| blob | 48146bdc6bdde8bff1a17533d2019d21cd0f535f |
1 /*
2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
60 /*
61 * Used in xfs_itruncate(). This is the maximum number of extents
62 * freed from a file in a single transaction.
63 */
64 #define XFS_ITRUNC_MAX_EXTENTS 2
72 #ifdef DEBUG
73 /*
74 * Make sure that the extents in the given memory buffer
75 * are valid.
76 */
77 STATIC void
82 xfs_exntfmt_t fmt)
83 {
85 xfs_bmbt_irec_t irec;
86 xfs_bmbt_rec_t rec;
95 else
99 }
100 }
102 #define xfs_validate_extents(ifp, nrecs, disk, fmt)
105 /*
106 * Check that none of the inode's in the buffer have a next
107 * unlinked field of 0.
108 */
109 #if defined(DEBUG)
110 void
114 {
126 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
127 bp);
129 }
130 }
131 }
132 #endif
134 /*
135 * This routine is called to map an inode number within a file
136 * system to the buffer containing the on-disk version of the
137 * inode. It returns a pointer to the buffer containing the
138 * on-disk inode in the bpp parameter, and in the dip parameter
139 * it returns a pointer to the on-disk inode within that buffer.
140 *
141 * If a non-zero error is returned, then the contents of bpp and
142 * dipp are undefined.
143 *
144 * Use xfs_imap() to determine the size and location of the
145 * buffer to read from disk.
146 */
147 STATIC int
151 xfs_ino_t ino,
155 {
157 xfs_imap_t imap;
162 /*
163 * Call the space management code to find the location of the
164 * inode on disk.
165 */
170 "xfs_inotobp: xfs_imap() returned an "
173 }
175 /*
176 * If the inode number maps to a block outside the bounds of the
177 * file system then return NULL rather than calling read_buf
178 * and panicing when we get an error from the driver.
179 */
183 "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
188 }
190 /*
191 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
192 * default to just a read_buf() call.
193 */
199 "xfs_inotobp: xfs_trans_read_buf() returned an "
202 }
204 di_ok =
208 XFS_RANDOM_ITOBP_INOTOBP))) {
212 "xfs_inotobp: XFS_TEST_ERROR() returned an "
215 }
219 /*
220 * Set *dipp to point to the on-disk inode in the buffer.
221 */
226 }
229 /*
230 * This routine is called to map an inode to the buffer containing
231 * the on-disk version of the inode. It returns a pointer to the
232 * buffer containing the on-disk inode in the bpp parameter, and in
233 * the dip parameter it returns a pointer to the on-disk inode within
234 * that buffer.
235 *
236 * If a non-zero error is returned, then the contents of bpp and
237 * dipp are undefined.
238 *
239 * If the inode is new and has not yet been initialized, use xfs_imap()
240 * to determine the size and location of the buffer to read from disk.
241 * If the inode has already been mapped to its buffer and read in once,
242 * then use the mapping information stored in the inode rather than
243 * calling xfs_imap(). This allows us to avoid the overhead of looking
244 * at the inode btree for small block file systems (see xfs_dilocate()).
245 * We can tell whether the inode has been mapped in before by comparing
246 * its disk block address to 0. Only uninitialized inodes will have
247 * 0 for the disk block address.
248 */
249 int
256 xfs_daddr_t bno,
257 uint imap_flags)
258 {
261 xfs_imap_t imap;
262 #ifdef __KERNEL__
265 #endif
268 /*
269 * Call the space management code to find the location of the
270 * inode on disk.
271 */
277 /*
278 * If the inode number maps to a block outside the bounds
279 * of the file system then return NULL rather than calling
280 * read_buf and panicing when we get an error from the
281 * driver.
282 */
285 #ifdef DEBUG
287 "(imap.im_blkno (0x%llx) "
288 "+ imap.im_len (0x%llx)) > "
289 " XFS_FSB_TO_BB(mp, "
296 }
298 /*
299 * Fill in the fields in the inode that will be used to
300 * map the inode to its buffer from now on.
301 */
306 /*
307 * We've already mapped the inode once, so just use the
308 * mapping that we saved the first time.
309 */
313 }
316 /*
317 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
318 * default to just a read_buf() call.
319 */
324 #ifdef DEBUG
326 "xfs_trans_read_buf() returned error %d, "
332 }
333 #ifdef __KERNEL__
334 /*
335 * Validate the magic number and version of every inode in the buffer
336 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
337 */
338 #ifdef DEBUG
341 #else
343 #endif
353 XFS_RANDOM_ITOBP_INOTOBP))) {
354 #ifdef DEBUG
359 #endif
364 }
365 }
370 /*
371 * Mark the buffer as an inode buffer now that it looks good
372 */
375 /*
376 * Set *dipp to point to the on-disk inode in the buffer.
377 */
381 }
383 /*
384 * Move inode type and inode format specific information from the
385 * on-disk inode to the in-core inode. For fifos, devs, and sockets
386 * this means set if_rdev to the proper value. For files, directories,
387 * and symlinks this means to bring in the in-line data or extent
388 * pointers. For a file in B-tree format, only the root is immediately
389 * brought in-core. The rest will be in-lined in if_extents when it
390 * is first referenced (see xfs_iread_extents()).
391 */
392 STATIC int
396 {
400 xfs_fsize_t di_size;
419 }
421 if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) {
429 }
440 }
450 /*
451 * no local regular files yet
452 */
455 "corrupt inode %Lu "
459 XFS_ERRLEVEL_LOW,
462 }
467 "corrupt inode %Lu "
472 XFS_ERRLEVEL_LOW,
475 }
490 }
496 }
499 }
521 }
526 }
528 }
530 /*
531 * The file is in-lined in the on-disk inode.
532 * If it fits into if_inline_data, then copy
533 * it there, otherwise allocate a buffer for it
534 * and copy the data there. Either way, set
535 * if_data to point at the data.
536 * If we allocate a buffer for the data, make
537 * sure that its size is a multiple of 4 and
538 * record the real size in i_real_bytes.
539 */
540 STATIC int
546 {
550 /*
551 * If the size is unreasonable, then something
552 * is wrong and we just bail out rather than crash in
553 * kmem_alloc() or memcpy() below.
554 */
557 "corrupt inode %Lu "
564 }
574 }
582 }
584 /*
585 * The file consists of a set of extents all
586 * of which fit into the on-disk inode.
587 * If there are few enough extents to fit into
588 * the if_inline_ext, then copy them there.
589 * Otherwise allocate a buffer for them and copy
590 * them into it. Either way, set if_extents
591 * to point at the extents.
592 */
593 STATIC int
598 {
609 /*
610 * If the number of extents is unreasonable, then something
611 * is wrong and we just bail out rather than crash in
612 * kmem_alloc() or memcpy() below.
613 */
621 }
628 else
638 ARCH_CONVERT);
640 ARCH_CONVERT);
641 }
643 whichfork);
649 XFS_ERRLEVEL_LOW,
652 }
653 }
656 }
658 /*
659 * The file has too many extents to fit into
660 * the inode, so they are in B-tree format.
661 * Allocate a buffer for the root of the B-tree
662 * and copy the root into it. The i_extents
663 * field will remain NULL until all of the
664 * extents are read in (when they are needed).
665 */
666 STATIC int
671 {
674 /* REFERENCED */
683 /*
684 * blow out if -- fork has less extents than can fit in
685 * fork (fork shouldn't be a btree format), root btree
686 * block has more records than can fit into the fork,
687 * or the number of extents is greater than the number of
688 * blocks.
689 */
700 }
705 /*
706 * Copy and convert from the on-disk structure
707 * to the in-memory structure.
708 */
715 }
717 /*
718 * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
719 * and native format
720 *
721 * buf = on-disk representation
722 * dip = native representation
723 * dir = direction - +ve -> disk to native
724 * -ve -> native to disk
725 */
726 void
728 xfs_caddr_t buf,
731 {
754 }
781 }
783 STATIC uint
786 __uint16_t di_flags)
787 {
815 }
818 }
820 uint
823 {
828 }
830 uint
833 {
836 }
838 /*
839 * Given a mount structure and an inode number, return a pointer
840 * to a newly allocated in-core inode corresponding to the given
841 * inode number.
842 *
843 * Initialize the inode's attributes and extent pointers if it
844 * already has them (it will not if the inode has no links).
845 */
846 int
850 xfs_ino_t ino,
852 xfs_daddr_t bno)
853 {
865 /*
866 * Get pointer's to the on-disk inode and the buffer containing it.
867 * If the inode number refers to a block outside the file system
868 * then xfs_itobp() will return NULL. In this case we should
869 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
870 * know that this is a new incore inode.
871 */
876 }
878 /*
879 * Initialize inode's trace buffers.
880 * Do this before xfs_iformat in case it adds entries.
881 */
882 #ifdef XFS_BMAP_TRACE
884 #endif
885 #ifdef XFS_BMBT_TRACE
887 #endif
888 #ifdef XFS_RW_TRACE
890 #endif
891 #ifdef XFS_ILOCK_TRACE
893 #endif
894 #ifdef XFS_DIR2_TRACE
896 #endif
898 /*
899 * If we got something that isn't an inode it means someone
900 * (nfs or dmi) has a stale handle.
901 */
905 #ifdef DEBUG
907 "dip->di_core.di_magic (0x%x) != "
910 XFS_DINODE_MAGIC);
913 }
915 /*
916 * If the on-disk inode is already linked to a directory
917 * entry, copy all of the inode into the in-core inode.
918 * xfs_iformat() handles copying in the inode format
919 * specific information.
920 * Otherwise, just get the truly permanent information.
921 */
929 #ifdef DEBUG
932 error);
935 }
941 /*
942 * Make sure to pull in the mode here as well in
943 * case the inode is released without being used.
944 * This ensures that xfs_inactive() will see that
945 * the inode is already free and not try to mess
946 * with the uninitialized part of it.
947 */
949 /*
950 * Initialize the per-fork minima and maxima for a new
951 * inode here. xfs_iformat will do it for old inodes.
952 */
955 }
959 /*
960 * The inode format changed when we moved the link count and
961 * made it 32 bits long. If this is an old format inode,
962 * convert it in memory to look like a new one. If it gets
963 * flushed to disk we will convert back before flushing or
964 * logging it. We zero out the new projid field and the old link
965 * count field. We'll handle clearing the pad field (the remains
966 * of the old uuid field) when we actually convert the inode to
967 * the new format. We don't change the version number so that we
968 * can distinguish this from a real new format inode.
969 */
974 }
978 /*
979 * Mark the buffer containing the inode as something to keep
980 * around for a while. This helps to keep recently accessed
981 * meta-data in-core longer.
982 */
985 /*
986 * Use xfs_trans_brelse() to release the buffer containing the
987 * on-disk inode, because it was acquired with xfs_trans_read_buf()
988 * in xfs_itobp() above. If tp is NULL, this is just a normal
989 * brelse(). If we're within a transaction, then xfs_trans_brelse()
990 * will only release the buffer if it is not dirty within the
991 * transaction. It will be OK to release the buffer in this case,
992 * because inodes on disk are never destroyed and we will be
993 * locking the new in-core inode before putting it in the hash
994 * table where other processes can find it. Thus we don't have
995 * to worry about the inode being changed just because we released
996 * the buffer.
997 */
1001 }
1003 /*
1004 * Read in extents from a btree-format inode.
1005 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1006 */
1007 int
1012 {
1015 xfs_extnum_t nextents;
1022 }
1027 /*
1028 * We know that the size is valid (it's checked in iformat_btree)
1029 */
1039 }
1042 }
1044 /*
1045 * Allocate an inode on disk and return a copy of its in-core version.
1046 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1047 * appropriately within the inode. The uid and gid for the inode are
1048 * set according to the contents of the given cred structure.
1049 *
1050 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1051 * has a free inode available, call xfs_iget()
1052 * to obtain the in-core version of the allocated inode. Finally,
1053 * fill in the inode and log its initial contents. In this case,
1054 * ialloc_context would be set to NULL and call_again set to false.
1055 *
1056 * If xfs_dialloc() does not have an available inode,
1057 * it will replenish its supply by doing an allocation. Since we can
1058 * only do one allocation within a transaction without deadlocks, we
1059 * must commit the current transaction before returning the inode itself.
1060 * In this case, therefore, we will set call_again to true and return.
1061 * The caller should then commit the current transaction, start a new
1062 * transaction, and call xfs_ialloc() again to actually get the inode.
1063 *
1064 * To ensure that some other process does not grab the inode that
1065 * was allocated during the first call to xfs_ialloc(), this routine
1066 * also returns the [locked] bp pointing to the head of the freelist
1067 * as ialloc_context. The caller should hold this buffer across
1068 * the commit and pass it back into this routine on the second call.
1069 */
1070 int
1074 mode_t mode,
1075 xfs_nlink_t nlink,
1076 xfs_dev_t rdev,
1078 xfs_prid_t prid,
1083 {
1084 xfs_ino_t ino;
1087 uint flags;
1090 /*
1091 * Call the space management code to pick
1092 * the on-disk inode to be allocated.
1093 */
1098 }
1102 }
1105 /*
1106 * Get the in-core inode with the lock held exclusively.
1107 * This is because we're setting fields here we need
1108 * to prevent others from looking at until we're done.
1109 */
1114 }
1127 /*
1128 * If the superblock version is up to where we support new format
1129 * inodes and this is currently an old format inode, then change
1130 * the inode version number now. This way we only do the conversion
1131 * here rather than here and in the flush/logging code.
1132 */
1136 /*
1137 * We've already zeroed the old link count, the projid field,
1138 * and the pad field.
1139 */
1140 }
1142 /*
1143 * Project ids won't be stored on disk if we are using a version 1 inode.
1144 */
1152 }
1153 }
1155 /*
1156 * If the group ID of the new file does not match the effective group
1157 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1158 * (and only if the irix_sgid_inherit compatibility variable is set).
1159 */
1164 }
1170 /*
1171 * di_gen will have been taken care of in xfs_iread.
1172 */
1199 }
1204 }
1208 }
1209 }
1211 xfs_inherit_noatime)
1214 xfs_inherit_nodump)
1217 xfs_inherit_sync)
1220 xfs_inherit_nosymlinks)
1225 }
1226 /* FALLTHROUGH */
1235 }
1236 /*
1237 * Attribute fork settings for new inode.
1238 */
1242 /*
1243 * Log the new values stuffed into the inode.
1244 */
1247 /* now that we have an i_mode we can set Linux inode ops (& unlock) */
1252 }
1254 /*
1255 * Check to make sure that there are no blocks allocated to the
1256 * file beyond the size of the file. We don't check this for
1257 * files with fixed size extents or real time extents, but we
1258 * at least do it for regular files.
1259 */
1260 #ifdef DEBUG
1261 void
1265 xfs_fsize_t isize)
1266 {
1267 xfs_fileoff_t map_first;
1279 /*
1280 * The filesystem could be shutting down, so bmapi may return
1281 * an error.
1282 */
1286 map_first),
1288 NULL))
1292 }
1295 /*
1296 * Calculate the last possible buffered byte in a file. This must
1297 * include data that was buffered beyond the EOF by the write code.
1298 * This also needs to deal with overflowing the xfs_fsize_t type
1299 * which can happen for sizes near the limit.
1300 *
1301 * We also need to take into account any blocks beyond the EOF. It
1302 * may be the case that they were buffered by a write which failed.
1303 * In that case the pages will still be in memory, but the inode size
1304 * will never have been updated.
1305 */
1306 xfs_fsize_t
1309 {
1311 xfs_fsize_t last_byte;
1312 xfs_fileoff_t last_block;
1313 xfs_fileoff_t size_last_block;
1319 /*
1320 * Only check for blocks beyond the EOF if the extents have
1321 * been read in. This eliminates the need for the inode lock,
1322 * and it also saves us from looking when it really isn't
1323 * necessary.
1324 */
1327 XFS_DATA_FORK);
1330 }
1333 }
1340 }
1344 }
1346 }
1348 #if defined(XFS_RW_TRACE)
1349 STATIC void
1354 xfs_fsize_t new_size,
1355 xfs_off_t toss_start,
1356 xfs_off_t toss_finish)
1357 {
1360 }
1379 }
1380 #else
1381 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1382 #endif
1384 /*
1385 * Start the truncation of the file to new_size. The new size
1386 * must be smaller than the current size. This routine will
1387 * clear the buffer and page caches of file data in the removed
1388 * range, and xfs_itruncate_finish() will remove the underlying
1389 * disk blocks.
1390 *
1391 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1392 * must NOT have the inode lock held at all. This is because we're
1393 * calling into the buffer/page cache code and we can't hold the
1394 * inode lock when we do so.
1395 *
1396 * We need to wait for any direct I/Os in flight to complete before we
1397 * proceed with the truncate. This is needed to prevent the extents
1398 * being read or written by the direct I/Os from being removed while the
1399 * I/O is in flight as there is no other method of synchronising
1400 * direct I/O with the truncate operation. Also, because we hold
1401 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1402 * started until the truncate completes and drops the lock. Essentially,
1403 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1404 * between direct I/Os and the truncate operation.
1405 *
1406 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1407 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1408 * in the case that the caller is locking things out of order and
1409 * may not be able to call xfs_itruncate_finish() with the inode lock
1410 * held without dropping the I/O lock. If the caller must drop the
1411 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1412 * must be called again with all the same restrictions as the initial
1413 * call.
1414 */
1415 void
1418 uint flags,
1419 xfs_fsize_t new_size)
1420 {
1421 xfs_fsize_t last_byte;
1422 xfs_off_t toss_start;
1436 /*
1437 * Call VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES() to get rid of pages and buffers
1438 * overlapping the region being removed. We have to use
1439 * the less efficient VOP_FLUSHINVAL_PAGES() in the case that the
1440 * caller may not be able to finish the truncate without
1441 * dropping the inode's I/O lock. Make sure
1442 * to catch any pages brought in by buffers overlapping
1443 * the EOF by searching out beyond the isize by our
1444 * block size. We round new_size up to a block boundary
1445 * so that we don't toss things on the same block as
1446 * new_size but before it.
1447 *
1448 * Before calling VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES(), make sure to
1449 * call remapf() over the same region if the file is mapped.
1450 * This frees up mapped file references to the pages in the
1451 * given range and for the VOP_FLUSHINVAL_PAGES() case it ensures
1452 * that we get the latest mapped changes flushed out.
1453 */
1457 /*
1458 * The place to start tossing is beyond our maximum
1459 * file size, so there is no way that the data extended
1460 * out there.
1461 */
1463 }
1466 last_byte);
1472 }
1473 }
1475 #ifdef DEBUG
1478 }
1479 #endif
1480 }
1482 /*
1483 * Shrink the file to the given new_size. The new
1484 * size must be smaller than the current size.
1485 * This will free up the underlying blocks
1486 * in the removed range after a call to xfs_itruncate_start()
1487 * or xfs_atruncate_start().
1488 *
1489 * The transaction passed to this routine must have made
1490 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1491 * This routine may commit the given transaction and
1492 * start new ones, so make sure everything involved in
1493 * the transaction is tidy before calling here.
1494 * Some transaction will be returned to the caller to be
1495 * committed. The incoming transaction must already include
1496 * the inode, and both inode locks must be held exclusively.
1497 * The inode must also be "held" within the transaction. On
1498 * return the inode will be "held" within the returned transaction.
1499 * This routine does NOT require any disk space to be reserved
1500 * for it within the transaction.
1501 *
1502 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1503 * and it indicates the fork which is to be truncated. For the
1504 * attribute fork we only support truncation to size 0.
1505 *
1506 * We use the sync parameter to indicate whether or not the first
1507 * transaction we perform might have to be synchronous. For the attr fork,
1508 * it needs to be so if the unlink of the inode is not yet known to be
1509 * permanent in the log. This keeps us from freeing and reusing the
1510 * blocks of the attribute fork before the unlink of the inode becomes
1511 * permanent.
1512 *
1513 * For the data fork, we normally have to run synchronously if we're
1514 * being called out of the inactive path or we're being called
1515 * out of the create path where we're truncating an existing file.
1516 * Either way, the truncate needs to be sync so blocks don't reappear
1517 * in the file with altered data in case of a crash. wsync filesystems
1518 * can run the first case async because anything that shrinks the inode
1519 * has to run sync so by the time we're called here from inactive, the
1520 * inode size is permanently set to 0.
1521 *
1522 * Calls from the truncate path always need to be sync unless we're
1523 * in a wsync filesystem and the file has already been unlinked.
1524 *
1525 * The caller is responsible for correctly setting the sync parameter.
1526 * It gets too hard for us to guess here which path we're being called
1527 * out of just based on inode state.
1528 */
1529 int
1533 xfs_fsize_t new_size,
1536 {
1537 xfs_fsblock_t first_block;
1538 xfs_fileoff_t first_unmap_block;
1539 xfs_fileoff_t last_block;
1545 xfs_bmap_free_t free_list;
1562 /*
1563 * We only support truncating the entire attribute fork.
1564 */
1567 }
1570 /*
1571 * The first thing we do is set the size to new_size permanently
1572 * on disk. This way we don't have to worry about anyone ever
1573 * being able to look at the data being freed even in the face
1574 * of a crash. What we're getting around here is the case where
1575 * we free a block, it is allocated to another file, it is written
1576 * to, and then we crash. If the new data gets written to the
1577 * file but the log buffers containing the free and reallocation
1578 * don't, then we'd end up with garbage in the blocks being freed.
1579 * As long as we make the new_size permanent before actually
1580 * freeing any blocks it doesn't matter if they get writtten to.
1581 *
1582 * The callers must signal into us whether or not the size
1583 * setting here must be synchronous. There are a few cases
1584 * where it doesn't have to be synchronous. Those cases
1585 * occur if the file is unlinked and we know the unlink is
1586 * permanent or if the blocks being truncated are guaranteed
1587 * to be beyond the inode eof (regardless of the link count)
1588 * and the eof value is permanent. Both of these cases occur
1589 * only on wsync-mounted filesystems. In those cases, we're
1590 * guaranteed that no user will ever see the data in the blocks
1591 * that are being truncated so the truncate can run async.
1592 * In the free beyond eof case, the file may wind up with
1593 * more blocks allocated to it than it needs if we crash
1594 * and that won't get fixed until the next time the file
1595 * is re-opened and closed but that's ok as that shouldn't
1596 * be too many blocks.
1597 *
1598 * However, we can't just make all wsync xactions run async
1599 * because there's one call out of the create path that needs
1600 * to run sync where it's truncating an existing file to size
1601 * 0 whose size is > 0.
1602 *
1603 * It's probably possible to come up with a test in this
1604 * routine that would correctly distinguish all the above
1605 * cases from the values of the function parameters and the
1606 * inode state but for sanity's sake, I've decided to let the
1607 * layers above just tell us. It's simpler to correctly figure
1608 * out in the layer above exactly under what conditions we
1609 * can run async and I think it's easier for others read and
1610 * follow the logic in case something has to be changed.
1611 * cscope is your friend -- rcc.
1612 *
1613 * The attribute fork is much simpler.
1614 *
1615 * For the attribute fork we allow the caller to tell us whether
1616 * the unlink of the inode that led to this call is yet permanent
1617 * in the on disk log. If it is not and we will be freeing extents
1618 * in this inode then we make the first transaction synchronous
1619 * to make sure that the unlink is permanent by the time we free
1620 * the blocks.
1621 */
1626 }
1631 }
1637 /*
1638 * Since it is possible for space to become allocated beyond
1639 * the end of the file (in a crash where the space is allocated
1640 * but the inode size is not yet updated), simply remove any
1641 * blocks which show up between the new EOF and the maximum
1642 * possible file size. If the first block to be removed is
1643 * beyond the maximum file size (ie it is the same as last_block),
1644 * then there is nothing to do.
1645 */
1653 }
1655 /*
1656 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1657 * will tell us whether it freed the entire range or
1658 * not. If this is a synchronous mount (wsync),
1659 * then we can tell bunmapi to keep all the
1660 * transactions asynchronous since the unlink
1661 * transaction that made this inode inactive has
1662 * already hit the disk. There's no danger of
1663 * the freed blocks being reused, there being a
1664 * crash, and the reused blocks suddenly reappearing
1665 * in this file with garbage in them once recovery
1666 * runs.
1667 */
1670 unmap_len,
1673 XFS_ITRUNC_MAX_EXTENTS,
1676 /*
1677 * If the bunmapi call encounters an error,
1678 * return to the caller where the transaction
1679 * can be properly aborted. We just need to
1680 * make sure we're not holding any resources
1681 * that we were not when we came in.
1682 */
1685 }
1687 /*
1688 * Duplicate the transaction that has the permanent
1689 * reservation and commit the old transaction.
1690 */
1695 /*
1696 * If the bmap finish call encounters an error,
1697 * return to the caller where the transaction
1698 * can be properly aborted. We just need to
1699 * make sure we're not holding any resources
1700 * that we were not when we came in.
1701 *
1702 * Aborting from this point might lose some
1703 * blocks in the file system, but oh well.
1704 */
1707 /*
1708 * If the passed in transaction committed
1709 * in xfs_bmap_finish(), then we want to
1710 * add the inode to this one before returning.
1711 * This keeps things simple for the higher
1712 * level code, because it always knows that
1713 * the inode is locked and held in the
1714 * transaction that returns to it whether
1715 * errors occur or not. We don't mark the
1716 * inode dirty so that this transaction can
1717 * be easily aborted if possible.
1718 */
1722 }
1724 }
1727 /*
1728 * The first xact was committed,
1729 * so add the inode to the new one.
1730 * Mark it dirty so it will be logged
1731 * and moved forward in the log as
1732 * part of every commit.
1733 */
1738 }
1743 XFS_TRANS_PERM_LOG_RES,
1744 XFS_ITRUNCATE_LOG_COUNT);
1745 /*
1746 * Add the inode being truncated to the next chained
1747 * transaction.
1748 */
1753 }
1754 /*
1755 * Only update the size in the case of the data fork, but
1756 * always re-log the inode so that our permanent transaction
1757 * can keep on rolling it forward in the log.
1758 */
1762 }
1772 }
1775 /*
1776 * xfs_igrow_start
1777 *
1778 * Do the first part of growing a file: zero any data in the last
1779 * block that is beyond the old EOF. We need to do this before
1780 * the inode is joined to the transaction to modify the i_size.
1781 * That way we can drop the inode lock and call into the buffer
1782 * cache to get the buffer mapping the EOF.
1783 */
1784 int
1787 xfs_fsize_t new_size,
1789 {
1796 /*
1797 * Zero any pages that may have been created by
1798 * xfs_write_file() beyond the end of the file
1799 * and any blocks between the old and new file sizes.
1800 */
1804 }
1806 /*
1807 * xfs_igrow_finish
1808 *
1809 * This routine is called to extend the size of a file.
1810 * The inode must have both the iolock and the ilock locked
1811 * for update and it must be a part of the current transaction.
1812 * The xfs_igrow_start() function must have been called previously.
1813 * If the change_flag is not zero, the inode change timestamp will
1814 * be updated.
1815 */
1816 void
1820 xfs_fsize_t new_size,
1822 {
1828 /*
1829 * Update the file size. Update the inode change timestamp
1830 * if change_flag set.
1831 */
1837 }
1840 /*
1841 * This is called when the inode's link count goes to 0.
1842 * We place the on-disk inode on a list in the AGI. It
1843 * will be pulled from this list when the inode is freed.
1844 */
1845 int
1849 {
1855 xfs_agnumber_t agno;
1856 xfs_daddr_t agdaddr;
1857 xfs_agino_t agino;
1872 /*
1873 * Get the agi buffer first. It ensures lock ordering
1874 * on the list.
1875 */
1880 }
1881 /*
1882 * Validate the magic number of the agi block.
1883 */
1885 agi_ok =
1889 XFS_RANDOM_IUNLINK))) {
1893 }
1894 /*
1895 * Get the index into the agi hash table for the
1896 * list this inode will go on.
1897 */
1905 /*
1906 * There is already another inode in the bucket we need
1907 * to add ourselves to. Add us at the front of the list.
1908 * Here we put the head pointer into our next pointer,
1909 * and then we fall through to point the head at us.
1910 */
1914 }
1917 /* both on-disk, don't endian flip twice */
1925 }
1927 /*
1928 * Point the bucket head pointer at the inode being inserted.
1929 */
1937 }
1939 /*
1940 * Pull the on-disk inode from the AGI unlinked list.
1941 */
1942 STATIC int
1946 {
1947 xfs_ino_t next_ino;
1953 xfs_agnumber_t agno;
1954 xfs_daddr_t agdaddr;
1955 xfs_agino_t agino;
1956 xfs_agino_t next_agino;
1964 /*
1965 * First pull the on-disk inode from the AGI unlinked list.
1966 */
1972 /*
1973 * Get the agi buffer first. It ensures lock ordering
1974 * on the list.
1975 */
1983 }
1984 /*
1985 * Validate the magic number of the agi block.
1986 */
1988 agi_ok =
1992 XFS_RANDOM_IUNLINK_REMOVE))) {
2000 }
2001 /*
2002 * Get the index into the agi hash table for the
2003 * list this inode will go on.
2004 */
2012 /*
2013 * We're at the head of the list. Get the inode's
2014 * on-disk buffer to see if there is anyone after us
2015 * on the list. Only modify our next pointer if it
2016 * is not already NULLAGINO. This saves us the overhead
2017 * of dealing with the buffer when there is no need to
2018 * change it.
2019 */
2026 }
2039 }
2040 /*
2041 * Point the bucket head pointer at the next inode.
2042 */
2051 /*
2052 * We need to search the list for the inode being freed.
2053 */
2057 /*
2058 * If the last inode wasn't the one pointing to
2059 * us, then release its buffer since we're not
2060 * going to do anything with it.
2061 */
2064 }
2073 }
2077 }
2078 /*
2079 * Now last_ibp points to the buffer previous to us on
2080 * the unlinked list. Pull us from the list.
2081 */
2088 }
2102 }
2103 /*
2104 * Point the previous inode on the list to the next inode.
2105 */
2113 }
2115 }
2118 {
2122 }
2124 STATIC void
2128 xfs_ino_t inum)
2129 {
2135 xfs_daddr_t blkno;
2152 }
2161 /*
2162 * Look for each inode in memory and attempt to lock it,
2163 * we can be racing with flush and tail pushing here.
2164 * any inode we get the locks on, add to an array of
2165 * inode items to process later.
2166 *
2167 * The get the buffer lock, we could beat a flush
2168 * or tail pushing thread to the lock here, in which
2169 * case they will go looking for the inode buffer
2170 * and fail, we need some other form of interlock
2171 * here.
2172 */
2180 }
2182 /* Inode not in memory or we found it already,
2183 * nothing to do
2184 */
2188 }
2193 }
2195 /* If we can get the locks then add it to the
2196 * list, otherwise by the time we get the bp lock
2197 * below it will already be attached to the
2198 * inode buffer.
2199 */
2201 /* This inode will already be locked - by us, lets
2202 * keep it that way.
2203 */
2213 }
2214 }
2217 }
2228 }
2231 }
2232 }
2235 }
2239 XFS_BUF_LOCK);
2252 pre_flushed++;
2253 }
2255 }
2266 }
2280 }
2281 }
2286 }
2289 }
2291 /*
2292 * This is called to return an inode to the inode free list.
2293 * The inode should already be truncated to 0 length and have
2294 * no pages associated with it. This routine also assumes that
2295 * the inode is already a part of the transaction.
2296 *
2297 * The on-disk copy of the inode will have been added to the list
2298 * of unlinked inodes in the AGI. We need to remove the inode from
2299 * that list atomically with respect to freeing it here.
2300 */
2301 int
2306 {
2309 xfs_ino_t first_ino;
2320 /*
2321 * Pull the on-disk inode from the AGI unlinked list.
2322 */
2326 }
2331 }
2340 /*
2341 * Bump the generation count so no one will be confused
2342 * by reincarnations of this inode.
2343 */
2349 }
2352 }
2354 /*
2355 * Reallocate the space for if_broot based on the number of records
2356 * being added or deleted as indicated in rec_diff. Move the records
2357 * and pointers in if_broot to fit the new size. When shrinking this
2358 * will eliminate holes between the records and pointers created by
2359 * the caller. When growing this will create holes to be filled in
2360 * by the caller.
2361 *
2362 * The caller must not request to add more records than would fit in
2363 * the on-disk inode root. If the if_broot is currently NULL, then
2364 * if we adding records one will be allocated. The caller must also
2365 * not request that the number of records go below zero, although
2366 * it can go to zero.
2367 *
2368 * ip -- the inode whose if_broot area is changing
2369 * ext_diff -- the change in the number of records, positive or negative,
2370 * requested for the if_broot array.
2371 */
2372 void
2377 {
2386 /*
2387 * Handle the degenerate case quietly.
2388 */
2391 }
2395 /*
2396 * If there wasn't any memory allocated before, just
2397 * allocate it now and get out.
2398 */
2402 KM_SLEEP);
2405 }
2407 /*
2408 * If there is already an existing if_broot, then we need
2409 * to realloc() it and shift the pointers to their new
2410 * location. The records don't change location because
2411 * they are kept butted up against the btree block header.
2412 */
2418 new_size,
2420 KM_SLEEP);
2430 }
2432 /*
2433 * rec_diff is less than 0. In this case, we are shrinking the
2434 * if_broot buffer. It must already exist. If we go to zero
2435 * records, just get rid of the root and clear the status bit.
2436 */
2443 else
2447 /*
2448 * First copy over the btree block header.
2449 */
2454 }
2456 /*
2457 * Only copy the records and pointers if there are any.
2458 */
2460 /*
2461 * First copy the records.
2462 */
2469 /*
2470 * Then copy the pointers.
2471 */
2477 }
2484 }
2487 /*
2488 * This is called when the amount of space needed for if_data
2489 * is increased or decreased. The change in size is indicated by
2490 * the number of bytes that need to be added or deleted in the
2491 * byte_diff parameter.
2492 *
2493 * If the amount of space needed has decreased below the size of the
2494 * inline buffer, then switch to using the inline buffer. Otherwise,
2495 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2496 * to what is needed.
2497 *
2498 * ip -- the inode whose if_data area is changing
2499 * byte_diff -- the change in the number of bytes, positive or negative,
2500 * requested for the if_data array.
2501 */
2502 void
2507 {
2514 }
2523 }
2527 /*
2528 * If the valid extents/data can fit in if_inline_ext/data,
2529 * copy them from the malloc'd vector and free it.
2530 */
2536 new_size);
2539 }
2542 /*
2543 * Stuck with malloc/realloc.
2544 * For inline data, the underlying buffer must be
2545 * a multiple of 4 bytes in size so that it can be
2546 * logged and stay on word boundaries. We enforce
2547 * that here.
2548 */
2554 /*
2555 * Only do the realloc if the underlying size
2556 * is really changing.
2557 */
2561 real_size,
2563 KM_SLEEP);
2564 }
2570 }
2571 }
2575 }
2580 /*
2581 * Map inode to disk block and offset.
2582 *
2583 * mp -- the mount point structure for the current file system
2584 * tp -- the current transaction
2585 * ino -- the inode number of the inode to be located
2586 * imap -- this structure is filled in with the information necessary
2587 * to retrieve the given inode from disk
2588 * flags -- flags to pass to xfs_dilocate indicating whether or not
2589 * lookups in the inode btree were OK or not
2590 */
2591 int
2595 xfs_ino_t ino,
2597 uint flags)
2598 {
2599 xfs_fsblock_t fsbno;
2609 }
2616 }
2618 void
2622 {
2629 }
2631 /*
2632 * If the format is local, then we can't have an extents
2633 * array so just look for an inline data array. If we're
2634 * not local then we may or may not have an extents list,
2635 * so check and free it up if we do.
2636 */
2644 }
2651 }
2658 }
2659 }
2661 /*
2662 * This is called free all the memory associated with an inode.
2663 * It must free the inode itself and any buffers allocated for
2664 * if_extents/if_data and if_broot. It must also free the lock
2665 * associated with the inode.
2666 */
2667 void
2670 {
2678 }
2684 #ifdef XFS_BMAP_TRACE
2686 #endif
2687 #ifdef XFS_BMBT_TRACE
2689 #endif
2690 #ifdef XFS_RW_TRACE
2692 #endif
2693 #ifdef XFS_ILOCK_TRACE
2695 #endif
2696 #ifdef XFS_DIR2_TRACE
2698 #endif
2700 /* XXXdpd should be able to assert this but shutdown
2701 * is leaving the AIL behind. */
2705 }
2707 }
2710 /*
2711 * Increment the pin count of the given buffer.
2712 * This value is protected by ipinlock spinlock in the mount structure.
2713 */
2714 void
2717 {
2721 }
2723 /*
2724 * Decrement the pin count of the given inode, and wake up
2725 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2726 * inode must have been previously pinned with a call to xfs_ipin().
2727 */
2728 void
2731 {
2737 /* make sync come back and flush this inode */
2743 }
2746 }
2747 }
2749 /*
2750 * This is called to wait for the given inode to be unpinned.
2751 * It will sleep until this happens. The caller must have the
2752 * inode locked in at least shared mode so that the buffer cannot
2753 * be subsequently pinned once someone is waiting for it to be
2754 * unpinned.
2755 */
2756 STATIC void
2759 {
2761 xfs_lsn_t lsn;
2767 }
2774 }
2776 /*
2777 * Give the log a push so we don't wait here too long.
2778 */
2782 }
2785 /*
2786 * xfs_iextents_copy()
2787 *
2788 * This is called to copy the REAL extents (as opposed to the delayed
2789 * allocation extents) from the inode into the given buffer. It
2790 * returns the number of bytes copied into the buffer.
2791 *
2792 * If there are no delayed allocation extents, then we can just
2793 * memcpy() the extents into the buffer. Otherwise, we need to
2794 * examine each extent in turn and skip those which are delayed.
2795 */
2796 int
2801 {
2805 #ifdef XFS_BMAP_TRACE
2807 #endif
2811 xfs_fsblock_t start_block;
2821 /*
2822 * There are some delayed allocation extents in the
2823 * inode, so copy the extents one at a time and skip
2824 * the delayed ones. There must be at least one
2825 * non-delayed extent.
2826 */
2833 /*
2834 * It's a delayed allocation extent, so skip it.
2835 */
2837 }
2839 /* Translate to on disk format */
2844 dest_ep++;
2845 copied++;
2846 }
2851 }
2853 /*
2854 * Each of the following cases stores data into the same region
2855 * of the on-disk inode, so only one of them can be valid at
2856 * any given time. While it is possible to have conflicting formats
2857 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2858 * in EXTENTS format, this can only happen when the fork has
2859 * changed formats after being modified but before being flushed.
2860 * In these cases, the format always takes precedence, because the
2861 * format indicates the current state of the fork.
2862 */
2863 /*ARGSUSED*/
2864 STATIC int
2871 {
2875 #ifdef XFS_TRANS_DEBUG
2877 #endif
2888 /*
2889 * This can happen if we gave up in iformat in an error path,
2890 * for the attribute fork.
2891 */
2895 }
2905 }
2911 }
2912 }
2926 whichfork);
2927 }
2936 XFS_BROOT_SIZE_ADJ));
2940 }
2947 }
2955 }
2961 }
2964 }
2966 /*
2967 * xfs_iflush() will write a modified inode's changes out to the
2968 * inode's on disk home. The caller must have the inode lock held
2969 * in at least shared mode and the inode flush semaphore must be
2970 * held as well. The inode lock will still be held upon return from
2971 * the call and the caller is free to unlock it.
2972 * The inode flush lock will be unlocked when the inode reaches the disk.
2973 * The flags indicate how the inode's buffer should be written out.
2974 */
2975 int
2978 uint flags)
2979 {
2985 /* REFERENCED */
3003 /*
3004 * If the inode isn't dirty, then just release the inode
3005 * flush lock and do nothing.
3006 */
3013 }
3015 /*
3016 * We can't flush the inode until it is unpinned, so
3017 * wait for it. We know noone new can pin it, because
3018 * we are holding the inode lock shared and you need
3019 * to hold it exclusively to pin the inode.
3020 */
3023 /*
3024 * This may have been unpinned because the filesystem is shutting
3025 * down forcibly. If that's the case we must not write this inode
3026 * to disk, because the log record didn't make it to disk!
3027 */
3034 }
3036 /*
3037 * Get the buffer containing the on-disk inode.
3038 */
3043 }
3045 /*
3046 * Decide how buffer will be flushed out. This is done before
3047 * the call to xfs_iflush_int because this field is zeroed by it.
3048 */
3050 /*
3051 * Flush out the inode buffer according to the directions
3052 * of the caller. In the cases where the caller has given
3053 * us a choice choose the non-delwri case. This is because
3054 * the inode is in the AIL and we need to get it out soon.
3055 */
3072 }
3090 }
3091 }
3093 /*
3094 * First flush out the inode that xfs_iflush was called with.
3095 */
3099 }
3101 /*
3102 * inode clustering:
3103 * see if other inodes can be gathered into this write
3104 */
3113 /*
3114 * Do an un-protected check to see if the inode is dirty and
3115 * is a candidate for flushing. These checks will be repeated
3116 * later after the appropriate locks are acquired.
3117 */
3124 }
3126 /*
3127 * Try to get locks. If any are unavailable,
3128 * then this inode cannot be flushed and is skipped.
3129 */
3131 /* get inode locks (just i_lock) */
3133 /* get inode flush lock */
3135 /* check if pinned */
3137 /* arriving here means that
3138 * this inode can be flushed.
3139 * first re-check that it's
3140 * dirty
3141 */
3146 clcount++;
3150 XFS_ILOCK_SHARED);
3152 }
3155 }
3158 }
3159 }
3161 }
3162 }
3168 }
3170 /*
3171 * If the buffer is pinned then push on the log so we won't
3172 * get stuck waiting in the write for too long.
3173 */
3176 }
3184 }
3187 corrupt_out:
3191 /*
3192 * Unlocks the flush lock
3193 */
3196 cluster_corrupt_out:
3197 /* Corruption detected in the clustering loop. Invalidate the
3198 * inode buffer and shut down the filesystem.
3199 */
3202 /*
3203 * Clean up the buffer. If it was B_DELWRI, just release it --
3204 * brelse can handle it with no problems. If not, shut down the
3205 * filesystem before releasing the buffer.
3206 */
3209 }
3214 /*
3215 * Just like incore_relse: if we have b_iodone functions,
3216 * mark the buffer as an error and call them. Otherwise
3217 * mark it as stale and brelse.
3218 */
3229 }
3230 }
3233 /*
3234 * Unlocks the flush lock
3235 */
3237 }
3240 STATIC int
3244 {
3248 #ifdef XFS_TRANS_DEBUG
3250 #endif
3262 /*
3263 * If the inode isn't dirty, then just release the inode
3264 * flush lock and do nothing.
3265 */
3270 }
3272 /* set *dip = inode's place in the buffer */
3275 /*
3276 * Clear i_update_core before copying out the data.
3277 * This is for coordination with our timestamp updates
3278 * that don't hold the inode lock. They will always
3279 * update the timestamps BEFORE setting i_update_core,
3280 * so if we clear i_update_core after they set it we
3281 * are guaranteed to see their updates to the timestamps.
3282 * I believe that this depends on strongly ordered memory
3283 * semantics, but we have that. We use the SYNCHRONIZE
3284 * macro to make sure that the compiler does not reorder
3285 * the i_update_core access below the data copy below.
3286 */
3290 /*
3291 * Make sure to get the latest atime from the Linux inode.
3292 */
3301 }
3308 }
3318 }
3329 }
3330 }
3333 XFS_RANDOM_IFLUSH_5)) {
3339 ip);
3341 }
3348 }
3349 /*
3350 * bump the flush iteration count, used to detect flushes which
3351 * postdate a log record during recovery.
3352 */
3356 /*
3357 * Copy the dirty parts of the inode into the on-disk
3358 * inode. We always copy out the core of the inode,
3359 * because if the inode is dirty at all the core must
3360 * be.
3361 */
3364 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3368 /*
3369 * If this is really an old format inode and the superblock version
3370 * has not been updated to support only new format inodes, then
3371 * convert back to the old inode format. If the superblock version
3372 * has been updated, then make the conversion permanent.
3373 */
3378 /*
3379 * Convert it back.
3380 */
3384 /*
3385 * The superblock version has already been bumped,
3386 * so just make the conversion to the new inode
3387 * format permanent.
3388 */
3397 }
3398 }
3402 }
3405 /*
3406 * The only error from xfs_iflush_fork is on the data fork.
3407 */
3409 }
3412 /*
3413 * We've recorded everything logged in the inode, so we'd
3414 * like to clear the ilf_fields bits so we don't log and
3415 * flush things unnecessarily. However, we can't stop
3416 * logging all this information until the data we've copied
3417 * into the disk buffer is written to disk. If we did we might
3418 * overwrite the copy of the inode in the log with all the
3419 * data after re-logging only part of it, and in the face of
3420 * a crash we wouldn't have all the data we need to recover.
3421 *
3422 * What we do is move the bits to the ili_last_fields field.
3423 * When logging the inode, these bits are moved back to the
3424 * ilf_fields field. In the xfs_iflush_done() routine we
3425 * clear ili_last_fields, since we know that the information
3426 * those bits represent is permanently on disk. As long as
3427 * the flush completes before the inode is logged again, then
3428 * both ilf_fields and ili_last_fields will be cleared.
3429 *
3430 * We can play with the ilf_fields bits here, because the inode
3431 * lock must be held exclusively in order to set bits there
3432 * and the flush lock protects the ili_last_fields bits.
3433 * Set ili_logged so the flush done
3434 * routine can tell whether or not to look in the AIL.
3435 * Also, store the current LSN of the inode so that we can tell
3436 * whether the item has moved in the AIL from xfs_iflush_done().
3437 * In order to read the lsn we need the AIL lock, because
3438 * it is a 64 bit value that cannot be read atomically.
3439 */
3450 /*
3451 * Attach the function xfs_iflush_done to the inode's
3452 * buffer. This will remove the inode from the AIL
3453 * and unlock the inode's flush lock when the inode is
3454 * completely written to disk.
3455 */
3462 /*
3463 * We're flushing an inode which is not in the AIL and has
3464 * not been logged but has i_update_core set. For this
3465 * case we can use a B_DELWRI flush and immediately drop
3466 * the inode flush lock because we can avoid the whole
3467 * AIL state thing. It's OK to drop the flush lock now,
3468 * because we've already locked the buffer and to do anything
3469 * you really need both.
3470 */
3475 }
3477 }
3481 corrupt_out:
3483 }
3486 /*
3487 * Flush all inactive inodes in mp.
3488 */
3489 void
3492 {
3496 again:
3503 /* Make sure we skip markers inserted by sync */
3507 }
3514 }
3520 out:
3522 }
3524 /*
3525 * xfs_iaccess: check accessibility of inode for mode.
3526 */
3527 int
3530 mode_t mode,
3532 {
3546 }
3548 /*
3549 * If there's an Access Control List it's used instead of
3550 * the mode bits.
3551 */
3559 }
3561 /*
3562 * If the DACs are ok we don't need any capability check.
3563 */
3566 /*
3567 * Read/write DACs are always overridable.
3568 * Executable DACs are overridable if at least one exec bit is set.
3569 */
3579 #ifdef NOISE
3583 }
3585 }
3587 /*
3588 * xfs_iroundup: round up argument to next power of two
3589 */
3590 uint
3592 uint v)
3593 {
3595 uint m;
3608 }
3611 }
3613 #ifdef XFS_ILOCK_TRACE
3616 void
3618 {
3627 }
3628 #endif
3630 /*
3631 * Return a pointer to the extent record at file index idx.
3632 */
3633 xfs_bmbt_rec_t *
3637 {
3652 }
3653 }
3655 /*
3656 * Insert new item(s) into the extent records for incore inode
3657 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3658 */
3659 void
3665 {
3674 }
3675 }
3677 /*
3678 * This is called when the amount of space required for incore file
3679 * extents needs to be increased. The ext_diff parameter stores the
3680 * number of new extents being added and the idx parameter contains
3681 * the extent index where the new extents will be added. If the new
3682 * extents are being appended, then we just need to (re)allocate and
3683 * initialize the space. Otherwise, if the new extents are being
3684 * inserted into the middle of the existing entries, a bit more work
3685 * is required to make room for the new extents to be inserted. The
3686 * caller is responsible for filling in the new extent entries upon
3687 * return.
3688 */
3689 void
3694 {
3703 /*
3704 * If the new number of extents (nextents + ext_diff)
3705 * fits inside the inode, then continue to use the inline
3706 * extent buffer.
3707 */
3714 }
3718 }
3719 /*
3720 * Otherwise use a linear (direct) extent list.
3721 * If the extents are currently inside the inode,
3722 * xfs_iext_realloc_direct will switch us from
3723 * inline to direct extent allocation mode.
3724 */
3732 }
3733 }
3734 /* Indirection array */
3747 }
3748 /* Extents fit in target extent page */
3756 }
3759 }
3760 /* Insert a new extent page */
3764 }
3765 /*
3766 * If extent(s) are being appended to the last page in
3767 * the indirection array and the new extent(s) don't fit
3768 * in the page, then erp is NULL and erp_idx is set to
3769 * the next index needed in the indirection array.
3770 */
3779 erp_idx++;
3780 }
3781 }
3782 }
3783 }
3785 }
3787 /*
3788 * This is called when incore extents are being added to the indirection
3789 * array and the new extents do not fit in the target extent list. The
3790 * erp_idx parameter contains the irec index for the target extent list
3791 * in the indirection array, and the idx parameter contains the extent
3792 * index within the list. The number of extents being added is stored
3793 * in the count parameter.
3794 *
3795 * |-------| |-------|
3796 * | | | | idx - number of extents before idx
3797 * | idx | | count |
3798 * | | | | count - number of extents being inserted at idx
3799 * |-------| |-------|
3800 * | count | | nex2 | nex2 - number of extents after idx + count
3801 * |-------| |-------|
3802 */
3803 void
3809 {
3823 /*
3824 * Save second part of target extent list
3825 * (all extents past */
3833 }
3835 /*
3836 * Add the new extents to the end of the target
3837 * list, then allocate new irec record(s) and
3838 * extent buffer(s) as needed to store the rest
3839 * of the new extents.
3840 */
3847 }
3849 erp_idx++;
3855 }
3857 /* Add nex2 extents back to indirection array */
3859 xfs_extnum_t ext_avail;
3865 /*
3866 * If nex2 extents fit in the current page, append
3867 * nex2_ep after the new extents.
3868 */
3871 }
3872 /*
3873 * Otherwise, check if space is available in the
3874 * next page.
3875 */
3879 erp_idx++;
3880 erp++;
3881 /* Create a hole for nex2 extents */
3884 }
3885 /*
3886 * Final choice, create a new extent page for
3887 * nex2 extents.
3888 */
3890 erp_idx++;
3892 }
3897 }
3898 }
3900 /*
3901 * This is called when the amount of space required for incore file
3902 * extents needs to be decreased. The ext_diff parameter stores the
3903 * number of extents to be removed and the idx parameter contains
3904 * the extent index where the extents will be removed from.
3905 *
3906 * If the amount of space needed has decreased below the linear
3907 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3908 * extent array. Otherwise, use kmem_realloc() to adjust the
3909 * size to what is needed.
3910 */
3911 void
3916 {
3932 }
3934 }
3936 /*
3937 * This removes ext_diff extents from the inline buffer, beginning
3938 * at extent index idx.
3939 */
3940 void
3945 {
3964 }
3965 }
3967 /*
3968 * This removes ext_diff extents from a linear (direct) extent list,
3969 * beginning at extent index idx. If the extents are being removed
3970 * from the end of the list (ie. truncate) then we just need to re-
3971 * allocate the list to remove the extra space. Otherwise, if the
3972 * extents are being removed from the middle of the existing extent
3973 * entries, then we first need to move the extent records beginning
3974 * at idx + ext_diff up in the list to overwrite the records being
3975 * removed, then remove the extra space via kmem_realloc.
3976 */
3977 void
3982 {
3994 }
3995 /* Move extents up in the list (if needed) */
4001 }
4004 /*
4005 * Reallocate the direct extent list. If the extents
4006 * will fit inside the inode then xfs_iext_realloc_direct
4007 * will switch from direct to inline extent allocation
4008 * mode for us.
4009 */
4012 }
4014 /*
4015 * This is called when incore extents are being removed from the
4016 * indirection array and the extents being removed span multiple extent
4017 * buffers. The idx parameter contains the file extent index where we
4018 * want to begin removing extents, and the count parameter contains
4019 * how many extents need to be removed.
4020 *
4021 * |-------| |-------|
4022 * | nex1 | | | nex1 - number of extents before idx
4023 * |-------| | count |
4024 * | | | | count - number of extents being removed at idx
4025 * | count | |-------|
4026 * | | | nex2 | nex2 - number of extents after idx + count
4027 * |-------| |-------|
4028 */
4029 void
4034 {
4053 /*
4054 * Check for deletion of entire list;
4055 * xfs_iext_irec_remove() updates extent offsets.
4056 */
4063 XFS_IEXT_BUFSZ);
4069 }
4070 }
4071 /* Move extents up (if needed) */
4076 }
4077 /* Zero out rest of page */
4080 /* Update remaining counters */
4085 erp_idx++;
4086 erp++;
4087 }
4090 }
4092 /*
4093 * Create, destroy, or resize a linear (direct) block of extents.
4094 */
4095 void
4099 {
4108 /* Free extent records */
4111 }
4112 /* Resize direct extent list and zero any new bytes */
4114 /* Check if extents will fit inside the inode */
4120 }
4123 }
4127 rnew_size,
4129 KM_SLEEP);
4130 }
4135 }
4136 }
4137 /*
4138 * Switch from the inline extent buffer to a direct
4139 * extent list. Be sure to include the inline extent
4140 * bytes in new_size.
4141 */
4146 }
4148 }
4151 }
4153 /*
4154 * Switch from linear (direct) extent records to inline buffer.
4155 */
4156 void
4160 {
4163 /*
4164 * The inline buffer was zeroed when we switched
4165 * from inline to direct extent allocation mode,
4166 * so we don't need to clear it here.
4167 */
4173 }
4175 /*
4176 * Switch from inline buffer to linear (direct) extent records.
4177 * new_size should already be rounded up to the next power of 2
4178 * by the caller (when appropriate), so use new_size as it is.
4179 * However, since new_size may be rounded up, we can't update
4180 * if_bytes here. It is the caller's responsibility to update
4181 * if_bytes upon return.
4182 */
4183 void
4187 {
4196 }
4198 }
4200 /*
4201 * Resize an extent indirection array to new_size bytes.
4202 */
4203 void
4207 {
4222 }
4223 }
4225 /*
4226 * Switch from indirection array to linear (direct) extent allocations.
4227 */
4228 void
4231 {
4251 }
4252 }
4254 /*
4255 * Free incore file extents.
4256 */
4257 void
4260 {
4268 }
4275 }
4279 }
4281 /*
4282 * Return a pointer to the extent record for file system block bno.
4283 */
4289 {
4304 }
4307 /* Find target extent list */
4315 }
4316 /* Binary search extent records */
4327 /* Convert back to file-based extent index */
4330 }
4333 }
4334 }
4335 /* Convert back to file-based extent index */
4338 }
4344 }
4345 }
4348 }
4350 /*
4351 * Return a pointer to the indirection array entry containing the
4352 * extent record for filesystem block bno. Store the index of the
4353 * target irec in *erp_idxp.
4354 */
4360 {
4384 }
4385 }
4388 }
4390 /*
4391 * Return a pointer to the indirection array entry containing the
4392 * extent record at file extent index *idxp. Store the index of the
4393 * target irec in *erp_idxp and store the page index of the target
4394 * extent record in *idxp.
4395 */
4396 xfs_ext_irec_t *
4402 {
4419 /* Binary search extent irec's */
4435 erp_idx++;
4441 }
4442 }
4446 }
4448 /*
4449 * Allocate and initialize an indirection array once the space needed
4450 * for incore extents increases above XFS_IEXT_BUFSZ.
4451 */
4452 void
4455 {
4473 }
4484 }
4486 /*
4487 * Allocate and initialize a new entry in the indirection array.
4488 */
4489 xfs_ext_irec_t *
4493 {
4501 /* Resize indirection array */
4504 /*
4505 * Move records down in the array so the
4506 * new page can use erp_idx.
4507 */
4511 }
4514 /* Initialize new extent record */
4524 }
4526 /*
4527 * Remove a record from the indirection array.
4528 */
4529 void
4533 {
4545 }
4546 /* Compact extent records */
4550 }
4551 /*
4552 * Manually free the last extent record from the indirection
4553 * array. A call to xfs_iext_realloc_indirect() with a size
4554 * of zero would result in a call to xfs_iext_destroy() which
4555 * would in turn call this function again, creating a nasty
4556 * infinite loop.
4557 */
4564 }
4566 }
4568 /*
4569 * This is called to clean up large amounts of unused memory allocated
4570 * by the indirection array. Before compacting anything though, verify
4571 * that the indirection array is still needed and switch back to the
4572 * linear extent list (or even the inline buffer) if possible. The
4573 * compaction policy is as follows:
4574 *
4575 * Full Compaction: Extents fit into a single page (or inline buffer)
4576 * Full Compaction: Extents occupy less than 10% of allocated space
4577 * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
4578 * No Compaction: Extents occupy at least 50% of allocated space
4579 */
4580 void
4583 {
4602 }
4603 }
4605 /*
4606 * Combine extents from neighboring extent pages.
4607 */
4608 void
4611 {
4627 /*
4628 * Free page before removing extent record
4629 * so er_extoffs don't get modified in
4630 * xfs_iext_irec_remove.
4631 */
4637 erp_idx++;
4638 }
4639 }
4640 }
4642 /*
4643 * Fully compact the extent records managed by the indirection array.
4644 */
4645 void
4648 {
4668 /* Remove next page */
4670 /*
4671 * Free page before removing extent record
4672 * so er_extoffs don't get modified in
4673 * xfs_iext_irec_remove.
4674 */
4681 /* Update next page */
4683 /* Move rest of page up to become next new page */
4690 }
4692 erp_idx++;
4695 else
4697 }
4701 }
4702 }
4704 /*
4705 * This is called to update the er_extoff field in the indirection
4706 * array when extents have been added or removed from one of the
4707 * extent lists. erp_idx contains the irec index to begin updating
4708 * at and ext_diff contains the number of extents that were added
4709 * or removed.
4710 */
4711 void
4716 {
4724 }
4725 }