| blob | 0349e714dc309daf7f1ec43cd0814f2f5eca3750 |
1 /*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
58 /*
59 * Used in xfs_itruncate(). This is the maximum number of extents
60 * freed from a file in a single transaction.
61 */
62 #define XFS_ITRUNC_MAX_EXTENTS 2
69 #ifdef DEBUG
70 /*
71 * Make sure that the extents in the given memory buffer
72 * are valid.
73 */
74 STATIC void
78 xfs_exntfmt_t fmt)
79 {
80 xfs_bmbt_irec_t irec;
81 xfs_bmbt_rec_host_t rec;
91 }
92 }
94 #define xfs_validate_extents(ifp, nrecs, fmt)
97 /*
98 * Check that none of the inode's in the buffer have a next
99 * unlinked field of 0.
100 */
101 #if defined(DEBUG)
102 void
106 {
118 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
119 bp);
121 }
122 }
123 }
124 #endif
126 /*
127 * This routine is called to map an inode number within a file
128 * system to the buffer containing the on-disk version of the
129 * inode. It returns a pointer to the buffer containing the
130 * on-disk inode in the bpp parameter, and in the dip parameter
131 * it returns a pointer to the on-disk inode within that buffer.
132 *
133 * If a non-zero error is returned, then the contents of bpp and
134 * dipp are undefined.
135 *
136 * Use xfs_imap() to determine the size and location of the
137 * buffer to read from disk.
138 */
139 STATIC int
143 xfs_ino_t ino,
147 {
149 xfs_imap_t imap;
154 /*
155 * Call the space management code to find the location of the
156 * inode on disk.
157 */
162 "xfs_inotobp: xfs_imap() returned an "
165 }
167 /*
168 * If the inode number maps to a block outside the bounds of the
169 * file system then return NULL rather than calling read_buf
170 * and panicing when we get an error from the driver.
171 */
175 "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
180 }
182 /*
183 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
184 * default to just a read_buf() call.
185 */
191 "xfs_inotobp: xfs_trans_read_buf() returned an "
194 }
196 di_ok =
200 XFS_RANDOM_ITOBP_INOTOBP))) {
204 "xfs_inotobp: XFS_TEST_ERROR() returned an "
207 }
211 /*
212 * Set *dipp to point to the on-disk inode in the buffer.
213 */
218 }
221 /*
222 * This routine is called to map an inode to the buffer containing
223 * the on-disk version of the inode. It returns a pointer to the
224 * buffer containing the on-disk inode in the bpp parameter, and in
225 * the dip parameter it returns a pointer to the on-disk inode within
226 * that buffer.
227 *
228 * If a non-zero error is returned, then the contents of bpp and
229 * dipp are undefined.
230 *
231 * If the inode is new and has not yet been initialized, use xfs_imap()
232 * to determine the size and location of the buffer to read from disk.
233 * If the inode has already been mapped to its buffer and read in once,
234 * then use the mapping information stored in the inode rather than
235 * calling xfs_imap(). This allows us to avoid the overhead of looking
236 * at the inode btree for small block file systems (see xfs_dilocate()).
237 * We can tell whether the inode has been mapped in before by comparing
238 * its disk block address to 0. Only uninitialized inodes will have
239 * 0 for the disk block address.
240 */
241 int
248 xfs_daddr_t bno,
249 uint imap_flags)
250 {
251 xfs_imap_t imap;
258 /*
259 * Call the space management code to find the location of the
260 * inode on disk.
261 */
267 /*
268 * If the inode number maps to a block outside the bounds
269 * of the file system then return NULL rather than calling
270 * read_buf and panicing when we get an error from the
271 * driver.
272 */
275 #ifdef DEBUG
277 "(imap.im_blkno (0x%llx) "
278 "+ imap.im_len (0x%llx)) > "
279 " XFS_FSB_TO_BB(mp, "
286 }
288 /*
289 * Fill in the fields in the inode that will be used to
290 * map the inode to its buffer from now on.
291 */
296 /*
297 * We've already mapped the inode once, so just use the
298 * mapping that we saved the first time.
299 */
303 }
306 /*
307 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
308 * default to just a read_buf() call.
309 */
313 #ifdef DEBUG
315 "xfs_trans_read_buf() returned error %d, "
321 }
323 /*
324 * Validate the magic number and version of every inode in the buffer
325 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
326 * No validation is done here in userspace (xfs_repair).
327 */
328 #if !defined(__KERNEL__)
330 #elif defined(DEBUG)
334 #endif
345 XFS_ERRTAG_ITOBP_INOTOBP,
346 XFS_RANDOM_ITOBP_INOTOBP))) {
350 }
351 #ifdef DEBUG
353 "Device %s - bad inode magic/vsn "
358 #endif
363 }
364 }
368 /*
369 * Mark the buffer as an inode buffer now that it looks good
370 */
373 /*
374 * Set *dipp to point to the on-disk inode in the buffer.
375 */
379 }
381 /*
382 * Move inode type and inode format specific information from the
383 * on-disk inode to the in-core inode. For fifos, devs, and sockets
384 * this means set if_rdev to the proper value. For files, directories,
385 * and symlinks this means to bring in the in-line data or extent
386 * pointers. For a file in B-tree format, only the root is immediately
387 * brought in-core. The rest will be in-lined in if_extents when it
388 * is first referenced (see xfs_iread_extents()).
389 */
390 STATIC int
394 {
398 xfs_fsize_t di_size;
416 }
426 }
437 }
448 /*
449 * no local regular files yet
450 */
453 "corrupt inode %Lu "
457 XFS_ERRLEVEL_LOW,
460 }
465 "corrupt inode %Lu "
470 XFS_ERRLEVEL_LOW,
473 }
488 }
494 }
497 }
519 }
524 }
526 }
528 /*
529 * The file is in-lined in the on-disk inode.
530 * If it fits into if_inline_data, then copy
531 * it there, otherwise allocate a buffer for it
532 * and copy the data there. Either way, set
533 * if_data to point at the data.
534 * If we allocate a buffer for the data, make
535 * sure that its size is a multiple of 4 and
536 * record the real size in i_real_bytes.
537 */
538 STATIC int
544 {
548 /*
549 * If the size is unreasonable, then something
550 * is wrong and we just bail out rather than crash in
551 * kmem_alloc() or memcpy() below.
552 */
555 "corrupt inode %Lu "
562 }
572 }
580 }
582 /*
583 * The file consists of a set of extents all
584 * of which fit into the on-disk inode.
585 * If there are few enough extents to fit into
586 * the if_inline_ext, then copy them there.
587 * Otherwise allocate a buffer for them and copy
588 * them into it. Either way, set if_extents
589 * to point at the extents.
590 */
591 STATIC int
596 {
607 /*
608 * If the number of extents is unreasonable, then something
609 * is wrong and we just bail out rather than crash in
610 * kmem_alloc() or memcpy() below.
611 */
619 }
626 else
637 }
644 XFS_ERRLEVEL_LOW,
647 }
648 }
651 }
653 /*
654 * The file has too many extents to fit into
655 * the inode, so they are in B-tree format.
656 * Allocate a buffer for the root of the B-tree
657 * and copy the root into it. The i_extents
658 * field will remain NULL until all of the
659 * extents are read in (when they are needed).
660 */
661 STATIC int
666 {
669 /* REFERENCED */
678 /*
679 * blow out if -- fork has less extents than can fit in
680 * fork (fork shouldn't be a btree format), root btree
681 * block has more records than can fit into the fork,
682 * or the number of extents is greater than the number of
683 * blocks.
684 */
695 }
700 /*
701 * Copy and convert from the on-disk structure
702 * to the in-memory structure.
703 */
710 }
712 void
716 {
745 }
747 void
751 {
780 }
782 STATIC uint
784 __uint16_t di_flags)
785 {
817 }
820 }
822 uint
825 {
830 }
832 uint
835 {
838 }
840 /*
841 * Given a mount structure and an inode number, return a pointer
842 * to a newly allocated in-core inode corresponding to the given
843 * inode number.
844 *
845 * Initialize the inode's attributes and extent pointers if it
846 * already has them (it will not if the inode has no links).
847 */
848 int
852 xfs_ino_t ino,
854 xfs_daddr_t bno,
855 uint imap_flags)
856 {
870 /*
871 * Get pointer's to the on-disk inode and the buffer containing it.
872 * If the inode number refers to a block outside the file system
873 * then xfs_itobp() will return NULL. In this case we should
874 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
875 * know that this is a new incore inode.
876 */
881 }
883 /*
884 * Initialize inode's trace buffers.
885 * Do this before xfs_iformat in case it adds entries.
886 */
887 #ifdef XFS_BMAP_TRACE
889 #endif
890 #ifdef XFS_BMBT_TRACE
892 #endif
893 #ifdef XFS_RW_TRACE
895 #endif
896 #ifdef XFS_ILOCK_TRACE
898 #endif
899 #ifdef XFS_DIR2_TRACE
901 #endif
903 /*
904 * If we got something that isn't an inode it means someone
905 * (nfs or dmi) has a stale handle.
906 */
910 #ifdef DEBUG
912 "dip->di_core.di_magic (0x%x) != "
915 XFS_DINODE_MAGIC);
918 }
920 /*
921 * If the on-disk inode is already linked to a directory
922 * entry, copy all of the inode into the in-core inode.
923 * xfs_iformat() handles copying in the inode format
924 * specific information.
925 * Otherwise, just get the truly permanent information.
926 */
933 #ifdef DEBUG
936 error);
939 }
945 /*
946 * Make sure to pull in the mode here as well in
947 * case the inode is released without being used.
948 * This ensures that xfs_inactive() will see that
949 * the inode is already free and not try to mess
950 * with the uninitialized part of it.
951 */
953 /*
954 * Initialize the per-fork minima and maxima for a new
955 * inode here. xfs_iformat will do it for old inodes.
956 */
959 }
963 /*
964 * The inode format changed when we moved the link count and
965 * made it 32 bits long. If this is an old format inode,
966 * convert it in memory to look like a new one. If it gets
967 * flushed to disk we will convert back before flushing or
968 * logging it. We zero out the new projid field and the old link
969 * count field. We'll handle clearing the pad field (the remains
970 * of the old uuid field) when we actually convert the inode to
971 * the new format. We don't change the version number so that we
972 * can distinguish this from a real new format inode.
973 */
978 }
983 /*
984 * Mark the buffer containing the inode as something to keep
985 * around for a while. This helps to keep recently accessed
986 * meta-data in-core longer.
987 */
990 /*
991 * Use xfs_trans_brelse() to release the buffer containing the
992 * on-disk inode, because it was acquired with xfs_trans_read_buf()
993 * in xfs_itobp() above. If tp is NULL, this is just a normal
994 * brelse(). If we're within a transaction, then xfs_trans_brelse()
995 * will only release the buffer if it is not dirty within the
996 * transaction. It will be OK to release the buffer in this case,
997 * because inodes on disk are never destroyed and we will be
998 * locking the new in-core inode before putting it in the hash
999 * table where other processes can find it. Thus we don't have
1000 * to worry about the inode being changed just because we released
1001 * the buffer.
1002 */
1006 }
1008 /*
1009 * Read in extents from a btree-format inode.
1010 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1011 */
1012 int
1017 {
1020 xfs_extnum_t nextents;
1027 }
1032 /*
1033 * We know that the size is valid (it's checked in iformat_btree)
1034 */
1044 }
1047 }
1049 /*
1050 * Allocate an inode on disk and return a copy of its in-core version.
1051 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1052 * appropriately within the inode. The uid and gid for the inode are
1053 * set according to the contents of the given cred structure.
1054 *
1055 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1056 * has a free inode available, call xfs_iget()
1057 * to obtain the in-core version of the allocated inode. Finally,
1058 * fill in the inode and log its initial contents. In this case,
1059 * ialloc_context would be set to NULL and call_again set to false.
1060 *
1061 * If xfs_dialloc() does not have an available inode,
1062 * it will replenish its supply by doing an allocation. Since we can
1063 * only do one allocation within a transaction without deadlocks, we
1064 * must commit the current transaction before returning the inode itself.
1065 * In this case, therefore, we will set call_again to true and return.
1066 * The caller should then commit the current transaction, start a new
1067 * transaction, and call xfs_ialloc() again to actually get the inode.
1068 *
1069 * To ensure that some other process does not grab the inode that
1070 * was allocated during the first call to xfs_ialloc(), this routine
1071 * also returns the [locked] bp pointing to the head of the freelist
1072 * as ialloc_context. The caller should hold this buffer across
1073 * the commit and pass it back into this routine on the second call.
1074 *
1075 * If we are allocating quota inodes, we do not have a parent inode
1076 * to attach to or associate with (i.e. pip == NULL) because they
1077 * are not linked into the directory structure - they are attached
1078 * directly to the superblock - and so have no parent.
1079 */
1080 int
1084 mode_t mode,
1085 xfs_nlink_t nlink,
1086 xfs_dev_t rdev,
1088 xfs_prid_t prid,
1093 {
1094 xfs_ino_t ino;
1097 uint flags;
1100 /*
1101 * Call the space management code to pick
1102 * the on-disk inode to be allocated.
1103 */
1108 }
1112 }
1115 /*
1116 * Get the in-core inode with the lock held exclusively.
1117 * This is because we're setting fields here we need
1118 * to prevent others from looking at until we're done.
1119 */
1124 }
1137 /*
1138 * If the superblock version is up to where we support new format
1139 * inodes and this is currently an old format inode, then change
1140 * the inode version number now. This way we only do the conversion
1141 * here rather than here and in the flush/logging code.
1142 */
1146 /*
1147 * We've already zeroed the old link count, the projid field,
1148 * and the pad field.
1149 */
1150 }
1152 /*
1153 * Project ids won't be stored on disk if we are using a version 1 inode.
1154 */
1162 }
1163 }
1165 /*
1166 * If the group ID of the new file does not match the effective group
1167 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1168 * (and only if the irix_sgid_inherit compatibility variable is set).
1169 */
1174 }
1181 /*
1182 * di_gen will have been taken care of in xfs_iread.
1183 */
1206 }
1207 /* fall through */
1218 }
1223 }
1227 }
1228 }
1230 xfs_inherit_noatime)
1233 xfs_inherit_nodump)
1236 xfs_inherit_sync)
1239 xfs_inherit_nosymlinks)
1244 xfs_inherit_nodefrag)
1249 }
1250 /* FALLTHROUGH */
1259 }
1260 /*
1261 * Attribute fork settings for new inode.
1262 */
1266 /*
1267 * Log the new values stuffed into the inode.
1268 */
1271 /* now that we have an i_mode we can setup inode ops and unlock */
1276 }
1278 /*
1279 * Check to make sure that there are no blocks allocated to the
1280 * file beyond the size of the file. We don't check this for
1281 * files with fixed size extents or real time extents, but we
1282 * at least do it for regular files.
1283 */
1284 #ifdef DEBUG
1285 void
1289 xfs_fsize_t isize)
1290 {
1291 xfs_fileoff_t map_first;
1303 /*
1304 * The filesystem could be shutting down, so bmapi may return
1305 * an error.
1306 */
1310 map_first),
1316 }
1319 /*
1320 * Calculate the last possible buffered byte in a file. This must
1321 * include data that was buffered beyond the EOF by the write code.
1322 * This also needs to deal with overflowing the xfs_fsize_t type
1323 * which can happen for sizes near the limit.
1324 *
1325 * We also need to take into account any blocks beyond the EOF. It
1326 * may be the case that they were buffered by a write which failed.
1327 * In that case the pages will still be in memory, but the inode size
1328 * will never have been updated.
1329 */
1330 xfs_fsize_t
1333 {
1335 xfs_fsize_t last_byte;
1336 xfs_fileoff_t last_block;
1337 xfs_fileoff_t size_last_block;
1343 /*
1344 * Only check for blocks beyond the EOF if the extents have
1345 * been read in. This eliminates the need for the inode lock,
1346 * and it also saves us from looking when it really isn't
1347 * necessary.
1348 */
1351 XFS_DATA_FORK);
1354 }
1357 }
1364 }
1368 }
1370 }
1372 #if defined(XFS_RW_TRACE)
1373 STATIC void
1378 xfs_fsize_t new_size,
1379 xfs_off_t toss_start,
1380 xfs_off_t toss_finish)
1381 {
1384 }
1403 }
1404 #else
1405 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1406 #endif
1408 /*
1409 * Start the truncation of the file to new_size. The new size
1410 * must be smaller than the current size. This routine will
1411 * clear the buffer and page caches of file data in the removed
1412 * range, and xfs_itruncate_finish() will remove the underlying
1413 * disk blocks.
1414 *
1415 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1416 * must NOT have the inode lock held at all. This is because we're
1417 * calling into the buffer/page cache code and we can't hold the
1418 * inode lock when we do so.
1419 *
1420 * We need to wait for any direct I/Os in flight to complete before we
1421 * proceed with the truncate. This is needed to prevent the extents
1422 * being read or written by the direct I/Os from being removed while the
1423 * I/O is in flight as there is no other method of synchronising
1424 * direct I/O with the truncate operation. Also, because we hold
1425 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1426 * started until the truncate completes and drops the lock. Essentially,
1427 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1428 * between direct I/Os and the truncate operation.
1429 *
1430 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1431 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1432 * in the case that the caller is locking things out of order and
1433 * may not be able to call xfs_itruncate_finish() with the inode lock
1434 * held without dropping the I/O lock. If the caller must drop the
1435 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1436 * must be called again with all the same restrictions as the initial
1437 * call.
1438 */
1439 int
1442 uint flags,
1443 xfs_fsize_t new_size)
1444 {
1445 xfs_fsize_t last_byte;
1446 xfs_off_t toss_start;
1461 /*
1462 * Call toss_pages or flushinval_pages to get rid of pages
1463 * overlapping the region being removed. We have to use
1464 * the less efficient flushinval_pages in the case that the
1465 * caller may not be able to finish the truncate without
1466 * dropping the inode's I/O lock. Make sure
1467 * to catch any pages brought in by buffers overlapping
1468 * the EOF by searching out beyond the isize by our
1469 * block size. We round new_size up to a block boundary
1470 * so that we don't toss things on the same block as
1471 * new_size but before it.
1472 *
1473 * Before calling toss_page or flushinval_pages, make sure to
1474 * call remapf() over the same region if the file is mapped.
1475 * This frees up mapped file references to the pages in the
1476 * given range and for the flushinval_pages case it ensures
1477 * that we get the latest mapped changes flushed out.
1478 */
1482 /*
1483 * The place to start tossing is beyond our maximum
1484 * file size, so there is no way that the data extended
1485 * out there.
1486 */
1488 }
1491 last_byte);
1499 }
1500 }
1502 #ifdef DEBUG
1505 }
1506 #endif
1508 }
1510 /*
1511 * Shrink the file to the given new_size. The new
1512 * size must be smaller than the current size.
1513 * This will free up the underlying blocks
1514 * in the removed range after a call to xfs_itruncate_start()
1515 * or xfs_atruncate_start().
1516 *
1517 * The transaction passed to this routine must have made
1518 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1519 * This routine may commit the given transaction and
1520 * start new ones, so make sure everything involved in
1521 * the transaction is tidy before calling here.
1522 * Some transaction will be returned to the caller to be
1523 * committed. The incoming transaction must already include
1524 * the inode, and both inode locks must be held exclusively.
1525 * The inode must also be "held" within the transaction. On
1526 * return the inode will be "held" within the returned transaction.
1527 * This routine does NOT require any disk space to be reserved
1528 * for it within the transaction.
1529 *
1530 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1531 * and it indicates the fork which is to be truncated. For the
1532 * attribute fork we only support truncation to size 0.
1533 *
1534 * We use the sync parameter to indicate whether or not the first
1535 * transaction we perform might have to be synchronous. For the attr fork,
1536 * it needs to be so if the unlink of the inode is not yet known to be
1537 * permanent in the log. This keeps us from freeing and reusing the
1538 * blocks of the attribute fork before the unlink of the inode becomes
1539 * permanent.
1540 *
1541 * For the data fork, we normally have to run synchronously if we're
1542 * being called out of the inactive path or we're being called
1543 * out of the create path where we're truncating an existing file.
1544 * Either way, the truncate needs to be sync so blocks don't reappear
1545 * in the file with altered data in case of a crash. wsync filesystems
1546 * can run the first case async because anything that shrinks the inode
1547 * has to run sync so by the time we're called here from inactive, the
1548 * inode size is permanently set to 0.
1549 *
1550 * Calls from the truncate path always need to be sync unless we're
1551 * in a wsync filesystem and the file has already been unlinked.
1552 *
1553 * The caller is responsible for correctly setting the sync parameter.
1554 * It gets too hard for us to guess here which path we're being called
1555 * out of just based on inode state.
1556 */
1557 int
1561 xfs_fsize_t new_size,
1564 {
1565 xfs_fsblock_t first_block;
1566 xfs_fileoff_t first_unmap_block;
1567 xfs_fileoff_t last_block;
1573 xfs_bmap_free_t free_list;
1590 /*
1591 * We only support truncating the entire attribute fork.
1592 */
1595 }
1598 /*
1599 * The first thing we do is set the size to new_size permanently
1600 * on disk. This way we don't have to worry about anyone ever
1601 * being able to look at the data being freed even in the face
1602 * of a crash. What we're getting around here is the case where
1603 * we free a block, it is allocated to another file, it is written
1604 * to, and then we crash. If the new data gets written to the
1605 * file but the log buffers containing the free and reallocation
1606 * don't, then we'd end up with garbage in the blocks being freed.
1607 * As long as we make the new_size permanent before actually
1608 * freeing any blocks it doesn't matter if they get writtten to.
1609 *
1610 * The callers must signal into us whether or not the size
1611 * setting here must be synchronous. There are a few cases
1612 * where it doesn't have to be synchronous. Those cases
1613 * occur if the file is unlinked and we know the unlink is
1614 * permanent or if the blocks being truncated are guaranteed
1615 * to be beyond the inode eof (regardless of the link count)
1616 * and the eof value is permanent. Both of these cases occur
1617 * only on wsync-mounted filesystems. In those cases, we're
1618 * guaranteed that no user will ever see the data in the blocks
1619 * that are being truncated so the truncate can run async.
1620 * In the free beyond eof case, the file may wind up with
1621 * more blocks allocated to it than it needs if we crash
1622 * and that won't get fixed until the next time the file
1623 * is re-opened and closed but that's ok as that shouldn't
1624 * be too many blocks.
1625 *
1626 * However, we can't just make all wsync xactions run async
1627 * because there's one call out of the create path that needs
1628 * to run sync where it's truncating an existing file to size
1629 * 0 whose size is > 0.
1630 *
1631 * It's probably possible to come up with a test in this
1632 * routine that would correctly distinguish all the above
1633 * cases from the values of the function parameters and the
1634 * inode state but for sanity's sake, I've decided to let the
1635 * layers above just tell us. It's simpler to correctly figure
1636 * out in the layer above exactly under what conditions we
1637 * can run async and I think it's easier for others read and
1638 * follow the logic in case something has to be changed.
1639 * cscope is your friend -- rcc.
1640 *
1641 * The attribute fork is much simpler.
1642 *
1643 * For the attribute fork we allow the caller to tell us whether
1644 * the unlink of the inode that led to this call is yet permanent
1645 * in the on disk log. If it is not and we will be freeing extents
1646 * in this inode then we make the first transaction synchronous
1647 * to make sure that the unlink is permanent by the time we free
1648 * the blocks.
1649 */
1652 /*
1653 * If we are not changing the file size then do
1654 * not update the on-disk file size - we may be
1655 * called from xfs_inactive_free_eofblocks(). If we
1656 * update the on-disk file size and then the system
1657 * crashes before the contents of the file are
1658 * flushed to disk then the files may be full of
1659 * holes (ie NULL files bug).
1660 */
1665 }
1666 }
1671 }
1677 /*
1678 * Since it is possible for space to become allocated beyond
1679 * the end of the file (in a crash where the space is allocated
1680 * but the inode size is not yet updated), simply remove any
1681 * blocks which show up between the new EOF and the maximum
1682 * possible file size. If the first block to be removed is
1683 * beyond the maximum file size (ie it is the same as last_block),
1684 * then there is nothing to do.
1685 */
1693 }
1695 /*
1696 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1697 * will tell us whether it freed the entire range or
1698 * not. If this is a synchronous mount (wsync),
1699 * then we can tell bunmapi to keep all the
1700 * transactions asynchronous since the unlink
1701 * transaction that made this inode inactive has
1702 * already hit the disk. There's no danger of
1703 * the freed blocks being reused, there being a
1704 * crash, and the reused blocks suddenly reappearing
1705 * in this file with garbage in them once recovery
1706 * runs.
1707 */
1713 XFS_ITRUNC_MAX_EXTENTS,
1717 /*
1718 * If the bunmapi call encounters an error,
1719 * return to the caller where the transaction
1720 * can be properly aborted. We just need to
1721 * make sure we're not holding any resources
1722 * that we were not when we came in.
1723 */
1726 }
1728 /*
1729 * Duplicate the transaction that has the permanent
1730 * reservation and commit the old transaction.
1731 */
1735 /*
1736 * If the bmap finish call encounters an error,
1737 * return to the caller where the transaction
1738 * can be properly aborted. We just need to
1739 * make sure we're not holding any resources
1740 * that we were not when we came in.
1741 *
1742 * Aborting from this point might lose some
1743 * blocks in the file system, but oh well.
1744 */
1747 /*
1748 * If the passed in transaction committed
1749 * in xfs_bmap_finish(), then we want to
1750 * add the inode to this one before returning.
1751 * This keeps things simple for the higher
1752 * level code, because it always knows that
1753 * the inode is locked and held in the
1754 * transaction that returns to it whether
1755 * errors occur or not. We don't mark the
1756 * inode dirty so that this transaction can
1757 * be easily aborted if possible.
1758 */
1762 }
1764 }
1767 /*
1768 * The first xact was committed,
1769 * so add the inode to the new one.
1770 * Mark it dirty so it will be logged
1771 * and moved forward in the log as
1772 * part of every commit.
1773 */
1778 }
1783 XFS_TRANS_PERM_LOG_RES,
1784 XFS_ITRUNCATE_LOG_COUNT);
1785 /*
1786 * Add the inode being truncated to the next chained
1787 * transaction.
1788 */
1793 }
1794 /*
1795 * Only update the size in the case of the data fork, but
1796 * always re-log the inode so that our permanent transaction
1797 * can keep on rolling it forward in the log.
1798 */
1801 /*
1802 * If we are not changing the file size then do
1803 * not update the on-disk file size - we may be
1804 * called from xfs_inactive_free_eofblocks(). If we
1805 * update the on-disk file size and then the system
1806 * crashes before the contents of the file are
1807 * flushed to disk then the files may be full of
1808 * holes (ie NULL files bug).
1809 */
1813 }
1814 }
1824 }
1827 /*
1828 * xfs_igrow_start
1829 *
1830 * Do the first part of growing a file: zero any data in the last
1831 * block that is beyond the old EOF. We need to do this before
1832 * the inode is joined to the transaction to modify the i_size.
1833 * That way we can drop the inode lock and call into the buffer
1834 * cache to get the buffer mapping the EOF.
1835 */
1836 int
1839 xfs_fsize_t new_size,
1841 {
1848 /*
1849 * Zero any pages that may have been created by
1850 * xfs_write_file() beyond the end of the file
1851 * and any blocks between the old and new file sizes.
1852 */
1856 }
1858 /*
1859 * xfs_igrow_finish
1860 *
1861 * This routine is called to extend the size of a file.
1862 * The inode must have both the iolock and the ilock locked
1863 * for update and it must be a part of the current transaction.
1864 * The xfs_igrow_start() function must have been called previously.
1865 * If the change_flag is not zero, the inode change timestamp will
1866 * be updated.
1867 */
1868 void
1872 xfs_fsize_t new_size,
1874 {
1880 /*
1881 * Update the file size. Update the inode change timestamp
1882 * if change_flag set.
1883 */
1890 }
1893 /*
1894 * This is called when the inode's link count goes to 0.
1895 * We place the on-disk inode on a list in the AGI. It
1896 * will be pulled from this list when the inode is freed.
1897 */
1898 int
1902 {
1908 xfs_agnumber_t agno;
1909 xfs_daddr_t agdaddr;
1910 xfs_agino_t agino;
1925 /*
1926 * Get the agi buffer first. It ensures lock ordering
1927 * on the list.
1928 */
1933 }
1934 /*
1935 * Validate the magic number of the agi block.
1936 */
1938 agi_ok =
1942 XFS_RANDOM_IUNLINK))) {
1946 }
1947 /*
1948 * Get the index into the agi hash table for the
1949 * list this inode will go on.
1950 */
1958 /*
1959 * There is already another inode in the bucket we need
1960 * to add ourselves to. Add us at the front of the list.
1961 * Here we put the head pointer into our next pointer,
1962 * and then we fall through to point the head at us.
1963 */
1967 }
1969 /* both on-disk, don't endian flip twice */
1977 }
1979 /*
1980 * Point the bucket head pointer at the inode being inserted.
1981 */
1989 }
1991 /*
1992 * Pull the on-disk inode from the AGI unlinked list.
1993 */
1994 STATIC int
1998 {
1999 xfs_ino_t next_ino;
2005 xfs_agnumber_t agno;
2006 xfs_daddr_t agdaddr;
2007 xfs_agino_t agino;
2008 xfs_agino_t next_agino;
2016 /*
2017 * First pull the on-disk inode from the AGI unlinked list.
2018 */
2024 /*
2025 * Get the agi buffer first. It ensures lock ordering
2026 * on the list.
2027 */
2035 }
2036 /*
2037 * Validate the magic number of the agi block.
2038 */
2040 agi_ok =
2044 XFS_RANDOM_IUNLINK_REMOVE))) {
2052 }
2053 /*
2054 * Get the index into the agi hash table for the
2055 * list this inode will go on.
2056 */
2064 /*
2065 * We're at the head of the list. Get the inode's
2066 * on-disk buffer to see if there is anyone after us
2067 * on the list. Only modify our next pointer if it
2068 * is not already NULLAGINO. This saves us the overhead
2069 * of dealing with the buffer when there is no need to
2070 * change it.
2071 */
2078 }
2091 }
2092 /*
2093 * Point the bucket head pointer at the next inode.
2094 */
2103 /*
2104 * We need to search the list for the inode being freed.
2105 */
2109 /*
2110 * If the last inode wasn't the one pointing to
2111 * us, then release its buffer since we're not
2112 * going to do anything with it.
2113 */
2116 }
2125 }
2129 }
2130 /*
2131 * Now last_ibp points to the buffer previous to us on
2132 * the unlinked list. Pull us from the list.
2133 */
2140 }
2154 }
2155 /*
2156 * Point the previous inode on the list to the next inode.
2157 */
2165 }
2167 }
2170 {
2174 }
2176 STATIC void
2180 xfs_ino_t inum)
2181 {
2187 xfs_daddr_t blkno;
2204 }
2213 /*
2214 * Look for each inode in memory and attempt to lock it,
2215 * we can be racing with flush and tail pushing here.
2216 * any inode we get the locks on, add to an array of
2217 * inode items to process later.
2218 *
2219 * The get the buffer lock, we could beat a flush
2220 * or tail pushing thread to the lock here, in which
2221 * case they will go looking for the inode buffer
2222 * and fail, we need some other form of interlock
2223 * here.
2224 */
2231 /* Inode not in memory or we found it already,
2232 * nothing to do
2233 */
2237 }
2242 }
2244 /* If we can get the locks then add it to the
2245 * list, otherwise by the time we get the bp lock
2246 * below it will already be attached to the
2247 * inode buffer.
2248 */
2250 /* This inode will already be locked - by us, lets
2251 * keep it that way.
2252 */
2261 }
2262 }
2265 }
2276 }
2279 }
2280 }
2282 }
2286 XFS_BUF_LOCK);
2299 pre_flushed++;
2300 }
2302 }
2313 }
2327 }
2328 }
2333 }
2337 }
2339 /*
2340 * This is called to return an inode to the inode free list.
2341 * The inode should already be truncated to 0 length and have
2342 * no pages associated with it. This routine also assumes that
2343 * the inode is already a part of the transaction.
2344 *
2345 * The on-disk copy of the inode will have been added to the list
2346 * of unlinked inodes in the AGI. We need to remove the inode from
2347 * that list atomically with respect to freeing it here.
2348 */
2349 int
2354 {
2357 xfs_ino_t first_ino;
2368 /*
2369 * Pull the on-disk inode from the AGI unlinked list.
2370 */
2374 }
2379 }
2388 /*
2389 * Bump the generation count so no one will be confused
2390 * by reincarnations of this inode.
2391 */
2397 }
2400 }
2402 /*
2403 * Reallocate the space for if_broot based on the number of records
2404 * being added or deleted as indicated in rec_diff. Move the records
2405 * and pointers in if_broot to fit the new size. When shrinking this
2406 * will eliminate holes between the records and pointers created by
2407 * the caller. When growing this will create holes to be filled in
2408 * by the caller.
2409 *
2410 * The caller must not request to add more records than would fit in
2411 * the on-disk inode root. If the if_broot is currently NULL, then
2412 * if we adding records one will be allocated. The caller must also
2413 * not request that the number of records go below zero, although
2414 * it can go to zero.
2415 *
2416 * ip -- the inode whose if_broot area is changing
2417 * ext_diff -- the change in the number of records, positive or negative,
2418 * requested for the if_broot array.
2419 */
2420 void
2425 {
2434 /*
2435 * Handle the degenerate case quietly.
2436 */
2439 }
2443 /*
2444 * If there wasn't any memory allocated before, just
2445 * allocate it now and get out.
2446 */
2450 KM_SLEEP);
2453 }
2455 /*
2456 * If there is already an existing if_broot, then we need
2457 * to realloc() it and shift the pointers to their new
2458 * location. The records don't change location because
2459 * they are kept butted up against the btree block header.
2460 */
2466 new_size,
2468 KM_SLEEP);
2478 }
2480 /*
2481 * rec_diff is less than 0. In this case, we are shrinking the
2482 * if_broot buffer. It must already exist. If we go to zero
2483 * records, just get rid of the root and clear the status bit.
2484 */
2491 else
2495 /*
2496 * First copy over the btree block header.
2497 */
2502 }
2504 /*
2505 * Only copy the records and pointers if there are any.
2506 */
2508 /*
2509 * First copy the records.
2510 */
2517 /*
2518 * Then copy the pointers.
2519 */
2525 }
2532 }
2535 /*
2536 * This is called when the amount of space needed for if_data
2537 * is increased or decreased. The change in size is indicated by
2538 * the number of bytes that need to be added or deleted in the
2539 * byte_diff parameter.
2540 *
2541 * If the amount of space needed has decreased below the size of the
2542 * inline buffer, then switch to using the inline buffer. Otherwise,
2543 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2544 * to what is needed.
2545 *
2546 * ip -- the inode whose if_data area is changing
2547 * byte_diff -- the change in the number of bytes, positive or negative,
2548 * requested for the if_data array.
2549 */
2550 void
2555 {
2562 }
2571 }
2575 /*
2576 * If the valid extents/data can fit in if_inline_ext/data,
2577 * copy them from the malloc'd vector and free it.
2578 */
2584 new_size);
2587 }
2590 /*
2591 * Stuck with malloc/realloc.
2592 * For inline data, the underlying buffer must be
2593 * a multiple of 4 bytes in size so that it can be
2594 * logged and stay on word boundaries. We enforce
2595 * that here.
2596 */
2602 /*
2603 * Only do the realloc if the underlying size
2604 * is really changing.
2605 */
2609 real_size,
2611 KM_SLEEP);
2612 }
2618 }
2619 }
2623 }
2628 /*
2629 * Map inode to disk block and offset.
2630 *
2631 * mp -- the mount point structure for the current file system
2632 * tp -- the current transaction
2633 * ino -- the inode number of the inode to be located
2634 * imap -- this structure is filled in with the information necessary
2635 * to retrieve the given inode from disk
2636 * flags -- flags to pass to xfs_dilocate indicating whether or not
2637 * lookups in the inode btree were OK or not
2638 */
2639 int
2643 xfs_ino_t ino,
2645 uint flags)
2646 {
2647 xfs_fsblock_t fsbno;
2657 }
2664 }
2666 void
2670 {
2677 }
2679 /*
2680 * If the format is local, then we can't have an extents
2681 * array so just look for an inline data array. If we're
2682 * not local then we may or may not have an extents list,
2683 * so check and free it up if we do.
2684 */
2692 }
2699 }
2706 }
2707 }
2709 /*
2710 * This is called free all the memory associated with an inode.
2711 * It must free the inode itself and any buffers allocated for
2712 * if_extents/if_data and if_broot. It must also free the lock
2713 * associated with the inode.
2714 */
2715 void
2718 {
2726 }
2732 #ifdef XFS_BMAP_TRACE
2734 #endif
2735 #ifdef XFS_BMBT_TRACE
2737 #endif
2738 #ifdef XFS_RW_TRACE
2740 #endif
2741 #ifdef XFS_ILOCK_TRACE
2743 #endif
2744 #ifdef XFS_DIR2_TRACE
2746 #endif
2748 /*
2749 * Only if we are shutting down the fs will we see an
2750 * inode still in the AIL. If it is there, we should remove
2751 * it to prevent a use-after-free from occurring.
2752 */
2763 else
2765 }
2767 }
2769 }
2772 /*
2773 * Increment the pin count of the given buffer.
2774 * This value is protected by ipinlock spinlock in the mount structure.
2775 */
2776 void
2779 {
2783 }
2785 /*
2786 * Decrement the pin count of the given inode, and wake up
2787 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2788 * inode must have been previously pinned with a call to xfs_ipin().
2789 */
2790 void
2793 {
2798 /*
2799 * If the inode is currently being reclaimed, the link between
2800 * the bhv_vnode and the xfs_inode will be broken after the
2801 * XFS_IRECLAIM* flag is set. Hence, if these flags are not
2802 * set, then we can move forward and mark the linux inode dirty
2803 * knowing that it is still valid as it won't freed until after
2804 * the bhv_vnode<->xfs_inode link is broken in xfs_reclaim. The
2805 * i_flags_lock is used to synchronise the setting of the
2806 * XFS_IRECLAIM* flags and the breaking of the link, and so we
2807 * can execute atomically w.r.t to reclaim by holding this lock
2808 * here.
2809 *
2810 * However, we still need to issue the unpin wakeup call as the
2811 * inode reclaim may be blocked waiting for the inode to become
2812 * unpinned.
2813 */
2823 /* 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 {
2889 xfs_fsblock_t start_block;
2899 /*
2900 * There are some delayed allocation extents in the
2901 * inode, so copy the extents one at a time and skip
2902 * the delayed ones. There must be at least one
2903 * non-delayed extent.
2904 */
2910 /*
2911 * It's a delayed allocation extent, so skip it.
2912 */
2914 }
2916 /* Translate to on disk format */
2919 dp++;
2920 copied++;
2921 }
2926 }
2928 /*
2929 * Each of the following cases stores data into the same region
2930 * of the on-disk inode, so only one of them can be valid at
2931 * any given time. While it is possible to have conflicting formats
2932 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2933 * in EXTENTS format, this can only happen when the fork has
2934 * changed formats after being modified but before being flushed.
2935 * In these cases, the format always takes precedence, because the
2936 * format indicates the current state of the fork.
2937 */
2938 /*ARGSUSED*/
2939 STATIC int
2946 {
2950 #ifdef XFS_TRANS_DEBUG
2952 #endif
2963 /*
2964 * This can happen if we gave up in iformat in an error path,
2965 * for the attribute fork.
2966 */
2970 }
2980 }
2994 whichfork);
2995 }
3004 XFS_BROOT_SIZE_ADJ));
3008 }
3015 }
3023 }
3029 }
3032 }
3034 /*
3035 * xfs_iflush() will write a modified inode's changes out to the
3036 * inode's on disk home. The caller must have the inode lock held
3037 * in at least shared mode and the inode flush semaphore must be
3038 * held as well. The inode lock will still be held upon return from
3039 * the call and the caller is free to unlock it.
3040 * The inode flush lock will be unlocked when the inode reaches the disk.
3041 * The flags indicate how the inode's buffer should be written out.
3042 */
3043 int
3046 uint flags)
3047 {
3053 /* REFERENCED */
3070 /*
3071 * If the inode isn't dirty, then just release the inode
3072 * flush lock and do nothing.
3073 */
3080 }
3082 /*
3083 * We can't flush the inode until it is unpinned, so
3084 * wait for it. We know noone new can pin it, because
3085 * we are holding the inode lock shared and you need
3086 * to hold it exclusively to pin the inode.
3087 */
3090 /*
3091 * This may have been unpinned because the filesystem is shutting
3092 * down forcibly. If that's the case we must not write this inode
3093 * to disk, because the log record didn't make it to disk!
3094 */
3101 }
3103 /*
3104 * Get the buffer containing the on-disk inode.
3105 */
3110 }
3112 /*
3113 * Decide how buffer will be flushed out. This is done before
3114 * the call to xfs_iflush_int because this field is zeroed by it.
3115 */
3117 /*
3118 * Flush out the inode buffer according to the directions
3119 * of the caller. In the cases where the caller has given
3120 * us a choice choose the non-delwri case. This is because
3121 * the inode is in the AIL and we need to get it out soon.
3122 */
3139 }
3157 }
3158 }
3160 /*
3161 * First flush out the inode that xfs_iflush was called with.
3162 */
3166 }
3168 /*
3169 * inode clustering:
3170 * see if other inodes can be gathered into this write
3171 */
3180 /*
3181 * Do an un-protected check to see if the inode is dirty and
3182 * is a candidate for flushing. These checks will be repeated
3183 * later after the appropriate locks are acquired.
3184 */
3191 }
3193 /*
3194 * Try to get locks. If any are unavailable,
3195 * then this inode cannot be flushed and is skipped.
3196 */
3198 /* get inode locks (just i_lock) */
3200 /* get inode flush lock */
3202 /* check if pinned */
3204 /* arriving here means that
3205 * this inode can be flushed.
3206 * first re-check that it's
3207 * dirty
3208 */
3213 clcount++;
3217 XFS_ILOCK_SHARED);
3219 }
3222 }
3225 }
3226 }
3228 }
3229 }
3235 }
3237 /*
3238 * If the buffer is pinned then push on the log so we won't
3239 * get stuck waiting in the write for too long.
3240 */
3243 }
3251 }
3254 corrupt_out:
3258 /*
3259 * Unlocks the flush lock
3260 */
3263 cluster_corrupt_out:
3264 /* Corruption detected in the clustering loop. Invalidate the
3265 * inode buffer and shut down the filesystem.
3266 */
3269 /*
3270 * Clean up the buffer. If it was B_DELWRI, just release it --
3271 * brelse can handle it with no problems. If not, shut down the
3272 * filesystem before releasing the buffer.
3273 */
3276 }
3281 /*
3282 * Just like incore_relse: if we have b_iodone functions,
3283 * mark the buffer as an error and call them. Otherwise
3284 * mark it as stale and brelse.
3285 */
3296 }
3297 }
3300 /*
3301 * Unlocks the flush lock
3302 */
3304 }
3307 STATIC int
3311 {
3315 #ifdef XFS_TRANS_DEBUG
3317 #endif
3329 /*
3330 * If the inode isn't dirty, then just release the inode
3331 * flush lock and do nothing.
3332 */
3337 }
3339 /* set *dip = inode's place in the buffer */
3342 /*
3343 * Clear i_update_core before copying out the data.
3344 * This is for coordination with our timestamp updates
3345 * that don't hold the inode lock. They will always
3346 * update the timestamps BEFORE setting i_update_core,
3347 * so if we clear i_update_core after they set it we
3348 * are guaranteed to see their updates to the timestamps.
3349 * I believe that this depends on strongly ordered memory
3350 * semantics, but we have that. We use the SYNCHRONIZE
3351 * macro to make sure that the compiler does not reorder
3352 * the i_update_core access below the data copy below.
3353 */
3357 /*
3358 * Make sure to get the latest atime from the Linux inode.
3359 */
3368 }
3375 }
3385 }
3396 }
3397 }
3400 XFS_RANDOM_IFLUSH_5)) {
3406 ip);
3408 }
3415 }
3416 /*
3417 * bump the flush iteration count, used to detect flushes which
3418 * postdate a log record during recovery.
3419 */
3423 /*
3424 * Copy the dirty parts of the inode into the on-disk
3425 * inode. We always copy out the core of the inode,
3426 * because if the inode is dirty at all the core must
3427 * be.
3428 */
3431 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3435 /*
3436 * If this is really an old format inode and the superblock version
3437 * has not been updated to support only new format inodes, then
3438 * convert back to the old inode format. If the superblock version
3439 * has been updated, then make the conversion permanent.
3440 */
3445 /*
3446 * Convert it back.
3447 */
3451 /*
3452 * The superblock version has already been bumped,
3453 * so just make the conversion to the new inode
3454 * format permanent.
3455 */
3464 }
3465 }
3469 }
3472 /*
3473 * The only error from xfs_iflush_fork is on the data fork.
3474 */
3476 }
3479 /*
3480 * We've recorded everything logged in the inode, so we'd
3481 * like to clear the ilf_fields bits so we don't log and
3482 * flush things unnecessarily. However, we can't stop
3483 * logging all this information until the data we've copied
3484 * into the disk buffer is written to disk. If we did we might
3485 * overwrite the copy of the inode in the log with all the
3486 * data after re-logging only part of it, and in the face of
3487 * a crash we wouldn't have all the data we need to recover.
3488 *
3489 * What we do is move the bits to the ili_last_fields field.
3490 * When logging the inode, these bits are moved back to the
3491 * ilf_fields field. In the xfs_iflush_done() routine we
3492 * clear ili_last_fields, since we know that the information
3493 * those bits represent is permanently on disk. As long as
3494 * the flush completes before the inode is logged again, then
3495 * both ilf_fields and ili_last_fields will be cleared.
3496 *
3497 * We can play with the ilf_fields bits here, because the inode
3498 * lock must be held exclusively in order to set bits there
3499 * and the flush lock protects the ili_last_fields bits.
3500 * Set ili_logged so the flush done
3501 * routine can tell whether or not to look in the AIL.
3502 * Also, store the current LSN of the inode so that we can tell
3503 * whether the item has moved in the AIL from xfs_iflush_done().
3504 * In order to read the lsn we need the AIL lock, because
3505 * it is a 64 bit value that cannot be read atomically.
3506 */
3517 /*
3518 * Attach the function xfs_iflush_done to the inode's
3519 * buffer. This will remove the inode from the AIL
3520 * and unlock the inode's flush lock when the inode is
3521 * completely written to disk.
3522 */
3529 /*
3530 * We're flushing an inode which is not in the AIL and has
3531 * not been logged but has i_update_core set. For this
3532 * case we can use a B_DELWRI flush and immediately drop
3533 * the inode flush lock because we can avoid the whole
3534 * AIL state thing. It's OK to drop the flush lock now,
3535 * because we've already locked the buffer and to do anything
3536 * you really need both.
3537 */
3542 }
3544 }
3548 corrupt_out:
3550 }
3553 /*
3554 * Flush all inactive inodes in mp.
3555 */
3556 void
3559 {
3563 again:
3570 /* Make sure we skip markers inserted by sync */
3574 }
3581 }
3587 out:
3589 }
3591 /*
3592 * xfs_iaccess: check accessibility of inode for mode.
3593 */
3594 int
3597 mode_t mode,
3599 {
3613 }
3615 /*
3616 * If there's an Access Control List it's used instead of
3617 * the mode bits.
3618 */
3626 }
3628 /*
3629 * If the DACs are ok we don't need any capability check.
3630 */
3633 /*
3634 * Read/write DACs are always overridable.
3635 * Executable DACs are overridable if at least one exec bit is set.
3636 */
3646 #ifdef NOISE
3650 }
3652 }
3654 /*
3655 * xfs_iroundup: round up argument to next power of two
3656 */
3657 uint
3659 uint v)
3660 {
3662 uint m;
3675 }
3678 }
3680 #ifdef XFS_ILOCK_TRACE
3683 void
3685 {
3694 }
3695 #endif
3697 /*
3698 * Return a pointer to the extent record at file index idx.
3699 */
3700 xfs_bmbt_rec_host_t *
3704 {
3719 }
3720 }
3722 /*
3723 * Insert new item(s) into the extent records for incore inode
3724 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3725 */
3726 void
3732 {
3739 }
3741 /*
3742 * This is called when the amount of space required for incore file
3743 * extents needs to be increased. The ext_diff parameter stores the
3744 * number of new extents being added and the idx parameter contains
3745 * the extent index where the new extents will be added. If the new
3746 * extents are being appended, then we just need to (re)allocate and
3747 * initialize the space. Otherwise, if the new extents are being
3748 * inserted into the middle of the existing entries, a bit more work
3749 * is required to make room for the new extents to be inserted. The
3750 * caller is responsible for filling in the new extent entries upon
3751 * return.
3752 */
3753 void
3758 {
3767 /*
3768 * If the new number of extents (nextents + ext_diff)
3769 * fits inside the inode, then continue to use the inline
3770 * extent buffer.
3771 */
3778 }
3782 }
3783 /*
3784 * Otherwise use a linear (direct) extent list.
3785 * If the extents are currently inside the inode,
3786 * xfs_iext_realloc_direct will switch us from
3787 * inline to direct extent allocation mode.
3788 */
3796 }
3797 }
3798 /* Indirection array */
3811 }
3812 /* Extents fit in target extent page */
3820 }
3823 }
3824 /* Insert a new extent page */
3828 }
3829 /*
3830 * If extent(s) are being appended to the last page in
3831 * the indirection array and the new extent(s) don't fit
3832 * in the page, then erp is NULL and erp_idx is set to
3833 * the next index needed in the indirection array.
3834 */
3843 erp_idx++;
3844 }
3845 }
3846 }
3847 }
3849 }
3851 /*
3852 * This is called when incore extents are being added to the indirection
3853 * array and the new extents do not fit in the target extent list. The
3854 * erp_idx parameter contains the irec index for the target extent list
3855 * in the indirection array, and the idx parameter contains the extent
3856 * index within the list. The number of extents being added is stored
3857 * in the count parameter.
3858 *
3859 * |-------| |-------|
3860 * | | | | idx - number of extents before idx
3861 * | idx | | count |
3862 * | | | | count - number of extents being inserted at idx
3863 * |-------| |-------|
3864 * | count | | nex2 | nex2 - number of extents after idx + count
3865 * |-------| |-------|
3866 */
3867 void
3873 {
3887 /*
3888 * Save second part of target extent list
3889 * (all extents past */
3897 }
3899 /*
3900 * Add the new extents to the end of the target
3901 * list, then allocate new irec record(s) and
3902 * extent buffer(s) as needed to store the rest
3903 * of the new extents.
3904 */
3911 }
3913 erp_idx++;
3919 }
3921 /* Add nex2 extents back to indirection array */
3923 xfs_extnum_t ext_avail;
3929 /*
3930 * If nex2 extents fit in the current page, append
3931 * nex2_ep after the new extents.
3932 */
3935 }
3936 /*
3937 * Otherwise, check if space is available in the
3938 * next page.
3939 */
3943 erp_idx++;
3944 erp++;
3945 /* Create a hole for nex2 extents */
3948 }
3949 /*
3950 * Final choice, create a new extent page for
3951 * nex2 extents.
3952 */
3954 erp_idx++;
3956 }
3961 }
3962 }
3964 /*
3965 * This is called when the amount of space required for incore file
3966 * extents needs to be decreased. The ext_diff parameter stores the
3967 * number of extents to be removed and the idx parameter contains
3968 * the extent index where the extents will be removed from.
3969 *
3970 * If the amount of space needed has decreased below the linear
3971 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3972 * extent array. Otherwise, use kmem_realloc() to adjust the
3973 * size to what is needed.
3974 */
3975 void
3980 {
3996 }
3998 }
4000 /*
4001 * This removes ext_diff extents from the inline buffer, beginning
4002 * at extent index idx.
4003 */
4004 void
4009 {
4028 }
4029 }
4031 /*
4032 * This removes ext_diff extents from a linear (direct) extent list,
4033 * beginning at extent index idx. If the extents are being removed
4034 * from the end of the list (ie. truncate) then we just need to re-
4035 * allocate the list to remove the extra space. Otherwise, if the
4036 * extents are being removed from the middle of the existing extent
4037 * entries, then we first need to move the extent records beginning
4038 * at idx + ext_diff up in the list to overwrite the records being
4039 * removed, then remove the extra space via kmem_realloc.
4040 */
4041 void
4046 {
4058 }
4059 /* Move extents up in the list (if needed) */
4065 }
4068 /*
4069 * Reallocate the direct extent list. If the extents
4070 * will fit inside the inode then xfs_iext_realloc_direct
4071 * will switch from direct to inline extent allocation
4072 * mode for us.
4073 */
4076 }
4078 /*
4079 * This is called when incore extents are being removed from the
4080 * indirection array and the extents being removed span multiple extent
4081 * buffers. The idx parameter contains the file extent index where we
4082 * want to begin removing extents, and the count parameter contains
4083 * how many extents need to be removed.
4084 *
4085 * |-------| |-------|
4086 * | nex1 | | | nex1 - number of extents before idx
4087 * |-------| | count |
4088 * | | | | count - number of extents being removed at idx
4089 * | count | |-------|
4090 * | | | nex2 | nex2 - number of extents after idx + count
4091 * |-------| |-------|
4092 */
4093 void
4098 {
4117 /*
4118 * Check for deletion of entire list;
4119 * xfs_iext_irec_remove() updates extent offsets.
4120 */
4127 XFS_IEXT_BUFSZ);
4133 }
4134 }
4135 /* Move extents up (if needed) */
4140 }
4141 /* Zero out rest of page */
4144 /* Update remaining counters */
4149 erp_idx++;
4150 erp++;
4151 }
4154 }
4156 /*
4157 * Create, destroy, or resize a linear (direct) block of extents.
4158 */
4159 void
4163 {
4172 /* Free extent records */
4175 }
4176 /* Resize direct extent list and zero any new bytes */
4178 /* Check if extents will fit inside the inode */
4184 }
4187 }
4191 rnew_size,
4193 KM_SLEEP);
4194 }
4199 }
4200 }
4201 /*
4202 * Switch from the inline extent buffer to a direct
4203 * extent list. Be sure to include the inline extent
4204 * bytes in new_size.
4205 */
4210 }
4212 }
4215 }
4217 /*
4218 * Switch from linear (direct) extent records to inline buffer.
4219 */
4220 void
4224 {
4227 /*
4228 * The inline buffer was zeroed when we switched
4229 * from inline to direct extent allocation mode,
4230 * so we don't need to clear it here.
4231 */
4237 }
4239 /*
4240 * Switch from inline buffer to linear (direct) extent records.
4241 * new_size should already be rounded up to the next power of 2
4242 * by the caller (when appropriate), so use new_size as it is.
4243 * However, since new_size may be rounded up, we can't update
4244 * if_bytes here. It is the caller's responsibility to update
4245 * if_bytes upon return.
4246 */
4247 void
4251 {
4259 }
4261 }
4263 /*
4264 * Resize an extent indirection array to new_size bytes.
4265 */
4266 void
4270 {
4285 }
4286 }
4288 /*
4289 * Switch from indirection array to linear (direct) extent allocations.
4290 */
4291 void
4294 {
4314 }
4315 }
4317 /*
4318 * Free incore file extents.
4319 */
4320 void
4323 {
4331 }
4338 }
4342 }
4344 /*
4345 * Return a pointer to the extent record for file system block bno.
4346 */
4352 {
4367 }
4370 /* Find target extent list */
4378 }
4379 /* Binary search extent records */
4390 /* Convert back to file-based extent index */
4393 }
4396 }
4397 }
4398 /* Convert back to file-based extent index */
4401 }
4407 }
4408 }
4411 }
4413 /*
4414 * Return a pointer to the indirection array entry containing the
4415 * extent record for filesystem block bno. Store the index of the
4416 * target irec in *erp_idxp.
4417 */
4423 {
4447 }
4448 }
4451 }
4453 /*
4454 * Return a pointer to the indirection array entry containing the
4455 * extent record at file extent index *idxp. Store the index of the
4456 * target irec in *erp_idxp and store the page index of the target
4457 * extent record in *idxp.
4458 */
4459 xfs_ext_irec_t *
4465 {
4482 /* Binary search extent irec's */
4498 erp_idx++;
4504 }
4505 }
4509 }
4511 /*
4512 * Allocate and initialize an indirection array once the space needed
4513 * for incore extents increases above XFS_IEXT_BUFSZ.
4514 */
4515 void
4518 {
4535 }
4546 }
4548 /*
4549 * Allocate and initialize a new entry in the indirection array.
4550 */
4551 xfs_ext_irec_t *
4555 {
4563 /* Resize indirection array */
4566 /*
4567 * Move records down in the array so the
4568 * new page can use erp_idx.
4569 */
4573 }
4576 /* Initialize new extent record */
4585 }
4587 /*
4588 * Remove a record from the indirection array.
4589 */
4590 void
4594 {
4606 }
4607 /* Compact extent records */
4611 }
4612 /*
4613 * Manually free the last extent record from the indirection
4614 * array. A call to xfs_iext_realloc_indirect() with a size
4615 * of zero would result in a call to xfs_iext_destroy() which
4616 * would in turn call this function again, creating a nasty
4617 * infinite loop.
4618 */
4625 }
4627 }
4629 /*
4630 * This is called to clean up large amounts of unused memory allocated
4631 * by the indirection array. Before compacting anything though, verify
4632 * that the indirection array is still needed and switch back to the
4633 * linear extent list (or even the inline buffer) if possible. The
4634 * compaction policy is as follows:
4635 *
4636 * Full Compaction: Extents fit into a single page (or inline buffer)
4637 * Full Compaction: Extents occupy less than 10% of allocated space
4638 * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
4639 * No Compaction: Extents occupy at least 50% of allocated space
4640 */
4641 void
4644 {
4663 }
4664 }
4666 /*
4667 * Combine extents from neighboring extent pages.
4668 */
4669 void
4672 {
4688 /*
4689 * Free page before removing extent record
4690 * so er_extoffs don't get modified in
4691 * xfs_iext_irec_remove.
4692 */
4698 erp_idx++;
4699 }
4700 }
4701 }
4703 /*
4704 * Fully compact the extent records managed by the indirection array.
4705 */
4706 void
4709 {
4729 /* Remove next page */
4731 /*
4732 * Free page before removing extent record
4733 * so er_extoffs don't get modified in
4734 * xfs_iext_irec_remove.
4735 */
4742 /* Update next page */
4744 /* Move rest of page up to become next new page */
4751 }
4753 erp_idx++;
4756 else
4758 }
4762 }
4763 }
4765 /*
4766 * This is called to update the er_extoff field in the indirection
4767 * array when extents have been added or removed from one of the
4768 * extent lists. erp_idx contains the irec index to begin updating
4769 * at and ext_diff contains the number of extents that were added
4770 * or removed.
4771 */
4772 void
4777 {
4785 }
4786 }