| blob | 6b7ac8bdcac24331d8629e9ba7d8e83ffb35a6fa |
1 /*
2 * Copyright (c) 2000-2003,2005 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 */
60 /*
61 * Used in xfs_itruncate(). This is the maximum number of extents
62 * freed from a file in a single transaction.
63 */
64 #define XFS_ITRUNC_MAX_EXTENTS 2
72 #ifdef DEBUG
73 /*
74 * Make sure that the extents in the given memory buffer
75 * are valid.
76 */
77 STATIC void
82 xfs_exntfmt_t fmt)
83 {
84 xfs_bmbt_irec_t irec;
85 xfs_bmbt_rec_t rec;
93 else
97 ep++;
98 }
99 }
101 #define xfs_validate_extents(ep, nrecs, disk, fmt)
104 /*
105 * Check that none of the inode's in the buffer have a next
106 * unlinked field of 0.
107 */
108 #if defined(DEBUG)
109 void
113 {
125 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
126 bp);
128 }
129 }
130 }
131 #endif
133 /*
134 * This routine is called to map an inode number within a file
135 * system to the buffer containing the on-disk version of the
136 * inode. It returns a pointer to the buffer containing the
137 * on-disk inode in the bpp parameter, and in the dip parameter
138 * it returns a pointer to the on-disk inode within that buffer.
139 *
140 * If a non-zero error is returned, then the contents of bpp and
141 * dipp are undefined.
142 *
143 * Use xfs_imap() to determine the size and location of the
144 * buffer to read from disk.
145 */
146 STATIC int
150 xfs_ino_t ino,
154 {
156 xfs_imap_t imap;
161 /*
162 * Call the space managment code to find the location of the
163 * inode on disk.
164 */
169 "xfs_inotobp: xfs_imap() returned an "
172 }
174 /*
175 * If the inode number maps to a block outside the bounds of the
176 * file system then return NULL rather than calling read_buf
177 * and panicing when we get an error from the driver.
178 */
182 "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
187 }
189 /*
190 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
191 * default to just a read_buf() call.
192 */
198 "xfs_inotobp: xfs_trans_read_buf() returned an "
201 }
203 di_ok =
207 XFS_RANDOM_ITOBP_INOTOBP))) {
211 "xfs_inotobp: XFS_TEST_ERROR() returned an "
214 }
218 /*
219 * Set *dipp to point to the on-disk inode in the buffer.
220 */
225 }
228 /*
229 * This routine is called to map an inode to the buffer containing
230 * the on-disk version of the inode. It returns a pointer to the
231 * buffer containing the on-disk inode in the bpp parameter, and in
232 * the dip parameter it returns a pointer to the on-disk inode within
233 * that buffer.
234 *
235 * If a non-zero error is returned, then the contents of bpp and
236 * dipp are undefined.
237 *
238 * If the inode is new and has not yet been initialized, use xfs_imap()
239 * to determine the size and location of the buffer to read from disk.
240 * If the inode has already been mapped to its buffer and read in once,
241 * then use the mapping information stored in the inode rather than
242 * calling xfs_imap(). This allows us to avoid the overhead of looking
243 * at the inode btree for small block file systems (see xfs_dilocate()).
244 * We can tell whether the inode has been mapped in before by comparing
245 * its disk block address to 0. Only uninitialized inodes will have
246 * 0 for the disk block address.
247 */
248 int
255 xfs_daddr_t bno)
256 {
259 xfs_imap_t imap;
260 #ifdef __KERNEL__
263 #endif
266 /*
267 * Call the space management code to find the location of the
268 * inode on disk.
269 */
274 }
276 /*
277 * If the inode number maps to a block outside the bounds
278 * of the file system then return NULL rather than calling
279 * read_buf and panicing when we get an error from the
280 * driver.
281 */
284 #ifdef DEBUG
286 "(imap.im_blkno (0x%llx) "
287 "+ imap.im_len (0x%llx)) > "
288 " XFS_FSB_TO_BB(mp, "
295 }
297 /*
298 * Fill in the fields in the inode that will be used to
299 * map the inode to its buffer from now on.
300 */
305 /*
306 * We've already mapped the inode once, so just use the
307 * mapping that we saved the first time.
308 */
312 }
315 /*
316 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
317 * default to just a read_buf() call.
318 */
323 #ifdef DEBUG
325 "xfs_trans_read_buf() returned error %d, "
331 }
332 #ifdef __KERNEL__
333 /*
334 * Validate the magic number and version of every inode in the buffer
335 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
336 */
337 #ifdef DEBUG
339 #else
341 #endif
351 XFS_RANDOM_ITOBP_INOTOBP))) {
352 #ifdef DEBUG
357 #endif
362 }
363 }
368 /*
369 * Mark the buffer as an inode buffer now that it looks good
370 */
373 /*
374 * Set *dipp to point to the on-disk inode in the buffer.
375 */
379 }
381 /*
382 * Move inode type and inode format specific information from the
383 * on-disk inode to the in-core inode. For fifos, devs, and sockets
384 * this means set if_rdev to the proper value. For files, directories,
385 * and symlinks this means to bring in the in-line data or extent
386 * pointers. For a file in B-tree format, only the root is immediately
387 * brought in-core. The rest will be in-lined in if_extents when it
388 * is first referenced (see xfs_iread_extents()).
389 */
390 STATIC int
394 {
398 xfs_fsize_t di_size;
408 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu."
418 }
420 if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) {
422 "corrupt dinode %Lu, forkoff = 0x%x."
429 }
440 }
450 /*
451 * no local regular files yet
452 */
458 XFS_ERRLEVEL_LOW,
461 }
470 XFS_ERRLEVEL_LOW,
473 }
488 }
494 }
497 }
519 }
524 }
526 }
528 /*
529 * The file is in-lined in the on-disk inode.
530 * If it fits into if_inline_data, then copy
531 * it there, otherwise allocate a buffer for it
532 * and copy the data there. Either way, set
533 * if_data to point at the data.
534 * If we allocate a buffer for the data, make
535 * sure that its size is a multiple of 4 and
536 * record the real size in i_real_bytes.
537 */
538 STATIC int
544 {
548 /*
549 * If the size is unreasonable, then something
550 * is wrong and we just bail out rather than crash in
551 * kmem_alloc() or memcpy() below.
552 */
561 }
571 }
579 }
581 /*
582 * The file consists of a set of extents all
583 * of which fit into the on-disk inode.
584 * If there are few enough extents to fit into
585 * the if_inline_ext, then copy them there.
586 * Otherwise allocate a buffer for them and copy
587 * them into it. Either way, set if_extents
588 * to point at the extents.
589 */
590 STATIC int
595 {
607 /*
608 * If the number of extents is unreasonable, then something
609 * is wrong and we just bail out rather than crash in
610 * kmem_alloc() or memcpy() below.
611 */
619 }
630 }
639 ARCH_CONVERT);
641 ARCH_CONVERT);
642 }
644 whichfork);
650 XFS_ERRLEVEL_LOW,
653 }
654 }
657 }
659 /*
660 * The file has too many extents to fit into
661 * the inode, so they are in B-tree format.
662 * Allocate a buffer for the root of the B-tree
663 * and copy the root into it. The i_extents
664 * field will remain NULL until all of the
665 * extents are read in (when they are needed).
666 */
667 STATIC int
672 {
675 /* REFERENCED */
684 /*
685 * blow out if -- fork has less extents than can fit in
686 * fork (fork shouldn't be a btree format), root btree
687 * block has more records than can fit into the fork,
688 * or the number of extents is greater than the number of
689 * blocks.
690 */
701 }
706 /*
707 * Copy and convert from the on-disk structure
708 * to the in-memory structure.
709 */
716 }
718 /*
719 * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
720 * and native format
721 *
722 * buf = on-disk representation
723 * dip = native representation
724 * dir = direction - +ve -> disk to native
725 * -ve -> native to disk
726 */
727 void
729 xfs_caddr_t buf,
732 {
755 }
782 }
784 STATIC uint
787 __uint16_t di_flags)
788 {
816 }
819 }
821 uint
824 {
829 }
831 uint
834 {
837 }
839 /*
840 * Given a mount structure and an inode number, return a pointer
841 * to a newly allocated in-core inode coresponding to the given
842 * inode number.
843 *
844 * Initialize the inode's attributes and extent pointers if it
845 * already has them (it will not if the inode has no links).
846 */
847 int
851 xfs_ino_t ino,
853 xfs_daddr_t bno)
854 {
866 /*
867 * Get pointer's to the on-disk inode and the buffer containing it.
868 * If the inode number refers to a block outside the file system
869 * then xfs_itobp() will return NULL. In this case we should
870 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
871 * know that this is a new incore inode.
872 */
878 }
880 /*
881 * Initialize inode's trace buffers.
882 * Do this before xfs_iformat in case it adds entries.
883 */
884 #ifdef XFS_BMAP_TRACE
886 #endif
887 #ifdef XFS_BMBT_TRACE
889 #endif
890 #ifdef XFS_RW_TRACE
892 #endif
893 #ifdef XFS_ILOCK_TRACE
895 #endif
896 #ifdef XFS_DIR2_TRACE
898 #endif
900 /*
901 * If we got something that isn't an inode it means someone
902 * (nfs or dmi) has a stale handle.
903 */
907 #ifdef DEBUG
909 "dip->di_core.di_magic (0x%x) != "
912 XFS_DINODE_MAGIC);
915 }
917 /*
918 * If the on-disk inode is already linked to a directory
919 * entry, copy all of the inode into the in-core inode.
920 * xfs_iformat() handles copying in the inode format
921 * specific information.
922 * Otherwise, just get the truly permanent information.
923 */
931 #ifdef DEBUG
934 error);
937 }
943 /*
944 * Make sure to pull in the mode here as well in
945 * case the inode is released without being used.
946 * This ensures that xfs_inactive() will see that
947 * the inode is already free and not try to mess
948 * with the uninitialized part of it.
949 */
951 /*
952 * Initialize the per-fork minima and maxima for a new
953 * inode here. xfs_iformat will do it for old inodes.
954 */
957 }
961 /*
962 * The inode format changed when we moved the link count and
963 * made it 32 bits long. If this is an old format inode,
964 * convert it in memory to look like a new one. If it gets
965 * flushed to disk we will convert back before flushing or
966 * logging it. We zero out the new projid field and the old link
967 * count field. We'll handle clearing the pad field (the remains
968 * of the old uuid field) when we actually convert the inode to
969 * the new format. We don't change the version number so that we
970 * can distinguish this from a real new format inode.
971 */
976 }
980 /*
981 * Mark the buffer containing the inode as something to keep
982 * around for a while. This helps to keep recently accessed
983 * meta-data in-core longer.
984 */
987 /*
988 * Use xfs_trans_brelse() to release the buffer containing the
989 * on-disk inode, because it was acquired with xfs_trans_read_buf()
990 * in xfs_itobp() above. If tp is NULL, this is just a normal
991 * brelse(). If we're within a transaction, then xfs_trans_brelse()
992 * will only release the buffer if it is not dirty within the
993 * transaction. It will be OK to release the buffer in this case,
994 * because inodes on disk are never destroyed and we will be
995 * locking the new in-core inode before putting it in the hash
996 * table where other processes can find it. Thus we don't have
997 * to worry about the inode being changed just because we released
998 * the buffer.
999 */
1003 }
1005 /*
1006 * Read in extents from a btree-format inode.
1007 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1008 */
1009 int
1014 {
1023 }
1026 /*
1027 * We know that the size is valid (it's checked in iformat_btree)
1028 */
1041 }
1045 }
1047 /*
1048 * Allocate an inode on disk and return a copy of its in-core version.
1049 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1050 * appropriately within the inode. The uid and gid for the inode are
1051 * set according to the contents of the given cred structure.
1052 *
1053 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1054 * has a free inode available, call xfs_iget()
1055 * to obtain the in-core version of the allocated inode. Finally,
1056 * fill in the inode and log its initial contents. In this case,
1057 * ialloc_context would be set to NULL and call_again set to false.
1058 *
1059 * If xfs_dialloc() does not have an available inode,
1060 * it will replenish its supply by doing an allocation. Since we can
1061 * only do one allocation within a transaction without deadlocks, we
1062 * must commit the current transaction before returning the inode itself.
1063 * In this case, therefore, we will set call_again to true and return.
1064 * The caller should then commit the current transaction, start a new
1065 * transaction, and call xfs_ialloc() again to actually get the inode.
1066 *
1067 * To ensure that some other process does not grab the inode that
1068 * was allocated during the first call to xfs_ialloc(), this routine
1069 * also returns the [locked] bp pointing to the head of the freelist
1070 * as ialloc_context. The caller should hold this buffer across
1071 * the commit and pass it back into this routine on the second call.
1072 */
1073 int
1077 mode_t mode,
1078 xfs_nlink_t nlink,
1079 xfs_dev_t rdev,
1081 xfs_prid_t prid,
1086 {
1087 xfs_ino_t ino;
1090 uint flags;
1093 /*
1094 * Call the space management code to pick
1095 * the on-disk inode to be allocated.
1096 */
1101 }
1105 }
1108 /*
1109 * Get the in-core inode with the lock held exclusively.
1110 * This is because we're setting fields here we need
1111 * to prevent others from looking at until we're done.
1112 */
1117 }
1130 /*
1131 * If the superblock version is up to where we support new format
1132 * inodes and this is currently an old format inode, then change
1133 * the inode version number now. This way we only do the conversion
1134 * here rather than here and in the flush/logging code.
1135 */
1139 /*
1140 * We've already zeroed the old link count, the projid field,
1141 * and the pad field.
1142 */
1143 }
1145 /*
1146 * Project ids won't be stored on disk if we are using a version 1 inode.
1147 */
1155 }
1156 }
1158 /*
1159 * If the group ID of the new file does not match the effective group
1160 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1161 * (and only if the irix_sgid_inherit compatibility variable is set).
1162 */
1167 }
1173 /*
1174 * di_gen will have been taken care of in xfs_iread.
1175 */
1202 }
1207 }
1211 }
1212 }
1214 xfs_inherit_noatime)
1217 xfs_inherit_nodump)
1220 xfs_inherit_sync)
1223 xfs_inherit_nosymlinks)
1228 }
1229 /* FALLTHROUGH */
1238 }
1239 /*
1240 * Attribute fork settings for new inode.
1241 */
1245 /*
1246 * Log the new values stuffed into the inode.
1247 */
1250 /* now that we have an i_mode we can set Linux inode ops (& unlock) */
1255 }
1257 /*
1258 * Check to make sure that there are no blocks allocated to the
1259 * file beyond the size of the file. We don't check this for
1260 * files with fixed size extents or real time extents, but we
1261 * at least do it for regular files.
1262 */
1263 #ifdef DEBUG
1264 void
1268 xfs_fsize_t isize)
1269 {
1270 xfs_fileoff_t map_first;
1282 /*
1283 * The filesystem could be shutting down, so bmapi may return
1284 * an error.
1285 */
1289 map_first),
1291 NULL))
1295 }
1298 /*
1299 * Calculate the last possible buffered byte in a file. This must
1300 * include data that was buffered beyond the EOF by the write code.
1301 * This also needs to deal with overflowing the xfs_fsize_t type
1302 * which can happen for sizes near the limit.
1303 *
1304 * We also need to take into account any blocks beyond the EOF. It
1305 * may be the case that they were buffered by a write which failed.
1306 * In that case the pages will still be in memory, but the inode size
1307 * will never have been updated.
1308 */
1309 xfs_fsize_t
1312 {
1314 xfs_fsize_t last_byte;
1315 xfs_fileoff_t last_block;
1316 xfs_fileoff_t size_last_block;
1322 /*
1323 * Only check for blocks beyond the EOF if the extents have
1324 * been read in. This eliminates the need for the inode lock,
1325 * and it also saves us from looking when it really isn't
1326 * necessary.
1327 */
1330 XFS_DATA_FORK);
1333 }
1336 }
1343 }
1347 }
1349 }
1351 #if defined(XFS_RW_TRACE)
1352 STATIC void
1357 xfs_fsize_t new_size,
1358 xfs_off_t toss_start,
1359 xfs_off_t toss_finish)
1360 {
1363 }
1382 }
1383 #else
1384 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1385 #endif
1387 /*
1388 * Start the truncation of the file to new_size. The new size
1389 * must be smaller than the current size. This routine will
1390 * clear the buffer and page caches of file data in the removed
1391 * range, and xfs_itruncate_finish() will remove the underlying
1392 * disk blocks.
1393 *
1394 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1395 * must NOT have the inode lock held at all. This is because we're
1396 * calling into the buffer/page cache code and we can't hold the
1397 * inode lock when we do so.
1398 *
1399 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1400 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1401 * in the case that the caller is locking things out of order and
1402 * may not be able to call xfs_itruncate_finish() with the inode lock
1403 * held without dropping the I/O lock. If the caller must drop the
1404 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1405 * must be called again with all the same restrictions as the initial
1406 * call.
1407 */
1408 void
1411 uint flags,
1412 xfs_fsize_t new_size)
1413 {
1414 xfs_fsize_t last_byte;
1415 xfs_off_t toss_start;
1426 /*
1427 * Call VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES() to get rid of pages and buffers
1428 * overlapping the region being removed. We have to use
1429 * the less efficient VOP_FLUSHINVAL_PAGES() in the case that the
1430 * caller may not be able to finish the truncate without
1431 * dropping the inode's I/O lock. Make sure
1432 * to catch any pages brought in by buffers overlapping
1433 * the EOF by searching out beyond the isize by our
1434 * block size. We round new_size up to a block boundary
1435 * so that we don't toss things on the same block as
1436 * new_size but before it.
1437 *
1438 * Before calling VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES(), make sure to
1439 * call remapf() over the same region if the file is mapped.
1440 * This frees up mapped file references to the pages in the
1441 * given range and for the VOP_FLUSHINVAL_PAGES() case it ensures
1442 * that we get the latest mapped changes flushed out.
1443 */
1447 /*
1448 * The place to start tossing is beyond our maximum
1449 * file size, so there is no way that the data extended
1450 * out there.
1451 */
1453 }
1456 last_byte);
1462 }
1463 }
1465 #ifdef DEBUG
1468 }
1469 #endif
1470 }
1472 /*
1473 * Shrink the file to the given new_size. The new
1474 * size must be smaller than the current size.
1475 * This will free up the underlying blocks
1476 * in the removed range after a call to xfs_itruncate_start()
1477 * or xfs_atruncate_start().
1478 *
1479 * The transaction passed to this routine must have made
1480 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1481 * This routine may commit the given transaction and
1482 * start new ones, so make sure everything involved in
1483 * the transaction is tidy before calling here.
1484 * Some transaction will be returned to the caller to be
1485 * committed. The incoming transaction must already include
1486 * the inode, and both inode locks must be held exclusively.
1487 * The inode must also be "held" within the transaction. On
1488 * return the inode will be "held" within the returned transaction.
1489 * This routine does NOT require any disk space to be reserved
1490 * for it within the transaction.
1491 *
1492 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1493 * and it indicates the fork which is to be truncated. For the
1494 * attribute fork we only support truncation to size 0.
1495 *
1496 * We use the sync parameter to indicate whether or not the first
1497 * transaction we perform might have to be synchronous. For the attr fork,
1498 * it needs to be so if the unlink of the inode is not yet known to be
1499 * permanent in the log. This keeps us from freeing and reusing the
1500 * blocks of the attribute fork before the unlink of the inode becomes
1501 * permanent.
1502 *
1503 * For the data fork, we normally have to run synchronously if we're
1504 * being called out of the inactive path or we're being called
1505 * out of the create path where we're truncating an existing file.
1506 * Either way, the truncate needs to be sync so blocks don't reappear
1507 * in the file with altered data in case of a crash. wsync filesystems
1508 * can run the first case async because anything that shrinks the inode
1509 * has to run sync so by the time we're called here from inactive, the
1510 * inode size is permanently set to 0.
1511 *
1512 * Calls from the truncate path always need to be sync unless we're
1513 * in a wsync filesystem and the file has already been unlinked.
1514 *
1515 * The caller is responsible for correctly setting the sync parameter.
1516 * It gets too hard for us to guess here which path we're being called
1517 * out of just based on inode state.
1518 */
1519 int
1523 xfs_fsize_t new_size,
1526 {
1527 xfs_fsblock_t first_block;
1528 xfs_fileoff_t first_unmap_block;
1529 xfs_fileoff_t last_block;
1535 xfs_bmap_free_t free_list;
1552 /*
1553 * We only support truncating the entire attribute fork.
1554 */
1557 }
1560 /*
1561 * The first thing we do is set the size to new_size permanently
1562 * on disk. This way we don't have to worry about anyone ever
1563 * being able to look at the data being freed even in the face
1564 * of a crash. What we're getting around here is the case where
1565 * we free a block, it is allocated to another file, it is written
1566 * to, and then we crash. If the new data gets written to the
1567 * file but the log buffers containing the free and reallocation
1568 * don't, then we'd end up with garbage in the blocks being freed.
1569 * As long as we make the new_size permanent before actually
1570 * freeing any blocks it doesn't matter if they get writtten to.
1571 *
1572 * The callers must signal into us whether or not the size
1573 * setting here must be synchronous. There are a few cases
1574 * where it doesn't have to be synchronous. Those cases
1575 * occur if the file is unlinked and we know the unlink is
1576 * permanent or if the blocks being truncated are guaranteed
1577 * to be beyond the inode eof (regardless of the link count)
1578 * and the eof value is permanent. Both of these cases occur
1579 * only on wsync-mounted filesystems. In those cases, we're
1580 * guaranteed that no user will ever see the data in the blocks
1581 * that are being truncated so the truncate can run async.
1582 * In the free beyond eof case, the file may wind up with
1583 * more blocks allocated to it than it needs if we crash
1584 * and that won't get fixed until the next time the file
1585 * is re-opened and closed but that's ok as that shouldn't
1586 * be too many blocks.
1587 *
1588 * However, we can't just make all wsync xactions run async
1589 * because there's one call out of the create path that needs
1590 * to run sync where it's truncating an existing file to size
1591 * 0 whose size is > 0.
1592 *
1593 * It's probably possible to come up with a test in this
1594 * routine that would correctly distinguish all the above
1595 * cases from the values of the function parameters and the
1596 * inode state but for sanity's sake, I've decided to let the
1597 * layers above just tell us. It's simpler to correctly figure
1598 * out in the layer above exactly under what conditions we
1599 * can run async and I think it's easier for others read and
1600 * follow the logic in case something has to be changed.
1601 * cscope is your friend -- rcc.
1602 *
1603 * The attribute fork is much simpler.
1604 *
1605 * For the attribute fork we allow the caller to tell us whether
1606 * the unlink of the inode that led to this call is yet permanent
1607 * in the on disk log. If it is not and we will be freeing extents
1608 * in this inode then we make the first transaction synchronous
1609 * to make sure that the unlink is permanent by the time we free
1610 * the blocks.
1611 */
1616 }
1621 }
1627 /*
1628 * Since it is possible for space to become allocated beyond
1629 * the end of the file (in a crash where the space is allocated
1630 * but the inode size is not yet updated), simply remove any
1631 * blocks which show up between the new EOF and the maximum
1632 * possible file size. If the first block to be removed is
1633 * beyond the maximum file size (ie it is the same as last_block),
1634 * then there is nothing to do.
1635 */
1643 }
1645 /*
1646 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1647 * will tell us whether it freed the entire range or
1648 * not. If this is a synchronous mount (wsync),
1649 * then we can tell bunmapi to keep all the
1650 * transactions asynchronous since the unlink
1651 * transaction that made this inode inactive has
1652 * already hit the disk. There's no danger of
1653 * the freed blocks being reused, there being a
1654 * crash, and the reused blocks suddenly reappearing
1655 * in this file with garbage in them once recovery
1656 * runs.
1657 */
1660 unmap_len,
1663 XFS_ITRUNC_MAX_EXTENTS,
1666 /*
1667 * If the bunmapi call encounters an error,
1668 * return to the caller where the transaction
1669 * can be properly aborted. We just need to
1670 * make sure we're not holding any resources
1671 * that we were not when we came in.
1672 */
1675 }
1677 /*
1678 * Duplicate the transaction that has the permanent
1679 * reservation and commit the old transaction.
1680 */
1685 /*
1686 * If the bmap finish call encounters an error,
1687 * return to the caller where the transaction
1688 * can be properly aborted. We just need to
1689 * make sure we're not holding any resources
1690 * that we were not when we came in.
1691 *
1692 * Aborting from this point might lose some
1693 * blocks in the file system, but oh well.
1694 */
1697 /*
1698 * If the passed in transaction committed
1699 * in xfs_bmap_finish(), then we want to
1700 * add the inode to this one before returning.
1701 * This keeps things simple for the higher
1702 * level code, because it always knows that
1703 * the inode is locked and held in the
1704 * transaction that returns to it whether
1705 * errors occur or not. We don't mark the
1706 * inode dirty so that this transaction can
1707 * be easily aborted if possible.
1708 */
1712 }
1714 }
1717 /*
1718 * The first xact was committed,
1719 * so add the inode to the new one.
1720 * Mark it dirty so it will be logged
1721 * and moved forward in the log as
1722 * part of every commit.
1723 */
1728 }
1733 XFS_TRANS_PERM_LOG_RES,
1734 XFS_ITRUNCATE_LOG_COUNT);
1735 /*
1736 * Add the inode being truncated to the next chained
1737 * transaction.
1738 */
1743 }
1744 /*
1745 * Only update the size in the case of the data fork, but
1746 * always re-log the inode so that our permanent transaction
1747 * can keep on rolling it forward in the log.
1748 */
1752 }
1762 }
1765 /*
1766 * xfs_igrow_start
1767 *
1768 * Do the first part of growing a file: zero any data in the last
1769 * block that is beyond the old EOF. We need to do this before
1770 * the inode is joined to the transaction to modify the i_size.
1771 * That way we can drop the inode lock and call into the buffer
1772 * cache to get the buffer mapping the EOF.
1773 */
1774 int
1777 xfs_fsize_t new_size,
1779 {
1786 /*
1787 * Zero any pages that may have been created by
1788 * xfs_write_file() beyond the end of the file
1789 * and any blocks between the old and new file sizes.
1790 */
1794 }
1796 /*
1797 * xfs_igrow_finish
1798 *
1799 * This routine is called to extend the size of a file.
1800 * The inode must have both the iolock and the ilock locked
1801 * for update and it must be a part of the current transaction.
1802 * The xfs_igrow_start() function must have been called previously.
1803 * If the change_flag is not zero, the inode change timestamp will
1804 * be updated.
1805 */
1806 void
1810 xfs_fsize_t new_size,
1812 {
1818 /*
1819 * Update the file size. Update the inode change timestamp
1820 * if change_flag set.
1821 */
1827 }
1830 /*
1831 * This is called when the inode's link count goes to 0.
1832 * We place the on-disk inode on a list in the AGI. It
1833 * will be pulled from this list when the inode is freed.
1834 */
1835 int
1839 {
1845 xfs_agnumber_t agno;
1846 xfs_daddr_t agdaddr;
1847 xfs_agino_t agino;
1862 /*
1863 * Get the agi buffer first. It ensures lock ordering
1864 * on the list.
1865 */
1870 }
1871 /*
1872 * Validate the magic number of the agi block.
1873 */
1875 agi_ok =
1879 XFS_RANDOM_IUNLINK))) {
1883 }
1884 /*
1885 * Get the index into the agi hash table for the
1886 * list this inode will go on.
1887 */
1895 /*
1896 * There is already another inode in the bucket we need
1897 * to add ourselves to. Add us at the front of the list.
1898 * Here we put the head pointer into our next pointer,
1899 * and then we fall through to point the head at us.
1900 */
1904 }
1907 /* both on-disk, don't endian flip twice */
1915 }
1917 /*
1918 * Point the bucket head pointer at the inode being inserted.
1919 */
1927 }
1929 /*
1930 * Pull the on-disk inode from the AGI unlinked list.
1931 */
1932 STATIC int
1936 {
1937 xfs_ino_t next_ino;
1943 xfs_agnumber_t agno;
1944 xfs_daddr_t agdaddr;
1945 xfs_agino_t agino;
1946 xfs_agino_t next_agino;
1954 /*
1955 * First pull the on-disk inode from the AGI unlinked list.
1956 */
1962 /*
1963 * Get the agi buffer first. It ensures lock ordering
1964 * on the list.
1965 */
1973 }
1974 /*
1975 * Validate the magic number of the agi block.
1976 */
1978 agi_ok =
1982 XFS_RANDOM_IUNLINK_REMOVE))) {
1990 }
1991 /*
1992 * Get the index into the agi hash table for the
1993 * list this inode will go on.
1994 */
2002 /*
2003 * We're at the head of the list. Get the inode's
2004 * on-disk buffer to see if there is anyone after us
2005 * on the list. Only modify our next pointer if it
2006 * is not already NULLAGINO. This saves us the overhead
2007 * of dealing with the buffer when there is no need to
2008 * change it.
2009 */
2016 }
2029 }
2030 /*
2031 * Point the bucket head pointer at the next inode.
2032 */
2041 /*
2042 * We need to search the list for the inode being freed.
2043 */
2047 /*
2048 * If the last inode wasn't the one pointing to
2049 * us, then release its buffer since we're not
2050 * going to do anything with it.
2051 */
2054 }
2063 }
2067 }
2068 /*
2069 * Now last_ibp points to the buffer previous to us on
2070 * the unlinked list. Pull us from the list.
2071 */
2078 }
2092 }
2093 /*
2094 * Point the previous inode on the list to the next inode.
2095 */
2103 }
2105 }
2108 {
2112 }
2114 STATIC void
2118 xfs_ino_t inum)
2119 {
2125 xfs_daddr_t blkno;
2142 }
2151 /*
2152 * Look for each inode in memory and attempt to lock it,
2153 * we can be racing with flush and tail pushing here.
2154 * any inode we get the locks on, add to an array of
2155 * inode items to process later.
2156 *
2157 * The get the buffer lock, we could beat a flush
2158 * or tail pushing thread to the lock here, in which
2159 * case they will go looking for the inode buffer
2160 * and fail, we need some other form of interlock
2161 * here.
2162 */
2170 }
2172 /* Inode not in memory or we found it already,
2173 * nothing to do
2174 */
2178 }
2183 }
2185 /* If we can get the locks then add it to the
2186 * list, otherwise by the time we get the bp lock
2187 * below it will already be attached to the
2188 * inode buffer.
2189 */
2191 /* This inode will already be locked - by us, lets
2192 * keep it that way.
2193 */
2203 }
2204 }
2207 }
2218 }
2221 }
2222 }
2225 }
2229 XFS_BUF_LOCK);
2242 pre_flushed++;
2243 }
2245 }
2256 }
2270 }
2271 }
2276 }
2279 }
2281 /*
2282 * This is called to return an inode to the inode free list.
2283 * The inode should already be truncated to 0 length and have
2284 * no pages associated with it. This routine also assumes that
2285 * the inode is already a part of the transaction.
2286 *
2287 * The on-disk copy of the inode will have been added to the list
2288 * of unlinked inodes in the AGI. We need to remove the inode from
2289 * that list atomically with respect to freeing it here.
2290 */
2291 int
2296 {
2299 xfs_ino_t first_ino;
2310 /*
2311 * Pull the on-disk inode from the AGI unlinked list.
2312 */
2316 }
2321 }
2330 /*
2331 * Bump the generation count so no one will be confused
2332 * by reincarnations of this inode.
2333 */
2339 }
2342 }
2344 /*
2345 * Reallocate the space for if_broot based on the number of records
2346 * being added or deleted as indicated in rec_diff. Move the records
2347 * and pointers in if_broot to fit the new size. When shrinking this
2348 * will eliminate holes between the records and pointers created by
2349 * the caller. When growing this will create holes to be filled in
2350 * by the caller.
2351 *
2352 * The caller must not request to add more records than would fit in
2353 * the on-disk inode root. If the if_broot is currently NULL, then
2354 * if we adding records one will be allocated. The caller must also
2355 * not request that the number of records go below zero, although
2356 * it can go to zero.
2357 *
2358 * ip -- the inode whose if_broot area is changing
2359 * ext_diff -- the change in the number of records, positive or negative,
2360 * requested for the if_broot array.
2361 */
2362 void
2367 {
2376 /*
2377 * Handle the degenerate case quietly.
2378 */
2381 }
2385 /*
2386 * If there wasn't any memory allocated before, just
2387 * allocate it now and get out.
2388 */
2392 KM_SLEEP);
2395 }
2397 /*
2398 * If there is already an existing if_broot, then we need
2399 * to realloc() it and shift the pointers to their new
2400 * location. The records don't change location because
2401 * they are kept butted up against the btree block header.
2402 */
2408 new_size,
2410 KM_SLEEP);
2420 }
2422 /*
2423 * rec_diff is less than 0. In this case, we are shrinking the
2424 * if_broot buffer. It must already exist. If we go to zero
2425 * records, just get rid of the root and clear the status bit.
2426 */
2433 else
2437 /*
2438 * First copy over the btree block header.
2439 */
2444 }
2446 /*
2447 * Only copy the records and pointers if there are any.
2448 */
2450 /*
2451 * First copy the records.
2452 */
2459 /*
2460 * Then copy the pointers.
2461 */
2467 }
2474 }
2477 /*
2478 * This is called when the amount of space needed for if_extents
2479 * is increased or decreased. The change in size is indicated by
2480 * the number of extents that need to be added or deleted in the
2481 * ext_diff parameter.
2482 *
2483 * If the amount of space needed has decreased below the size of the
2484 * inline buffer, then switch to using the inline buffer. Otherwise,
2485 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2486 * to what is needed.
2487 *
2488 * ip -- the inode whose if_extents area is changing
2489 * ext_diff -- the change in the number of extents, positive or negative,
2490 * requested for the if_extents array.
2491 */
2492 void
2497 {
2501 uint rnew_size;
2505 }
2516 }
2520 /*
2521 * If the valid extents can fit in if_inline_ext,
2522 * copy them from the malloc'd vector and free it.
2523 */
2525 /*
2526 * For now, empty files are format EXTENTS,
2527 * so the if_extents pointer is null.
2528 */
2534 }
2536 }
2542 /*
2543 * Stuck with malloc/realloc.
2544 */
2553 rnew_size,
2555 KM_NOFS);
2556 }
2557 }
2560 }
2563 /*
2564 * This is called when the amount of space needed for if_data
2565 * is increased or decreased. The change in size is indicated by
2566 * the number of bytes that need to be added or deleted in the
2567 * byte_diff parameter.
2568 *
2569 * If the amount of space needed has decreased below the size of the
2570 * inline buffer, then switch to using the inline buffer. Otherwise,
2571 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2572 * to what is needed.
2573 *
2574 * ip -- the inode whose if_data area is changing
2575 * byte_diff -- the change in the number of bytes, positive or negative,
2576 * requested for the if_data array.
2577 */
2578 void
2583 {
2590 }
2599 }
2603 /*
2604 * If the valid extents/data can fit in if_inline_ext/data,
2605 * copy them from the malloc'd vector and free it.
2606 */
2612 new_size);
2615 }
2618 /*
2619 * Stuck with malloc/realloc.
2620 * For inline data, the underlying buffer must be
2621 * a multiple of 4 bytes in size so that it can be
2622 * logged and stay on word boundaries. We enforce
2623 * that here.
2624 */
2630 /*
2631 * Only do the realloc if the underlying size
2632 * is really changing.
2633 */
2637 real_size,
2639 KM_SLEEP);
2640 }
2646 }
2647 }
2651 }
2656 /*
2657 * Map inode to disk block and offset.
2658 *
2659 * mp -- the mount point structure for the current file system
2660 * tp -- the current transaction
2661 * ino -- the inode number of the inode to be located
2662 * imap -- this structure is filled in with the information necessary
2663 * to retrieve the given inode from disk
2664 * flags -- flags to pass to xfs_dilocate indicating whether or not
2665 * lookups in the inode btree were OK or not
2666 */
2667 int
2671 xfs_ino_t ino,
2673 uint flags)
2674 {
2675 xfs_fsblock_t fsbno;
2685 }
2692 }
2694 void
2698 {
2705 }
2707 /*
2708 * If the format is local, then we can't have an extents
2709 * array so just look for an inline data array. If we're
2710 * not local then we may or may not have an extents list,
2711 * so check and free it up if we do.
2712 */
2720 }
2728 }
2735 }
2736 }
2738 /*
2739 * This is called free all the memory associated with an inode.
2740 * It must free the inode itself and any buffers allocated for
2741 * if_extents/if_data and if_broot. It must also free the lock
2742 * associated with the inode.
2743 */
2744 void
2747 {
2755 }
2761 #ifdef XFS_BMAP_TRACE
2763 #endif
2764 #ifdef XFS_BMBT_TRACE
2766 #endif
2767 #ifdef XFS_RW_TRACE
2769 #endif
2770 #ifdef XFS_ILOCK_TRACE
2772 #endif
2773 #ifdef XFS_DIR2_TRACE
2775 #endif
2777 /* XXXdpd should be able to assert this but shutdown
2778 * is leaving the AIL behind. */
2782 }
2784 }
2787 /*
2788 * Increment the pin count of the given buffer.
2789 * This value is protected by ipinlock spinlock in the mount structure.
2790 */
2791 void
2794 {
2798 }
2800 /*
2801 * Decrement the pin count of the given inode, and wake up
2802 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2803 * inode must have been previoulsy pinned with a call to xfs_ipin().
2804 */
2805 void
2808 {
2814 /* make sync come back and flush this inode */
2820 }
2823 }
2824 }
2826 /*
2827 * This is called to wait for the given inode to be unpinned.
2828 * It will sleep until this happens. The caller must have the
2829 * inode locked in at least shared mode so that the buffer cannot
2830 * be subsequently pinned once someone is waiting for it to be
2831 * unpinned.
2832 */
2833 STATIC void
2836 {
2838 xfs_lsn_t lsn;
2844 }
2851 }
2853 /*
2854 * Give the log a push so we don't wait here too long.
2855 */
2859 }
2862 /*
2863 * xfs_iextents_copy()
2864 *
2865 * This is called to copy the REAL extents (as opposed to the delayed
2866 * allocation extents) from the inode into the given buffer. It
2867 * returns the number of bytes copied into the buffer.
2868 *
2869 * If there are no delayed allocation extents, then we can just
2870 * memcpy() the extents into the buffer. Otherwise, we need to
2871 * examine each extent in turn and skip those which are delayed.
2872 */
2873 int
2878 {
2882 #ifdef XFS_BMAP_TRACE
2884 #endif
2888 xfs_fsblock_t start_block;
2898 /*
2899 * There are some delayed allocation extents in the
2900 * inode, so copy the extents one at a time and skip
2901 * the delayed ones. There must be at least one
2902 * non-delayed extent.
2903 */
2910 /*
2911 * It's a delayed allocation extent, so skip it.
2912 */
2913 ep++;
2915 }
2917 /* Translate to on disk format */
2922 dest_ep++;
2923 ep++;
2924 copied++;
2925 }
2930 }
2932 /*
2933 * Each of the following cases stores data into the same region
2934 * of the on-disk inode, so only one of them can be valid at
2935 * any given time. While it is possible to have conflicting formats
2936 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2937 * in EXTENTS format, this can only happen when the fork has
2938 * changed formats after being modified but before being flushed.
2939 * In these cases, the format always takes precedence, because the
2940 * format indicates the current state of the fork.
2941 */
2942 /*ARGSUSED*/
2943 STATIC int
2950 {
2954 #ifdef XFS_TRANS_DEBUG
2956 #endif
2967 /*
2968 * This can happen if we gave up in iformat in an error path,
2969 * for the attribute fork.
2970 */
2974 }
2984 }
2990 }
2991 }
3003 whichfork);
3004 }
3013 XFS_BROOT_SIZE_ADJ));
3017 }
3024 }
3032 }
3038 }
3041 }
3043 /*
3044 * xfs_iflush() will write a modified inode's changes out to the
3045 * inode's on disk home. The caller must have the inode lock held
3046 * in at least shared mode and the inode flush semaphore must be
3047 * held as well. The inode lock will still be held upon return from
3048 * the call and the caller is free to unlock it.
3049 * The inode flush lock will be unlocked when the inode reaches the disk.
3050 * The flags indicate how the inode's buffer should be written out.
3051 */
3052 int
3055 uint flags)
3056 {
3062 /* REFERENCED */
3080 /*
3081 * If the inode isn't dirty, then just release the inode
3082 * flush lock and do nothing.
3083 */
3090 }
3092 /*
3093 * We can't flush the inode until it is unpinned, so
3094 * wait for it. We know noone new can pin it, because
3095 * we are holding the inode lock shared and you need
3096 * to hold it exclusively to pin the inode.
3097 */
3100 /*
3101 * This may have been unpinned because the filesystem is shutting
3102 * down forcibly. If that's the case we must not write this inode
3103 * to disk, because the log record didn't make it to disk!
3104 */
3111 }
3113 /*
3114 * Get the buffer containing the on-disk inode.
3115 */
3120 }
3122 /*
3123 * Decide how buffer will be flushed out. This is done before
3124 * the call to xfs_iflush_int because this field is zeroed by it.
3125 */
3127 /*
3128 * Flush out the inode buffer according to the directions
3129 * of the caller. In the cases where the caller has given
3130 * us a choice choose the non-delwri case. This is because
3131 * the inode is in the AIL and we need to get it out soon.
3132 */
3149 }
3167 }
3168 }
3170 /*
3171 * First flush out the inode that xfs_iflush was called with.
3172 */
3176 }
3178 /*
3179 * inode clustering:
3180 * see if other inodes can be gathered into this write
3181 */
3190 /*
3191 * Do an un-protected check to see if the inode is dirty and
3192 * is a candidate for flushing. These checks will be repeated
3193 * later after the appropriate locks are acquired.
3194 */
3201 }
3203 /*
3204 * Try to get locks. If any are unavailable,
3205 * then this inode cannot be flushed and is skipped.
3206 */
3208 /* get inode locks (just i_lock) */
3210 /* get inode flush lock */
3212 /* check if pinned */
3214 /* arriving here means that
3215 * this inode can be flushed.
3216 * first re-check that it's
3217 * dirty
3218 */
3223 clcount++;
3227 XFS_ILOCK_SHARED);
3229 }
3232 }
3235 }
3236 }
3238 }
3239 }
3245 }
3247 /*
3248 * If the buffer is pinned then push on the log so we won't
3249 * get stuck waiting in the write for too long.
3250 */
3253 }
3261 }
3264 corrupt_out:
3268 /*
3269 * Unlocks the flush lock
3270 */
3273 cluster_corrupt_out:
3274 /* Corruption detected in the clustering loop. Invalidate the
3275 * inode buffer and shut down the filesystem.
3276 */
3279 /*
3280 * Clean up the buffer. If it was B_DELWRI, just release it --
3281 * brelse can handle it with no problems. If not, shut down the
3282 * filesystem before releasing the buffer.
3283 */
3286 }
3291 /*
3292 * Just like incore_relse: if we have b_iodone functions,
3293 * mark the buffer as an error and call them. Otherwise
3294 * mark it as stale and brelse.
3295 */
3306 }
3307 }
3310 /*
3311 * Unlocks the flush lock
3312 */
3314 }
3317 STATIC int
3321 {
3325 #ifdef XFS_TRANS_DEBUG
3327 #endif
3339 /*
3340 * If the inode isn't dirty, then just release the inode
3341 * flush lock and do nothing.
3342 */
3347 }
3349 /* set *dip = inode's place in the buffer */
3352 /*
3353 * Clear i_update_core before copying out the data.
3354 * This is for coordination with our timestamp updates
3355 * that don't hold the inode lock. They will always
3356 * update the timestamps BEFORE setting i_update_core,
3357 * so if we clear i_update_core after they set it we
3358 * are guaranteed to see their updates to the timestamps.
3359 * I believe that this depends on strongly ordered memory
3360 * semantics, but we have that. We use the SYNCHRONIZE
3361 * macro to make sure that the compiler does not reorder
3362 * the i_update_core access below the data copy below.
3363 */
3367 /*
3368 * Make sure to get the latest atime from the Linux inode.
3369 */
3378 }
3385 }
3395 }
3406 }
3407 }
3410 XFS_RANDOM_IFLUSH_5)) {
3416 ip);
3418 }
3425 }
3426 /*
3427 * bump the flush iteration count, used to detect flushes which
3428 * postdate a log record during recovery.
3429 */
3433 /*
3434 * Copy the dirty parts of the inode into the on-disk
3435 * inode. We always copy out the core of the inode,
3436 * because if the inode is dirty at all the core must
3437 * be.
3438 */
3441 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3445 /*
3446 * If this is really an old format inode and the superblock version
3447 * has not been updated to support only new format inodes, then
3448 * convert back to the old inode format. If the superblock version
3449 * has been updated, then make the conversion permanent.
3450 */
3455 /*
3456 * Convert it back.
3457 */
3461 /*
3462 * The superblock version has already been bumped,
3463 * so just make the conversion to the new inode
3464 * format permanent.
3465 */
3474 }
3475 }
3479 }
3482 /*
3483 * The only error from xfs_iflush_fork is on the data fork.
3484 */
3486 }
3489 /*
3490 * We've recorded everything logged in the inode, so we'd
3491 * like to clear the ilf_fields bits so we don't log and
3492 * flush things unnecessarily. However, we can't stop
3493 * logging all this information until the data we've copied
3494 * into the disk buffer is written to disk. If we did we might
3495 * overwrite the copy of the inode in the log with all the
3496 * data after re-logging only part of it, and in the face of
3497 * a crash we wouldn't have all the data we need to recover.
3498 *
3499 * What we do is move the bits to the ili_last_fields field.
3500 * When logging the inode, these bits are moved back to the
3501 * ilf_fields field. In the xfs_iflush_done() routine we
3502 * clear ili_last_fields, since we know that the information
3503 * those bits represent is permanently on disk. As long as
3504 * the flush completes before the inode is logged again, then
3505 * both ilf_fields and ili_last_fields will be cleared.
3506 *
3507 * We can play with the ilf_fields bits here, because the inode
3508 * lock must be held exclusively in order to set bits there
3509 * and the flush lock protects the ili_last_fields bits.
3510 * Set ili_logged so the flush done
3511 * routine can tell whether or not to look in the AIL.
3512 * Also, store the current LSN of the inode so that we can tell
3513 * whether the item has moved in the AIL from xfs_iflush_done().
3514 * In order to read the lsn we need the AIL lock, because
3515 * it is a 64 bit value that cannot be read atomically.
3516 */
3527 /*
3528 * Attach the function xfs_iflush_done to the inode's
3529 * buffer. This will remove the inode from the AIL
3530 * and unlock the inode's flush lock when the inode is
3531 * completely written to disk.
3532 */
3539 /*
3540 * We're flushing an inode which is not in the AIL and has
3541 * not been logged but has i_update_core set. For this
3542 * case we can use a B_DELWRI flush and immediately drop
3543 * the inode flush lock because we can avoid the whole
3544 * AIL state thing. It's OK to drop the flush lock now,
3545 * because we've already locked the buffer and to do anything
3546 * you really need both.
3547 */
3552 }
3554 }
3558 corrupt_out:
3560 }
3563 /*
3564 * Flush all inactive inodes in mp.
3565 */
3566 void
3569 {
3573 again:
3580 /* Make sure we skip markers inserted by sync */
3584 }
3591 }
3597 out:
3599 }
3601 /*
3602 * xfs_iaccess: check accessibility of inode for mode.
3603 */
3604 int
3607 mode_t mode,
3609 {
3623 }
3625 /*
3626 * If there's an Access Control List it's used instead of
3627 * the mode bits.
3628 */
3636 }
3638 /*
3639 * If the DACs are ok we don't need any capability check.
3640 */
3643 /*
3644 * Read/write DACs are always overridable.
3645 * Executable DACs are overridable if at least one exec bit is set.
3646 */
3656 #ifdef NOISE
3660 }
3662 }
3664 /*
3665 * xfs_iroundup: round up argument to next power of two
3666 */
3667 uint
3669 uint v)
3670 {
3672 uint m;
3685 }
3688 }
3690 #ifdef XFS_ILOCK_TRACE
3693 void
3695 {
3704 }
3705 #endif