1331 */
1339 /*
1340 * There is already another inode in the bucket we need
1341 * to add ourselves to. Add us at the front of the list.
1342 * Here we put the head pointer into our next pointer,
1343 * and then we fall through to point the head at us.
1344 */
1357 }
1359 /*
1360 * Point the bucket head pointer at the inode being inserted.
1361 */
1369 }
1371 /*
1372 * Pull the on-disk inode from the AGI unlinked list.
1373 */
1374 STATIC int
1378 {
1379 xfs_ino_t next_ino;
1385 xfs_agnumber_t agno;
1386 xfs_agino_t agino;
1387 xfs_agino_t next_agino;
1397 /*
1398 * Get the agi buffer first. It ensures lock ordering
1399 * on the list.
1400 */
1407 /*
1408 * Get the index into the agi hash table for the
1409 * list this inode will go on.
1410 */
1418 /*
1419 * We're at the head of the list. Get the inode's
1420 * on-disk buffer to see if there is anyone after us
1421 * on the list. Only modify our next pointer if it
1422 * is not already NULLAGINO. This saves us the overhead
1423 * of dealing with the buffer when there is no need to
1424 * change it.
1425 */
1431 }
1444 }
1445 /*
1446 * Point the bucket head pointer at the next inode.
1447 */
1456 /*
1457 * We need to search the list for the inode being freed.
1458 */
1462 /*
1463 * If the last inode wasn't the one pointing to
1464 * us, then release its buffer since we're not
1465 * going to do anything with it.
1466 */
1469 }
1478 }
1482 }
1483 /*
1484 * Now last_ibp points to the buffer previous to us on
1485 * the unlinked list. Pull us from the list.
1486 */
1492 }
1506 }
1507 /*
1508 * Point the previous inode on the list to the next inode.
1509 */
1517 }
1519 }
1521 /*
1522 * A big issue when freeing the inode cluster is is that we _cannot_ skip any
1523 * inodes that are in memory - they all must be marked stale and attached to
1524 * the cluster buffer.
1525 */
1526 STATIC int
1530 xfs_ino_t inum)
1531 {
1537 xfs_daddr_t blkno;
1554 }
1560 /*
1561 * We obtain and lock the backing buffer first in the process
1562 * here, as we have to ensure that any dirty inode that we
1563 * can't get the flush lock on is attached to the buffer.
1564 * If we scan the in-memory inodes first, then buffer IO can
1565 * complete before we get a lock on it, and hence we may fail
1566 * to mark all the active inodes on the buffer stale.
1567 */
1570 XBF_LOCK);
1574 /*
1575 * Walk the inodes already attached to the buffer and mark them
1576 * stale. These will all have the flush locks held, so an
1577 * in-memory inode walk can't lock them. By marking them all
1578 * stale first, we will not attempt to lock them in the loop
1579 * below as the XFS_ISTALE flag will be set.
1580 */
1591 }
1593 }
1596 /*
1597 * For each inode in memory attempt to add it to the inode
1598 * buffer and set it up for being staled on buffer IO
1599 * completion. This is safe as we've locked out tail pushing
1600 * and flushing by locking the buffer.
1601 *
1602 * We have already marked every inode that was part of a
1603 * transaction stale above, which means there is no point in
1604 * even trying to lock them.
1605 */
1607 retry:
1612 /* Inode not in memory, nothing to do */
1616 }
1618 /*
1619 * because this is an RCU protected lookup, we could
1620 * find a recently freed or even reallocated inode
1621 * during the lookup. We need to check under the
1622 * i_flags_lock for a valid inode here. Skip it if it
1623 * is not valid, the wrong inode or stale.
1624 */
1631 }
1634 /*
1635 * Don't try to lock/unlock the current inode, but we
1636 * _cannot_ skip the other inodes that we did not find
1637 * in the list attached to the buffer and are not
1638 * already marked stale. If we can't lock it, back off
1639 * and retry.
1640 */
1646 }
1652 /*
1653 * we don't need to attach clean inodes or those only
1654 * with unlogged changes (which we throw away, anyway).
1655 */
1663 }
1676 }
1680 }
1684 }
1686 /*
1687 * This is called to return an inode to the inode free list.
1688 * The inode should already be truncated to 0 length and have
1689 * no pages associated with it. This routine also assumes that
1690 * the inode is already a part of the transaction.
1691 *
1692 * The on-disk copy of the inode will have been added to the list
1693 * of unlinked inodes in the AGI. We need to remove the inode from
1694 * that list atomically with respect to freeing it here.
1695 */
1696 int
1701 {
1704 xfs_ino_t first_ino;
1715 /*
1716 * Pull the on-disk inode from the AGI unlinked list.
1717 */
1721 }
1726 }
1733 /*
1734 * Bump the generation count so no one will be confused
1735 * by reincarnations of this inode.
1736 */
1745 /*
1746 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
1747 * from picking up this inode when it is reclaimed (its incore state
1748 * initialzed but not flushed to disk yet). The in-core di_mode is
1749 * already cleared and a corresponding transaction logged.
1750 * The hack here just synchronizes the in-core to on-disk
1751 * di_mode value in advance before the actual inode sync to disk.
1752 * This is OK because the inode is already unlinked and would never
1753 * change its di_mode again for this inode generation.
1754 * This is a temporary hack that would require a proper fix
1755 * in the future.
1756 */
1761 }
1764 }
1766 /*
1767 * Reallocate the space for if_broot based on the number of records
1768 * being added or deleted as indicated in rec_diff. Move the records
1769 * and pointers in if_broot to fit the new size. When shrinking this
1770 * will eliminate holes between the records and pointers created by
1771 * the caller. When growing this will create holes to be filled in
1772 * by the caller.
1773 *
1774 * The caller must not request to add more records than would fit in
1775 * the on-disk inode root. If the if_broot is currently NULL, then
1776 * if we adding records one will be allocated. The caller must also
1777 * not request that the number of records go below zero, although
1778 * it can go to zero.
1779 *
1780 * ip -- the inode whose if_broot area is changing
1781 * ext_diff -- the change in the number of records, positive or negative,
1782 * requested for the if_broot array.
1783 */
1784 void
1789 {
1799 /*
1800 * Handle the degenerate case quietly.
1801 */
1804 }
1808 /*
1809 * If there wasn't any memory allocated before, just
1810 * allocate it now and get out.
1811 */
1817 }
1819 /*
1820 * If there is already an existing if_broot, then we need
1821 * to realloc() it and shift the pointers to their new
1822 * location. The records don't change location because
1823 * they are kept butted up against the btree block header.
1824 */
1840 }
1842 /*
1843 * rec_diff is less than 0. In this case, we are shrinking the
1844 * if_broot buffer. It must already exist. If we go to zero
1845 * records, just get rid of the root and clear the status bit.
1846 */
1853 else
1857 /*
1858 * First copy over the btree block header.
1859 */
1864 }
1866 /*
1867 * Only copy the records and pointers if there are any.
1868 */
1870 /*
1871 * First copy the records.
1872 */
1877 /*
1878 * Then copy the pointers.
1879 */
1885 }
1892 }
1895 /*
1896 * This is called when the amount of space needed for if_data
1897 * is increased or decreased. The change in size is indicated by
1898 * the number of bytes that need to be added or deleted in the
1899 * byte_diff parameter.
1900 *
1901 * If the amount of space needed has decreased below the size of the
1902 * inline buffer, then switch to using the inline buffer. Otherwise,
1903 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
1904 * to what is needed.
1905 *
1906 * ip -- the inode whose if_data area is changing
1907 * byte_diff -- the change in the number of bytes, positive or negative,
1908 * requested for the if_data array.
1909 */
1910 void
1915 {
1922 }
1931 }
1935 /*
1936 * If the valid extents/data can fit in if_inline_ext/data,
1937 * copy them from the malloc'd vector and free it.
1938 */
1944 new_size);
1947 }
1950 /*
1951 * Stuck with malloc/realloc.
1952 * For inline data, the underlying buffer must be
1953 * a multiple of 4 bytes in size so that it can be
1954 * logged and stay on word boundaries. We enforce
1955 * that here.
1956 */
1963 /*
1964 * Only do the realloc if the underlying size
1965 * is really changing.
1966 */
1970 real_size,
1973 }
1980 }
1981 }
1985 }
1987 void
1991 {
1998 }
2000 /*
2001 * If the format is local, then we can't have an extents
2002 * array so just look for an inline data array. If we're
2003 * not local then we may or may not have an extents list,
2004 * so check and free it up if we do.
2005 */
2013 }
2020 }
2027 }
2028 }
2030 /*
2031 * This is called to unpin an inode. The caller must have the inode locked
2032 * in at least shared mode so that the buffer cannot be subsequently pinned
2033 * once someone is waiting for it to be unpinned.
2034 */
2035 static void
2038 {
2043 /* Give the log a push to start the unpinning I/O */
2046 }
2048 static void
2051 {
2063 }
2065 void
2068 {
2071 }
2073 /*
2074 * xfs_iextents_copy()
2075 *
2076 * This is called to copy the REAL extents (as opposed to the delayed
2077 * allocation extents) from the inode into the given buffer. It
2078 * returns the number of bytes copied into the buffer.
2079 *
2080 * If there are no delayed allocation extents, then we can just
2081 * memcpy() the extents into the buffer. Otherwise, we need to
2082 * examine each extent in turn and skip those which are delayed.
2083 */
2084 int
2089 {
2094 xfs_fsblock_t start_block;
2104 /*
2105 * There are some delayed allocation extents in the
2106 * inode, so copy the extents one at a time and skip
2107 * the delayed ones. There must be at least one
2108 * non-delayed extent.
2109 */
2115 /*
2116 * It's a delayed allocation extent, so skip it.
2117 */
2119 }
2121 /* Translate to on disk format */
2124 dp++;
2125 copied++;
2126 }
2131 }
2133 /*
2134 * Each of the following cases stores data into the same region
2135 * of the on-disk inode, so only one of them can be valid at
2136 * any given time. While it is possible to have conflicting formats
2137 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2138 * in EXTENTS format, this can only happen when the fork has
2139 * changed formats after being modified but before being flushed.
2140 * In these cases, the format always takes precedence, because the
2141 * format indicates the current state of the fork.
2142 */
2143 /*ARGSUSED*/
2144 STATIC void
2151 {
2155 #ifdef XFS_TRANS_DEBUG
2157 #endif
2168 /*
2169 * This can happen if we gave up in iformat in an error path,
2170 * for the attribute fork.
2171 */
2175 }
2185 }
2196 whichfork);
2197 }
2206 XFS_BROOT_SIZE_ADJ));
2210 }
2217 }
2226 }
2232 }
2233 }
2235 STATIC int
2239 {
2263 /* really need a gang lookup range call here */
2274 /*
2275 * because this is an RCU protected lookup, we could find a
2276 * recently freed or even reallocated inode during the lookup.
2277 * We need to check under the i_flags_lock for a valid inode
2278 * here. Skip it if it is not valid or the wrong inode.
2279 */
2285 }
2288 /*
2289 * Do an un-protected check to see if the inode is dirty and
2290 * is a candidate for flushing. These checks will be repeated
2291 * later after the appropriate locks are acquired.
2292 */
2296 /*
2297 * Try to get locks. If any are unavailable or it is pinned,
2298 * then this inode cannot be flushed and is skipped.
2299 */
2306 }
2311 }
2313 /*
2314 * arriving here means that this inode can be flushed. First
2315 * re-check that it's dirty before flushing.
2316 */
2323 }
2324 clcount++;
2327 }
2329 }
2334 }
2336 out_free:
2339 out_put:
2344 cluster_corrupt_out:
2345 /*
2346 * Corruption detected in the clustering loop. Invalidate the
2347 * inode buffer and shut down the filesystem.
2348 */
2350 /*
2351 * Clean up the buffer. If it was B_DELWRI, just release it --
2352 * brelse can handle it with no problems. If not, shut down the
2353 * filesystem before releasing the buffer.
2354 */
2362 /*
2363 * Just like incore_relse: if we have b_iodone functions,
2364 * mark the buffer as an error and call them. Otherwise
2365 * mark it as stale and brelse.
2366 */
2375 }
2376 }
2378 /*
2379 * Unlocks the flush lock
2380 */
2385 }
2387 /*
2388 * xfs_iflush() will write a modified inode's changes out to the
2389 * inode's on disk home. The caller must have the inode lock held
2390 * in at least shared mode and the inode flush completion must be
2391 * active as well. The inode lock will still be held upon return from
2392 * the call and the caller is free to unlock it.
2393 * The inode flush will be completed when the inode reaches the disk.
2394 * The flags indicate how the inode's buffer should be written out.
2395 */
2396 int
2399 uint flags)
2400 {
2417 /*
2418 * We can't flush the inode until it is unpinned, so wait for it if we
2419 * are allowed to block. We know no one new can pin it, because we are
2420 * holding the inode lock shared and you need to hold it exclusively to
2421 * pin the inode.
2422 *
2423 * If we are not allowed to block, force the log out asynchronously so
2424 * that when we come back the inode will be unpinned. If other inodes
2425 * in the same cluster are dirty, they will probably write the inode
2426 * out for us if they occur after the log force completes.
2427 */
2432 }
2435 /*
2436 * For stale inodes we cannot rely on the backing buffer remaining
2437 * stale in cache for the remaining life of the stale inode and so
2438 * xfs_itobp() below may give us a buffer that no longer contains
2439 * inodes below. We have to check this after ensuring the inode is
2440 * unpinned so that it is safe to reclaim the stale inode after the
2441 * flush call.
2442 */
2446 }
2448 /*
2449 * This may have been unpinned because the filesystem is shutting
2450 * down forcibly. If that's the case we must not write this inode
2451 * to disk, because the log record didn't make it to disk!
2452 */
2459 }
2461 /*
2462 * Get the buffer containing the on-disk inode.
2463 */
2469 }
2471 /*
2472 * First flush out the inode that xfs_iflush was called with.
2473 */
2478 /*
2479 * If the buffer is pinned then push on the log now so we won't
2480 * get stuck waiting in the write for too long.
2481 */
2485 /*
2486 * inode clustering:
2487 * see if other inodes can be gathered into this write
2488 */
2495 else
2501 corrupt_out:
2504 cluster_corrupt_out:
2505 /*
2506 * Unlocks the flush lock
2507 */
2510 }
2513 STATIC int
2517 {
2521 #ifdef XFS_TRANS_DEBUG
2523 #endif
2533 /* set *dip = inode's place in the buffer */
2536 /*
2537 * Clear i_update_core before copying out the data.
2538 * This is for coordination with our timestamp updates
2539 * that don't hold the inode lock. They will always
2540 * update the timestamps BEFORE setting i_update_core,
2541 * so if we clear i_update_core after they set it we
2542 * are guaranteed to see their updates to the timestamps.
2543 * I believe that this depends on strongly ordered memory
2544 * semantics, but we have that. We use the SYNCHRONIZE
2545 * macro to make sure that the compiler does not reorder
2546 * the i_update_core access below the data copy below.
2547 */
2551 /*
2552 * Make sure to get the latest timestamps from the Linux inode.
2553 */
2562 }
2569 }
2579 }
2590 }
2591 }
2594 XFS_RANDOM_IFLUSH_5)) {
2596 "%s: detected corrupt incore inode %Lu, "
2602 }
2609 }
2610 /*
2611 * bump the flush iteration count, used to detect flushes which
2612 * postdate a log record during recovery.
2613 */
2617 /*
2618 * Copy the dirty parts of the inode into the on-disk
2619 * inode. We always copy out the core of the inode,
2620 * because if the inode is dirty at all the core must
2621 * be.
2622 */
2625 /* Wrap, we never let the log put out DI_MAX_FLUSH */
2629 /*
2630 * If this is really an old format inode and the superblock version
2631 * has not been updated to support only new format inodes, then
2632 * convert back to the old inode format. If the superblock version
2633 * has been updated, then make the conversion permanent.
2634 */
2638 /*
2639 * Convert it back.
2640 */
2644 /*
2645 * The superblock version has already been bumped,
2646 * so just make the conversion to the new inode
2647 * format permanent.
2648 */
2657 }
2658 }
2665 /*
2666 * We've recorded everything logged in the inode, so we'd
2667 * like to clear the ilf_fields bits so we don't log and
2668 * flush things unnecessarily. However, we can't stop
2669 * logging all this information until the data we've copied
2670 * into the disk buffer is written to disk. If we did we might
2671 * overwrite the copy of the inode in the log with all the
2672 * data after re-logging only part of it, and in the face of
2673 * a crash we wouldn't have all the data we need to recover.
2674 *
2675 * What we do is move the bits to the ili_last_fields field.
2676 * When logging the inode, these bits are moved back to the
2677 * ilf_fields field. In the xfs_iflush_done() routine we
2678 * clear ili_last_fields, since we know that the information
2679 * those bits represent is permanently on disk. As long as
2680 * the flush completes before the inode is logged again, then
2681 * both ilf_fields and ili_last_fields will be cleared.
2682 *
2683 * We can play with the ilf_fields bits here, because the inode
2684 * lock must be held exclusively in order to set bits there
2685 * and the flush lock protects the ili_last_fields bits.
2686 * Set ili_logged so the flush done
2687 * routine can tell whether or not to look in the AIL.
2688 * Also, store the current LSN of the inode so that we can tell
2689 * whether the item has moved in the AIL from xfs_iflush_done().
2690 * In order to read the lsn we need the AIL lock, because
2691 * it is a 64 bit value that cannot be read atomically.
2692 */
2701 /*
2702 * Attach the function xfs_iflush_done to the inode's
2703 * buffer. This will remove the inode from the AIL
2704 * and unlock the inode's flush lock when the inode is
2705 * completely written to disk.
2706 */
2712 /*
2713 * We're flushing an inode which is not in the AIL and has
2714 * not been logged but has i_update_core set. For this
2715 * case we can use a B_DELWRI flush and immediately drop
2716 * the inode flush lock because we can avoid the whole
2717 * AIL state thing. It's OK to drop the flush lock now,
2718 * because we've already locked the buffer and to do anything
2719 * you really need both.
2720 */
2725 }
2727 }
2731 corrupt_out:
2733 }
2735 void
2738 {
2751 }
2754 }
2756 /*
2757 * Return a pointer to the extent record at file index idx.
2758 */
2759 xfs_bmbt_rec_host_t *
2763 {
2780 }
2781 }
2783 /*
2784 * Insert new item(s) into the extent records for incore inode
2785 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
2786 */
2787 void
2794 {
2804 }
2806 /*
2807 * This is called when the amount of space required for incore file
2808 * extents needs to be increased. The ext_diff parameter stores the
2809 * number of new extents being added and the idx parameter contains
2810 * the extent index where the new extents will be added. If the new
2811 * extents are being appended, then we just need to (re)allocate and
2812 * initialize the space. Otherwise, if the new extents are being
2813 * inserted into the middle of the existing entries, a bit more work
2814 * is required to make room for the new extents to be inserted. The
2815 * caller is responsible for filling in the new extent entries upon
2816 * return.
2817 */
2818 void
2823 {
2832 /*
2833 * If the new number of extents (nextents + ext_diff)
2834 * fits inside the inode, then continue to use the inline
2835 * extent buffer.
2836 */
2843 }
2846 }
2847 /*
2848 * Otherwise use a linear (direct) extent list.
2849 * If the extents are currently inside the inode,
2850 * xfs_iext_realloc_direct will switch us from
2851 * inline to direct extent allocation mode.
2852 */
2860 }
2861 }
2862 /* Indirection array */
2875 }
2876 /* Extents fit in target extent page */
2884 }
2887 }
2888 /* Insert a new extent page */
2892 }
2893 /*
2894 * If extent(s) are being appended to the last page in
2895 * the indirection array and the new extent(s) don't fit
2896 * in the page, then erp is NULL and erp_idx is set to
2897 * the next index needed in the indirection array.
2898 */
2907 erp_idx++;
2908 }
2909 }
2910 }
2911 }
2913 }
2915 /*
2916 * This is called when incore extents are being added to the indirection
2917 * array and the new extents do not fit in the target extent list. The
2918 * erp_idx parameter contains the irec index for the target extent list
2919 * in the indirection array, and the idx parameter contains the extent
2920 * index within the list. The number of extents being added is stored
2921 * in the count parameter.
2922 *
2923 * |-------| |-------|
2924 * | | | | idx - number of extents before idx
2925 * | idx | | count |
2926 * | | | | count - number of extents being inserted at idx
2927 * |-------| |-------|
2928 * | count | | nex2 | nex2 - number of extents after idx + count
2929 * |-------| |-------|
2930 */
2931 void
2937 {
2951 /*
2952 * Save second part of target extent list
2953 * (all extents past */
2961 }
2963 /*
2964 * Add the new extents to the end of the target
2965 * list, then allocate new irec record(s) and
2966 * extent buffer(s) as needed to store the rest
2967 * of the new extents.
2968 */
2975 }
2977 erp_idx++;
2983 }
2985 /* Add nex2 extents back to indirection array */
2987 xfs_extnum_t ext_avail;
2993 /*
2994 * If nex2 extents fit in the current page, append
2995 * nex2_ep after the new extents.
2996 */
2999 }
3000 /*
3001 * Otherwise, check if space is available in the
3002 * next page.
3003 */
3007 erp_idx++;
3008 erp++;
3009 /* Create a hole for nex2 extents */
3012 }
3013 /*
3014 * Final choice, create a new extent page for
3015 * nex2 extents.
3016 */
3018 erp_idx++;
3020 }
3025 }
3026 }
3028 /*
3029 * This is called when the amount of space required for incore file
3030 * extents needs to be decreased. The ext_diff parameter stores the
3031 * number of extents to be removed and the idx parameter contains
3032 * the extent index where the extents will be removed from.
3033 *
3034 * If the amount of space needed has decreased below the linear
3035 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3036 * extent array. Otherwise, use kmem_realloc() to adjust the
3037 * size to what is needed.
3038 */
3039 void
3045 {
3064 }
3066 }
3068 /*
3069 * This removes ext_diff extents from the inline buffer, beginning
3070 * at extent index idx.
3071 */
3072 void
3077 {
3096 }
3097 }
3099 /*
3100 * This removes ext_diff extents from a linear (direct) extent list,
3101 * beginning at extent index idx. If the extents are being removed
3102 * from the end of the list (ie. truncate) then we just need to re-
3103 * allocate the list to remove the extra space. Otherwise, if the
3104 * extents are being removed from the middle of the existing extent
3105 * entries, then we first need to move the extent records beginning
3106 * at idx + ext_diff up in the list to overwrite the records being
3107 * removed, then remove the extra space via kmem_realloc.
3108 */
3109 void
3114 {
3126 }
3127 /* Move extents up in the list (if needed) */
3133 }
3136 /*
3137 * Reallocate the direct extent list. If the extents
3138 * will fit inside the inode then xfs_iext_realloc_direct
3139 * will switch from direct to inline extent allocation
3140 * mode for us.
3141 */
3144 }
3146 /*
3147 * This is called when incore extents are being removed from the
3148 * indirection array and the extents being removed span multiple extent
3149 * buffers. The idx parameter contains the file extent index where we
3150 * want to begin removing extents, and the count parameter contains
3151 * how many extents need to be removed.
3152 *
3153 * |-------| |-------|
3154 * | nex1 | | | nex1 - number of extents before idx
3155 * |-------| | count |
3156 * | | | | count - number of extents being removed at idx
3157 * | count | |-------|
3158 * | | | nex2 | nex2 - number of extents after idx + count
3159 * |-------| |-------|
3160 */
3161 void
3166 {
3183 /*
3184 * Check for deletion of entire list;
3185 * xfs_iext_irec_remove() updates extent offsets.
3186 */
3193 XFS_IEXT_BUFSZ);
3199 }
3200 }
3201 /* Move extents up (if needed) */
3206 }
3207 /* Zero out rest of page */
3210 /* Update remaining counters */
3215 erp_idx++;
3216 erp++;
3217 }
3220 }
3222 /*
3223 * Create, destroy, or resize a linear (direct) block of extents.
3224 */
3225 void
3229 {
3238 /* Free extent records */
3241 }
3242 /* Resize direct extent list and zero any new bytes */
3244 /* Check if extents will fit inside the inode */
3250 }
3253 }
3257 rnew_size,
3259 }
3264 }
3265 }
3266 /*
3267 * Switch from the inline extent buffer to a direct
3268 * extent list. Be sure to include the inline extent
3269 * bytes in new_size.
3270 */
3275 }
3277 }
3280 }
3282 /*
3283 * Switch from linear (direct) extent records to inline buffer.
3284 */
3285 void
3289 {
3292 /*
3293 * The inline buffer was zeroed when we switched
3294 * from inline to direct extent allocation mode,
3295 * so we don't need to clear it here.
3296 */
3302 }
3304 /*
3305 * Switch from inline buffer to linear (direct) extent records.
3306 * new_size should already be rounded up to the next power of 2
3307 * by the caller (when appropriate), so use new_size as it is.
3308 * However, since new_size may be rounded up, we can't update
3309 * if_bytes here. It is the caller's responsibility to update
3310 * if_bytes upon return.
3311 */
3312 void
3316 {
3324 }
3326 }
3328 /*
3329 * Resize an extent indirection array to new_size bytes.
3330 */
3331 STATIC void
3335 {
3350 }
3351 }
3353 /*
3354 * Switch from indirection array to linear (direct) extent allocations.
3355 */
3356 STATIC void
3359 {
3379 }
3380 }
3382 /*
3383 * Free incore file extents.
3384 */
3385 void
3388 {
3396 }
3403 }
3407 }
3409 /*
3410 * Return a pointer to the extent record for file system block bno.
3411 */
3417 {
3432 }
3435 /* Find target extent list */
3443 }
3444 /* Binary search extent records */
3455 /* Convert back to file-based extent index */
3458 }
3461 }
3462 }
3463 /* Convert back to file-based extent index */
3466 }
3472 }
3473 }
3476 }
3478 /*
3479 * Return a pointer to the indirection array entry containing the
3480 * extent record for filesystem block bno. Store the index of the
3481 * target irec in *erp_idxp.
3482 */
3488 {
3512 }
3513 }
3516 }
3518 /*
3519 * Return a pointer to the indirection array entry containing the
3520 * extent record at file extent index *idxp. Store the index of the
3521 * target irec in *erp_idxp and store the page index of the target
3522 * extent record in *idxp.
3523 */
3524 xfs_ext_irec_t *
3530 {
3549 /* Binary search extent irec's */
3565 erp_idx++;
3571 }
3572 }
3576 }
3578 /*
3579 * Allocate and initialize an indirection array once the space needed
3580 * for incore extents increases above XFS_IEXT_BUFSZ.
3581 */
3582 void
3585 {
3601 }
3612 }
3614 /*
3615 * Allocate and initialize a new entry in the indirection array.
3616 */
3617 xfs_ext_irec_t *
3621 {
3629 /* Resize indirection array */
3632 /*
3633 * Move records down in the array so the
3634 * new page can use erp_idx.
3635 */
3639 }
3642 /* Initialize new extent record */
3651 }
3653 /*
3654 * Remove a record from the indirection array.
3655 */
3656 void
3660 {
3672 }
3673 /* Compact extent records */
3677 }
3678 /*
3679 * Manually free the last extent record from the indirection
3680 * array. A call to xfs_iext_realloc_indirect() with a size
3681 * of zero would result in a call to xfs_iext_destroy() which
3682 * would in turn call this function again, creating a nasty
3683 * infinite loop.
3684 */
3690 }
3692 }
3694 /*
3695 * This is called to clean up large amounts of unused memory allocated
3696 * by the indirection array. Before compacting anything though, verify
3697 * that the indirection array is still needed and switch back to the
3698 * linear extent list (or even the inline buffer) if possible. The
3699 * compaction policy is as follows:
3700 *
3701 * Full Compaction: Extents fit into a single page (or inline buffer)
3702 * Partial Compaction: Extents occupy less than 50% of allocated space
3703 * No Compaction: Extents occupy at least 50% of allocated space
3704 */
3705 void
3708 {
3725 }
3726 }
3728 /*
3729 * Combine extents from neighboring extent pages.
3730 */
3731 void
3734 {
3750 /*
3751 * Free page before removing extent record
3752 * so er_extoffs don't get modified in
3753 * xfs_iext_irec_remove.
3754 */
3760 erp_idx++;
3761 }
3762 }
3763 }
3765 /*
3766 * This is called to update the er_extoff field in the indirection
3767 * array when extents have been added or removed from one of the
3768 * extent lists. erp_idx contains the irec index to begin updating
3769 * at and ext_diff contains the number of extents that were added
3770 * or removed.
3771 */
3772 void
3777 {
3785 }
3786 }