| blob | 5fa0adb7e1737ad0ca4451409f412d73e1b5ce2f |
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
70 #ifdef DEBUG
71 /*
72 * Make sure that the extents in the given memory buffer
73 * are valid.
74 */
75 STATIC void
80 xfs_exntfmt_t fmt)
81 {
83 xfs_bmbt_irec_t irec;
84 xfs_bmbt_rec_t rec;
93 else
97 }
98 }
100 #define xfs_validate_extents(ifp, nrecs, disk, fmt)
103 /*
104 * Check that none of the inode's in the buffer have a next
105 * unlinked field of 0.
106 */
107 #if defined(DEBUG)
108 void
112 {
124 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
125 bp);
127 }
128 }
129 }
130 #endif
132 /*
133 * This routine is called to map an inode number within a file
134 * system to the buffer containing the on-disk version of the
135 * inode. It returns a pointer to the buffer containing the
136 * on-disk inode in the bpp parameter, and in the dip parameter
137 * it returns a pointer to the on-disk inode within that buffer.
138 *
139 * If a non-zero error is returned, then the contents of bpp and
140 * dipp are undefined.
141 *
142 * Use xfs_imap() to determine the size and location of the
143 * buffer to read from disk.
144 */
145 STATIC int
149 xfs_ino_t ino,
153 {
155 xfs_imap_t imap;
160 /*
161 * Call the space management code to find the location of the
162 * inode on disk.
163 */
168 "xfs_inotobp: xfs_imap() returned an "
171 }
173 /*
174 * If the inode number maps to a block outside the bounds of the
175 * file system then return NULL rather than calling read_buf
176 * and panicing when we get an error from the driver.
177 */
181 "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
186 }
188 /*
189 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
190 * default to just a read_buf() call.
191 */
197 "xfs_inotobp: xfs_trans_read_buf() returned an "
200 }
202 di_ok =
206 XFS_RANDOM_ITOBP_INOTOBP))) {
210 "xfs_inotobp: XFS_TEST_ERROR() returned an "
213 }
217 /*
218 * Set *dipp to point to the on-disk inode in the buffer.
219 */
224 }
227 /*
228 * This routine is called to map an inode to the buffer containing
229 * the on-disk version of the inode. It returns a pointer to the
230 * buffer containing the on-disk inode in the bpp parameter, and in
231 * the dip parameter it returns a pointer to the on-disk inode within
232 * that buffer.
233 *
234 * If a non-zero error is returned, then the contents of bpp and
235 * dipp are undefined.
236 *
237 * If the inode is new and has not yet been initialized, use xfs_imap()
238 * to determine the size and location of the buffer to read from disk.
239 * If the inode has already been mapped to its buffer and read in once,
240 * then use the mapping information stored in the inode rather than
241 * calling xfs_imap(). This allows us to avoid the overhead of looking
242 * at the inode btree for small block file systems (see xfs_dilocate()).
243 * We can tell whether the inode has been mapped in before by comparing
244 * its disk block address to 0. Only uninitialized inodes will have
245 * 0 for the disk block address.
246 */
247 int
254 xfs_daddr_t bno,
255 uint imap_flags)
256 {
257 xfs_imap_t imap;
264 /*
265 * Call the space management code to find the location of the
266 * inode on disk.
267 */
273 /*
274 * If the inode number maps to a block outside the bounds
275 * of the file system then return NULL rather than calling
276 * read_buf and panicing when we get an error from the
277 * driver.
278 */
281 #ifdef DEBUG
283 "(imap.im_blkno (0x%llx) "
284 "+ imap.im_len (0x%llx)) > "
285 " XFS_FSB_TO_BB(mp, "
292 }
294 /*
295 * Fill in the fields in the inode that will be used to
296 * map the inode to its buffer from now on.
297 */
302 /*
303 * We've already mapped the inode once, so just use the
304 * mapping that we saved the first time.
305 */
309 }
312 /*
313 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
314 * default to just a read_buf() call.
315 */
319 #ifdef DEBUG
321 "xfs_trans_read_buf() returned error %d, "
327 }
329 /*
330 * Validate the magic number and version of every inode in the buffer
331 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
332 * No validation is done here in userspace (xfs_repair).
333 */
334 #if !defined(__KERNEL__)
336 #elif defined(DEBUG)
341 #endif
352 XFS_RANDOM_ITOBP_INOTOBP))) {
353 #ifdef DEBUG
356 "Device %s - bad inode magic/vsn "
361 #endif
366 }
367 }
371 /*
372 * Mark the buffer as an inode buffer now that it looks good
373 */
376 /*
377 * Set *dipp to point to the on-disk inode in the buffer.
378 */
382 }
384 /*
385 * Move inode type and inode format specific information from the
386 * on-disk inode to the in-core inode. For fifos, devs, and sockets
387 * this means set if_rdev to the proper value. For files, directories,
388 * and symlinks this means to bring in the in-line data or extent
389 * pointers. For a file in B-tree format, only the root is immediately
390 * brought in-core. The rest will be in-lined in if_extents when it
391 * is first referenced (see xfs_iread_extents()).
392 */
393 STATIC int
397 {
401 xfs_fsize_t di_size;
420 }
422 if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) {
430 }
441 }
451 /*
452 * no local regular files yet
453 */
456 "corrupt inode %Lu "
460 XFS_ERRLEVEL_LOW,
463 }
468 "corrupt inode %Lu "
473 XFS_ERRLEVEL_LOW,
476 }
491 }
497 }
500 }
522 }
527 }
529 }
531 /*
532 * The file is in-lined in the on-disk inode.
533 * If it fits into if_inline_data, then copy
534 * it there, otherwise allocate a buffer for it
535 * and copy the data there. Either way, set
536 * if_data to point at the data.
537 * If we allocate a buffer for the data, make
538 * sure that its size is a multiple of 4 and
539 * record the real size in i_real_bytes.
540 */
541 STATIC int
547 {
551 /*
552 * If the size is unreasonable, then something
553 * is wrong and we just bail out rather than crash in
554 * kmem_alloc() or memcpy() below.
555 */
558 "corrupt inode %Lu "
565 }
575 }
583 }
585 /*
586 * The file consists of a set of extents all
587 * of which fit into the on-disk inode.
588 * If there are few enough extents to fit into
589 * the if_inline_ext, then copy them there.
590 * Otherwise allocate a buffer for them and copy
591 * them into it. Either way, set if_extents
592 * to point at the extents.
593 */
594 STATIC int
599 {
610 /*
611 * If the number of extents is unreasonable, then something
612 * is wrong and we just bail out rather than crash in
613 * kmem_alloc() or memcpy() below.
614 */
622 }
629 else
639 ARCH_CONVERT);
641 ARCH_CONVERT);
642 }
644 whichfork);
650 XFS_ERRLEVEL_LOW,
653 }
654 }
657 }
659 /*
660 * The file has too many extents to fit into
661 * the inode, so they are in B-tree format.
662 * Allocate a buffer for the root of the B-tree
663 * and copy the root into it. The i_extents
664 * field will remain NULL until all of the
665 * extents are read in (when they are needed).
666 */
667 STATIC int
672 {
675 /* REFERENCED */
684 /*
685 * blow out if -- fork has less extents than can fit in
686 * fork (fork shouldn't be a btree format), root btree
687 * block has more records than can fit into the fork,
688 * or the number of extents is greater than the number of
689 * blocks.
690 */
701 }
706 /*
707 * Copy and convert from the on-disk structure
708 * to the in-memory structure.
709 */
716 }
718 /*
719 * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
720 * and native format
721 *
722 * buf = on-disk representation
723 * dip = native representation
724 * dir = direction - +ve -> disk to native
725 * -ve -> native to disk
726 */
727 void
729 xfs_caddr_t buf,
732 {
755 }
782 }
784 STATIC uint
786 __uint16_t di_flags)
787 {
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 {
867 /*
868 * Get pointer's to the on-disk inode and the buffer containing it.
869 * If the inode number refers to a block outside the file system
870 * then xfs_itobp() will return NULL. In this case we should
871 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
872 * know that this is a new incore inode.
873 */
878 }
880 /*
881 * Initialize inode's trace buffers.
882 * Do this before xfs_iformat in case it adds entries.
883 */
884 #ifdef XFS_BMAP_TRACE
886 #endif
887 #ifdef XFS_BMBT_TRACE
889 #endif
890 #ifdef XFS_RW_TRACE
892 #endif
893 #ifdef XFS_ILOCK_TRACE
895 #endif
896 #ifdef XFS_DIR2_TRACE
898 #endif
900 /*
901 * If we got something that isn't an inode it means someone
902 * (nfs or dmi) has a stale handle.
903 */
907 #ifdef DEBUG
909 "dip->di_core.di_magic (0x%x) != "
912 XFS_DINODE_MAGIC);
915 }
917 /*
918 * If the on-disk inode is already linked to a directory
919 * entry, copy all of the inode into the in-core inode.
920 * xfs_iformat() handles copying in the inode format
921 * specific information.
922 * Otherwise, just get the truly permanent information.
923 */
931 #ifdef DEBUG
934 error);
937 }
943 /*
944 * Make sure to pull in the mode here as well in
945 * case the inode is released without being used.
946 * This ensures that xfs_inactive() will see that
947 * the inode is already free and not try to mess
948 * with the uninitialized part of it.
949 */
951 /*
952 * Initialize the per-fork minima and maxima for a new
953 * inode here. xfs_iformat will do it for old inodes.
954 */
957 }
961 /*
962 * The inode format changed when we moved the link count and
963 * made it 32 bits long. If this is an old format inode,
964 * convert it in memory to look like a new one. If it gets
965 * flushed to disk we will convert back before flushing or
966 * logging it. We zero out the new projid field and the old link
967 * count field. We'll handle clearing the pad field (the remains
968 * of the old uuid field) when we actually convert the inode to
969 * the new format. We don't change the version number so that we
970 * can distinguish this from a real new format inode.
971 */
976 }
980 /*
981 * Mark the buffer containing the inode as something to keep
982 * around for a while. This helps to keep recently accessed
983 * meta-data in-core longer.
984 */
987 /*
988 * Use xfs_trans_brelse() to release the buffer containing the
989 * on-disk inode, because it was acquired with xfs_trans_read_buf()
990 * in xfs_itobp() above. If tp is NULL, this is just a normal
991 * brelse(). If we're within a transaction, then xfs_trans_brelse()
992 * will only release the buffer if it is not dirty within the
993 * transaction. It will be OK to release the buffer in this case,
994 * because inodes on disk are never destroyed and we will be
995 * locking the new in-core inode before putting it in the hash
996 * table where other processes can find it. Thus we don't have
997 * to worry about the inode being changed just because we released
998 * the buffer.
999 */
1003 }
1005 /*
1006 * Read in extents from a btree-format inode.
1007 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1008 */
1009 int
1014 {
1017 xfs_extnum_t nextents;
1024 }
1029 /*
1030 * We know that the size is valid (it's checked in iformat_btree)
1031 */
1041 }
1044 }
1046 /*
1047 * Allocate an inode on disk and return a copy of its in-core version.
1048 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1049 * appropriately within the inode. The uid and gid for the inode are
1050 * set according to the contents of the given cred structure.
1051 *
1052 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1053 * has a free inode available, call xfs_iget()
1054 * to obtain the in-core version of the allocated inode. Finally,
1055 * fill in the inode and log its initial contents. In this case,
1056 * ialloc_context would be set to NULL and call_again set to false.
1057 *
1058 * If xfs_dialloc() does not have an available inode,
1059 * it will replenish its supply by doing an allocation. Since we can
1060 * only do one allocation within a transaction without deadlocks, we
1061 * must commit the current transaction before returning the inode itself.
1062 * In this case, therefore, we will set call_again to true and return.
1063 * The caller should then commit the current transaction, start a new
1064 * transaction, and call xfs_ialloc() again to actually get the inode.
1065 *
1066 * To ensure that some other process does not grab the inode that
1067 * was allocated during the first call to xfs_ialloc(), this routine
1068 * also returns the [locked] bp pointing to the head of the freelist
1069 * as ialloc_context. The caller should hold this buffer across
1070 * the commit and pass it back into this routine on the second call.
1071 */
1072 int
1076 mode_t mode,
1077 xfs_nlink_t nlink,
1078 xfs_dev_t rdev,
1080 xfs_prid_t prid,
1085 {
1086 xfs_ino_t ino;
1089 uint flags;
1092 /*
1093 * Call the space management code to pick
1094 * the on-disk inode to be allocated.
1095 */
1100 }
1104 }
1107 /*
1108 * Get the in-core inode with the lock held exclusively.
1109 * This is because we're setting fields here we need
1110 * to prevent others from looking at until we're done.
1111 */
1116 }
1129 /*
1130 * If the superblock version is up to where we support new format
1131 * inodes and this is currently an old format inode, then change
1132 * the inode version number now. This way we only do the conversion
1133 * here rather than here and in the flush/logging code.
1134 */
1138 /*
1139 * We've already zeroed the old link count, the projid field,
1140 * and the pad field.
1141 */
1142 }
1144 /*
1145 * Project ids won't be stored on disk if we are using a version 1 inode.
1146 */
1154 }
1155 }
1157 /*
1158 * If the group ID of the new file does not match the effective group
1159 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1160 * (and only if the irix_sgid_inherit compatibility variable is set).
1161 */
1166 }
1172 /*
1173 * di_gen will have been taken care of in xfs_iread.
1174 */
1201 }
1206 }
1210 }
1211 }
1213 xfs_inherit_noatime)
1216 xfs_inherit_nodump)
1219 xfs_inherit_sync)
1222 xfs_inherit_nosymlinks)
1227 xfs_inherit_nodefrag)
1230 }
1231 /* FALLTHROUGH */
1240 }
1241 /*
1242 * Attribute fork settings for new inode.
1243 */
1247 /*
1248 * Log the new values stuffed into the inode.
1249 */
1252 /* now that we have an i_mode we can setup inode ops and unlock */
1257 }
1259 /*
1260 * Check to make sure that there are no blocks allocated to the
1261 * file beyond the size of the file. We don't check this for
1262 * files with fixed size extents or real time extents, but we
1263 * at least do it for regular files.
1264 */
1265 #ifdef DEBUG
1266 void
1270 xfs_fsize_t isize)
1271 {
1272 xfs_fileoff_t map_first;
1284 /*
1285 * The filesystem could be shutting down, so bmapi may return
1286 * an error.
1287 */
1291 map_first),
1297 }
1300 /*
1301 * Calculate the last possible buffered byte in a file. This must
1302 * include data that was buffered beyond the EOF by the write code.
1303 * This also needs to deal with overflowing the xfs_fsize_t type
1304 * which can happen for sizes near the limit.
1305 *
1306 * We also need to take into account any blocks beyond the EOF. It
1307 * may be the case that they were buffered by a write which failed.
1308 * In that case the pages will still be in memory, but the inode size
1309 * will never have been updated.
1310 */
1311 xfs_fsize_t
1314 {
1316 xfs_fsize_t last_byte;
1317 xfs_fileoff_t last_block;
1318 xfs_fileoff_t size_last_block;
1324 /*
1325 * Only check for blocks beyond the EOF if the extents have
1326 * been read in. This eliminates the need for the inode lock,
1327 * and it also saves us from looking when it really isn't
1328 * necessary.
1329 */
1332 XFS_DATA_FORK);
1335 }
1338 }
1345 }
1349 }
1351 }
1353 #if defined(XFS_RW_TRACE)
1354 STATIC void
1359 xfs_fsize_t new_size,
1360 xfs_off_t toss_start,
1361 xfs_off_t toss_finish)
1362 {
1365 }
1384 }
1385 #else
1386 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1387 #endif
1389 /*
1390 * Start the truncation of the file to new_size. The new size
1391 * must be smaller than the current size. This routine will
1392 * clear the buffer and page caches of file data in the removed
1393 * range, and xfs_itruncate_finish() will remove the underlying
1394 * disk blocks.
1395 *
1396 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1397 * must NOT have the inode lock held at all. This is because we're
1398 * calling into the buffer/page cache code and we can't hold the
1399 * inode lock when we do so.
1400 *
1401 * We need to wait for any direct I/Os in flight to complete before we
1402 * proceed with the truncate. This is needed to prevent the extents
1403 * being read or written by the direct I/Os from being removed while the
1404 * I/O is in flight as there is no other method of synchronising
1405 * direct I/O with the truncate operation. Also, because we hold
1406 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1407 * started until the truncate completes and drops the lock. Essentially,
1408 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1409 * between direct I/Os and the truncate operation.
1410 *
1411 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1412 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1413 * in the case that the caller is locking things out of order and
1414 * may not be able to call xfs_itruncate_finish() with the inode lock
1415 * held without dropping the I/O lock. If the caller must drop the
1416 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1417 * must be called again with all the same restrictions as the initial
1418 * call.
1419 */
1420 void
1423 uint flags,
1424 xfs_fsize_t new_size)
1425 {
1426 xfs_fsize_t last_byte;
1427 xfs_off_t toss_start;
1441 /*
1442 * Call toss_pages or flushinval_pages to get rid of pages
1443 * overlapping the region being removed. We have to use
1444 * the less efficient flushinval_pages in the case that the
1445 * caller may not be able to finish the truncate without
1446 * dropping the inode's I/O lock. Make sure
1447 * to catch any pages brought in by buffers overlapping
1448 * the EOF by searching out beyond the isize by our
1449 * block size. We round new_size up to a block boundary
1450 * so that we don't toss things on the same block as
1451 * new_size but before it.
1452 *
1453 * Before calling toss_page or flushinval_pages, make sure to
1454 * call remapf() over the same region if the file is mapped.
1455 * This frees up mapped file references to the pages in the
1456 * given range and for the flushinval_pages case it ensures
1457 * that we get the latest mapped changes flushed out.
1458 */
1462 /*
1463 * The place to start tossing is beyond our maximum
1464 * file size, so there is no way that the data extended
1465 * out there.
1466 */
1468 }
1471 last_byte);
1477 }
1478 }
1480 #ifdef DEBUG
1483 }
1484 #endif
1485 }
1487 /*
1488 * Shrink the file to the given new_size. The new
1489 * size must be smaller than the current size.
1490 * This will free up the underlying blocks
1491 * in the removed range after a call to xfs_itruncate_start()
1492 * or xfs_atruncate_start().
1493 *
1494 * The transaction passed to this routine must have made
1495 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1496 * This routine may commit the given transaction and
1497 * start new ones, so make sure everything involved in
1498 * the transaction is tidy before calling here.
1499 * Some transaction will be returned to the caller to be
1500 * committed. The incoming transaction must already include
1501 * the inode, and both inode locks must be held exclusively.
1502 * The inode must also be "held" within the transaction. On
1503 * return the inode will be "held" within the returned transaction.
1504 * This routine does NOT require any disk space to be reserved
1505 * for it within the transaction.
1506 *
1507 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1508 * and it indicates the fork which is to be truncated. For the
1509 * attribute fork we only support truncation to size 0.
1510 *
1511 * We use the sync parameter to indicate whether or not the first
1512 * transaction we perform might have to be synchronous. For the attr fork,
1513 * it needs to be so if the unlink of the inode is not yet known to be
1514 * permanent in the log. This keeps us from freeing and reusing the
1515 * blocks of the attribute fork before the unlink of the inode becomes
1516 * permanent.
1517 *
1518 * For the data fork, we normally have to run synchronously if we're
1519 * being called out of the inactive path or we're being called
1520 * out of the create path where we're truncating an existing file.
1521 * Either way, the truncate needs to be sync so blocks don't reappear
1522 * in the file with altered data in case of a crash. wsync filesystems
1523 * can run the first case async because anything that shrinks the inode
1524 * has to run sync so by the time we're called here from inactive, the
1525 * inode size is permanently set to 0.
1526 *
1527 * Calls from the truncate path always need to be sync unless we're
1528 * in a wsync filesystem and the file has already been unlinked.
1529 *
1530 * The caller is responsible for correctly setting the sync parameter.
1531 * It gets too hard for us to guess here which path we're being called
1532 * out of just based on inode state.
1533 */
1534 int
1538 xfs_fsize_t new_size,
1541 {
1542 xfs_fsblock_t first_block;
1543 xfs_fileoff_t first_unmap_block;
1544 xfs_fileoff_t last_block;
1550 xfs_bmap_free_t free_list;
1567 /*
1568 * We only support truncating the entire attribute fork.
1569 */
1572 }
1575 /*
1576 * The first thing we do is set the size to new_size permanently
1577 * on disk. This way we don't have to worry about anyone ever
1578 * being able to look at the data being freed even in the face
1579 * of a crash. What we're getting around here is the case where
1580 * we free a block, it is allocated to another file, it is written
1581 * to, and then we crash. If the new data gets written to the
1582 * file but the log buffers containing the free and reallocation
1583 * don't, then we'd end up with garbage in the blocks being freed.
1584 * As long as we make the new_size permanent before actually
1585 * freeing any blocks it doesn't matter if they get writtten to.
1586 *
1587 * The callers must signal into us whether or not the size
1588 * setting here must be synchronous. There are a few cases
1589 * where it doesn't have to be synchronous. Those cases
1590 * occur if the file is unlinked and we know the unlink is
1591 * permanent or if the blocks being truncated are guaranteed
1592 * to be beyond the inode eof (regardless of the link count)
1593 * and the eof value is permanent. Both of these cases occur
1594 * only on wsync-mounted filesystems. In those cases, we're
1595 * guaranteed that no user will ever see the data in the blocks
1596 * that are being truncated so the truncate can run async.
1597 * In the free beyond eof case, the file may wind up with
1598 * more blocks allocated to it than it needs if we crash
1599 * and that won't get fixed until the next time the file
1600 * is re-opened and closed but that's ok as that shouldn't
1601 * be too many blocks.
1602 *
1603 * However, we can't just make all wsync xactions run async
1604 * because there's one call out of the create path that needs
1605 * to run sync where it's truncating an existing file to size
1606 * 0 whose size is > 0.
1607 *
1608 * It's probably possible to come up with a test in this
1609 * routine that would correctly distinguish all the above
1610 * cases from the values of the function parameters and the
1611 * inode state but for sanity's sake, I've decided to let the
1612 * layers above just tell us. It's simpler to correctly figure
1613 * out in the layer above exactly under what conditions we
1614 * can run async and I think it's easier for others read and
1615 * follow the logic in case something has to be changed.
1616 * cscope is your friend -- rcc.
1617 *
1618 * The attribute fork is much simpler.
1619 *
1620 * For the attribute fork we allow the caller to tell us whether
1621 * the unlink of the inode that led to this call is yet permanent
1622 * in the on disk log. If it is not and we will be freeing extents
1623 * in this inode then we make the first transaction synchronous
1624 * to make sure that the unlink is permanent by the time we free
1625 * the blocks.
1626 */
1631 }
1636 }
1642 /*
1643 * Since it is possible for space to become allocated beyond
1644 * the end of the file (in a crash where the space is allocated
1645 * but the inode size is not yet updated), simply remove any
1646 * blocks which show up between the new EOF and the maximum
1647 * possible file size. If the first block to be removed is
1648 * beyond the maximum file size (ie it is the same as last_block),
1649 * then there is nothing to do.
1650 */
1658 }
1660 /*
1661 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1662 * will tell us whether it freed the entire range or
1663 * not. If this is a synchronous mount (wsync),
1664 * then we can tell bunmapi to keep all the
1665 * transactions asynchronous since the unlink
1666 * transaction that made this inode inactive has
1667 * already hit the disk. There's no danger of
1668 * the freed blocks being reused, there being a
1669 * crash, and the reused blocks suddenly reappearing
1670 * in this file with garbage in them once recovery
1671 * runs.
1672 */
1678 XFS_ITRUNC_MAX_EXTENTS,
1682 /*
1683 * If the bunmapi call encounters an error,
1684 * return to the caller where the transaction
1685 * can be properly aborted. We just need to
1686 * make sure we're not holding any resources
1687 * that we were not when we came in.
1688 */
1691 }
1693 /*
1694 * Duplicate the transaction that has the permanent
1695 * reservation and commit the old transaction.
1696 */
1701 /*
1702 * If the bmap finish call encounters an error,
1703 * return to the caller where the transaction
1704 * can be properly aborted. We just need to
1705 * make sure we're not holding any resources
1706 * that we were not when we came in.
1707 *
1708 * Aborting from this point might lose some
1709 * blocks in the file system, but oh well.
1710 */
1713 /*
1714 * If the passed in transaction committed
1715 * in xfs_bmap_finish(), then we want to
1716 * add the inode to this one before returning.
1717 * This keeps things simple for the higher
1718 * level code, because it always knows that
1719 * the inode is locked and held in the
1720 * transaction that returns to it whether
1721 * errors occur or not. We don't mark the
1722 * inode dirty so that this transaction can
1723 * be easily aborted if possible.
1724 */
1728 }
1730 }
1733 /*
1734 * The first xact was committed,
1735 * so add the inode to the new one.
1736 * Mark it dirty so it will be logged
1737 * and moved forward in the log as
1738 * part of every commit.
1739 */
1744 }
1749 XFS_TRANS_PERM_LOG_RES,
1750 XFS_ITRUNCATE_LOG_COUNT);
1751 /*
1752 * Add the inode being truncated to the next chained
1753 * transaction.
1754 */
1759 }
1760 /*
1761 * Only update the size in the case of the data fork, but
1762 * always re-log the inode so that our permanent transaction
1763 * can keep on rolling it forward in the log.
1764 */
1768 }
1778 }
1781 /*
1782 * xfs_igrow_start
1783 *
1784 * Do the first part of growing a file: zero any data in the last
1785 * block that is beyond the old EOF. We need to do this before
1786 * the inode is joined to the transaction to modify the i_size.
1787 * That way we can drop the inode lock and call into the buffer
1788 * cache to get the buffer mapping the EOF.
1789 */
1790 int
1793 xfs_fsize_t new_size,
1795 {
1802 /*
1803 * Zero any pages that may have been created by
1804 * xfs_write_file() beyond the end of the file
1805 * and any blocks between the old and new file sizes.
1806 */
1810 }
1812 /*
1813 * xfs_igrow_finish
1814 *
1815 * This routine is called to extend the size of a file.
1816 * The inode must have both the iolock and the ilock locked
1817 * for update and it must be a part of the current transaction.
1818 * The xfs_igrow_start() function must have been called previously.
1819 * If the change_flag is not zero, the inode change timestamp will
1820 * be updated.
1821 */
1822 void
1826 xfs_fsize_t new_size,
1828 {
1834 /*
1835 * Update the file size. Update the inode change timestamp
1836 * if change_flag set.
1837 */
1843 }
1846 /*
1847 * This is called when the inode's link count goes to 0.
1848 * We place the on-disk inode on a list in the AGI. It
1849 * will be pulled from this list when the inode is freed.
1850 */
1851 int
1855 {
1861 xfs_agnumber_t agno;
1862 xfs_daddr_t agdaddr;
1863 xfs_agino_t agino;
1878 /*
1879 * Get the agi buffer first. It ensures lock ordering
1880 * on the list.
1881 */
1886 }
1887 /*
1888 * Validate the magic number of the agi block.
1889 */
1891 agi_ok =
1895 XFS_RANDOM_IUNLINK))) {
1899 }
1900 /*
1901 * Get the index into the agi hash table for the
1902 * list this inode will go on.
1903 */
1911 /*
1912 * There is already another inode in the bucket we need
1913 * to add ourselves to. Add us at the front of the list.
1914 * Here we put the head pointer into our next pointer,
1915 * and then we fall through to point the head at us.
1916 */
1920 }
1923 /* both on-disk, don't endian flip twice */
1931 }
1933 /*
1934 * Point the bucket head pointer at the inode being inserted.
1935 */
1943 }
1945 /*
1946 * Pull the on-disk inode from the AGI unlinked list.
1947 */
1948 STATIC int
1952 {
1953 xfs_ino_t next_ino;
1959 xfs_agnumber_t agno;
1960 xfs_daddr_t agdaddr;
1961 xfs_agino_t agino;
1962 xfs_agino_t next_agino;
1970 /*
1971 * First pull the on-disk inode from the AGI unlinked list.
1972 */
1978 /*
1979 * Get the agi buffer first. It ensures lock ordering
1980 * on the list.
1981 */
1989 }
1990 /*
1991 * Validate the magic number of the agi block.
1992 */
1994 agi_ok =
1998 XFS_RANDOM_IUNLINK_REMOVE))) {
2006 }
2007 /*
2008 * Get the index into the agi hash table for the
2009 * list this inode will go on.
2010 */
2018 /*
2019 * We're at the head of the list. Get the inode's
2020 * on-disk buffer to see if there is anyone after us
2021 * on the list. Only modify our next pointer if it
2022 * is not already NULLAGINO. This saves us the overhead
2023 * of dealing with the buffer when there is no need to
2024 * change it.
2025 */
2032 }
2045 }
2046 /*
2047 * Point the bucket head pointer at the next inode.
2048 */
2057 /*
2058 * We need to search the list for the inode being freed.
2059 */
2063 /*
2064 * If the last inode wasn't the one pointing to
2065 * us, then release its buffer since we're not
2066 * going to do anything with it.
2067 */
2070 }
2079 }
2083 }
2084 /*
2085 * Now last_ibp points to the buffer previous to us on
2086 * the unlinked list. Pull us from the list.
2087 */
2094 }
2108 }
2109 /*
2110 * Point the previous inode on the list to the next inode.
2111 */
2119 }
2121 }
2124 {
2128 }
2130 STATIC void
2134 xfs_ino_t inum)
2135 {
2141 xfs_daddr_t blkno;
2158 }
2167 /*
2168 * Look for each inode in memory and attempt to lock it,
2169 * we can be racing with flush and tail pushing here.
2170 * any inode we get the locks on, add to an array of
2171 * inode items to process later.
2172 *
2173 * The get the buffer lock, we could beat a flush
2174 * or tail pushing thread to the lock here, in which
2175 * case they will go looking for the inode buffer
2176 * and fail, we need some other form of interlock
2177 * here.
2178 */
2186 }
2188 /* Inode not in memory or we found it already,
2189 * nothing to do
2190 */
2194 }
2199 }
2201 /* If we can get the locks then add it to the
2202 * list, otherwise by the time we get the bp lock
2203 * below it will already be attached to the
2204 * inode buffer.
2205 */
2207 /* This inode will already be locked - by us, lets
2208 * keep it that way.
2209 */
2219 }
2220 }
2223 }
2234 }
2237 }
2238 }
2241 }
2245 XFS_BUF_LOCK);
2258 pre_flushed++;
2259 }
2261 }
2272 }
2286 }
2287 }
2292 }
2295 }
2297 /*
2298 * This is called to return an inode to the inode free list.
2299 * The inode should already be truncated to 0 length and have
2300 * no pages associated with it. This routine also assumes that
2301 * the inode is already a part of the transaction.
2302 *
2303 * The on-disk copy of the inode will have been added to the list
2304 * of unlinked inodes in the AGI. We need to remove the inode from
2305 * that list atomically with respect to freeing it here.
2306 */
2307 int
2312 {
2315 xfs_ino_t first_ino;
2326 /*
2327 * Pull the on-disk inode from the AGI unlinked list.
2328 */
2332 }
2337 }
2346 /*
2347 * Bump the generation count so no one will be confused
2348 * by reincarnations of this inode.
2349 */
2355 }
2358 }
2360 /*
2361 * Reallocate the space for if_broot based on the number of records
2362 * being added or deleted as indicated in rec_diff. Move the records
2363 * and pointers in if_broot to fit the new size. When shrinking this
2364 * will eliminate holes between the records and pointers created by
2365 * the caller. When growing this will create holes to be filled in
2366 * by the caller.
2367 *
2368 * The caller must not request to add more records than would fit in
2369 * the on-disk inode root. If the if_broot is currently NULL, then
2370 * if we adding records one will be allocated. The caller must also
2371 * not request that the number of records go below zero, although
2372 * it can go to zero.
2373 *
2374 * ip -- the inode whose if_broot area is changing
2375 * ext_diff -- the change in the number of records, positive or negative,
2376 * requested for the if_broot array.
2377 */
2378 void
2383 {
2392 /*
2393 * Handle the degenerate case quietly.
2394 */
2397 }
2401 /*
2402 * If there wasn't any memory allocated before, just
2403 * allocate it now and get out.
2404 */
2408 KM_SLEEP);
2411 }
2413 /*
2414 * If there is already an existing if_broot, then we need
2415 * to realloc() it and shift the pointers to their new
2416 * location. The records don't change location because
2417 * they are kept butted up against the btree block header.
2418 */
2424 new_size,
2426 KM_SLEEP);
2436 }
2438 /*
2439 * rec_diff is less than 0. In this case, we are shrinking the
2440 * if_broot buffer. It must already exist. If we go to zero
2441 * records, just get rid of the root and clear the status bit.
2442 */
2449 else
2453 /*
2454 * First copy over the btree block header.
2455 */
2460 }
2462 /*
2463 * Only copy the records and pointers if there are any.
2464 */
2466 /*
2467 * First copy the records.
2468 */
2475 /*
2476 * Then copy the pointers.
2477 */
2483 }
2490 }
2493 /*
2494 * This is called when the amount of space needed for if_data
2495 * is increased or decreased. The change in size is indicated by
2496 * the number of bytes that need to be added or deleted in the
2497 * byte_diff parameter.
2498 *
2499 * If the amount of space needed has decreased below the size of the
2500 * inline buffer, then switch to using the inline buffer. Otherwise,
2501 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2502 * to what is needed.
2503 *
2504 * ip -- the inode whose if_data area is changing
2505 * byte_diff -- the change in the number of bytes, positive or negative,
2506 * requested for the if_data array.
2507 */
2508 void
2513 {
2520 }
2529 }
2533 /*
2534 * If the valid extents/data can fit in if_inline_ext/data,
2535 * copy them from the malloc'd vector and free it.
2536 */
2542 new_size);
2545 }
2548 /*
2549 * Stuck with malloc/realloc.
2550 * For inline data, the underlying buffer must be
2551 * a multiple of 4 bytes in size so that it can be
2552 * logged and stay on word boundaries. We enforce
2553 * that here.
2554 */
2560 /*
2561 * Only do the realloc if the underlying size
2562 * is really changing.
2563 */
2567 real_size,
2569 KM_SLEEP);
2570 }
2576 }
2577 }
2581 }
2586 /*
2587 * Map inode to disk block and offset.
2588 *
2589 * mp -- the mount point structure for the current file system
2590 * tp -- the current transaction
2591 * ino -- the inode number of the inode to be located
2592 * imap -- this structure is filled in with the information necessary
2593 * to retrieve the given inode from disk
2594 * flags -- flags to pass to xfs_dilocate indicating whether or not
2595 * lookups in the inode btree were OK or not
2596 */
2597 int
2601 xfs_ino_t ino,
2603 uint flags)
2604 {
2605 xfs_fsblock_t fsbno;
2615 }
2622 }
2624 void
2628 {
2635 }
2637 /*
2638 * If the format is local, then we can't have an extents
2639 * array so just look for an inline data array. If we're
2640 * not local then we may or may not have an extents list,
2641 * so check and free it up if we do.
2642 */
2650 }
2657 }
2664 }
2665 }
2667 /*
2668 * This is called free all the memory associated with an inode.
2669 * It must free the inode itself and any buffers allocated for
2670 * if_extents/if_data and if_broot. It must also free the lock
2671 * associated with the inode.
2672 */
2673 void
2676 {
2684 }
2690 #ifdef XFS_BMAP_TRACE
2692 #endif
2693 #ifdef XFS_BMBT_TRACE
2695 #endif
2696 #ifdef XFS_RW_TRACE
2698 #endif
2699 #ifdef XFS_ILOCK_TRACE
2701 #endif
2702 #ifdef XFS_DIR2_TRACE
2704 #endif
2706 /* XXXdpd should be able to assert this but shutdown
2707 * is leaving the AIL behind. */
2711 }
2713 }
2716 /*
2717 * Increment the pin count of the given buffer.
2718 * This value is protected by ipinlock spinlock in the mount structure.
2719 */
2720 void
2723 {
2727 }
2729 /*
2730 * Decrement the pin count of the given inode, and wake up
2731 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2732 * inode must have been previously pinned with a call to xfs_ipin().
2733 */
2734 void
2737 {
2741 /*
2742 * If the inode is currently being reclaimed, the
2743 * linux inode _and_ the xfs vnode may have been
2744 * freed so we cannot reference either of them safely.
2745 * Hence we should not try to do anything to them
2746 * if the xfs inode is currently in the reclaim
2747 * path.
2748 *
2749 * However, we still need to issue the unpin wakeup
2750 * call as the inode reclaim may be blocked waiting for
2751 * the inode to become unpinned.
2752 */
2756 /* make sync come back and flush this inode */
2763 }
2764 }
2766 }
2767 }
2769 /*
2770 * This is called to wait for the given inode to be unpinned.
2771 * It will sleep until this happens. The caller must have the
2772 * inode locked in at least shared mode so that the buffer cannot
2773 * be subsequently pinned once someone is waiting for it to be
2774 * unpinned.
2775 */
2776 STATIC void
2779 {
2781 xfs_lsn_t lsn;
2787 }
2794 }
2796 /*
2797 * Give the log a push so we don't wait here too long.
2798 */
2802 }
2805 /*
2806 * xfs_iextents_copy()
2807 *
2808 * This is called to copy the REAL extents (as opposed to the delayed
2809 * allocation extents) from the inode into the given buffer. It
2810 * returns the number of bytes copied into the buffer.
2811 *
2812 * If there are no delayed allocation extents, then we can just
2813 * memcpy() the extents into the buffer. Otherwise, we need to
2814 * examine each extent in turn and skip those which are delayed.
2815 */
2816 int
2821 {
2825 #ifdef XFS_BMAP_TRACE
2827 #endif
2831 xfs_fsblock_t start_block;
2841 /*
2842 * There are some delayed allocation extents in the
2843 * inode, so copy the extents one at a time and skip
2844 * the delayed ones. There must be at least one
2845 * non-delayed extent.
2846 */
2853 /*
2854 * It's a delayed allocation extent, so skip it.
2855 */
2857 }
2859 /* Translate to on disk format */
2864 dest_ep++;
2865 copied++;
2866 }
2871 }
2873 /*
2874 * Each of the following cases stores data into the same region
2875 * of the on-disk inode, so only one of them can be valid at
2876 * any given time. While it is possible to have conflicting formats
2877 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2878 * in EXTENTS format, this can only happen when the fork has
2879 * changed formats after being modified but before being flushed.
2880 * In these cases, the format always takes precedence, because the
2881 * format indicates the current state of the fork.
2882 */
2883 /*ARGSUSED*/
2884 STATIC int
2891 {
2895 #ifdef XFS_TRANS_DEBUG
2897 #endif
2908 /*
2909 * This can happen if we gave up in iformat in an error path,
2910 * for the attribute fork.
2911 */
2915 }
2925 }
2939 whichfork);
2940 }
2949 XFS_BROOT_SIZE_ADJ));
2953 }
2960 }
2968 }
2974 }
2977 }
2979 /*
2980 * xfs_iflush() will write a modified inode's changes out to the
2981 * inode's on disk home. The caller must have the inode lock held
2982 * in at least shared mode and the inode flush semaphore must be
2983 * held as well. The inode lock will still be held upon return from
2984 * the call and the caller is free to unlock it.
2985 * The inode flush lock will be unlocked when the inode reaches the disk.
2986 * The flags indicate how the inode's buffer should be written out.
2987 */
2988 int
2991 uint flags)
2992 {
2998 /* REFERENCED */
3016 /*
3017 * If the inode isn't dirty, then just release the inode
3018 * flush lock and do nothing.
3019 */
3026 }
3028 /*
3029 * We can't flush the inode until it is unpinned, so
3030 * wait for it. We know noone new can pin it, because
3031 * we are holding the inode lock shared and you need
3032 * to hold it exclusively to pin the inode.
3033 */
3036 /*
3037 * This may have been unpinned because the filesystem is shutting
3038 * down forcibly. If that's the case we must not write this inode
3039 * to disk, because the log record didn't make it to disk!
3040 */
3047 }
3049 /*
3050 * Get the buffer containing the on-disk inode.
3051 */
3056 }
3058 /*
3059 * Decide how buffer will be flushed out. This is done before
3060 * the call to xfs_iflush_int because this field is zeroed by it.
3061 */
3063 /*
3064 * Flush out the inode buffer according to the directions
3065 * of the caller. In the cases where the caller has given
3066 * us a choice choose the non-delwri case. This is because
3067 * the inode is in the AIL and we need to get it out soon.
3068 */
3085 }
3103 }
3104 }
3106 /*
3107 * First flush out the inode that xfs_iflush was called with.
3108 */
3112 }
3114 /*
3115 * inode clustering:
3116 * see if other inodes can be gathered into this write
3117 */
3126 /*
3127 * Do an un-protected check to see if the inode is dirty and
3128 * is a candidate for flushing. These checks will be repeated
3129 * later after the appropriate locks are acquired.
3130 */
3137 }
3139 /*
3140 * Try to get locks. If any are unavailable,
3141 * then this inode cannot be flushed and is skipped.
3142 */
3144 /* get inode locks (just i_lock) */
3146 /* get inode flush lock */
3148 /* check if pinned */
3150 /* arriving here means that
3151 * this inode can be flushed.
3152 * first re-check that it's
3153 * dirty
3154 */
3159 clcount++;
3163 XFS_ILOCK_SHARED);
3165 }
3168 }
3171 }
3172 }
3174 }
3175 }
3181 }
3183 /*
3184 * If the buffer is pinned then push on the log so we won't
3185 * get stuck waiting in the write for too long.
3186 */
3189 }
3197 }
3200 corrupt_out:
3204 /*
3205 * Unlocks the flush lock
3206 */
3209 cluster_corrupt_out:
3210 /* Corruption detected in the clustering loop. Invalidate the
3211 * inode buffer and shut down the filesystem.
3212 */
3215 /*
3216 * Clean up the buffer. If it was B_DELWRI, just release it --
3217 * brelse can handle it with no problems. If not, shut down the
3218 * filesystem before releasing the buffer.
3219 */
3222 }
3227 /*
3228 * Just like incore_relse: if we have b_iodone functions,
3229 * mark the buffer as an error and call them. Otherwise
3230 * mark it as stale and brelse.
3231 */
3242 }
3243 }
3246 /*
3247 * Unlocks the flush lock
3248 */
3250 }
3253 STATIC int
3257 {
3261 #ifdef XFS_TRANS_DEBUG
3263 #endif
3275 /*
3276 * If the inode isn't dirty, then just release the inode
3277 * flush lock and do nothing.
3278 */
3283 }
3285 /* set *dip = inode's place in the buffer */
3288 /*
3289 * Clear i_update_core before copying out the data.
3290 * This is for coordination with our timestamp updates
3291 * that don't hold the inode lock. They will always
3292 * update the timestamps BEFORE setting i_update_core,
3293 * so if we clear i_update_core after they set it we
3294 * are guaranteed to see their updates to the timestamps.
3295 * I believe that this depends on strongly ordered memory
3296 * semantics, but we have that. We use the SYNCHRONIZE
3297 * macro to make sure that the compiler does not reorder
3298 * the i_update_core access below the data copy below.
3299 */
3303 /*
3304 * Make sure to get the latest atime from the Linux inode.
3305 */
3314 }
3321 }
3331 }
3342 }
3343 }
3346 XFS_RANDOM_IFLUSH_5)) {
3352 ip);
3354 }
3361 }
3362 /*
3363 * bump the flush iteration count, used to detect flushes which
3364 * postdate a log record during recovery.
3365 */
3369 /*
3370 * Copy the dirty parts of the inode into the on-disk
3371 * inode. We always copy out the core of the inode,
3372 * because if the inode is dirty at all the core must
3373 * be.
3374 */
3377 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3381 /*
3382 * If this is really an old format inode and the superblock version
3383 * has not been updated to support only new format inodes, then
3384 * convert back to the old inode format. If the superblock version
3385 * has been updated, then make the conversion permanent.
3386 */
3391 /*
3392 * Convert it back.
3393 */
3397 /*
3398 * The superblock version has already been bumped,
3399 * so just make the conversion to the new inode
3400 * format permanent.
3401 */
3410 }
3411 }
3415 }
3418 /*
3419 * The only error from xfs_iflush_fork is on the data fork.
3420 */
3422 }
3425 /*
3426 * We've recorded everything logged in the inode, so we'd
3427 * like to clear the ilf_fields bits so we don't log and
3428 * flush things unnecessarily. However, we can't stop
3429 * logging all this information until the data we've copied
3430 * into the disk buffer is written to disk. If we did we might
3431 * overwrite the copy of the inode in the log with all the
3432 * data after re-logging only part of it, and in the face of
3433 * a crash we wouldn't have all the data we need to recover.
3434 *
3435 * What we do is move the bits to the ili_last_fields field.
3436 * When logging the inode, these bits are moved back to the
3437 * ilf_fields field. In the xfs_iflush_done() routine we
3438 * clear ili_last_fields, since we know that the information
3439 * those bits represent is permanently on disk. As long as
3440 * the flush completes before the inode is logged again, then
3441 * both ilf_fields and ili_last_fields will be cleared.
3442 *
3443 * We can play with the ilf_fields bits here, because the inode
3444 * lock must be held exclusively in order to set bits there
3445 * and the flush lock protects the ili_last_fields bits.
3446 * Set ili_logged so the flush done
3447 * routine can tell whether or not to look in the AIL.
3448 * Also, store the current LSN of the inode so that we can tell
3449 * whether the item has moved in the AIL from xfs_iflush_done().
3450 * In order to read the lsn we need the AIL lock, because
3451 * it is a 64 bit value that cannot be read atomically.
3452 */
3463 /*
3464 * Attach the function xfs_iflush_done to the inode's
3465 * buffer. This will remove the inode from the AIL
3466 * and unlock the inode's flush lock when the inode is
3467 * completely written to disk.
3468 */
3475 /*
3476 * We're flushing an inode which is not in the AIL and has
3477 * not been logged but has i_update_core set. For this
3478 * case we can use a B_DELWRI flush and immediately drop
3479 * the inode flush lock because we can avoid the whole
3480 * AIL state thing. It's OK to drop the flush lock now,
3481 * because we've already locked the buffer and to do anything
3482 * you really need both.
3483 */
3488 }
3490 }
3494 corrupt_out:
3496 }
3499 /*
3500 * Flush all inactive inodes in mp.
3501 */
3502 void
3505 {
3509 again:
3516 /* Make sure we skip markers inserted by sync */
3520 }
3527 }
3533 out:
3535 }
3537 /*
3538 * xfs_iaccess: check accessibility of inode for mode.
3539 */
3540 int
3543 mode_t mode,
3545 {
3559 }
3561 /*
3562 * If there's an Access Control List it's used instead of
3563 * the mode bits.
3564 */
3572 }
3574 /*
3575 * If the DACs are ok we don't need any capability check.
3576 */
3579 /*
3580 * Read/write DACs are always overridable.
3581 * Executable DACs are overridable if at least one exec bit is set.
3582 */
3592 #ifdef NOISE
3596 }
3598 }
3600 /*
3601 * xfs_iroundup: round up argument to next power of two
3602 */
3603 uint
3605 uint v)
3606 {
3608 uint m;
3621 }
3624 }
3626 #ifdef XFS_ILOCK_TRACE
3629 void
3631 {
3640 }
3641 #endif
3643 /*
3644 * Return a pointer to the extent record at file index idx.
3645 */
3646 xfs_bmbt_rec_t *
3650 {
3665 }
3666 }
3668 /*
3669 * Insert new item(s) into the extent records for incore inode
3670 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3671 */
3672 void
3678 {
3687 }
3688 }
3690 /*
3691 * This is called when the amount of space required for incore file
3692 * extents needs to be increased. The ext_diff parameter stores the
3693 * number of new extents being added and the idx parameter contains
3694 * the extent index where the new extents will be added. If the new
3695 * extents are being appended, then we just need to (re)allocate and
3696 * initialize the space. Otherwise, if the new extents are being
3697 * inserted into the middle of the existing entries, a bit more work
3698 * is required to make room for the new extents to be inserted. The
3699 * caller is responsible for filling in the new extent entries upon
3700 * return.
3701 */
3702 void
3707 {
3716 /*
3717 * If the new number of extents (nextents + ext_diff)
3718 * fits inside the inode, then continue to use the inline
3719 * extent buffer.
3720 */
3727 }
3731 }
3732 /*
3733 * Otherwise use a linear (direct) extent list.
3734 * If the extents are currently inside the inode,
3735 * xfs_iext_realloc_direct will switch us from
3736 * inline to direct extent allocation mode.
3737 */
3745 }
3746 }
3747 /* Indirection array */
3760 }
3761 /* Extents fit in target extent page */
3769 }
3772 }
3773 /* Insert a new extent page */
3777 }
3778 /*
3779 * If extent(s) are being appended to the last page in
3780 * the indirection array and the new extent(s) don't fit
3781 * in the page, then erp is NULL and erp_idx is set to
3782 * the next index needed in the indirection array.
3783 */
3792 erp_idx++;
3793 }
3794 }
3795 }
3796 }
3798 }
3800 /*
3801 * This is called when incore extents are being added to the indirection
3802 * array and the new extents do not fit in the target extent list. The
3803 * erp_idx parameter contains the irec index for the target extent list
3804 * in the indirection array, and the idx parameter contains the extent
3805 * index within the list. The number of extents being added is stored
3806 * in the count parameter.
3807 *
3808 * |-------| |-------|
3809 * | | | | idx - number of extents before idx
3810 * | idx | | count |
3811 * | | | | count - number of extents being inserted at idx
3812 * |-------| |-------|
3813 * | count | | nex2 | nex2 - number of extents after idx + count
3814 * |-------| |-------|
3815 */
3816 void
3822 {
3836 /*
3837 * Save second part of target extent list
3838 * (all extents past */
3846 }
3848 /*
3849 * Add the new extents to the end of the target
3850 * list, then allocate new irec record(s) and
3851 * extent buffer(s) as needed to store the rest
3852 * of the new extents.
3853 */
3860 }
3862 erp_idx++;
3868 }
3870 /* Add nex2 extents back to indirection array */
3872 xfs_extnum_t ext_avail;
3878 /*
3879 * If nex2 extents fit in the current page, append
3880 * nex2_ep after the new extents.
3881 */
3884 }
3885 /*
3886 * Otherwise, check if space is available in the
3887 * next page.
3888 */
3892 erp_idx++;
3893 erp++;
3894 /* Create a hole for nex2 extents */
3897 }
3898 /*
3899 * Final choice, create a new extent page for
3900 * nex2 extents.
3901 */
3903 erp_idx++;
3905 }
3910 }
3911 }
3913 /*
3914 * This is called when the amount of space required for incore file
3915 * extents needs to be decreased. The ext_diff parameter stores the
3916 * number of extents to be removed and the idx parameter contains
3917 * the extent index where the extents will be removed from.
3918 *
3919 * If the amount of space needed has decreased below the linear
3920 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3921 * extent array. Otherwise, use kmem_realloc() to adjust the
3922 * size to what is needed.
3923 */
3924 void
3929 {
3945 }
3947 }
3949 /*
3950 * This removes ext_diff extents from the inline buffer, beginning
3951 * at extent index idx.
3952 */
3953 void
3958 {
3977 }
3978 }
3980 /*
3981 * This removes ext_diff extents from a linear (direct) extent list,
3982 * beginning at extent index idx. If the extents are being removed
3983 * from the end of the list (ie. truncate) then we just need to re-
3984 * allocate the list to remove the extra space. Otherwise, if the
3985 * extents are being removed from the middle of the existing extent
3986 * entries, then we first need to move the extent records beginning
3987 * at idx + ext_diff up in the list to overwrite the records being
3988 * removed, then remove the extra space via kmem_realloc.
3989 */
3990 void
3995 {
4007 }
4008 /* Move extents up in the list (if needed) */
4014 }
4017 /*
4018 * Reallocate the direct extent list. If the extents
4019 * will fit inside the inode then xfs_iext_realloc_direct
4020 * will switch from direct to inline extent allocation
4021 * mode for us.
4022 */
4025 }
4027 /*
4028 * This is called when incore extents are being removed from the
4029 * indirection array and the extents being removed span multiple extent
4030 * buffers. The idx parameter contains the file extent index where we
4031 * want to begin removing extents, and the count parameter contains
4032 * how many extents need to be removed.
4033 *
4034 * |-------| |-------|
4035 * | nex1 | | | nex1 - number of extents before idx
4036 * |-------| | count |
4037 * | | | | count - number of extents being removed at idx
4038 * | count | |-------|
4039 * | | | nex2 | nex2 - number of extents after idx + count
4040 * |-------| |-------|
4041 */
4042 void
4047 {
4066 /*
4067 * Check for deletion of entire list;
4068 * xfs_iext_irec_remove() updates extent offsets.
4069 */
4076 XFS_IEXT_BUFSZ);
4082 }
4083 }
4084 /* Move extents up (if needed) */
4089 }
4090 /* Zero out rest of page */
4093 /* Update remaining counters */
4098 erp_idx++;
4099 erp++;
4100 }
4103 }
4105 /*
4106 * Create, destroy, or resize a linear (direct) block of extents.
4107 */
4108 void
4112 {
4121 /* Free extent records */
4124 }
4125 /* Resize direct extent list and zero any new bytes */
4127 /* Check if extents will fit inside the inode */
4133 }
4136 }
4140 rnew_size,
4142 KM_SLEEP);
4143 }
4148 }
4149 }
4150 /*
4151 * Switch from the inline extent buffer to a direct
4152 * extent list. Be sure to include the inline extent
4153 * bytes in new_size.
4154 */
4159 }
4161 }
4164 }
4166 /*
4167 * Switch from linear (direct) extent records to inline buffer.
4168 */
4169 void
4173 {
4176 /*
4177 * The inline buffer was zeroed when we switched
4178 * from inline to direct extent allocation mode,
4179 * so we don't need to clear it here.
4180 */
4186 }
4188 /*
4189 * Switch from inline buffer to linear (direct) extent records.
4190 * new_size should already be rounded up to the next power of 2
4191 * by the caller (when appropriate), so use new_size as it is.
4192 * However, since new_size may be rounded up, we can't update
4193 * if_bytes here. It is the caller's responsibility to update
4194 * if_bytes upon return.
4195 */
4196 void
4200 {
4209 }
4211 }
4213 /*
4214 * Resize an extent indirection array to new_size bytes.
4215 */
4216 void
4220 {
4235 }
4236 }
4238 /*
4239 * Switch from indirection array to linear (direct) extent allocations.
4240 */
4241 void
4244 {
4264 }
4265 }
4267 /*
4268 * Free incore file extents.
4269 */
4270 void
4273 {
4281 }
4288 }
4292 }
4294 /*
4295 * Return a pointer to the extent record for file system block bno.
4296 */
4302 {
4317 }
4320 /* Find target extent list */
4328 }
4329 /* Binary search extent records */
4340 /* Convert back to file-based extent index */
4343 }
4346 }
4347 }
4348 /* Convert back to file-based extent index */
4351 }
4357 }
4358 }
4361 }
4363 /*
4364 * Return a pointer to the indirection array entry containing the
4365 * extent record for filesystem block bno. Store the index of the
4366 * target irec in *erp_idxp.
4367 */
4373 {
4397 }
4398 }
4401 }
4403 /*
4404 * Return a pointer to the indirection array entry containing the
4405 * extent record at file extent index *idxp. Store the index of the
4406 * target irec in *erp_idxp and store the page index of the target
4407 * extent record in *idxp.
4408 */
4409 xfs_ext_irec_t *
4415 {
4432 /* Binary search extent irec's */
4448 erp_idx++;
4454 }
4455 }
4459 }
4461 /*
4462 * Allocate and initialize an indirection array once the space needed
4463 * for incore extents increases above XFS_IEXT_BUFSZ.
4464 */
4465 void
4468 {
4486 }
4497 }
4499 /*
4500 * Allocate and initialize a new entry in the indirection array.
4501 */
4502 xfs_ext_irec_t *
4506 {
4514 /* Resize indirection array */
4517 /*
4518 * Move records down in the array so the
4519 * new page can use erp_idx.
4520 */
4524 }
4527 /* Initialize new extent record */
4537 }
4539 /*
4540 * Remove a record from the indirection array.
4541 */
4542 void
4546 {
4558 }
4559 /* Compact extent records */
4563 }
4564 /*
4565 * Manually free the last extent record from the indirection
4566 * array. A call to xfs_iext_realloc_indirect() with a size
4567 * of zero would result in a call to xfs_iext_destroy() which
4568 * would in turn call this function again, creating a nasty
4569 * infinite loop.
4570 */
4577 }
4579 }
4581 /*
4582 * This is called to clean up large amounts of unused memory allocated
4583 * by the indirection array. Before compacting anything though, verify
4584 * that the indirection array is still needed and switch back to the
4585 * linear extent list (or even the inline buffer) if possible. The
4586 * compaction policy is as follows:
4587 *
4588 * Full Compaction: Extents fit into a single page (or inline buffer)
4589 * Full Compaction: Extents occupy less than 10% of allocated space
4590 * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
4591 * No Compaction: Extents occupy at least 50% of allocated space
4592 */
4593 void
4596 {
4615 }
4616 }
4618 /*
4619 * Combine extents from neighboring extent pages.
4620 */
4621 void
4624 {
4640 /*
4641 * Free page before removing extent record
4642 * so er_extoffs don't get modified in
4643 * xfs_iext_irec_remove.
4644 */
4650 erp_idx++;
4651 }
4652 }
4653 }
4655 /*
4656 * Fully compact the extent records managed by the indirection array.
4657 */
4658 void
4661 {
4681 /* Remove next page */
4683 /*
4684 * Free page before removing extent record
4685 * so er_extoffs don't get modified in
4686 * xfs_iext_irec_remove.
4687 */
4694 /* Update next page */
4696 /* Move rest of page up to become next new page */
4703 }
4705 erp_idx++;
4708 else
4710 }
4714 }
4715 }
4717 /*
4718 * This is called to update the er_extoff field in the indirection
4719 * array when extents have been added or removed from one of the
4720 * extent lists. erp_idx contains the irec index to begin updating
4721 * at and ext_diff contains the number of extents that were added
4722 * or removed.
4723 */
4724 void
4729 {
4737 }
4738 }