| blob | 34bdf5909687b30d0619e3e9f6d42feee4de4098 |
1 /*
2 * Copyright (c) 2000-2003 Silicon Graphics, Inc. All Rights Reserved.
3 *
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it would be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
11 *
12 * Further, this software is distributed without any warranty that it is
13 * free of the rightful claim of any third person regarding infringement
14 * or the like. Any license provided herein, whether implied or
15 * otherwise, applies only to this software file. Patent licenses, if
16 * any, provided herein do not apply to combinations of this program with
17 * other software, or any other product whatsoever.
18 *
19 * You should have received a copy of the GNU General Public License along
20 * with this program; if not, write the Free Software Foundation, Inc., 59
21 * Temple Place - Suite 330, Boston MA 02111-1307, USA.
22 *
23 * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24 * Mountain View, CA 94043, or:
25 *
26 * http://www.sgi.com
27 *
28 * For further information regarding this notice, see:
29 *
30 * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
31 */
75 /*
76 * Used in xfs_itruncate(). This is the maximum number of extents
77 * freed from a file in a single transaction.
78 */
79 #define XFS_ITRUNC_MAX_EXTENTS 2
87 #ifdef DEBUG
88 /*
89 * Make sure that the extents in the given memory buffer
90 * are valid.
91 */
92 STATIC void
97 xfs_exntfmt_t fmt)
98 {
99 xfs_bmbt_irec_t irec;
100 xfs_bmbt_rec_t rec;
108 else
112 ep++;
113 }
114 }
116 #define xfs_validate_extents(ep, nrecs, disk, fmt)
119 /*
120 * Check that none of the inode's in the buffer have a next
121 * unlinked field of 0.
122 */
123 #if defined(DEBUG)
124 void
128 {
140 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
141 bp);
143 }
144 }
145 }
146 #endif
148 /*
149 * This routine is called to map an inode number within a file
150 * system to the buffer containing the on-disk version of the
151 * inode. It returns a pointer to the buffer containing the
152 * on-disk inode in the bpp parameter, and in the dip parameter
153 * it returns a pointer to the on-disk inode within that buffer.
154 *
155 * If a non-zero error is returned, then the contents of bpp and
156 * dipp are undefined.
157 *
158 * Use xfs_imap() to determine the size and location of the
159 * buffer to read from disk.
160 */
161 STATIC int
165 xfs_ino_t ino,
169 {
171 xfs_imap_t imap;
176 /*
177 * Call the space managment code to find the location of the
178 * inode on disk.
179 */
184 "xfs_inotobp: xfs_imap() returned an "
187 }
189 /*
190 * If the inode number maps to a block outside the bounds of the
191 * file system then return NULL rather than calling read_buf
192 * and panicing when we get an error from the driver.
193 */
197 "xfs_inotobp: inode number (%d + %d) maps to a block outside the bounds "
201 }
203 /*
204 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
205 * default to just a read_buf() call.
206 */
212 "xfs_inotobp: xfs_trans_read_buf() returned an "
215 }
217 di_ok =
221 XFS_RANDOM_ITOBP_INOTOBP))) {
225 "xfs_inotobp: XFS_TEST_ERROR() returned an "
228 }
232 /*
233 * Set *dipp to point to the on-disk inode in the buffer.
234 */
239 }
242 /*
243 * This routine is called to map an inode to the buffer containing
244 * the on-disk version of the inode. It returns a pointer to the
245 * buffer containing the on-disk inode in the bpp parameter, and in
246 * the dip parameter it returns a pointer to the on-disk inode within
247 * that buffer.
248 *
249 * If a non-zero error is returned, then the contents of bpp and
250 * dipp are undefined.
251 *
252 * If the inode is new and has not yet been initialized, use xfs_imap()
253 * to determine the size and location of the buffer to read from disk.
254 * If the inode has already been mapped to its buffer and read in once,
255 * then use the mapping information stored in the inode rather than
256 * calling xfs_imap(). This allows us to avoid the overhead of looking
257 * at the inode btree for small block file systems (see xfs_dilocate()).
258 * We can tell whether the inode has been mapped in before by comparing
259 * its disk block address to 0. Only uninitialized inodes will have
260 * 0 for the disk block address.
261 */
262 int
269 xfs_daddr_t bno)
270 {
273 xfs_imap_t imap;
274 #ifdef __KERNEL__
277 #endif
280 /*
281 * Call the space management code to find the location of the
282 * inode on disk.
283 */
288 }
290 /*
291 * If the inode number maps to a block outside the bounds
292 * of the file system then return NULL rather than calling
293 * read_buf and panicing when we get an error from the
294 * driver.
295 */
298 #ifdef DEBUG
300 "(imap.im_blkno (0x%llx) "
301 "+ imap.im_len (0x%llx)) > "
302 " XFS_FSB_TO_BB(mp, "
309 }
311 /*
312 * Fill in the fields in the inode that will be used to
313 * map the inode to its buffer from now on.
314 */
319 /*
320 * We've already mapped the inode once, so just use the
321 * mapping that we saved the first time.
322 */
326 }
329 /*
330 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
331 * default to just a read_buf() call.
332 */
337 #ifdef DEBUG
339 "xfs_trans_read_buf() returned error %d, "
345 }
346 #ifdef __KERNEL__
347 /*
348 * Validate the magic number and version of every inode in the buffer
349 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
350 */
351 #ifdef DEBUG
353 #else
355 #endif
365 XFS_RANDOM_ITOBP_INOTOBP))) {
366 #ifdef DEBUG
371 #endif
376 }
377 }
382 /*
383 * Mark the buffer as an inode buffer now that it looks good
384 */
387 /*
388 * Set *dipp to point to the on-disk inode in the buffer.
389 */
393 }
395 /*
396 * Move inode type and inode format specific information from the
397 * on-disk inode to the in-core inode. For fifos, devs, and sockets
398 * this means set if_rdev to the proper value. For files, directories,
399 * and symlinks this means to bring in the in-line data or extent
400 * pointers. For a file in B-tree format, only the root is immediately
401 * brought in-core. The rest will be in-lined in if_extents when it
402 * is first referenced (see xfs_iread_extents()).
403 */
404 STATIC int
408 {
412 xfs_fsize_t di_size;
422 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu."
432 }
434 if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) {
436 "corrupt dinode %Lu, forkoff = 0x%x."
443 }
454 }
464 /*
465 * no local regular files yet
466 */
472 XFS_ERRLEVEL_LOW,
475 }
484 XFS_ERRLEVEL_LOW,
487 }
502 }
508 }
511 }
533 }
538 }
540 }
542 /*
543 * The file is in-lined in the on-disk inode.
544 * If it fits into if_inline_data, then copy
545 * it there, otherwise allocate a buffer for it
546 * and copy the data there. Either way, set
547 * if_data to point at the data.
548 * If we allocate a buffer for the data, make
549 * sure that its size is a multiple of 4 and
550 * record the real size in i_real_bytes.
551 */
552 STATIC int
558 {
562 /*
563 * If the size is unreasonable, then something
564 * is wrong and we just bail out rather than crash in
565 * kmem_alloc() or memcpy() below.
566 */
575 }
585 }
593 }
595 /*
596 * The file consists of a set of extents all
597 * of which fit into the on-disk inode.
598 * If there are few enough extents to fit into
599 * the if_inline_ext, then copy them there.
600 * Otherwise allocate a buffer for them and copy
601 * them into it. Either way, set if_extents
602 * to point at the extents.
603 */
604 STATIC int
609 {
621 /*
622 * If the number of extents is unreasonable, then something
623 * is wrong and we just bail out rather than crash in
624 * kmem_alloc() or memcpy() below.
625 */
633 }
644 }
653 ARCH_CONVERT);
655 ARCH_CONVERT);
656 }
658 whichfork);
664 XFS_ERRLEVEL_LOW,
667 }
668 }
671 }
673 /*
674 * The file has too many extents to fit into
675 * the inode, so they are in B-tree format.
676 * Allocate a buffer for the root of the B-tree
677 * and copy the root into it. The i_extents
678 * field will remain NULL until all of the
679 * extents are read in (when they are needed).
680 */
681 STATIC int
686 {
689 /* REFERENCED */
698 /*
699 * blow out if -- fork has less extents than can fit in
700 * fork (fork shouldn't be a btree format), root btree
701 * block has more records than can fit into the fork,
702 * or the number of extents is greater than the number of
703 * blocks.
704 */
715 }
720 /*
721 * Copy and convert from the on-disk structure
722 * to the in-memory structure.
723 */
730 }
732 /*
733 * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
734 * and native format
735 *
736 * buf = on-disk representation
737 * dip = native representation
738 * dir = direction - +ve -> disk to native
739 * -ve -> native to disk
740 */
741 void
743 xfs_caddr_t buf,
746 {
769 }
796 }
798 STATIC uint
801 __uint16_t di_flags)
802 {
826 }
829 }
831 uint
834 {
839 }
841 uint
844 {
847 }
849 /*
850 * Given a mount structure and an inode number, return a pointer
851 * to a newly allocated in-core inode coresponding to the given
852 * inode number.
853 *
854 * Initialize the inode's attributes and extent pointers if it
855 * already has them (it will not if the inode has no links).
856 */
857 int
861 xfs_ino_t ino,
863 xfs_daddr_t bno)
864 {
876 /*
877 * Get pointer's to the on-disk inode and the buffer containing it.
878 * If the inode number refers to a block outside the file system
879 * then xfs_itobp() will return NULL. In this case we should
880 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
881 * know that this is a new incore inode.
882 */
888 }
890 /*
891 * Initialize inode's trace buffers.
892 * Do this before xfs_iformat in case it adds entries.
893 */
894 #ifdef XFS_BMAP_TRACE
896 #endif
897 #ifdef XFS_BMBT_TRACE
899 #endif
900 #ifdef XFS_RW_TRACE
902 #endif
903 #ifdef XFS_ILOCK_TRACE
905 #endif
906 #ifdef XFS_DIR2_TRACE
908 #endif
910 /*
911 * If we got something that isn't an inode it means someone
912 * (nfs or dmi) has a stale handle.
913 */
917 #ifdef DEBUG
919 "dip->di_core.di_magic (0x%x) != "
922 XFS_DINODE_MAGIC);
925 }
927 /*
928 * If the on-disk inode is already linked to a directory
929 * entry, copy all of the inode into the in-core inode.
930 * xfs_iformat() handles copying in the inode format
931 * specific information.
932 * Otherwise, just get the truly permanent information.
933 */
941 #ifdef DEBUG
944 error);
947 }
953 /*
954 * Make sure to pull in the mode here as well in
955 * case the inode is released without being used.
956 * This ensures that xfs_inactive() will see that
957 * the inode is already free and not try to mess
958 * with the uninitialized part of it.
959 */
961 /*
962 * Initialize the per-fork minima and maxima for a new
963 * inode here. xfs_iformat will do it for old inodes.
964 */
967 }
971 /*
972 * The inode format changed when we moved the link count and
973 * made it 32 bits long. If this is an old format inode,
974 * convert it in memory to look like a new one. If it gets
975 * flushed to disk we will convert back before flushing or
976 * logging it. We zero out the new projid field and the old link
977 * count field. We'll handle clearing the pad field (the remains
978 * of the old uuid field) when we actually convert the inode to
979 * the new format. We don't change the version number so that we
980 * can distinguish this from a real new format inode.
981 */
986 }
990 /*
991 * Mark the buffer containing the inode as something to keep
992 * around for a while. This helps to keep recently accessed
993 * meta-data in-core longer.
994 */
997 /*
998 * Use xfs_trans_brelse() to release the buffer containing the
999 * on-disk inode, because it was acquired with xfs_trans_read_buf()
1000 * in xfs_itobp() above. If tp is NULL, this is just a normal
1001 * brelse(). If we're within a transaction, then xfs_trans_brelse()
1002 * will only release the buffer if it is not dirty within the
1003 * transaction. It will be OK to release the buffer in this case,
1004 * because inodes on disk are never destroyed and we will be
1005 * locking the new in-core inode before putting it in the hash
1006 * table where other processes can find it. Thus we don't have
1007 * to worry about the inode being changed just because we released
1008 * the buffer.
1009 */
1013 }
1015 /*
1016 * Read in extents from a btree-format inode.
1017 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1018 */
1019 int
1024 {
1033 }
1036 /*
1037 * We know that the size is valid (it's checked in iformat_btree)
1038 */
1051 }
1055 }
1057 /*
1058 * Allocate an inode on disk and return a copy of its in-core version.
1059 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1060 * appropriately within the inode. The uid and gid for the inode are
1061 * set according to the contents of the given cred structure.
1062 *
1063 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1064 * has a free inode available, call xfs_iget()
1065 * to obtain the in-core version of the allocated inode. Finally,
1066 * fill in the inode and log its initial contents. In this case,
1067 * ialloc_context would be set to NULL and call_again set to false.
1068 *
1069 * If xfs_dialloc() does not have an available inode,
1070 * it will replenish its supply by doing an allocation. Since we can
1071 * only do one allocation within a transaction without deadlocks, we
1072 * must commit the current transaction before returning the inode itself.
1073 * In this case, therefore, we will set call_again to true and return.
1074 * The caller should then commit the current transaction, start a new
1075 * transaction, and call xfs_ialloc() again to actually get the inode.
1076 *
1077 * To ensure that some other process does not grab the inode that
1078 * was allocated during the first call to xfs_ialloc(), this routine
1079 * also returns the [locked] bp pointing to the head of the freelist
1080 * as ialloc_context. The caller should hold this buffer across
1081 * the commit and pass it back into this routine on the second call.
1082 */
1083 int
1087 mode_t mode,
1088 xfs_nlink_t nlink,
1089 xfs_dev_t rdev,
1091 xfs_prid_t prid,
1096 {
1097 xfs_ino_t ino;
1100 uint flags;
1103 /*
1104 * Call the space management code to pick
1105 * the on-disk inode to be allocated.
1106 */
1111 }
1115 }
1118 /*
1119 * Get the in-core inode with the lock held exclusively.
1120 * This is because we're setting fields here we need
1121 * to prevent others from looking at until we're done.
1122 */
1127 }
1141 /*
1142 * If the superblock version is up to where we support new format
1143 * inodes and this is currently an old format inode, then change
1144 * the inode version number now. This way we only do the conversion
1145 * here rather than here and in the flush/logging code.
1146 */
1150 /*
1151 * We've already zeroed the old link count, the projid field,
1152 * and the pad field.
1153 */
1154 }
1156 /*
1157 * Project ids won't be stored on disk if we are using a version 1 inode.
1158 */
1166 }
1167 }
1169 /*
1170 * If the group ID of the new file does not match the effective group
1171 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1172 * (and only if the irix_sgid_inherit compatibility variable is set).
1173 */
1178 }
1184 /*
1185 * di_gen will have been taken care of in xfs_iread.
1186 */
1214 }
1215 }
1217 xfs_inherit_noatime)
1220 xfs_inherit_nodump)
1223 xfs_inherit_sync)
1226 xfs_inherit_nosymlinks)
1231 }
1232 /* FALLTHROUGH */
1241 }
1242 /*
1243 * Attribute fork settings for new inode.
1244 */
1248 /*
1249 * Log the new values stuffed into the inode.
1250 */
1253 /* now that we have a v_type we can set Linux inode ops (& unlock) */
1258 }
1260 /*
1261 * Check to make sure that there are no blocks allocated to the
1262 * file beyond the size of the file. We don't check this for
1263 * files with fixed size extents or real time extents, but we
1264 * at least do it for regular files.
1265 */
1266 #ifdef DEBUG
1267 void
1271 xfs_fsize_t isize)
1272 {
1273 xfs_fileoff_t map_first;
1285 /*
1286 * The filesystem could be shutting down, so bmapi may return
1287 * an error.
1288 */
1292 map_first),
1294 NULL))
1298 }
1301 /*
1302 * Calculate the last possible buffered byte in a file. This must
1303 * include data that was buffered beyond the EOF by the write code.
1304 * This also needs to deal with overflowing the xfs_fsize_t type
1305 * which can happen for sizes near the limit.
1306 *
1307 * We also need to take into account any blocks beyond the EOF. It
1308 * may be the case that they were buffered by a write which failed.
1309 * In that case the pages will still be in memory, but the inode size
1310 * will never have been updated.
1311 */
1312 xfs_fsize_t
1315 {
1317 xfs_fsize_t last_byte;
1318 xfs_fileoff_t last_block;
1319 xfs_fileoff_t size_last_block;
1325 /*
1326 * Only check for blocks beyond the EOF if the extents have
1327 * been read in. This eliminates the need for the inode lock,
1328 * and it also saves us from looking when it really isn't
1329 * necessary.
1330 */
1333 XFS_DATA_FORK);
1336 }
1339 }
1346 }
1350 }
1352 }
1354 #if defined(XFS_RW_TRACE)
1355 STATIC void
1360 xfs_fsize_t new_size,
1361 xfs_off_t toss_start,
1362 xfs_off_t toss_finish)
1363 {
1366 }
1385 }
1386 #else
1387 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1388 #endif
1390 /*
1391 * Start the truncation of the file to new_size. The new size
1392 * must be smaller than the current size. This routine will
1393 * clear the buffer and page caches of file data in the removed
1394 * range, and xfs_itruncate_finish() will remove the underlying
1395 * disk blocks.
1396 *
1397 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1398 * must NOT have the inode lock held at all. This is because we're
1399 * calling into the buffer/page cache code and we can't hold the
1400 * inode lock when we do so.
1401 *
1402 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1403 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1404 * in the case that the caller is locking things out of order and
1405 * may not be able to call xfs_itruncate_finish() with the inode lock
1406 * held without dropping the I/O lock. If the caller must drop the
1407 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1408 * must be called again with all the same restrictions as the initial
1409 * call.
1410 */
1411 void
1414 uint flags,
1415 xfs_fsize_t new_size)
1416 {
1417 xfs_fsize_t last_byte;
1418 xfs_off_t toss_start;
1429 /*
1430 * Call VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES() to get rid of pages and buffers
1431 * overlapping the region being removed. We have to use
1432 * the less efficient VOP_FLUSHINVAL_PAGES() in the case that the
1433 * caller may not be able to finish the truncate without
1434 * dropping the inode's I/O lock. Make sure
1435 * to catch any pages brought in by buffers overlapping
1436 * the EOF by searching out beyond the isize by our
1437 * block size. We round new_size up to a block boundary
1438 * so that we don't toss things on the same block as
1439 * new_size but before it.
1440 *
1441 * Before calling VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES(), make sure to
1442 * call remapf() over the same region if the file is mapped.
1443 * This frees up mapped file references to the pages in the
1444 * given range and for the VOP_FLUSHINVAL_PAGES() case it ensures
1445 * that we get the latest mapped changes flushed out.
1446 */
1450 /*
1451 * The place to start tossing is beyond our maximum
1452 * file size, so there is no way that the data extended
1453 * out there.
1454 */
1456 }
1459 last_byte);
1465 }
1466 }
1468 #ifdef DEBUG
1471 }
1472 #endif
1473 }
1475 /*
1476 * Shrink the file to the given new_size. The new
1477 * size must be smaller than the current size.
1478 * This will free up the underlying blocks
1479 * in the removed range after a call to xfs_itruncate_start()
1480 * or xfs_atruncate_start().
1481 *
1482 * The transaction passed to this routine must have made
1483 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1484 * This routine may commit the given transaction and
1485 * start new ones, so make sure everything involved in
1486 * the transaction is tidy before calling here.
1487 * Some transaction will be returned to the caller to be
1488 * committed. The incoming transaction must already include
1489 * the inode, and both inode locks must be held exclusively.
1490 * The inode must also be "held" within the transaction. On
1491 * return the inode will be "held" within the returned transaction.
1492 * This routine does NOT require any disk space to be reserved
1493 * for it within the transaction.
1494 *
1495 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1496 * and it indicates the fork which is to be truncated. For the
1497 * attribute fork we only support truncation to size 0.
1498 *
1499 * We use the sync parameter to indicate whether or not the first
1500 * transaction we perform might have to be synchronous. For the attr fork,
1501 * it needs to be so if the unlink of the inode is not yet known to be
1502 * permanent in the log. This keeps us from freeing and reusing the
1503 * blocks of the attribute fork before the unlink of the inode becomes
1504 * permanent.
1505 *
1506 * For the data fork, we normally have to run synchronously if we're
1507 * being called out of the inactive path or we're being called
1508 * out of the create path where we're truncating an existing file.
1509 * Either way, the truncate needs to be sync so blocks don't reappear
1510 * in the file with altered data in case of a crash. wsync filesystems
1511 * can run the first case async because anything that shrinks the inode
1512 * has to run sync so by the time we're called here from inactive, the
1513 * inode size is permanently set to 0.
1514 *
1515 * Calls from the truncate path always need to be sync unless we're
1516 * in a wsync filesystem and the file has already been unlinked.
1517 *
1518 * The caller is responsible for correctly setting the sync parameter.
1519 * It gets too hard for us to guess here which path we're being called
1520 * out of just based on inode state.
1521 */
1522 int
1526 xfs_fsize_t new_size,
1529 {
1530 xfs_fsblock_t first_block;
1531 xfs_fileoff_t first_unmap_block;
1532 xfs_fileoff_t last_block;
1538 xfs_bmap_free_t free_list;
1555 /*
1556 * We only support truncating the entire attribute fork.
1557 */
1560 }
1563 /*
1564 * The first thing we do is set the size to new_size permanently
1565 * on disk. This way we don't have to worry about anyone ever
1566 * being able to look at the data being freed even in the face
1567 * of a crash. What we're getting around here is the case where
1568 * we free a block, it is allocated to another file, it is written
1569 * to, and then we crash. If the new data gets written to the
1570 * file but the log buffers containing the free and reallocation
1571 * don't, then we'd end up with garbage in the blocks being freed.
1572 * As long as we make the new_size permanent before actually
1573 * freeing any blocks it doesn't matter if they get writtten to.
1574 *
1575 * The callers must signal into us whether or not the size
1576 * setting here must be synchronous. There are a few cases
1577 * where it doesn't have to be synchronous. Those cases
1578 * occur if the file is unlinked and we know the unlink is
1579 * permanent or if the blocks being truncated are guaranteed
1580 * to be beyond the inode eof (regardless of the link count)
1581 * and the eof value is permanent. Both of these cases occur
1582 * only on wsync-mounted filesystems. In those cases, we're
1583 * guaranteed that no user will ever see the data in the blocks
1584 * that are being truncated so the truncate can run async.
1585 * In the free beyond eof case, the file may wind up with
1586 * more blocks allocated to it than it needs if we crash
1587 * and that won't get fixed until the next time the file
1588 * is re-opened and closed but that's ok as that shouldn't
1589 * be too many blocks.
1590 *
1591 * However, we can't just make all wsync xactions run async
1592 * because there's one call out of the create path that needs
1593 * to run sync where it's truncating an existing file to size
1594 * 0 whose size is > 0.
1595 *
1596 * It's probably possible to come up with a test in this
1597 * routine that would correctly distinguish all the above
1598 * cases from the values of the function parameters and the
1599 * inode state but for sanity's sake, I've decided to let the
1600 * layers above just tell us. It's simpler to correctly figure
1601 * out in the layer above exactly under what conditions we
1602 * can run async and I think it's easier for others read and
1603 * follow the logic in case something has to be changed.
1604 * cscope is your friend -- rcc.
1605 *
1606 * The attribute fork is much simpler.
1607 *
1608 * For the attribute fork we allow the caller to tell us whether
1609 * the unlink of the inode that led to this call is yet permanent
1610 * in the on disk log. If it is not and we will be freeing extents
1611 * in this inode then we make the first transaction synchronous
1612 * to make sure that the unlink is permanent by the time we free
1613 * the blocks.
1614 */
1619 }
1624 }
1630 /*
1631 * Since it is possible for space to become allocated beyond
1632 * the end of the file (in a crash where the space is allocated
1633 * but the inode size is not yet updated), simply remove any
1634 * blocks which show up between the new EOF and the maximum
1635 * possible file size. If the first block to be removed is
1636 * beyond the maximum file size (ie it is the same as last_block),
1637 * then there is nothing to do.
1638 */
1646 }
1648 /*
1649 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1650 * will tell us whether it freed the entire range or
1651 * not. If this is a synchronous mount (wsync),
1652 * then we can tell bunmapi to keep all the
1653 * transactions asynchronous since the unlink
1654 * transaction that made this inode inactive has
1655 * already hit the disk. There's no danger of
1656 * the freed blocks being reused, there being a
1657 * crash, and the reused blocks suddenly reappearing
1658 * in this file with garbage in them once recovery
1659 * runs.
1660 */
1663 unmap_len,
1666 XFS_ITRUNC_MAX_EXTENTS,
1669 /*
1670 * If the bunmapi call encounters an error,
1671 * return to the caller where the transaction
1672 * can be properly aborted. We just need to
1673 * make sure we're not holding any resources
1674 * that we were not when we came in.
1675 */
1678 }
1680 /*
1681 * Duplicate the transaction that has the permanent
1682 * reservation and commit the old transaction.
1683 */
1688 /*
1689 * If the bmap finish call encounters an error,
1690 * return to the caller where the transaction
1691 * can be properly aborted. We just need to
1692 * make sure we're not holding any resources
1693 * that we were not when we came in.
1694 *
1695 * Aborting from this point might lose some
1696 * blocks in the file system, but oh well.
1697 */
1700 /*
1701 * If the passed in transaction committed
1702 * in xfs_bmap_finish(), then we want to
1703 * add the inode to this one before returning.
1704 * This keeps things simple for the higher
1705 * level code, because it always knows that
1706 * the inode is locked and held in the
1707 * transaction that returns to it whether
1708 * errors occur or not. We don't mark the
1709 * inode dirty so that this transaction can
1710 * be easily aborted if possible.
1711 */
1715 }
1717 }
1720 /*
1721 * The first xact was committed,
1722 * so add the inode to the new one.
1723 * Mark it dirty so it will be logged
1724 * and moved forward in the log as
1725 * part of every commit.
1726 */
1731 }
1736 XFS_TRANS_PERM_LOG_RES,
1737 XFS_ITRUNCATE_LOG_COUNT);
1738 /*
1739 * Add the inode being truncated to the next chained
1740 * transaction.
1741 */
1746 }
1747 /*
1748 * Only update the size in the case of the data fork, but
1749 * always re-log the inode so that our permanent transaction
1750 * can keep on rolling it forward in the log.
1751 */
1755 }
1765 }
1768 /*
1769 * xfs_igrow_start
1770 *
1771 * Do the first part of growing a file: zero any data in the last
1772 * block that is beyond the old EOF. We need to do this before
1773 * the inode is joined to the transaction to modify the i_size.
1774 * That way we can drop the inode lock and call into the buffer
1775 * cache to get the buffer mapping the EOF.
1776 */
1777 int
1780 xfs_fsize_t new_size,
1782 {
1783 xfs_fsize_t isize;
1792 /*
1793 * Zero any pages that may have been created by
1794 * xfs_write_file() beyond the end of the file
1795 * and any blocks between the old and new file sizes.
1796 */
1798 new_size);
1800 }
1802 /*
1803 * xfs_igrow_finish
1804 *
1805 * This routine is called to extend the size of a file.
1806 * The inode must have both the iolock and the ilock locked
1807 * for update and it must be a part of the current transaction.
1808 * The xfs_igrow_start() function must have been called previously.
1809 * If the change_flag is not zero, the inode change timestamp will
1810 * be updated.
1811 */
1812 void
1816 xfs_fsize_t new_size,
1818 {
1824 /*
1825 * Update the file size. Update the inode change timestamp
1826 * if change_flag set.
1827 */
1833 }
1836 /*
1837 * This is called when the inode's link count goes to 0.
1838 * We place the on-disk inode on a list in the AGI. It
1839 * will be pulled from this list when the inode is freed.
1840 */
1841 int
1845 {
1851 xfs_agnumber_t agno;
1852 xfs_daddr_t agdaddr;
1853 xfs_agino_t agino;
1868 /*
1869 * Get the agi buffer first. It ensures lock ordering
1870 * on the list.
1871 */
1876 }
1877 /*
1878 * Validate the magic number of the agi block.
1879 */
1881 agi_ok =
1885 XFS_RANDOM_IUNLINK))) {
1889 }
1890 /*
1891 * Get the index into the agi hash table for the
1892 * list this inode will go on.
1893 */
1901 /*
1902 * There is already another inode in the bucket we need
1903 * to add ourselves to. Add us at the front of the list.
1904 * Here we put the head pointer into our next pointer,
1905 * and then we fall through to point the head at us.
1906 */
1910 }
1913 /* both on-disk, don't endian flip twice */
1921 }
1923 /*
1924 * Point the bucket head pointer at the inode being inserted.
1925 */
1933 }
1935 /*
1936 * Pull the on-disk inode from the AGI unlinked list.
1937 */
1938 STATIC int
1942 {
1943 xfs_ino_t next_ino;
1949 xfs_agnumber_t agno;
1950 xfs_daddr_t agdaddr;
1951 xfs_agino_t agino;
1952 xfs_agino_t next_agino;
1960 /*
1961 * First pull the on-disk inode from the AGI unlinked list.
1962 */
1968 /*
1969 * Get the agi buffer first. It ensures lock ordering
1970 * on the list.
1971 */
1979 }
1980 /*
1981 * Validate the magic number of the agi block.
1982 */
1984 agi_ok =
1988 XFS_RANDOM_IUNLINK_REMOVE))) {
1996 }
1997 /*
1998 * Get the index into the agi hash table for the
1999 * list this inode will go on.
2000 */
2008 /*
2009 * We're at the head of the list. Get the inode's
2010 * on-disk buffer to see if there is anyone after us
2011 * on the list. Only modify our next pointer if it
2012 * is not already NULLAGINO. This saves us the overhead
2013 * of dealing with the buffer when there is no need to
2014 * change it.
2015 */
2022 }
2035 }
2036 /*
2037 * Point the bucket head pointer at the next inode.
2038 */
2047 /*
2048 * We need to search the list for the inode being freed.
2049 */
2053 /*
2054 * If the last inode wasn't the one pointing to
2055 * us, then release its buffer since we're not
2056 * going to do anything with it.
2057 */
2060 }
2069 }
2073 }
2074 /*
2075 * Now last_ibp points to the buffer previous to us on
2076 * the unlinked list. Pull us from the list.
2077 */
2084 }
2098 }
2099 /*
2100 * Point the previous inode on the list to the next inode.
2101 */
2109 }
2111 }
2114 {
2118 }
2120 STATIC void
2124 xfs_ino_t inum)
2125 {
2131 xfs_daddr_t blkno;
2148 }
2157 /*
2158 * Look for each inode in memory and attempt to lock it,
2159 * we can be racing with flush and tail pushing here.
2160 * any inode we get the locks on, add to an array of
2161 * inode items to process later.
2162 *
2163 * The get the buffer lock, we could beat a flush
2164 * or tail pushing thread to the lock here, in which
2165 * case they will go looking for the inode buffer
2166 * and fail, we need some other form of interlock
2167 * here.
2168 */
2176 }
2178 /* Inode not in memory or we found it already,
2179 * nothing to do
2180 */
2184 }
2189 }
2191 /* If we can get the locks then add it to the
2192 * list, otherwise by the time we get the bp lock
2193 * below it will already be attached to the
2194 * inode buffer.
2195 */
2197 /* This inode will already be locked - by us, lets
2198 * keep it that way.
2199 */
2209 }
2210 }
2213 }
2224 }
2227 }
2228 }
2231 }
2235 XFS_BUF_LOCK);
2248 pre_flushed++;
2249 }
2251 }
2262 }
2276 }
2277 }
2282 }
2285 }
2287 /*
2288 * This is called to return an inode to the inode free list.
2289 * The inode should already be truncated to 0 length and have
2290 * no pages associated with it. This routine also assumes that
2291 * the inode is already a part of the transaction.
2292 *
2293 * The on-disk copy of the inode will have been added to the list
2294 * of unlinked inodes in the AGI. We need to remove the inode from
2295 * that list atomically with respect to freeing it here.
2296 */
2297 int
2302 {
2305 xfs_ino_t first_ino;
2316 /*
2317 * Pull the on-disk inode from the AGI unlinked list.
2318 */
2322 }
2327 }
2336 /*
2337 * Bump the generation count so no one will be confused
2338 * by reincarnations of this inode.
2339 */
2345 }
2348 }
2350 /*
2351 * Reallocate the space for if_broot based on the number of records
2352 * being added or deleted as indicated in rec_diff. Move the records
2353 * and pointers in if_broot to fit the new size. When shrinking this
2354 * will eliminate holes between the records and pointers created by
2355 * the caller. When growing this will create holes to be filled in
2356 * by the caller.
2357 *
2358 * The caller must not request to add more records than would fit in
2359 * the on-disk inode root. If the if_broot is currently NULL, then
2360 * if we adding records one will be allocated. The caller must also
2361 * not request that the number of records go below zero, although
2362 * it can go to zero.
2363 *
2364 * ip -- the inode whose if_broot area is changing
2365 * ext_diff -- the change in the number of records, positive or negative,
2366 * requested for the if_broot array.
2367 */
2368 void
2373 {
2382 /*
2383 * Handle the degenerate case quietly.
2384 */
2387 }
2391 /*
2392 * If there wasn't any memory allocated before, just
2393 * allocate it now and get out.
2394 */
2398 KM_SLEEP);
2401 }
2403 /*
2404 * If there is already an existing if_broot, then we need
2405 * to realloc() it and shift the pointers to their new
2406 * location. The records don't change location because
2407 * they are kept butted up against the btree block header.
2408 */
2414 new_size,
2416 KM_SLEEP);
2426 }
2428 /*
2429 * rec_diff is less than 0. In this case, we are shrinking the
2430 * if_broot buffer. It must already exist. If we go to zero
2431 * records, just get rid of the root and clear the status bit.
2432 */
2439 else
2443 /*
2444 * First copy over the btree block header.
2445 */
2450 }
2452 /*
2453 * Only copy the records and pointers if there are any.
2454 */
2456 /*
2457 * First copy the records.
2458 */
2465 /*
2466 * Then copy the pointers.
2467 */
2473 }
2480 }
2483 /*
2484 * This is called when the amount of space needed for if_extents
2485 * is increased or decreased. The change in size is indicated by
2486 * the number of extents that need to be added or deleted in the
2487 * ext_diff parameter.
2488 *
2489 * If the amount of space needed has decreased below the size of the
2490 * inline buffer, then switch to using the inline buffer. Otherwise,
2491 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2492 * to what is needed.
2493 *
2494 * ip -- the inode whose if_extents area is changing
2495 * ext_diff -- the change in the number of extents, positive or negative,
2496 * requested for the if_extents array.
2497 */
2498 void
2503 {
2507 uint rnew_size;
2511 }
2522 }
2526 /*
2527 * If the valid extents can fit in if_inline_ext,
2528 * copy them from the malloc'd vector and free it.
2529 */
2531 /*
2532 * For now, empty files are format EXTENTS,
2533 * so the if_extents pointer is null.
2534 */
2540 }
2542 }
2548 /*
2549 * Stuck with malloc/realloc.
2550 */
2559 rnew_size,
2561 KM_NOFS);
2562 }
2563 }
2566 }
2569 /*
2570 * This is called when the amount of space needed for if_data
2571 * is increased or decreased. The change in size is indicated by
2572 * the number of bytes that need to be added or deleted in the
2573 * byte_diff parameter.
2574 *
2575 * If the amount of space needed has decreased below the size of the
2576 * inline buffer, then switch to using the inline buffer. Otherwise,
2577 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2578 * to what is needed.
2579 *
2580 * ip -- the inode whose if_data area is changing
2581 * byte_diff -- the change in the number of bytes, positive or negative,
2582 * requested for the if_data array.
2583 */
2584 void
2589 {
2596 }
2605 }
2609 /*
2610 * If the valid extents/data can fit in if_inline_ext/data,
2611 * copy them from the malloc'd vector and free it.
2612 */
2618 new_size);
2621 }
2624 /*
2625 * Stuck with malloc/realloc.
2626 * For inline data, the underlying buffer must be
2627 * a multiple of 4 bytes in size so that it can be
2628 * logged and stay on word boundaries. We enforce
2629 * that here.
2630 */
2636 /*
2637 * Only do the realloc if the underlying size
2638 * is really changing.
2639 */
2643 real_size,
2645 KM_SLEEP);
2646 }
2652 }
2653 }
2657 }
2662 /*
2663 * Map inode to disk block and offset.
2664 *
2665 * mp -- the mount point structure for the current file system
2666 * tp -- the current transaction
2667 * ino -- the inode number of the inode to be located
2668 * imap -- this structure is filled in with the information necessary
2669 * to retrieve the given inode from disk
2670 * flags -- flags to pass to xfs_dilocate indicating whether or not
2671 * lookups in the inode btree were OK or not
2672 */
2673 int
2677 xfs_ino_t ino,
2679 uint flags)
2680 {
2681 xfs_fsblock_t fsbno;
2691 }
2698 }
2700 void
2704 {
2711 }
2713 /*
2714 * If the format is local, then we can't have an extents
2715 * array so just look for an inline data array. If we're
2716 * not local then we may or may not have an extents list,
2717 * so check and free it up if we do.
2718 */
2726 }
2734 }
2741 }
2742 }
2744 /*
2745 * This is called free all the memory associated with an inode.
2746 * It must free the inode itself and any buffers allocated for
2747 * if_extents/if_data and if_broot. It must also free the lock
2748 * associated with the inode.
2749 */
2750 void
2753 {
2761 }
2767 #ifdef XFS_BMAP_TRACE
2769 #endif
2770 #ifdef XFS_BMBT_TRACE
2772 #endif
2773 #ifdef XFS_RW_TRACE
2775 #endif
2776 #ifdef XFS_ILOCK_TRACE
2778 #endif
2779 #ifdef XFS_DIR2_TRACE
2781 #endif
2783 /* XXXdpd should be able to assert this but shutdown
2784 * is leaving the AIL behind. */
2788 }
2790 }
2793 /*
2794 * Increment the pin count of the given buffer.
2795 * This value is protected by ipinlock spinlock in the mount structure.
2796 */
2797 void
2800 {
2804 }
2806 /*
2807 * Decrement the pin count of the given inode, and wake up
2808 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2809 * inode must have been previoulsy pinned with a call to xfs_ipin().
2810 */
2811 void
2814 {
2820 /* make sync come back and flush this inode */
2826 }
2829 }
2830 }
2832 /*
2833 * This is called to wait for the given inode to be unpinned.
2834 * It will sleep until this happens. The caller must have the
2835 * inode locked in at least shared mode so that the buffer cannot
2836 * be subsequently pinned once someone is waiting for it to be
2837 * unpinned.
2838 */
2839 STATIC void
2842 {
2844 xfs_lsn_t lsn;
2850 }
2857 }
2859 /*
2860 * Give the log a push so we don't wait here too long.
2861 */
2865 }
2868 /*
2869 * xfs_iextents_copy()
2870 *
2871 * This is called to copy the REAL extents (as opposed to the delayed
2872 * allocation extents) from the inode into the given buffer. It
2873 * returns the number of bytes copied into the buffer.
2874 *
2875 * If there are no delayed allocation extents, then we can just
2876 * memcpy() the extents into the buffer. Otherwise, we need to
2877 * examine each extent in turn and skip those which are delayed.
2878 */
2879 int
2884 {
2888 #ifdef XFS_BMAP_TRACE
2890 #endif
2894 xfs_fsblock_t start_block;
2904 /*
2905 * There are some delayed allocation extents in the
2906 * inode, so copy the extents one at a time and skip
2907 * the delayed ones. There must be at least one
2908 * non-delayed extent.
2909 */
2916 /*
2917 * It's a delayed allocation extent, so skip it.
2918 */
2919 ep++;
2921 }
2923 /* Translate to on disk format */
2928 dest_ep++;
2929 ep++;
2930 copied++;
2931 }
2936 }
2938 /*
2939 * Each of the following cases stores data into the same region
2940 * of the on-disk inode, so only one of them can be valid at
2941 * any given time. While it is possible to have conflicting formats
2942 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2943 * in EXTENTS format, this can only happen when the fork has
2944 * changed formats after being modified but before being flushed.
2945 * In these cases, the format always takes precedence, because the
2946 * format indicates the current state of the fork.
2947 */
2948 /*ARGSUSED*/
2949 STATIC int
2956 {
2960 #ifdef XFS_TRANS_DEBUG
2962 #endif
2973 /*
2974 * This can happen if we gave up in iformat in an error path,
2975 * for the attribute fork.
2976 */
2980 }
2990 }
2996 }
2997 }
3009 whichfork);
3010 }
3019 XFS_BROOT_SIZE_ADJ));
3023 }
3030 }
3038 }
3044 }
3047 }
3049 /*
3050 * xfs_iflush() will write a modified inode's changes out to the
3051 * inode's on disk home. The caller must have the inode lock held
3052 * in at least shared mode and the inode flush semaphore must be
3053 * held as well. The inode lock will still be held upon return from
3054 * the call and the caller is free to unlock it.
3055 * The inode flush lock will be unlocked when the inode reaches the disk.
3056 * The flags indicate how the inode's buffer should be written out.
3057 */
3058 int
3061 uint flags)
3062 {
3068 /* REFERENCED */
3086 /*
3087 * If the inode isn't dirty, then just release the inode
3088 * flush lock and do nothing.
3089 */
3096 }
3098 /*
3099 * We can't flush the inode until it is unpinned, so
3100 * wait for it. We know noone new can pin it, because
3101 * we are holding the inode lock shared and you need
3102 * to hold it exclusively to pin the inode.
3103 */
3106 /*
3107 * This may have been unpinned because the filesystem is shutting
3108 * down forcibly. If that's the case we must not write this inode
3109 * to disk, because the log record didn't make it to disk!
3110 */
3117 }
3119 /*
3120 * Get the buffer containing the on-disk inode.
3121 */
3126 }
3128 /*
3129 * Decide how buffer will be flushed out. This is done before
3130 * the call to xfs_iflush_int because this field is zeroed by it.
3131 */
3133 /*
3134 * Flush out the inode buffer according to the directions
3135 * of the caller. In the cases where the caller has given
3136 * us a choice choose the non-delwri case. This is because
3137 * the inode is in the AIL and we need to get it out soon.
3138 */
3155 }
3173 }
3174 }
3176 /*
3177 * First flush out the inode that xfs_iflush was called with.
3178 */
3182 }
3184 /*
3185 * inode clustering:
3186 * see if other inodes can be gathered into this write
3187 */
3196 /*
3197 * Do an un-protected check to see if the inode is dirty and
3198 * is a candidate for flushing. These checks will be repeated
3199 * later after the appropriate locks are acquired.
3200 */
3207 }
3209 /*
3210 * Try to get locks. If any are unavailable,
3211 * then this inode cannot be flushed and is skipped.
3212 */
3214 /* get inode locks (just i_lock) */
3216 /* get inode flush lock */
3218 /* check if pinned */
3220 /* arriving here means that
3221 * this inode can be flushed.
3222 * first re-check that it's
3223 * dirty
3224 */
3229 clcount++;
3233 XFS_ILOCK_SHARED);
3235 }
3238 }
3241 }
3242 }
3244 }
3245 }
3251 }
3253 /*
3254 * If the buffer is pinned then push on the log so we won't
3255 * get stuck waiting in the write for too long.
3256 */
3259 }
3267 }
3270 corrupt_out:
3274 /*
3275 * Unlocks the flush lock
3276 */
3279 cluster_corrupt_out:
3280 /* Corruption detected in the clustering loop. Invalidate the
3281 * inode buffer and shut down the filesystem.
3282 */
3285 /*
3286 * Clean up the buffer. If it was B_DELWRI, just release it --
3287 * brelse can handle it with no problems. If not, shut down the
3288 * filesystem before releasing the buffer.
3289 */
3292 }
3297 /*
3298 * Just like incore_relse: if we have b_iodone functions,
3299 * mark the buffer as an error and call them. Otherwise
3300 * mark it as stale and brelse.
3301 */
3312 }
3313 }
3316 /*
3317 * Unlocks the flush lock
3318 */
3320 }
3323 STATIC int
3327 {
3331 #ifdef XFS_TRANS_DEBUG
3333 #endif
3345 /*
3346 * If the inode isn't dirty, then just release the inode
3347 * flush lock and do nothing.
3348 */
3353 }
3355 /* set *dip = inode's place in the buffer */
3358 /*
3359 * Clear i_update_core before copying out the data.
3360 * This is for coordination with our timestamp updates
3361 * that don't hold the inode lock. They will always
3362 * update the timestamps BEFORE setting i_update_core,
3363 * so if we clear i_update_core after they set it we
3364 * are guaranteed to see their updates to the timestamps.
3365 * I believe that this depends on strongly ordered memory
3366 * semantics, but we have that. We use the SYNCHRONIZE
3367 * macro to make sure that the compiler does not reorder
3368 * the i_update_core access below the data copy below.
3369 */
3379 }
3386 }
3396 }
3407 }
3408 }
3411 XFS_RANDOM_IFLUSH_5)) {
3417 ip);
3419 }
3426 }
3427 /*
3428 * bump the flush iteration count, used to detect flushes which
3429 * postdate a log record during recovery.
3430 */
3434 /*
3435 * Copy the dirty parts of the inode into the on-disk
3436 * inode. We always copy out the core of the inode,
3437 * because if the inode is dirty at all the core must
3438 * be.
3439 */
3442 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3446 /*
3447 * If this is really an old format inode and the superblock version
3448 * has not been updated to support only new format inodes, then
3449 * convert back to the old inode format. If the superblock version
3450 * has been updated, then make the conversion permanent.
3451 */
3456 /*
3457 * Convert it back.
3458 */
3462 /*
3463 * The superblock version has already been bumped,
3464 * so just make the conversion to the new inode
3465 * format permanent.
3466 */
3475 }
3476 }
3480 }
3483 /*
3484 * The only error from xfs_iflush_fork is on the data fork.
3485 */
3487 }
3490 /*
3491 * We've recorded everything logged in the inode, so we'd
3492 * like to clear the ilf_fields bits so we don't log and
3493 * flush things unnecessarily. However, we can't stop
3494 * logging all this information until the data we've copied
3495 * into the disk buffer is written to disk. If we did we might
3496 * overwrite the copy of the inode in the log with all the
3497 * data after re-logging only part of it, and in the face of
3498 * a crash we wouldn't have all the data we need to recover.
3499 *
3500 * What we do is move the bits to the ili_last_fields field.
3501 * When logging the inode, these bits are moved back to the
3502 * ilf_fields field. In the xfs_iflush_done() routine we
3503 * clear ili_last_fields, since we know that the information
3504 * those bits represent is permanently on disk. As long as
3505 * the flush completes before the inode is logged again, then
3506 * both ilf_fields and ili_last_fields will be cleared.
3507 *
3508 * We can play with the ilf_fields bits here, because the inode
3509 * lock must be held exclusively in order to set bits there
3510 * and the flush lock protects the ili_last_fields bits.
3511 * Set ili_logged so the flush done
3512 * routine can tell whether or not to look in the AIL.
3513 * Also, store the current LSN of the inode so that we can tell
3514 * whether the item has moved in the AIL from xfs_iflush_done().
3515 * In order to read the lsn we need the AIL lock, because
3516 * it is a 64 bit value that cannot be read atomically.
3517 */
3528 /*
3529 * Attach the function xfs_iflush_done to the inode's
3530 * buffer. This will remove the inode from the AIL
3531 * and unlock the inode's flush lock when the inode is
3532 * completely written to disk.
3533 */
3540 /*
3541 * We're flushing an inode which is not in the AIL and has
3542 * not been logged but has i_update_core set. For this
3543 * case we can use a B_DELWRI flush and immediately drop
3544 * the inode flush lock because we can avoid the whole
3545 * AIL state thing. It's OK to drop the flush lock now,
3546 * because we've already locked the buffer and to do anything
3547 * you really need both.
3548 */
3553 }
3555 }
3559 corrupt_out:
3561 }
3564 /*
3565 * Flush all inactive inodes in mp.
3566 */
3567 void
3570 {
3574 again:
3581 /* Make sure we skip markers inserted by sync */
3585 }
3592 }
3598 out:
3600 }
3602 /*
3603 * xfs_iaccess: check accessibility of inode for mode.
3604 */
3605 int
3608 mode_t mode,
3610 {
3624 }
3626 /*
3627 * If there's an Access Control List it's used instead of
3628 * the mode bits.
3629 */
3637 }
3639 /*
3640 * If the DACs are ok we don't need any capability check.
3641 */
3644 /*
3645 * Read/write DACs are always overridable.
3646 * Executable DACs are overridable if at least one exec bit is set.
3647 */
3657 #ifdef NOISE
3661 }
3663 }
3665 /*
3666 * xfs_iroundup: round up argument to next power of two
3667 */
3668 uint
3670 uint v)
3671 {
3673 uint m;
3686 }
3689 }
3691 /*
3692 * Change the requested timestamp in the given inode.
3693 * We don't lock across timestamp updates, and we don't log them but
3694 * we do record the fact that there is dirty information in core.
3695 *
3696 * NOTE -- callers MUST combine XFS_ICHGTIME_MOD or XFS_ICHGTIME_CHG
3697 * with XFS_ICHGTIME_ACC to be sure that access time
3698 * update will take. Calling first with XFS_ICHGTIME_ACC
3699 * and then XFS_ICHGTIME_MOD may fail to modify the access
3700 * timestamp if the filesystem is mounted noacctm.
3701 */
3702 void
3705 {
3706 timespec_t tv;
3710 /*
3711 * We're not supposed to change timestamps in readonly-mounted
3712 * filesystems. Throw it away if anyone asks us.
3713 */
3717 /*
3718 * Don't update access timestamps on reads if mounted "noatime"
3719 * Throw it away if anyone asks us.
3720 */
3731 }
3736 }
3741 }
3743 /*
3744 * We update the i_update_core field _after_ changing
3745 * the timestamps in order to coordinate properly with
3746 * xfs_iflush() so that we don't lose timestamp updates.
3747 * This keeps us from having to hold the inode lock
3748 * while doing this. We use the SYNCHRONIZE macro to
3749 * ensure that the compiler does not reorder the update
3750 * of i_update_core above the timestamp updates above.
3751 */
3756 }
3758 #ifdef XFS_ILOCK_TRACE
3761 void
3763 {
3772 }
3773 #endif