| blob | bc8c8c7f9039d507978d727a25c446b661af4d56 |
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 * called from bwrite on xfs inode buffers
150 */
151 void
153 {
178 }
187 }
188 }
191 }
193 /*
194 * This routine is called to map an inode number within a file
195 * system to the buffer containing the on-disk version of the
196 * inode. It returns a pointer to the buffer containing the
197 * on-disk inode in the bpp parameter, and in the dip parameter
198 * it returns a pointer to the on-disk inode within that buffer.
199 *
200 * If a non-zero error is returned, then the contents of bpp and
201 * dipp are undefined.
202 *
203 * Use xfs_imap() to determine the size and location of the
204 * buffer to read from disk.
205 */
206 int
210 xfs_ino_t ino,
214 {
216 xfs_imap_t imap;
221 /*
222 * Call the space managment code to find the location of the
223 * inode on disk.
224 */
229 "xfs_inotobp: xfs_imap() returned an "
232 }
234 /*
235 * If the inode number maps to a block outside the bounds of the
236 * file system then return NULL rather than calling read_buf
237 * and panicing when we get an error from the driver.
238 */
242 "xfs_inotobp: inode number (%d + %d) maps to a block outside the bounds "
246 }
248 /*
249 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
250 * default to just a read_buf() call.
251 */
257 "xfs_inotobp: xfs_trans_read_buf() returned an "
260 }
262 di_ok =
266 XFS_RANDOM_ITOBP_INOTOBP))) {
270 "xfs_inotobp: XFS_TEST_ERROR() returned an "
273 }
277 /*
278 * Set *dipp to point to the on-disk inode in the buffer.
279 */
284 }
287 /*
288 * This routine is called to map an inode to the buffer containing
289 * the on-disk version of the inode. It returns a pointer to the
290 * buffer containing the on-disk inode in the bpp parameter, and in
291 * the dip parameter it returns a pointer to the on-disk inode within
292 * that buffer.
293 *
294 * If a non-zero error is returned, then the contents of bpp and
295 * dipp are undefined.
296 *
297 * If the inode is new and has not yet been initialized, use xfs_imap()
298 * to determine the size and location of the buffer to read from disk.
299 * If the inode has already been mapped to its buffer and read in once,
300 * then use the mapping information stored in the inode rather than
301 * calling xfs_imap(). This allows us to avoid the overhead of looking
302 * at the inode btree for small block file systems (see xfs_dilocate()).
303 * We can tell whether the inode has been mapped in before by comparing
304 * its disk block address to 0. Only uninitialized inodes will have
305 * 0 for the disk block address.
306 */
307 int
314 xfs_daddr_t bno)
315 {
318 xfs_imap_t imap;
319 #ifdef __KERNEL__
322 #endif
325 /*
326 * Call the space management code to find the location of the
327 * inode on disk.
328 */
333 }
335 /*
336 * If the inode number maps to a block outside the bounds
337 * of the file system then return NULL rather than calling
338 * read_buf and panicing when we get an error from the
339 * driver.
340 */
343 #ifdef DEBUG
345 "(imap.im_blkno (0x%llx) "
346 "+ imap.im_len (0x%llx)) > "
347 " XFS_FSB_TO_BB(mp, "
354 }
356 /*
357 * Fill in the fields in the inode that will be used to
358 * map the inode to its buffer from now on.
359 */
364 /*
365 * We've already mapped the inode once, so just use the
366 * mapping that we saved the first time.
367 */
371 }
374 /*
375 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
376 * default to just a read_buf() call.
377 */
382 #ifdef DEBUG
384 "xfs_trans_read_buf() returned error %d, "
390 }
391 #ifdef __KERNEL__
392 /*
393 * Validate the magic number and version of every inode in the buffer
394 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
395 */
396 #ifdef DEBUG
398 #else
400 #endif
410 XFS_RANDOM_ITOBP_INOTOBP))) {
411 #ifdef DEBUG
416 #endif
421 }
422 }
427 /*
428 * Mark the buffer as an inode buffer now that it looks good
429 */
432 /*
433 * Set *dipp to point to the on-disk inode in the buffer.
434 */
438 }
440 /*
441 * Move inode type and inode format specific information from the
442 * on-disk inode to the in-core inode. For fifos, devs, and sockets
443 * this means set if_rdev to the proper value. For files, directories,
444 * and symlinks this means to bring in the in-line data or extent
445 * pointers. For a file in B-tree format, only the root is immediately
446 * brought in-core. The rest will be in-lined in if_extents when it
447 * is first referenced (see xfs_iread_extents()).
448 */
449 STATIC int
453 {
457 xfs_fsize_t di_size;
467 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu."
477 }
479 if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) {
481 "corrupt dinode %Lu, forkoff = 0x%x."
488 }
499 }
509 /*
510 * no local regular files yet
511 */
517 XFS_ERRLEVEL_LOW,
520 }
529 XFS_ERRLEVEL_LOW,
532 }
547 }
553 }
556 }
578 }
583 }
585 }
587 /*
588 * The file is in-lined in the on-disk inode.
589 * If it fits into if_inline_data, then copy
590 * it there, otherwise allocate a buffer for it
591 * and copy the data there. Either way, set
592 * if_data to point at the data.
593 * If we allocate a buffer for the data, make
594 * sure that its size is a multiple of 4 and
595 * record the real size in i_real_bytes.
596 */
597 STATIC int
603 {
607 /*
608 * If the size is unreasonable, then something
609 * is wrong and we just bail out rather than crash in
610 * kmem_alloc() or memcpy() below.
611 */
620 }
630 }
638 }
640 /*
641 * The file consists of a set of extents all
642 * of which fit into the on-disk inode.
643 * If there are few enough extents to fit into
644 * the if_inline_ext, then copy them there.
645 * Otherwise allocate a buffer for them and copy
646 * them into it. Either way, set if_extents
647 * to point at the extents.
648 */
649 STATIC int
654 {
666 /*
667 * If the number of extents is unreasonable, then something
668 * is wrong and we just bail out rather than crash in
669 * kmem_alloc() or memcpy() below.
670 */
678 }
689 }
698 ARCH_CONVERT);
700 ARCH_CONVERT);
701 }
703 whichfork);
709 XFS_ERRLEVEL_LOW,
712 }
713 }
716 }
718 /*
719 * The file has too many extents to fit into
720 * the inode, so they are in B-tree format.
721 * Allocate a buffer for the root of the B-tree
722 * and copy the root into it. The i_extents
723 * field will remain NULL until all of the
724 * extents are read in (when they are needed).
725 */
726 STATIC int
731 {
734 /* REFERENCED */
743 /*
744 * blow out if -- fork has less extents than can fit in
745 * fork (fork shouldn't be a btree format), root btree
746 * block has more records than can fit into the fork,
747 * or the number of extents is greater than the number of
748 * blocks.
749 */
760 }
765 /*
766 * Copy and convert from the on-disk structure
767 * to the in-memory structure.
768 */
775 }
777 /*
778 * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
779 * and native format
780 *
781 * buf = on-disk representation
782 * dip = native representation
783 * dir = direction - +ve -> disk to native
784 * -ve -> native to disk
785 */
786 void
788 xfs_caddr_t buf,
791 {
814 }
841 }
843 STATIC uint
846 __uint16_t di_flags)
847 {
871 }
874 }
876 uint
879 {
884 }
886 uint
889 {
892 }
894 /*
895 * Given a mount structure and an inode number, return a pointer
896 * to a newly allocated in-core inode coresponding to the given
897 * inode number.
898 *
899 * Initialize the inode's attributes and extent pointers if it
900 * already has them (it will not if the inode has no links).
901 */
902 int
906 xfs_ino_t ino,
908 xfs_daddr_t bno)
909 {
921 /*
922 * Get pointer's to the on-disk inode and the buffer containing it.
923 * If the inode number refers to a block outside the file system
924 * then xfs_itobp() will return NULL. In this case we should
925 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
926 * know that this is a new incore inode.
927 */
933 }
935 /*
936 * Initialize inode's trace buffers.
937 * Do this before xfs_iformat in case it adds entries.
938 */
939 #ifdef XFS_BMAP_TRACE
941 #endif
942 #ifdef XFS_BMBT_TRACE
944 #endif
945 #ifdef XFS_RW_TRACE
947 #endif
948 #ifdef XFS_ILOCK_TRACE
950 #endif
951 #ifdef XFS_DIR2_TRACE
953 #endif
955 /*
956 * If we got something that isn't an inode it means someone
957 * (nfs or dmi) has a stale handle.
958 */
962 #ifdef DEBUG
964 "dip->di_core.di_magic (0x%x) != "
967 XFS_DINODE_MAGIC);
970 }
972 /*
973 * If the on-disk inode is already linked to a directory
974 * entry, copy all of the inode into the in-core inode.
975 * xfs_iformat() handles copying in the inode format
976 * specific information.
977 * Otherwise, just get the truly permanent information.
978 */
986 #ifdef DEBUG
989 error);
992 }
998 /*
999 * Make sure to pull in the mode here as well in
1000 * case the inode is released without being used.
1001 * This ensures that xfs_inactive() will see that
1002 * the inode is already free and not try to mess
1003 * with the uninitialized part of it.
1004 */
1006 /*
1007 * Initialize the per-fork minima and maxima for a new
1008 * inode here. xfs_iformat will do it for old inodes.
1009 */
1012 }
1016 /*
1017 * The inode format changed when we moved the link count and
1018 * made it 32 bits long. If this is an old format inode,
1019 * convert it in memory to look like a new one. If it gets
1020 * flushed to disk we will convert back before flushing or
1021 * logging it. We zero out the new projid field and the old link
1022 * count field. We'll handle clearing the pad field (the remains
1023 * of the old uuid field) when we actually convert the inode to
1024 * the new format. We don't change the version number so that we
1025 * can distinguish this from a real new format inode.
1026 */
1031 }
1035 /*
1036 * Mark the buffer containing the inode as something to keep
1037 * around for a while. This helps to keep recently accessed
1038 * meta-data in-core longer.
1039 */
1042 /*
1043 * Use xfs_trans_brelse() to release the buffer containing the
1044 * on-disk inode, because it was acquired with xfs_trans_read_buf()
1045 * in xfs_itobp() above. If tp is NULL, this is just a normal
1046 * brelse(). If we're within a transaction, then xfs_trans_brelse()
1047 * will only release the buffer if it is not dirty within the
1048 * transaction. It will be OK to release the buffer in this case,
1049 * because inodes on disk are never destroyed and we will be
1050 * locking the new in-core inode before putting it in the hash
1051 * table where other processes can find it. Thus we don't have
1052 * to worry about the inode being changed just because we released
1053 * the buffer.
1054 */
1058 }
1060 /*
1061 * Read in extents from a btree-format inode.
1062 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1063 */
1064 int
1069 {
1078 }
1081 /*
1082 * We know that the size is valid (it's checked in iformat_btree)
1083 */
1096 }
1100 }
1102 /*
1103 * Allocate an inode on disk and return a copy of its in-core version.
1104 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1105 * appropriately within the inode. The uid and gid for the inode are
1106 * set according to the contents of the given cred structure.
1107 *
1108 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1109 * has a free inode available, call xfs_iget()
1110 * to obtain the in-core version of the allocated inode. Finally,
1111 * fill in the inode and log its initial contents. In this case,
1112 * ialloc_context would be set to NULL and call_again set to false.
1113 *
1114 * If xfs_dialloc() does not have an available inode,
1115 * it will replenish its supply by doing an allocation. Since we can
1116 * only do one allocation within a transaction without deadlocks, we
1117 * must commit the current transaction before returning the inode itself.
1118 * In this case, therefore, we will set call_again to true and return.
1119 * The caller should then commit the current transaction, start a new
1120 * transaction, and call xfs_ialloc() again to actually get the inode.
1121 *
1122 * To ensure that some other process does not grab the inode that
1123 * was allocated during the first call to xfs_ialloc(), this routine
1124 * also returns the [locked] bp pointing to the head of the freelist
1125 * as ialloc_context. The caller should hold this buffer across
1126 * the commit and pass it back into this routine on the second call.
1127 */
1128 int
1132 mode_t mode,
1133 xfs_nlink_t nlink,
1134 xfs_dev_t rdev,
1136 xfs_prid_t prid,
1141 {
1142 xfs_ino_t ino;
1145 uint flags;
1148 /*
1149 * Call the space management code to pick
1150 * the on-disk inode to be allocated.
1151 */
1156 }
1160 }
1163 /*
1164 * Get the in-core inode with the lock held exclusively.
1165 * This is because we're setting fields here we need
1166 * to prevent others from looking at until we're done.
1167 */
1172 }
1186 /*
1187 * If the superblock version is up to where we support new format
1188 * inodes and this is currently an old format inode, then change
1189 * the inode version number now. This way we only do the conversion
1190 * here rather than here and in the flush/logging code.
1191 */
1195 /*
1196 * We've already zeroed the old link count, the projid field,
1197 * and the pad field.
1198 */
1199 }
1201 /*
1202 * Project ids won't be stored on disk if we are using a version 1 inode.
1203 */
1211 }
1212 }
1214 /*
1215 * If the group ID of the new file does not match the effective group
1216 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1217 * (and only if the irix_sgid_inherit compatibility variable is set).
1218 */
1223 }
1229 /*
1230 * di_gen will have been taken care of in xfs_iread.
1231 */
1256 }
1257 }
1259 xfs_inherit_noatime)
1262 xfs_inherit_nodump)
1265 xfs_inherit_sync)
1268 xfs_inherit_nosymlinks)
1270 }
1271 /* FALLTHROUGH */
1280 }
1281 /*
1282 * Attribute fork settings for new inode.
1283 */
1287 /*
1288 * Log the new values stuffed into the inode.
1289 */
1292 /* now that we have a v_type we can set Linux inode ops (& unlock) */
1297 }
1299 /*
1300 * Check to make sure that there are no blocks allocated to the
1301 * file beyond the size of the file. We don't check this for
1302 * files with fixed size extents or real time extents, but we
1303 * at least do it for regular files.
1304 */
1305 #ifdef DEBUG
1306 void
1310 xfs_fsize_t isize)
1311 {
1312 xfs_fileoff_t map_first;
1324 /*
1325 * The filesystem could be shutting down, so bmapi may return
1326 * an error.
1327 */
1331 map_first),
1333 NULL))
1337 }
1340 /*
1341 * Calculate the last possible buffered byte in a file. This must
1342 * include data that was buffered beyond the EOF by the write code.
1343 * This also needs to deal with overflowing the xfs_fsize_t type
1344 * which can happen for sizes near the limit.
1345 *
1346 * We also need to take into account any blocks beyond the EOF. It
1347 * may be the case that they were buffered by a write which failed.
1348 * In that case the pages will still be in memory, but the inode size
1349 * will never have been updated.
1350 */
1351 xfs_fsize_t
1354 {
1356 xfs_fsize_t last_byte;
1357 xfs_fileoff_t last_block;
1358 xfs_fileoff_t size_last_block;
1364 /*
1365 * Only check for blocks beyond the EOF if the extents have
1366 * been read in. This eliminates the need for the inode lock,
1367 * and it also saves us from looking when it really isn't
1368 * necessary.
1369 */
1372 XFS_DATA_FORK);
1375 }
1378 }
1385 }
1389 }
1391 }
1393 #if defined(XFS_RW_TRACE)
1394 STATIC void
1399 xfs_fsize_t new_size,
1400 xfs_off_t toss_start,
1401 xfs_off_t toss_finish)
1402 {
1405 }
1424 }
1425 #else
1426 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1427 #endif
1429 /*
1430 * Start the truncation of the file to new_size. The new size
1431 * must be smaller than the current size. This routine will
1432 * clear the buffer and page caches of file data in the removed
1433 * range, and xfs_itruncate_finish() will remove the underlying
1434 * disk blocks.
1435 *
1436 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1437 * must NOT have the inode lock held at all. This is because we're
1438 * calling into the buffer/page cache code and we can't hold the
1439 * inode lock when we do so.
1440 *
1441 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1442 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1443 * in the case that the caller is locking things out of order and
1444 * may not be able to call xfs_itruncate_finish() with the inode lock
1445 * held without dropping the I/O lock. If the caller must drop the
1446 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1447 * must be called again with all the same restrictions as the initial
1448 * call.
1449 */
1450 void
1453 uint flags,
1454 xfs_fsize_t new_size)
1455 {
1456 xfs_fsize_t last_byte;
1457 xfs_off_t toss_start;
1468 /*
1469 * Call VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES() to get rid of pages and buffers
1470 * overlapping the region being removed. We have to use
1471 * the less efficient VOP_FLUSHINVAL_PAGES() in the case that the
1472 * caller may not be able to finish the truncate without
1473 * dropping the inode's I/O lock. Make sure
1474 * to catch any pages brought in by buffers overlapping
1475 * the EOF by searching out beyond the isize by our
1476 * block size. We round new_size up to a block boundary
1477 * so that we don't toss things on the same block as
1478 * new_size but before it.
1479 *
1480 * Before calling VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES(), make sure to
1481 * call remapf() over the same region if the file is mapped.
1482 * This frees up mapped file references to the pages in the
1483 * given range and for the VOP_FLUSHINVAL_PAGES() case it ensures
1484 * that we get the latest mapped changes flushed out.
1485 */
1489 /*
1490 * The place to start tossing is beyond our maximum
1491 * file size, so there is no way that the data extended
1492 * out there.
1493 */
1495 }
1498 last_byte);
1504 }
1505 }
1507 #ifdef DEBUG
1510 }
1511 #endif
1512 }
1514 /*
1515 * Shrink the file to the given new_size. The new
1516 * size must be smaller than the current size.
1517 * This will free up the underlying blocks
1518 * in the removed range after a call to xfs_itruncate_start()
1519 * or xfs_atruncate_start().
1520 *
1521 * The transaction passed to this routine must have made
1522 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1523 * This routine may commit the given transaction and
1524 * start new ones, so make sure everything involved in
1525 * the transaction is tidy before calling here.
1526 * Some transaction will be returned to the caller to be
1527 * committed. The incoming transaction must already include
1528 * the inode, and both inode locks must be held exclusively.
1529 * The inode must also be "held" within the transaction. On
1530 * return the inode will be "held" within the returned transaction.
1531 * This routine does NOT require any disk space to be reserved
1532 * for it within the transaction.
1533 *
1534 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1535 * and it indicates the fork which is to be truncated. For the
1536 * attribute fork we only support truncation to size 0.
1537 *
1538 * We use the sync parameter to indicate whether or not the first
1539 * transaction we perform might have to be synchronous. For the attr fork,
1540 * it needs to be so if the unlink of the inode is not yet known to be
1541 * permanent in the log. This keeps us from freeing and reusing the
1542 * blocks of the attribute fork before the unlink of the inode becomes
1543 * permanent.
1544 *
1545 * For the data fork, we normally have to run synchronously if we're
1546 * being called out of the inactive path or we're being called
1547 * out of the create path where we're truncating an existing file.
1548 * Either way, the truncate needs to be sync so blocks don't reappear
1549 * in the file with altered data in case of a crash. wsync filesystems
1550 * can run the first case async because anything that shrinks the inode
1551 * has to run sync so by the time we're called here from inactive, the
1552 * inode size is permanently set to 0.
1553 *
1554 * Calls from the truncate path always need to be sync unless we're
1555 * in a wsync filesystem and the file has already been unlinked.
1556 *
1557 * The caller is responsible for correctly setting the sync parameter.
1558 * It gets too hard for us to guess here which path we're being called
1559 * out of just based on inode state.
1560 */
1561 int
1565 xfs_fsize_t new_size,
1568 {
1569 xfs_fsblock_t first_block;
1570 xfs_fileoff_t first_unmap_block;
1571 xfs_fileoff_t last_block;
1577 xfs_bmap_free_t free_list;
1594 /*
1595 * We only support truncating the entire attribute fork.
1596 */
1599 }
1602 /*
1603 * The first thing we do is set the size to new_size permanently
1604 * on disk. This way we don't have to worry about anyone ever
1605 * being able to look at the data being freed even in the face
1606 * of a crash. What we're getting around here is the case where
1607 * we free a block, it is allocated to another file, it is written
1608 * to, and then we crash. If the new data gets written to the
1609 * file but the log buffers containing the free and reallocation
1610 * don't, then we'd end up with garbage in the blocks being freed.
1611 * As long as we make the new_size permanent before actually
1612 * freeing any blocks it doesn't matter if they get writtten to.
1613 *
1614 * The callers must signal into us whether or not the size
1615 * setting here must be synchronous. There are a few cases
1616 * where it doesn't have to be synchronous. Those cases
1617 * occur if the file is unlinked and we know the unlink is
1618 * permanent or if the blocks being truncated are guaranteed
1619 * to be beyond the inode eof (regardless of the link count)
1620 * and the eof value is permanent. Both of these cases occur
1621 * only on wsync-mounted filesystems. In those cases, we're
1622 * guaranteed that no user will ever see the data in the blocks
1623 * that are being truncated so the truncate can run async.
1624 * In the free beyond eof case, the file may wind up with
1625 * more blocks allocated to it than it needs if we crash
1626 * and that won't get fixed until the next time the file
1627 * is re-opened and closed but that's ok as that shouldn't
1628 * be too many blocks.
1629 *
1630 * However, we can't just make all wsync xactions run async
1631 * because there's one call out of the create path that needs
1632 * to run sync where it's truncating an existing file to size
1633 * 0 whose size is > 0.
1634 *
1635 * It's probably possible to come up with a test in this
1636 * routine that would correctly distinguish all the above
1637 * cases from the values of the function parameters and the
1638 * inode state but for sanity's sake, I've decided to let the
1639 * layers above just tell us. It's simpler to correctly figure
1640 * out in the layer above exactly under what conditions we
1641 * can run async and I think it's easier for others read and
1642 * follow the logic in case something has to be changed.
1643 * cscope is your friend -- rcc.
1644 *
1645 * The attribute fork is much simpler.
1646 *
1647 * For the attribute fork we allow the caller to tell us whether
1648 * the unlink of the inode that led to this call is yet permanent
1649 * in the on disk log. If it is not and we will be freeing extents
1650 * in this inode then we make the first transaction synchronous
1651 * to make sure that the unlink is permanent by the time we free
1652 * the blocks.
1653 */
1658 }
1663 }
1669 /*
1670 * Since it is possible for space to become allocated beyond
1671 * the end of the file (in a crash where the space is allocated
1672 * but the inode size is not yet updated), simply remove any
1673 * blocks which show up between the new EOF and the maximum
1674 * possible file size. If the first block to be removed is
1675 * beyond the maximum file size (ie it is the same as last_block),
1676 * then there is nothing to do.
1677 */
1685 }
1687 /*
1688 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1689 * will tell us whether it freed the entire range or
1690 * not. If this is a synchronous mount (wsync),
1691 * then we can tell bunmapi to keep all the
1692 * transactions asynchronous since the unlink
1693 * transaction that made this inode inactive has
1694 * already hit the disk. There's no danger of
1695 * the freed blocks being reused, there being a
1696 * crash, and the reused blocks suddenly reappearing
1697 * in this file with garbage in them once recovery
1698 * runs.
1699 */
1702 unmap_len,
1705 XFS_ITRUNC_MAX_EXTENTS,
1708 /*
1709 * If the bunmapi call encounters an error,
1710 * return to the caller where the transaction
1711 * can be properly aborted. We just need to
1712 * make sure we're not holding any resources
1713 * that we were not when we came in.
1714 */
1717 }
1719 /*
1720 * Duplicate the transaction that has the permanent
1721 * reservation and commit the old transaction.
1722 */
1727 /*
1728 * If the bmap finish call encounters an error,
1729 * return to the caller where the transaction
1730 * can be properly aborted. We just need to
1731 * make sure we're not holding any resources
1732 * that we were not when we came in.
1733 *
1734 * Aborting from this point might lose some
1735 * blocks in the file system, but oh well.
1736 */
1739 /*
1740 * If the passed in transaction committed
1741 * in xfs_bmap_finish(), then we want to
1742 * add the inode to this one before returning.
1743 * This keeps things simple for the higher
1744 * level code, because it always knows that
1745 * the inode is locked and held in the
1746 * transaction that returns to it whether
1747 * errors occur or not. We don't mark the
1748 * inode dirty so that this transaction can
1749 * be easily aborted if possible.
1750 */
1754 }
1756 }
1759 /*
1760 * The first xact was committed,
1761 * so add the inode to the new one.
1762 * Mark it dirty so it will be logged
1763 * and moved forward in the log as
1764 * part of every commit.
1765 */
1770 }
1775 XFS_TRANS_PERM_LOG_RES,
1776 XFS_ITRUNCATE_LOG_COUNT);
1777 /*
1778 * Add the inode being truncated to the next chained
1779 * transaction.
1780 */
1785 }
1786 /*
1787 * Only update the size in the case of the data fork, but
1788 * always re-log the inode so that our permanent transaction
1789 * can keep on rolling it forward in the log.
1790 */
1794 }
1804 }
1807 /*
1808 * xfs_igrow_start
1809 *
1810 * Do the first part of growing a file: zero any data in the last
1811 * block that is beyond the old EOF. We need to do this before
1812 * the inode is joined to the transaction to modify the i_size.
1813 * That way we can drop the inode lock and call into the buffer
1814 * cache to get the buffer mapping the EOF.
1815 */
1816 int
1819 xfs_fsize_t new_size,
1821 {
1822 xfs_fsize_t isize;
1831 /*
1832 * Zero any pages that may have been created by
1833 * xfs_write_file() beyond the end of the file
1834 * and any blocks between the old and new file sizes.
1835 */
1837 new_size);
1839 }
1841 /*
1842 * xfs_igrow_finish
1843 *
1844 * This routine is called to extend the size of a file.
1845 * The inode must have both the iolock and the ilock locked
1846 * for update and it must be a part of the current transaction.
1847 * The xfs_igrow_start() function must have been called previously.
1848 * If the change_flag is not zero, the inode change timestamp will
1849 * be updated.
1850 */
1851 void
1855 xfs_fsize_t new_size,
1857 {
1863 /*
1864 * Update the file size. Update the inode change timestamp
1865 * if change_flag set.
1866 */
1872 }
1875 /*
1876 * This is called when the inode's link count goes to 0.
1877 * We place the on-disk inode on a list in the AGI. It
1878 * will be pulled from this list when the inode is freed.
1879 */
1880 int
1884 {
1890 xfs_agnumber_t agno;
1891 xfs_daddr_t agdaddr;
1892 xfs_agino_t agino;
1907 /*
1908 * Get the agi buffer first. It ensures lock ordering
1909 * on the list.
1910 */
1915 }
1916 /*
1917 * Validate the magic number of the agi block.
1918 */
1920 agi_ok =
1924 XFS_RANDOM_IUNLINK))) {
1928 }
1929 /*
1930 * Get the index into the agi hash table for the
1931 * list this inode will go on.
1932 */
1940 /*
1941 * There is already another inode in the bucket we need
1942 * to add ourselves to. Add us at the front of the list.
1943 * Here we put the head pointer into our next pointer,
1944 * and then we fall through to point the head at us.
1945 */
1949 }
1952 /* both on-disk, don't endian flip twice */
1960 }
1962 /*
1963 * Point the bucket head pointer at the inode being inserted.
1964 */
1972 }
1974 /*
1975 * Pull the on-disk inode from the AGI unlinked list.
1976 */
1977 STATIC int
1981 {
1982 xfs_ino_t next_ino;
1988 xfs_agnumber_t agno;
1989 xfs_daddr_t agdaddr;
1990 xfs_agino_t agino;
1991 xfs_agino_t next_agino;
1999 /*
2000 * First pull the on-disk inode from the AGI unlinked list.
2001 */
2007 /*
2008 * Get the agi buffer first. It ensures lock ordering
2009 * on the list.
2010 */
2018 }
2019 /*
2020 * Validate the magic number of the agi block.
2021 */
2023 agi_ok =
2027 XFS_RANDOM_IUNLINK_REMOVE))) {
2035 }
2036 /*
2037 * Get the index into the agi hash table for the
2038 * list this inode will go on.
2039 */
2047 /*
2048 * We're at the head of the list. Get the inode's
2049 * on-disk buffer to see if there is anyone after us
2050 * on the list. Only modify our next pointer if it
2051 * is not already NULLAGINO. This saves us the overhead
2052 * of dealing with the buffer when there is no need to
2053 * change it.
2054 */
2061 }
2074 }
2075 /*
2076 * Point the bucket head pointer at the next inode.
2077 */
2086 /*
2087 * We need to search the list for the inode being freed.
2088 */
2092 /*
2093 * If the last inode wasn't the one pointing to
2094 * us, then release its buffer since we're not
2095 * going to do anything with it.
2096 */
2099 }
2108 }
2112 }
2113 /*
2114 * Now last_ibp points to the buffer previous to us on
2115 * the unlinked list. Pull us from the list.
2116 */
2123 }
2137 }
2138 /*
2139 * Point the previous inode on the list to the next inode.
2140 */
2148 }
2150 }
2153 {
2157 }
2159 void
2163 xfs_ino_t inum)
2164 {
2170 xfs_daddr_t blkno;
2187 }
2196 /*
2197 * Look for each inode in memory and attempt to lock it,
2198 * we can be racing with flush and tail pushing here.
2199 * any inode we get the locks on, add to an array of
2200 * inode items to process later.
2201 *
2202 * The get the buffer lock, we could beat a flush
2203 * or tail pushing thread to the lock here, in which
2204 * case they will go looking for the inode buffer
2205 * and fail, we need some other form of interlock
2206 * here.
2207 */
2215 }
2217 /* Inode not in memory or we found it already,
2218 * nothing to do
2219 */
2223 }
2228 }
2230 /* If we can get the locks then add it to the
2231 * list, otherwise by the time we get the bp lock
2232 * below it will already be attached to the
2233 * inode buffer.
2234 */
2236 /* This inode will already be locked - by us, lets
2237 * keep it that way.
2238 */
2248 }
2249 }
2252 }
2263 }
2266 }
2267 }
2270 }
2274 XFS_BUF_LOCK);
2287 pre_flushed++;
2288 }
2290 }
2301 }
2315 }
2316 }
2321 }
2324 }
2326 /*
2327 * This is called to return an inode to the inode free list.
2328 * The inode should already be truncated to 0 length and have
2329 * no pages associated with it. This routine also assumes that
2330 * the inode is already a part of the transaction.
2331 *
2332 * The on-disk copy of the inode will have been added to the list
2333 * of unlinked inodes in the AGI. We need to remove the inode from
2334 * that list atomically with respect to freeing it here.
2335 */
2336 int
2341 {
2344 xfs_ino_t first_ino;
2355 /*
2356 * Pull the on-disk inode from the AGI unlinked list.
2357 */
2361 }
2366 }
2375 /*
2376 * Bump the generation count so no one will be confused
2377 * by reincarnations of this inode.
2378 */
2384 }
2387 }
2389 /*
2390 * Reallocate the space for if_broot based on the number of records
2391 * being added or deleted as indicated in rec_diff. Move the records
2392 * and pointers in if_broot to fit the new size. When shrinking this
2393 * will eliminate holes between the records and pointers created by
2394 * the caller. When growing this will create holes to be filled in
2395 * by the caller.
2396 *
2397 * The caller must not request to add more records than would fit in
2398 * the on-disk inode root. If the if_broot is currently NULL, then
2399 * if we adding records one will be allocated. The caller must also
2400 * not request that the number of records go below zero, although
2401 * it can go to zero.
2402 *
2403 * ip -- the inode whose if_broot area is changing
2404 * ext_diff -- the change in the number of records, positive or negative,
2405 * requested for the if_broot array.
2406 */
2407 void
2412 {
2421 /*
2422 * Handle the degenerate case quietly.
2423 */
2426 }
2430 /*
2431 * If there wasn't any memory allocated before, just
2432 * allocate it now and get out.
2433 */
2437 KM_SLEEP);
2440 }
2442 /*
2443 * If there is already an existing if_broot, then we need
2444 * to realloc() it and shift the pointers to their new
2445 * location. The records don't change location because
2446 * they are kept butted up against the btree block header.
2447 */
2453 new_size,
2455 KM_SLEEP);
2465 }
2467 /*
2468 * rec_diff is less than 0. In this case, we are shrinking the
2469 * if_broot buffer. It must already exist. If we go to zero
2470 * records, just get rid of the root and clear the status bit.
2471 */
2478 else
2482 /*
2483 * First copy over the btree block header.
2484 */
2489 }
2491 /*
2492 * Only copy the records and pointers if there are any.
2493 */
2495 /*
2496 * First copy the records.
2497 */
2504 /*
2505 * Then copy the pointers.
2506 */
2512 }
2519 }
2522 /*
2523 * This is called when the amount of space needed for if_extents
2524 * is increased or decreased. The change in size is indicated by
2525 * the number of extents that need to be added or deleted in the
2526 * ext_diff parameter.
2527 *
2528 * If the amount of space needed has decreased below the size of the
2529 * inline buffer, then switch to using the inline buffer. Otherwise,
2530 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2531 * to what is needed.
2532 *
2533 * ip -- the inode whose if_extents area is changing
2534 * ext_diff -- the change in the number of extents, positive or negative,
2535 * requested for the if_extents array.
2536 */
2537 void
2542 {
2546 uint rnew_size;
2550 }
2561 }
2565 /*
2566 * If the valid extents can fit in if_inline_ext,
2567 * copy them from the malloc'd vector and free it.
2568 */
2570 /*
2571 * For now, empty files are format EXTENTS,
2572 * so the if_extents pointer is null.
2573 */
2579 }
2581 }
2587 /*
2588 * Stuck with malloc/realloc.
2589 */
2598 rnew_size,
2600 KM_NOFS);
2601 }
2602 }
2605 }
2608 /*
2609 * This is called when the amount of space needed for if_data
2610 * is increased or decreased. The change in size is indicated by
2611 * the number of bytes that need to be added or deleted in the
2612 * byte_diff parameter.
2613 *
2614 * If the amount of space needed has decreased below the size of the
2615 * inline buffer, then switch to using the inline buffer. Otherwise,
2616 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2617 * to what is needed.
2618 *
2619 * ip -- the inode whose if_data area is changing
2620 * byte_diff -- the change in the number of bytes, positive or negative,
2621 * requested for the if_data array.
2622 */
2623 void
2628 {
2635 }
2644 }
2648 /*
2649 * If the valid extents/data can fit in if_inline_ext/data,
2650 * copy them from the malloc'd vector and free it.
2651 */
2657 new_size);
2660 }
2663 /*
2664 * Stuck with malloc/realloc.
2665 * For inline data, the underlying buffer must be
2666 * a multiple of 4 bytes in size so that it can be
2667 * logged and stay on word boundaries. We enforce
2668 * that here.
2669 */
2675 /*
2676 * Only do the realloc if the underlying size
2677 * is really changing.
2678 */
2682 real_size,
2684 KM_SLEEP);
2685 }
2691 }
2692 }
2696 }
2701 /*
2702 * Map inode to disk block and offset.
2703 *
2704 * mp -- the mount point structure for the current file system
2705 * tp -- the current transaction
2706 * ino -- the inode number of the inode to be located
2707 * imap -- this structure is filled in with the information necessary
2708 * to retrieve the given inode from disk
2709 * flags -- flags to pass to xfs_dilocate indicating whether or not
2710 * lookups in the inode btree were OK or not
2711 */
2712 int
2716 xfs_ino_t ino,
2718 uint flags)
2719 {
2720 xfs_fsblock_t fsbno;
2730 }
2737 }
2739 void
2743 {
2750 }
2752 /*
2753 * If the format is local, then we can't have an extents
2754 * array so just look for an inline data array. If we're
2755 * not local then we may or may not have an extents list,
2756 * so check and free it up if we do.
2757 */
2765 }
2773 }
2780 }
2781 }
2783 /*
2784 * This is called free all the memory associated with an inode.
2785 * It must free the inode itself and any buffers allocated for
2786 * if_extents/if_data and if_broot. It must also free the lock
2787 * associated with the inode.
2788 */
2789 void
2792 {
2800 }
2806 #ifdef XFS_BMAP_TRACE
2808 #endif
2809 #ifdef XFS_BMBT_TRACE
2811 #endif
2812 #ifdef XFS_RW_TRACE
2814 #endif
2815 #ifdef XFS_ILOCK_TRACE
2817 #endif
2818 #ifdef XFS_DIR2_TRACE
2820 #endif
2822 /* XXXdpd should be able to assert this but shutdown
2823 * is leaving the AIL behind. */
2827 }
2829 }
2832 /*
2833 * Increment the pin count of the given buffer.
2834 * This value is protected by ipinlock spinlock in the mount structure.
2835 */
2836 void
2839 {
2843 }
2845 /*
2846 * Decrement the pin count of the given inode, and wake up
2847 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2848 * inode must have been previoulsy pinned with a call to xfs_ipin().
2849 */
2850 void
2853 {
2859 /* make sync come back and flush this inode */
2865 }
2868 }
2869 }
2871 /*
2872 * This is called to wait for the given inode to be unpinned.
2873 * It will sleep until this happens. The caller must have the
2874 * inode locked in at least shared mode so that the buffer cannot
2875 * be subsequently pinned once someone is waiting for it to be
2876 * unpinned.
2877 */
2878 void
2881 {
2883 xfs_lsn_t lsn;
2889 }
2896 }
2898 /*
2899 * Give the log a push so we don't wait here too long.
2900 */
2904 }
2907 /*
2908 * xfs_iextents_copy()
2909 *
2910 * This is called to copy the REAL extents (as opposed to the delayed
2911 * allocation extents) from the inode into the given buffer. It
2912 * returns the number of bytes copied into the buffer.
2913 *
2914 * If there are no delayed allocation extents, then we can just
2915 * memcpy() the extents into the buffer. Otherwise, we need to
2916 * examine each extent in turn and skip those which are delayed.
2917 */
2918 int
2923 {
2927 #ifdef XFS_BMAP_TRACE
2929 #endif
2933 xfs_fsblock_t start_block;
2943 /*
2944 * There are some delayed allocation extents in the
2945 * inode, so copy the extents one at a time and skip
2946 * the delayed ones. There must be at least one
2947 * non-delayed extent.
2948 */
2955 /*
2956 * It's a delayed allocation extent, so skip it.
2957 */
2958 ep++;
2960 }
2962 /* Translate to on disk format */
2967 dest_ep++;
2968 ep++;
2969 copied++;
2970 }
2975 }
2977 /*
2978 * Each of the following cases stores data into the same region
2979 * of the on-disk inode, so only one of them can be valid at
2980 * any given time. While it is possible to have conflicting formats
2981 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2982 * in EXTENTS format, this can only happen when the fork has
2983 * changed formats after being modified but before being flushed.
2984 * In these cases, the format always takes precedence, because the
2985 * format indicates the current state of the fork.
2986 */
2987 /*ARGSUSED*/
2988 STATIC int
2995 {
2999 #ifdef XFS_TRANS_DEBUG
3001 #endif
3012 /*
3013 * This can happen if we gave up in iformat in an error path,
3014 * for the attribute fork.
3015 */
3019 }
3029 }
3035 }
3036 }
3048 whichfork);
3049 }
3058 XFS_BROOT_SIZE_ADJ));
3062 }
3069 }
3077 }
3083 }
3086 }
3088 /*
3089 * xfs_iflush() will write a modified inode's changes out to the
3090 * inode's on disk home. The caller must have the inode lock held
3091 * in at least shared mode and the inode flush semaphore must be
3092 * held as well. The inode lock will still be held upon return from
3093 * the call and the caller is free to unlock it.
3094 * The inode flush lock will be unlocked when the inode reaches the disk.
3095 * The flags indicate how the inode's buffer should be written out.
3096 */
3097 int
3100 uint flags)
3101 {
3107 /* REFERENCED */
3125 /*
3126 * If the inode isn't dirty, then just release the inode
3127 * flush lock and do nothing.
3128 */
3135 }
3137 /*
3138 * We can't flush the inode until it is unpinned, so
3139 * wait for it. We know noone new can pin it, because
3140 * we are holding the inode lock shared and you need
3141 * to hold it exclusively to pin the inode.
3142 */
3145 /*
3146 * This may have been unpinned because the filesystem is shutting
3147 * down forcibly. If that's the case we must not write this inode
3148 * to disk, because the log record didn't make it to disk!
3149 */
3156 }
3158 /*
3159 * Get the buffer containing the on-disk inode.
3160 */
3165 }
3167 /*
3168 * Decide how buffer will be flushed out. This is done before
3169 * the call to xfs_iflush_int because this field is zeroed by it.
3170 */
3172 /*
3173 * Flush out the inode buffer according to the directions
3174 * of the caller. In the cases where the caller has given
3175 * us a choice choose the non-delwri case. This is because
3176 * the inode is in the AIL and we need to get it out soon.
3177 */
3194 }
3212 }
3213 }
3215 /*
3216 * First flush out the inode that xfs_iflush was called with.
3217 */
3221 }
3223 /*
3224 * inode clustering:
3225 * see if other inodes can be gathered into this write
3226 */
3235 /*
3236 * Do an un-protected check to see if the inode is dirty and
3237 * is a candidate for flushing. These checks will be repeated
3238 * later after the appropriate locks are acquired.
3239 */
3246 }
3248 /*
3249 * Try to get locks. If any are unavailable,
3250 * then this inode cannot be flushed and is skipped.
3251 */
3253 /* get inode locks (just i_lock) */
3255 /* get inode flush lock */
3257 /* check if pinned */
3259 /* arriving here means that
3260 * this inode can be flushed.
3261 * first re-check that it's
3262 * dirty
3263 */
3268 clcount++;
3272 XFS_ILOCK_SHARED);
3274 }
3277 }
3280 }
3281 }
3283 }
3284 }
3290 }
3292 /*
3293 * If the buffer is pinned then push on the log so we won't
3294 * get stuck waiting in the write for too long.
3295 */
3298 }
3306 }
3309 corrupt_out:
3313 /*
3314 * Unlocks the flush lock
3315 */
3318 cluster_corrupt_out:
3319 /* Corruption detected in the clustering loop. Invalidate the
3320 * inode buffer and shut down the filesystem.
3321 */
3324 /*
3325 * Clean up the buffer. If it was B_DELWRI, just release it --
3326 * brelse can handle it with no problems. If not, shut down the
3327 * filesystem before releasing the buffer.
3328 */
3331 }
3336 /*
3337 * Just like incore_relse: if we have b_iodone functions,
3338 * mark the buffer as an error and call them. Otherwise
3339 * mark it as stale and brelse.
3340 */
3351 }
3352 }
3355 /*
3356 * Unlocks the flush lock
3357 */
3359 }
3362 STATIC int
3366 {
3370 #ifdef XFS_TRANS_DEBUG
3372 #endif
3384 /*
3385 * If the inode isn't dirty, then just release the inode
3386 * flush lock and do nothing.
3387 */
3392 }
3394 /* set *dip = inode's place in the buffer */
3397 /*
3398 * Clear i_update_core before copying out the data.
3399 * This is for coordination with our timestamp updates
3400 * that don't hold the inode lock. They will always
3401 * update the timestamps BEFORE setting i_update_core,
3402 * so if we clear i_update_core after they set it we
3403 * are guaranteed to see their updates to the timestamps.
3404 * I believe that this depends on strongly ordered memory
3405 * semantics, but we have that. We use the SYNCHRONIZE
3406 * macro to make sure that the compiler does not reorder
3407 * the i_update_core access below the data copy below.
3408 */
3418 }
3425 }
3435 }
3446 }
3447 }
3450 XFS_RANDOM_IFLUSH_5)) {
3456 ip);
3458 }
3465 }
3466 /*
3467 * bump the flush iteration count, used to detect flushes which
3468 * postdate a log record during recovery.
3469 */
3473 /*
3474 * Copy the dirty parts of the inode into the on-disk
3475 * inode. We always copy out the core of the inode,
3476 * because if the inode is dirty at all the core must
3477 * be.
3478 */
3481 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3485 /*
3486 * If this is really an old format inode and the superblock version
3487 * has not been updated to support only new format inodes, then
3488 * convert back to the old inode format. If the superblock version
3489 * has been updated, then make the conversion permanent.
3490 */
3495 /*
3496 * Convert it back.
3497 */
3501 /*
3502 * The superblock version has already been bumped,
3503 * so just make the conversion to the new inode
3504 * format permanent.
3505 */
3514 }
3515 }
3519 }
3522 /*
3523 * The only error from xfs_iflush_fork is on the data fork.
3524 */
3526 }
3529 /*
3530 * We've recorded everything logged in the inode, so we'd
3531 * like to clear the ilf_fields bits so we don't log and
3532 * flush things unnecessarily. However, we can't stop
3533 * logging all this information until the data we've copied
3534 * into the disk buffer is written to disk. If we did we might
3535 * overwrite the copy of the inode in the log with all the
3536 * data after re-logging only part of it, and in the face of
3537 * a crash we wouldn't have all the data we need to recover.
3538 *
3539 * What we do is move the bits to the ili_last_fields field.
3540 * When logging the inode, these bits are moved back to the
3541 * ilf_fields field. In the xfs_iflush_done() routine we
3542 * clear ili_last_fields, since we know that the information
3543 * those bits represent is permanently on disk. As long as
3544 * the flush completes before the inode is logged again, then
3545 * both ilf_fields and ili_last_fields will be cleared.
3546 *
3547 * We can play with the ilf_fields bits here, because the inode
3548 * lock must be held exclusively in order to set bits there
3549 * and the flush lock protects the ili_last_fields bits.
3550 * Set ili_logged so the flush done
3551 * routine can tell whether or not to look in the AIL.
3552 * Also, store the current LSN of the inode so that we can tell
3553 * whether the item has moved in the AIL from xfs_iflush_done().
3554 * In order to read the lsn we need the AIL lock, because
3555 * it is a 64 bit value that cannot be read atomically.
3556 */
3567 /*
3568 * Attach the function xfs_iflush_done to the inode's
3569 * buffer. This will remove the inode from the AIL
3570 * and unlock the inode's flush lock when the inode is
3571 * completely written to disk.
3572 */
3579 /*
3580 * We're flushing an inode which is not in the AIL and has
3581 * not been logged but has i_update_core set. For this
3582 * case we can use a B_DELWRI flush and immediately drop
3583 * the inode flush lock because we can avoid the whole
3584 * AIL state thing. It's OK to drop the flush lock now,
3585 * because we've already locked the buffer and to do anything
3586 * you really need both.
3587 */
3592 }
3594 }
3598 corrupt_out:
3600 }
3603 /*
3604 * Flush all inactive inodes in mp. Return true if no user references
3605 * were found, false otherwise.
3606 */
3607 int
3611 {
3616 vmap_t vmap;
3626 }
3628 /* Make sure we skip markers inserted by sync */
3632 }
3634 /*
3635 * It's up to our caller to purge the root
3636 * and quota vnodes later.
3637 */
3645 }
3656 }
3661 }
3662 /*
3663 * Ignore busy inodes but continue flushing
3664 * others.
3665 */
3668 }
3669 /*
3670 * Sample vp mapping while holding mp locked on MP
3671 * systems, so we don't purge a reclaimed or
3672 * nonexistent vnode. We break from the loop
3673 * since we know that we modify
3674 * it by pulling ourselves from it in xfs_reclaim()
3675 * called via vn_purge() below. Set ip to the next
3676 * entry in the list anyway so we'll know below
3677 * whether we reached the end or not.
3678 */
3687 /*
3688 * We need to distinguish between when we exit the loop
3689 * after a purge and when we simply hit the end of the
3690 * list. We can't use the (ip == mp->m_inodes) test,
3691 * because when we purge an inode at the start of the list
3692 * the next inode on the list becomes mp->m_inodes. That
3693 * would cause such a test to bail out early. The purged
3694 * variable tells us how we got out of the loop.
3695 */
3698 }
3699 }
3702 }
3705 /*
3706 * xfs_iaccess: check accessibility of inode for mode.
3707 */
3708 int
3711 mode_t mode,
3713 {
3727 }
3729 /*
3730 * If there's an Access Control List it's used instead of
3731 * the mode bits.
3732 */
3740 }
3742 /*
3743 * If the DACs are ok we don't need any capability check.
3744 */
3747 /*
3748 * Read/write DACs are always overridable.
3749 * Executable DACs are overridable if at least one exec bit is set.
3750 */
3760 #ifdef NOISE
3764 }
3766 }
3768 /*
3769 * xfs_iroundup: round up argument to next power of two
3770 */
3771 uint
3773 uint v)
3774 {
3776 uint m;
3789 }
3792 }
3794 /*
3795 * Change the requested timestamp in the given inode.
3796 * We don't lock across timestamp updates, and we don't log them but
3797 * we do record the fact that there is dirty information in core.
3798 *
3799 * NOTE -- callers MUST combine XFS_ICHGTIME_MOD or XFS_ICHGTIME_CHG
3800 * with XFS_ICHGTIME_ACC to be sure that access time
3801 * update will take. Calling first with XFS_ICHGTIME_ACC
3802 * and then XFS_ICHGTIME_MOD may fail to modify the access
3803 * timestamp if the filesystem is mounted noacctm.
3804 */
3805 void
3808 {
3809 timespec_t tv;
3813 /*
3814 * We're not supposed to change timestamps in readonly-mounted
3815 * filesystems. Throw it away if anyone asks us.
3816 */
3820 /*
3821 * Don't update access timestamps on reads if mounted "noatime"
3822 * Throw it away if anyone asks us.
3823 */
3834 }
3839 }
3844 }
3846 /*
3847 * We update the i_update_core field _after_ changing
3848 * the timestamps in order to coordinate properly with
3849 * xfs_iflush() so that we don't lose timestamp updates.
3850 * This keeps us from having to hold the inode lock
3851 * while doing this. We use the SYNCHRONIZE macro to
3852 * ensure that the compiler does not reorder the update
3853 * of i_update_core above the timestamp updates above.
3854 */
3859 }
3861 #ifdef XFS_ILOCK_TRACE
3864 void
3866 {
3875 }
3876 #endif