| blob | 1f8ecff8553a3b205525d3998abc0968f25bd2e5 |
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)
340 #endif
351 XFS_ERRTAG_ITOBP_INOTOBP,
352 XFS_RANDOM_ITOBP_INOTOBP))) {
356 }
357 #ifdef DEBUG
359 "Device %s - bad inode magic/vsn "
364 #endif
369 }
370 }
374 /*
375 * Mark the buffer as an inode buffer now that it looks good
376 */
379 /*
380 * Set *dipp to point to the on-disk inode in the buffer.
381 */
385 }
387 /*
388 * Move inode type and inode format specific information from the
389 * on-disk inode to the in-core inode. For fifos, devs, and sockets
390 * this means set if_rdev to the proper value. For files, directories,
391 * and symlinks this means to bring in the in-line data or extent
392 * pointers. For a file in B-tree format, only the root is immediately
393 * brought in-core. The rest will be in-lined in if_extents when it
394 * is first referenced (see xfs_iread_extents()).
395 */
396 STATIC int
400 {
404 xfs_fsize_t di_size;
423 }
425 if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) {
433 }
444 }
454 /*
455 * no local regular files yet
456 */
459 "corrupt inode %Lu "
463 XFS_ERRLEVEL_LOW,
466 }
471 "corrupt inode %Lu "
476 XFS_ERRLEVEL_LOW,
479 }
494 }
500 }
503 }
525 }
530 }
532 }
534 /*
535 * The file is in-lined in the on-disk inode.
536 * If it fits into if_inline_data, then copy
537 * it there, otherwise allocate a buffer for it
538 * and copy the data there. Either way, set
539 * if_data to point at the data.
540 * If we allocate a buffer for the data, make
541 * sure that its size is a multiple of 4 and
542 * record the real size in i_real_bytes.
543 */
544 STATIC int
550 {
554 /*
555 * If the size is unreasonable, then something
556 * is wrong and we just bail out rather than crash in
557 * kmem_alloc() or memcpy() below.
558 */
561 "corrupt inode %Lu "
568 }
578 }
586 }
588 /*
589 * The file consists of a set of extents all
590 * of which fit into the on-disk inode.
591 * If there are few enough extents to fit into
592 * the if_inline_ext, then copy them there.
593 * Otherwise allocate a buffer for them and copy
594 * them into it. Either way, set if_extents
595 * to point at the extents.
596 */
597 STATIC int
602 {
613 /*
614 * If the number of extents is unreasonable, then something
615 * is wrong and we just bail out rather than crash in
616 * kmem_alloc() or memcpy() below.
617 */
625 }
632 else
642 ARCH_CONVERT);
644 ARCH_CONVERT);
645 }
647 whichfork);
653 XFS_ERRLEVEL_LOW,
656 }
657 }
660 }
662 /*
663 * The file has too many extents to fit into
664 * the inode, so they are in B-tree format.
665 * Allocate a buffer for the root of the B-tree
666 * and copy the root into it. The i_extents
667 * field will remain NULL until all of the
668 * extents are read in (when they are needed).
669 */
670 STATIC int
675 {
678 /* REFERENCED */
687 /*
688 * blow out if -- fork has less extents than can fit in
689 * fork (fork shouldn't be a btree format), root btree
690 * block has more records than can fit into the fork,
691 * or the number of extents is greater than the number of
692 * blocks.
693 */
704 }
709 /*
710 * Copy and convert from the on-disk structure
711 * to the in-memory structure.
712 */
719 }
721 /*
722 * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
723 * and native format
724 *
725 * buf = on-disk representation
726 * dip = native representation
727 * dir = direction - +ve -> disk to native
728 * -ve -> native to disk
729 */
730 void
732 xfs_caddr_t buf,
735 {
758 }
785 }
787 STATIC uint
789 __uint16_t di_flags)
790 {
820 }
823 }
825 uint
828 {
833 }
835 uint
838 {
841 }
843 /*
844 * Given a mount structure and an inode number, return a pointer
845 * to a newly allocated in-core inode corresponding to the given
846 * inode number.
847 *
848 * Initialize the inode's attributes and extent pointers if it
849 * already has them (it will not if the inode has no links).
850 */
851 int
855 xfs_ino_t ino,
857 xfs_daddr_t bno)
858 {
870 /*
871 * Get pointer's to the on-disk inode and the buffer containing it.
872 * If the inode number refers to a block outside the file system
873 * then xfs_itobp() will return NULL. In this case we should
874 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
875 * know that this is a new incore inode.
876 */
881 }
883 /*
884 * Initialize inode's trace buffers.
885 * Do this before xfs_iformat in case it adds entries.
886 */
887 #ifdef XFS_BMAP_TRACE
889 #endif
890 #ifdef XFS_BMBT_TRACE
892 #endif
893 #ifdef XFS_RW_TRACE
895 #endif
896 #ifdef XFS_ILOCK_TRACE
898 #endif
899 #ifdef XFS_DIR2_TRACE
901 #endif
903 /*
904 * If we got something that isn't an inode it means someone
905 * (nfs or dmi) has a stale handle.
906 */
910 #ifdef DEBUG
912 "dip->di_core.di_magic (0x%x) != "
915 XFS_DINODE_MAGIC);
918 }
920 /*
921 * If the on-disk inode is already linked to a directory
922 * entry, copy all of the inode into the in-core inode.
923 * xfs_iformat() handles copying in the inode format
924 * specific information.
925 * Otherwise, just get the truly permanent information.
926 */
934 #ifdef DEBUG
937 error);
940 }
946 /*
947 * Make sure to pull in the mode here as well in
948 * case the inode is released without being used.
949 * This ensures that xfs_inactive() will see that
950 * the inode is already free and not try to mess
951 * with the uninitialized part of it.
952 */
954 /*
955 * Initialize the per-fork minima and maxima for a new
956 * inode here. xfs_iformat will do it for old inodes.
957 */
960 }
964 /*
965 * The inode format changed when we moved the link count and
966 * made it 32 bits long. If this is an old format inode,
967 * convert it in memory to look like a new one. If it gets
968 * flushed to disk we will convert back before flushing or
969 * logging it. We zero out the new projid field and the old link
970 * count field. We'll handle clearing the pad field (the remains
971 * of the old uuid field) when we actually convert the inode to
972 * the new format. We don't change the version number so that we
973 * can distinguish this from a real new format inode.
974 */
979 }
983 /*
984 * Mark the buffer containing the inode as something to keep
985 * around for a while. This helps to keep recently accessed
986 * meta-data in-core longer.
987 */
990 /*
991 * Use xfs_trans_brelse() to release the buffer containing the
992 * on-disk inode, because it was acquired with xfs_trans_read_buf()
993 * in xfs_itobp() above. If tp is NULL, this is just a normal
994 * brelse(). If we're within a transaction, then xfs_trans_brelse()
995 * will only release the buffer if it is not dirty within the
996 * transaction. It will be OK to release the buffer in this case,
997 * because inodes on disk are never destroyed and we will be
998 * locking the new in-core inode before putting it in the hash
999 * table where other processes can find it. Thus we don't have
1000 * to worry about the inode being changed just because we released
1001 * the buffer.
1002 */
1006 }
1008 /*
1009 * Read in extents from a btree-format inode.
1010 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1011 */
1012 int
1017 {
1020 xfs_extnum_t nextents;
1027 }
1032 /*
1033 * We know that the size is valid (it's checked in iformat_btree)
1034 */
1044 }
1047 }
1049 /*
1050 * Allocate an inode on disk and return a copy of its in-core version.
1051 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1052 * appropriately within the inode. The uid and gid for the inode are
1053 * set according to the contents of the given cred structure.
1054 *
1055 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1056 * has a free inode available, call xfs_iget()
1057 * to obtain the in-core version of the allocated inode. Finally,
1058 * fill in the inode and log its initial contents. In this case,
1059 * ialloc_context would be set to NULL and call_again set to false.
1060 *
1061 * If xfs_dialloc() does not have an available inode,
1062 * it will replenish its supply by doing an allocation. Since we can
1063 * only do one allocation within a transaction without deadlocks, we
1064 * must commit the current transaction before returning the inode itself.
1065 * In this case, therefore, we will set call_again to true and return.
1066 * The caller should then commit the current transaction, start a new
1067 * transaction, and call xfs_ialloc() again to actually get the inode.
1068 *
1069 * To ensure that some other process does not grab the inode that
1070 * was allocated during the first call to xfs_ialloc(), this routine
1071 * also returns the [locked] bp pointing to the head of the freelist
1072 * as ialloc_context. The caller should hold this buffer across
1073 * the commit and pass it back into this routine on the second call.
1074 */
1075 int
1079 mode_t mode,
1080 xfs_nlink_t nlink,
1081 xfs_dev_t rdev,
1083 xfs_prid_t prid,
1088 {
1089 xfs_ino_t ino;
1092 uint flags;
1095 /*
1096 * Call the space management code to pick
1097 * the on-disk inode to be allocated.
1098 */
1103 }
1107 }
1110 /*
1111 * Get the in-core inode with the lock held exclusively.
1112 * This is because we're setting fields here we need
1113 * to prevent others from looking at until we're done.
1114 */
1119 }
1132 /*
1133 * If the superblock version is up to where we support new format
1134 * inodes and this is currently an old format inode, then change
1135 * the inode version number now. This way we only do the conversion
1136 * here rather than here and in the flush/logging code.
1137 */
1141 /*
1142 * We've already zeroed the old link count, the projid field,
1143 * and the pad field.
1144 */
1145 }
1147 /*
1148 * Project ids won't be stored on disk if we are using a version 1 inode.
1149 */
1157 }
1158 }
1160 /*
1161 * If the group ID of the new file does not match the effective group
1162 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1163 * (and only if the irix_sgid_inherit compatibility variable is set).
1164 */
1169 }
1175 /*
1176 * di_gen will have been taken care of in xfs_iread.
1177 */
1204 }
1209 }
1213 }
1214 }
1216 xfs_inherit_noatime)
1219 xfs_inherit_nodump)
1222 xfs_inherit_sync)
1225 xfs_inherit_nosymlinks)
1230 xfs_inherit_nodefrag)
1233 }
1234 /* FALLTHROUGH */
1243 }
1244 /*
1245 * Attribute fork settings for new inode.
1246 */
1250 /*
1251 * Log the new values stuffed into the inode.
1252 */
1255 /* now that we have an i_mode we can setup inode ops and unlock */
1260 }
1262 /*
1263 * Check to make sure that there are no blocks allocated to the
1264 * file beyond the size of the file. We don't check this for
1265 * files with fixed size extents or real time extents, but we
1266 * at least do it for regular files.
1267 */
1268 #ifdef DEBUG
1269 void
1273 xfs_fsize_t isize)
1274 {
1275 xfs_fileoff_t map_first;
1287 /*
1288 * The filesystem could be shutting down, so bmapi may return
1289 * an error.
1290 */
1294 map_first),
1300 }
1303 /*
1304 * Calculate the last possible buffered byte in a file. This must
1305 * include data that was buffered beyond the EOF by the write code.
1306 * This also needs to deal with overflowing the xfs_fsize_t type
1307 * which can happen for sizes near the limit.
1308 *
1309 * We also need to take into account any blocks beyond the EOF. It
1310 * may be the case that they were buffered by a write which failed.
1311 * In that case the pages will still be in memory, but the inode size
1312 * will never have been updated.
1313 */
1314 xfs_fsize_t
1317 {
1319 xfs_fsize_t last_byte;
1320 xfs_fileoff_t last_block;
1321 xfs_fileoff_t size_last_block;
1327 /*
1328 * Only check for blocks beyond the EOF if the extents have
1329 * been read in. This eliminates the need for the inode lock,
1330 * and it also saves us from looking when it really isn't
1331 * necessary.
1332 */
1335 XFS_DATA_FORK);
1338 }
1341 }
1348 }
1352 }
1354 }
1356 #if defined(XFS_RW_TRACE)
1357 STATIC void
1362 xfs_fsize_t new_size,
1363 xfs_off_t toss_start,
1364 xfs_off_t toss_finish)
1365 {
1368 }
1387 }
1388 #else
1389 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1390 #endif
1392 /*
1393 * Start the truncation of the file to new_size. The new size
1394 * must be smaller than the current size. This routine will
1395 * clear the buffer and page caches of file data in the removed
1396 * range, and xfs_itruncate_finish() will remove the underlying
1397 * disk blocks.
1398 *
1399 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1400 * must NOT have the inode lock held at all. This is because we're
1401 * calling into the buffer/page cache code and we can't hold the
1402 * inode lock when we do so.
1403 *
1404 * We need to wait for any direct I/Os in flight to complete before we
1405 * proceed with the truncate. This is needed to prevent the extents
1406 * being read or written by the direct I/Os from being removed while the
1407 * I/O is in flight as there is no other method of synchronising
1408 * direct I/O with the truncate operation. Also, because we hold
1409 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1410 * started until the truncate completes and drops the lock. Essentially,
1411 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1412 * between direct I/Os and the truncate operation.
1413 *
1414 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1415 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1416 * in the case that the caller is locking things out of order and
1417 * may not be able to call xfs_itruncate_finish() with the inode lock
1418 * held without dropping the I/O lock. If the caller must drop the
1419 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1420 * must be called again with all the same restrictions as the initial
1421 * call.
1422 */
1423 void
1426 uint flags,
1427 xfs_fsize_t new_size)
1428 {
1429 xfs_fsize_t last_byte;
1430 xfs_off_t toss_start;
1444 /*
1445 * Call toss_pages or flushinval_pages to get rid of pages
1446 * overlapping the region being removed. We have to use
1447 * the less efficient flushinval_pages in the case that the
1448 * caller may not be able to finish the truncate without
1449 * dropping the inode's I/O lock. Make sure
1450 * to catch any pages brought in by buffers overlapping
1451 * the EOF by searching out beyond the isize by our
1452 * block size. We round new_size up to a block boundary
1453 * so that we don't toss things on the same block as
1454 * new_size but before it.
1455 *
1456 * Before calling toss_page or flushinval_pages, make sure to
1457 * call remapf() over the same region if the file is mapped.
1458 * This frees up mapped file references to the pages in the
1459 * given range and for the flushinval_pages case it ensures
1460 * that we get the latest mapped changes flushed out.
1461 */
1465 /*
1466 * The place to start tossing is beyond our maximum
1467 * file size, so there is no way that the data extended
1468 * out there.
1469 */
1471 }
1474 last_byte);
1480 }
1481 }
1483 #ifdef DEBUG
1486 }
1487 #endif
1488 }
1490 /*
1491 * Shrink the file to the given new_size. The new
1492 * size must be smaller than the current size.
1493 * This will free up the underlying blocks
1494 * in the removed range after a call to xfs_itruncate_start()
1495 * or xfs_atruncate_start().
1496 *
1497 * The transaction passed to this routine must have made
1498 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1499 * This routine may commit the given transaction and
1500 * start new ones, so make sure everything involved in
1501 * the transaction is tidy before calling here.
1502 * Some transaction will be returned to the caller to be
1503 * committed. The incoming transaction must already include
1504 * the inode, and both inode locks must be held exclusively.
1505 * The inode must also be "held" within the transaction. On
1506 * return the inode will be "held" within the returned transaction.
1507 * This routine does NOT require any disk space to be reserved
1508 * for it within the transaction.
1509 *
1510 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1511 * and it indicates the fork which is to be truncated. For the
1512 * attribute fork we only support truncation to size 0.
1513 *
1514 * We use the sync parameter to indicate whether or not the first
1515 * transaction we perform might have to be synchronous. For the attr fork,
1516 * it needs to be so if the unlink of the inode is not yet known to be
1517 * permanent in the log. This keeps us from freeing and reusing the
1518 * blocks of the attribute fork before the unlink of the inode becomes
1519 * permanent.
1520 *
1521 * For the data fork, we normally have to run synchronously if we're
1522 * being called out of the inactive path or we're being called
1523 * out of the create path where we're truncating an existing file.
1524 * Either way, the truncate needs to be sync so blocks don't reappear
1525 * in the file with altered data in case of a crash. wsync filesystems
1526 * can run the first case async because anything that shrinks the inode
1527 * has to run sync so by the time we're called here from inactive, the
1528 * inode size is permanently set to 0.
1529 *
1530 * Calls from the truncate path always need to be sync unless we're
1531 * in a wsync filesystem and the file has already been unlinked.
1532 *
1533 * The caller is responsible for correctly setting the sync parameter.
1534 * It gets too hard for us to guess here which path we're being called
1535 * out of just based on inode state.
1536 */
1537 int
1541 xfs_fsize_t new_size,
1544 {
1545 xfs_fsblock_t first_block;
1546 xfs_fileoff_t first_unmap_block;
1547 xfs_fileoff_t last_block;
1553 xfs_bmap_free_t free_list;
1570 /*
1571 * We only support truncating the entire attribute fork.
1572 */
1575 }
1578 /*
1579 * The first thing we do is set the size to new_size permanently
1580 * on disk. This way we don't have to worry about anyone ever
1581 * being able to look at the data being freed even in the face
1582 * of a crash. What we're getting around here is the case where
1583 * we free a block, it is allocated to another file, it is written
1584 * to, and then we crash. If the new data gets written to the
1585 * file but the log buffers containing the free and reallocation
1586 * don't, then we'd end up with garbage in the blocks being freed.
1587 * As long as we make the new_size permanent before actually
1588 * freeing any blocks it doesn't matter if they get writtten to.
1589 *
1590 * The callers must signal into us whether or not the size
1591 * setting here must be synchronous. There are a few cases
1592 * where it doesn't have to be synchronous. Those cases
1593 * occur if the file is unlinked and we know the unlink is
1594 * permanent or if the blocks being truncated are guaranteed
1595 * to be beyond the inode eof (regardless of the link count)
1596 * and the eof value is permanent. Both of these cases occur
1597 * only on wsync-mounted filesystems. In those cases, we're
1598 * guaranteed that no user will ever see the data in the blocks
1599 * that are being truncated so the truncate can run async.
1600 * In the free beyond eof case, the file may wind up with
1601 * more blocks allocated to it than it needs if we crash
1602 * and that won't get fixed until the next time the file
1603 * is re-opened and closed but that's ok as that shouldn't
1604 * be too many blocks.
1605 *
1606 * However, we can't just make all wsync xactions run async
1607 * because there's one call out of the create path that needs
1608 * to run sync where it's truncating an existing file to size
1609 * 0 whose size is > 0.
1610 *
1611 * It's probably possible to come up with a test in this
1612 * routine that would correctly distinguish all the above
1613 * cases from the values of the function parameters and the
1614 * inode state but for sanity's sake, I've decided to let the
1615 * layers above just tell us. It's simpler to correctly figure
1616 * out in the layer above exactly under what conditions we
1617 * can run async and I think it's easier for others read and
1618 * follow the logic in case something has to be changed.
1619 * cscope is your friend -- rcc.
1620 *
1621 * The attribute fork is much simpler.
1622 *
1623 * For the attribute fork we allow the caller to tell us whether
1624 * the unlink of the inode that led to this call is yet permanent
1625 * in the on disk log. If it is not and we will be freeing extents
1626 * in this inode then we make the first transaction synchronous
1627 * to make sure that the unlink is permanent by the time we free
1628 * the blocks.
1629 */
1634 }
1639 }
1645 /*
1646 * Since it is possible for space to become allocated beyond
1647 * the end of the file (in a crash where the space is allocated
1648 * but the inode size is not yet updated), simply remove any
1649 * blocks which show up between the new EOF and the maximum
1650 * possible file size. If the first block to be removed is
1651 * beyond the maximum file size (ie it is the same as last_block),
1652 * then there is nothing to do.
1653 */
1661 }
1663 /*
1664 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1665 * will tell us whether it freed the entire range or
1666 * not. If this is a synchronous mount (wsync),
1667 * then we can tell bunmapi to keep all the
1668 * transactions asynchronous since the unlink
1669 * transaction that made this inode inactive has
1670 * already hit the disk. There's no danger of
1671 * the freed blocks being reused, there being a
1672 * crash, and the reused blocks suddenly reappearing
1673 * in this file with garbage in them once recovery
1674 * runs.
1675 */
1681 XFS_ITRUNC_MAX_EXTENTS,
1685 /*
1686 * If the bunmapi call encounters an error,
1687 * return to the caller where the transaction
1688 * can be properly aborted. We just need to
1689 * make sure we're not holding any resources
1690 * that we were not when we came in.
1691 */
1694 }
1696 /*
1697 * Duplicate the transaction that has the permanent
1698 * reservation and commit the old transaction.
1699 */
1704 /*
1705 * If the bmap finish call encounters an error,
1706 * return to the caller where the transaction
1707 * can be properly aborted. We just need to
1708 * make sure we're not holding any resources
1709 * that we were not when we came in.
1710 *
1711 * Aborting from this point might lose some
1712 * blocks in the file system, but oh well.
1713 */
1716 /*
1717 * If the passed in transaction committed
1718 * in xfs_bmap_finish(), then we want to
1719 * add the inode to this one before returning.
1720 * This keeps things simple for the higher
1721 * level code, because it always knows that
1722 * the inode is locked and held in the
1723 * transaction that returns to it whether
1724 * errors occur or not. We don't mark the
1725 * inode dirty so that this transaction can
1726 * be easily aborted if possible.
1727 */
1731 }
1733 }
1736 /*
1737 * The first xact was committed,
1738 * so add the inode to the new one.
1739 * Mark it dirty so it will be logged
1740 * and moved forward in the log as
1741 * part of every commit.
1742 */
1747 }
1752 XFS_TRANS_PERM_LOG_RES,
1753 XFS_ITRUNCATE_LOG_COUNT);
1754 /*
1755 * Add the inode being truncated to the next chained
1756 * transaction.
1757 */
1762 }
1763 /*
1764 * Only update the size in the case of the data fork, but
1765 * always re-log the inode so that our permanent transaction
1766 * can keep on rolling it forward in the log.
1767 */
1771 }
1781 }
1784 /*
1785 * xfs_igrow_start
1786 *
1787 * Do the first part of growing a file: zero any data in the last
1788 * block that is beyond the old EOF. We need to do this before
1789 * the inode is joined to the transaction to modify the i_size.
1790 * That way we can drop the inode lock and call into the buffer
1791 * cache to get the buffer mapping the EOF.
1792 */
1793 int
1796 xfs_fsize_t new_size,
1798 {
1805 /*
1806 * Zero any pages that may have been created by
1807 * xfs_write_file() beyond the end of the file
1808 * and any blocks between the old and new file sizes.
1809 */
1813 }
1815 /*
1816 * xfs_igrow_finish
1817 *
1818 * This routine is called to extend the size of a file.
1819 * The inode must have both the iolock and the ilock locked
1820 * for update and it must be a part of the current transaction.
1821 * The xfs_igrow_start() function must have been called previously.
1822 * If the change_flag is not zero, the inode change timestamp will
1823 * be updated.
1824 */
1825 void
1829 xfs_fsize_t new_size,
1831 {
1837 /*
1838 * Update the file size. Update the inode change timestamp
1839 * if change_flag set.
1840 */
1846 }
1849 /*
1850 * This is called when the inode's link count goes to 0.
1851 * We place the on-disk inode on a list in the AGI. It
1852 * will be pulled from this list when the inode is freed.
1853 */
1854 int
1858 {
1864 xfs_agnumber_t agno;
1865 xfs_daddr_t agdaddr;
1866 xfs_agino_t agino;
1881 /*
1882 * Get the agi buffer first. It ensures lock ordering
1883 * on the list.
1884 */
1889 }
1890 /*
1891 * Validate the magic number of the agi block.
1892 */
1894 agi_ok =
1898 XFS_RANDOM_IUNLINK))) {
1902 }
1903 /*
1904 * Get the index into the agi hash table for the
1905 * list this inode will go on.
1906 */
1914 /*
1915 * There is already another inode in the bucket we need
1916 * to add ourselves to. Add us at the front of the list.
1917 * Here we put the head pointer into our next pointer,
1918 * and then we fall through to point the head at us.
1919 */
1923 }
1926 /* both on-disk, don't endian flip twice */
1934 }
1936 /*
1937 * Point the bucket head pointer at the inode being inserted.
1938 */
1946 }
1948 /*
1949 * Pull the on-disk inode from the AGI unlinked list.
1950 */
1951 STATIC int
1955 {
1956 xfs_ino_t next_ino;
1962 xfs_agnumber_t agno;
1963 xfs_daddr_t agdaddr;
1964 xfs_agino_t agino;
1965 xfs_agino_t next_agino;
1973 /*
1974 * First pull the on-disk inode from the AGI unlinked list.
1975 */
1981 /*
1982 * Get the agi buffer first. It ensures lock ordering
1983 * on the list.
1984 */
1992 }
1993 /*
1994 * Validate the magic number of the agi block.
1995 */
1997 agi_ok =
2001 XFS_RANDOM_IUNLINK_REMOVE))) {
2009 }
2010 /*
2011 * Get the index into the agi hash table for the
2012 * list this inode will go on.
2013 */
2021 /*
2022 * We're at the head of the list. Get the inode's
2023 * on-disk buffer to see if there is anyone after us
2024 * on the list. Only modify our next pointer if it
2025 * is not already NULLAGINO. This saves us the overhead
2026 * of dealing with the buffer when there is no need to
2027 * change it.
2028 */
2035 }
2048 }
2049 /*
2050 * Point the bucket head pointer at the next inode.
2051 */
2060 /*
2061 * We need to search the list for the inode being freed.
2062 */
2066 /*
2067 * If the last inode wasn't the one pointing to
2068 * us, then release its buffer since we're not
2069 * going to do anything with it.
2070 */
2073 }
2082 }
2086 }
2087 /*
2088 * Now last_ibp points to the buffer previous to us on
2089 * the unlinked list. Pull us from the list.
2090 */
2097 }
2111 }
2112 /*
2113 * Point the previous inode on the list to the next inode.
2114 */
2122 }
2124 }
2127 {
2131 }
2133 STATIC void
2137 xfs_ino_t inum)
2138 {
2144 xfs_daddr_t blkno;
2161 }
2170 /*
2171 * Look for each inode in memory and attempt to lock it,
2172 * we can be racing with flush and tail pushing here.
2173 * any inode we get the locks on, add to an array of
2174 * inode items to process later.
2175 *
2176 * The get the buffer lock, we could beat a flush
2177 * or tail pushing thread to the lock here, in which
2178 * case they will go looking for the inode buffer
2179 * and fail, we need some other form of interlock
2180 * here.
2181 */
2189 }
2191 /* Inode not in memory or we found it already,
2192 * nothing to do
2193 */
2197 }
2202 }
2204 /* If we can get the locks then add it to the
2205 * list, otherwise by the time we get the bp lock
2206 * below it will already be attached to the
2207 * inode buffer.
2208 */
2210 /* This inode will already be locked - by us, lets
2211 * keep it that way.
2212 */
2222 }
2223 }
2226 }
2237 }
2240 }
2241 }
2244 }
2248 XFS_BUF_LOCK);
2261 pre_flushed++;
2262 }
2264 }
2275 }
2289 }
2290 }
2295 }
2298 }
2300 /*
2301 * This is called to return an inode to the inode free list.
2302 * The inode should already be truncated to 0 length and have
2303 * no pages associated with it. This routine also assumes that
2304 * the inode is already a part of the transaction.
2305 *
2306 * The on-disk copy of the inode will have been added to the list
2307 * of unlinked inodes in the AGI. We need to remove the inode from
2308 * that list atomically with respect to freeing it here.
2309 */
2310 int
2315 {
2318 xfs_ino_t first_ino;
2329 /*
2330 * Pull the on-disk inode from the AGI unlinked list.
2331 */
2335 }
2340 }
2349 /*
2350 * Bump the generation count so no one will be confused
2351 * by reincarnations of this inode.
2352 */
2358 }
2361 }
2363 /*
2364 * Reallocate the space for if_broot based on the number of records
2365 * being added or deleted as indicated in rec_diff. Move the records
2366 * and pointers in if_broot to fit the new size. When shrinking this
2367 * will eliminate holes between the records and pointers created by
2368 * the caller. When growing this will create holes to be filled in
2369 * by the caller.
2370 *
2371 * The caller must not request to add more records than would fit in
2372 * the on-disk inode root. If the if_broot is currently NULL, then
2373 * if we adding records one will be allocated. The caller must also
2374 * not request that the number of records go below zero, although
2375 * it can go to zero.
2376 *
2377 * ip -- the inode whose if_broot area is changing
2378 * ext_diff -- the change in the number of records, positive or negative,
2379 * requested for the if_broot array.
2380 */
2381 void
2386 {
2395 /*
2396 * Handle the degenerate case quietly.
2397 */
2400 }
2404 /*
2405 * If there wasn't any memory allocated before, just
2406 * allocate it now and get out.
2407 */
2411 KM_SLEEP);
2414 }
2416 /*
2417 * If there is already an existing if_broot, then we need
2418 * to realloc() it and shift the pointers to their new
2419 * location. The records don't change location because
2420 * they are kept butted up against the btree block header.
2421 */
2427 new_size,
2429 KM_SLEEP);
2439 }
2441 /*
2442 * rec_diff is less than 0. In this case, we are shrinking the
2443 * if_broot buffer. It must already exist. If we go to zero
2444 * records, just get rid of the root and clear the status bit.
2445 */
2452 else
2456 /*
2457 * First copy over the btree block header.
2458 */
2463 }
2465 /*
2466 * Only copy the records and pointers if there are any.
2467 */
2469 /*
2470 * First copy the records.
2471 */
2478 /*
2479 * Then copy the pointers.
2480 */
2486 }
2493 }
2496 /*
2497 * This is called when the amount of space needed for if_data
2498 * is increased or decreased. The change in size is indicated by
2499 * the number of bytes that need to be added or deleted in the
2500 * byte_diff parameter.
2501 *
2502 * If the amount of space needed has decreased below the size of the
2503 * inline buffer, then switch to using the inline buffer. Otherwise,
2504 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2505 * to what is needed.
2506 *
2507 * ip -- the inode whose if_data area is changing
2508 * byte_diff -- the change in the number of bytes, positive or negative,
2509 * requested for the if_data array.
2510 */
2511 void
2516 {
2523 }
2532 }
2536 /*
2537 * If the valid extents/data can fit in if_inline_ext/data,
2538 * copy them from the malloc'd vector and free it.
2539 */
2545 new_size);
2548 }
2551 /*
2552 * Stuck with malloc/realloc.
2553 * For inline data, the underlying buffer must be
2554 * a multiple of 4 bytes in size so that it can be
2555 * logged and stay on word boundaries. We enforce
2556 * that here.
2557 */
2563 /*
2564 * Only do the realloc if the underlying size
2565 * is really changing.
2566 */
2570 real_size,
2572 KM_SLEEP);
2573 }
2579 }
2580 }
2584 }
2589 /*
2590 * Map inode to disk block and offset.
2591 *
2592 * mp -- the mount point structure for the current file system
2593 * tp -- the current transaction
2594 * ino -- the inode number of the inode to be located
2595 * imap -- this structure is filled in with the information necessary
2596 * to retrieve the given inode from disk
2597 * flags -- flags to pass to xfs_dilocate indicating whether or not
2598 * lookups in the inode btree were OK or not
2599 */
2600 int
2604 xfs_ino_t ino,
2606 uint flags)
2607 {
2608 xfs_fsblock_t fsbno;
2618 }
2625 }
2627 void
2631 {
2638 }
2640 /*
2641 * If the format is local, then we can't have an extents
2642 * array so just look for an inline data array. If we're
2643 * not local then we may or may not have an extents list,
2644 * so check and free it up if we do.
2645 */
2653 }
2660 }
2667 }
2668 }
2670 /*
2671 * This is called free all the memory associated with an inode.
2672 * It must free the inode itself and any buffers allocated for
2673 * if_extents/if_data and if_broot. It must also free the lock
2674 * associated with the inode.
2675 */
2676 void
2679 {
2687 }
2693 #ifdef XFS_BMAP_TRACE
2695 #endif
2696 #ifdef XFS_BMBT_TRACE
2698 #endif
2699 #ifdef XFS_RW_TRACE
2701 #endif
2702 #ifdef XFS_ILOCK_TRACE
2704 #endif
2705 #ifdef XFS_DIR2_TRACE
2707 #endif
2709 /* XXXdpd should be able to assert this but shutdown
2710 * is leaving the AIL behind. */
2714 }
2716 }
2719 /*
2720 * Increment the pin count of the given buffer.
2721 * This value is protected by ipinlock spinlock in the mount structure.
2722 */
2723 void
2726 {
2730 }
2732 /*
2733 * Decrement the pin count of the given inode, and wake up
2734 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2735 * inode must have been previously pinned with a call to xfs_ipin().
2736 */
2737 void
2740 {
2744 /*
2745 * If the inode is currently being reclaimed, the
2746 * linux inode _and_ the xfs vnode may have been
2747 * freed so we cannot reference either of them safely.
2748 * Hence we should not try to do anything to them
2749 * if the xfs inode is currently in the reclaim
2750 * path.
2751 *
2752 * However, we still need to issue the unpin wakeup
2753 * call as the inode reclaim may be blocked waiting for
2754 * the inode to become unpinned.
2755 */
2759 /* make sync come back and flush this inode */
2766 }
2767 }
2769 }
2770 }
2772 /*
2773 * This is called to wait for the given inode to be unpinned.
2774 * It will sleep until this happens. The caller must have the
2775 * inode locked in at least shared mode so that the buffer cannot
2776 * be subsequently pinned once someone is waiting for it to be
2777 * unpinned.
2778 */
2779 STATIC void
2782 {
2784 xfs_lsn_t lsn;
2790 }
2797 }
2799 /*
2800 * Give the log a push so we don't wait here too long.
2801 */
2805 }
2808 /*
2809 * xfs_iextents_copy()
2810 *
2811 * This is called to copy the REAL extents (as opposed to the delayed
2812 * allocation extents) from the inode into the given buffer. It
2813 * returns the number of bytes copied into the buffer.
2814 *
2815 * If there are no delayed allocation extents, then we can just
2816 * memcpy() the extents into the buffer. Otherwise, we need to
2817 * examine each extent in turn and skip those which are delayed.
2818 */
2819 int
2824 {
2828 #ifdef XFS_BMAP_TRACE
2830 #endif
2834 xfs_fsblock_t start_block;
2844 /*
2845 * There are some delayed allocation extents in the
2846 * inode, so copy the extents one at a time and skip
2847 * the delayed ones. There must be at least one
2848 * non-delayed extent.
2849 */
2856 /*
2857 * It's a delayed allocation extent, so skip it.
2858 */
2860 }
2862 /* Translate to on disk format */
2867 dest_ep++;
2868 copied++;
2869 }
2874 }
2876 /*
2877 * Each of the following cases stores data into the same region
2878 * of the on-disk inode, so only one of them can be valid at
2879 * any given time. While it is possible to have conflicting formats
2880 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2881 * in EXTENTS format, this can only happen when the fork has
2882 * changed formats after being modified but before being flushed.
2883 * In these cases, the format always takes precedence, because the
2884 * format indicates the current state of the fork.
2885 */
2886 /*ARGSUSED*/
2887 STATIC int
2894 {
2898 #ifdef XFS_TRANS_DEBUG
2900 #endif
2911 /*
2912 * This can happen if we gave up in iformat in an error path,
2913 * for the attribute fork.
2914 */
2918 }
2928 }
2942 whichfork);
2943 }
2952 XFS_BROOT_SIZE_ADJ));
2956 }
2963 }
2971 }
2977 }
2980 }
2982 /*
2983 * xfs_iflush() will write a modified inode's changes out to the
2984 * inode's on disk home. The caller must have the inode lock held
2985 * in at least shared mode and the inode flush semaphore must be
2986 * held as well. The inode lock will still be held upon return from
2987 * the call and the caller is free to unlock it.
2988 * The inode flush lock will be unlocked when the inode reaches the disk.
2989 * The flags indicate how the inode's buffer should be written out.
2990 */
2991 int
2994 uint flags)
2995 {
3001 /* REFERENCED */
3019 /*
3020 * If the inode isn't dirty, then just release the inode
3021 * flush lock and do nothing.
3022 */
3029 }
3031 /*
3032 * We can't flush the inode until it is unpinned, so
3033 * wait for it. We know noone new can pin it, because
3034 * we are holding the inode lock shared and you need
3035 * to hold it exclusively to pin the inode.
3036 */
3039 /*
3040 * This may have been unpinned because the filesystem is shutting
3041 * down forcibly. If that's the case we must not write this inode
3042 * to disk, because the log record didn't make it to disk!
3043 */
3050 }
3052 /*
3053 * Get the buffer containing the on-disk inode.
3054 */
3059 }
3061 /*
3062 * Decide how buffer will be flushed out. This is done before
3063 * the call to xfs_iflush_int because this field is zeroed by it.
3064 */
3066 /*
3067 * Flush out the inode buffer according to the directions
3068 * of the caller. In the cases where the caller has given
3069 * us a choice choose the non-delwri case. This is because
3070 * the inode is in the AIL and we need to get it out soon.
3071 */
3088 }
3106 }
3107 }
3109 /*
3110 * First flush out the inode that xfs_iflush was called with.
3111 */
3115 }
3117 /*
3118 * inode clustering:
3119 * see if other inodes can be gathered into this write
3120 */
3129 /*
3130 * Do an un-protected check to see if the inode is dirty and
3131 * is a candidate for flushing. These checks will be repeated
3132 * later after the appropriate locks are acquired.
3133 */
3140 }
3142 /*
3143 * Try to get locks. If any are unavailable,
3144 * then this inode cannot be flushed and is skipped.
3145 */
3147 /* get inode locks (just i_lock) */
3149 /* get inode flush lock */
3151 /* check if pinned */
3153 /* arriving here means that
3154 * this inode can be flushed.
3155 * first re-check that it's
3156 * dirty
3157 */
3162 clcount++;
3166 XFS_ILOCK_SHARED);
3168 }
3171 }
3174 }
3175 }
3177 }
3178 }
3184 }
3186 /*
3187 * If the buffer is pinned then push on the log so we won't
3188 * get stuck waiting in the write for too long.
3189 */
3192 }
3200 }
3203 corrupt_out:
3207 /*
3208 * Unlocks the flush lock
3209 */
3212 cluster_corrupt_out:
3213 /* Corruption detected in the clustering loop. Invalidate the
3214 * inode buffer and shut down the filesystem.
3215 */
3218 /*
3219 * Clean up the buffer. If it was B_DELWRI, just release it --
3220 * brelse can handle it with no problems. If not, shut down the
3221 * filesystem before releasing the buffer.
3222 */
3225 }
3230 /*
3231 * Just like incore_relse: if we have b_iodone functions,
3232 * mark the buffer as an error and call them. Otherwise
3233 * mark it as stale and brelse.
3234 */
3245 }
3246 }
3249 /*
3250 * Unlocks the flush lock
3251 */
3253 }
3256 STATIC int
3260 {
3264 #ifdef XFS_TRANS_DEBUG
3266 #endif
3278 /*
3279 * If the inode isn't dirty, then just release the inode
3280 * flush lock and do nothing.
3281 */
3286 }
3288 /* set *dip = inode's place in the buffer */
3291 /*
3292 * Clear i_update_core before copying out the data.
3293 * This is for coordination with our timestamp updates
3294 * that don't hold the inode lock. They will always
3295 * update the timestamps BEFORE setting i_update_core,
3296 * so if we clear i_update_core after they set it we
3297 * are guaranteed to see their updates to the timestamps.
3298 * I believe that this depends on strongly ordered memory
3299 * semantics, but we have that. We use the SYNCHRONIZE
3300 * macro to make sure that the compiler does not reorder
3301 * the i_update_core access below the data copy below.
3302 */
3306 /*
3307 * Make sure to get the latest atime from the Linux inode.
3308 */
3317 }
3324 }
3334 }
3345 }
3346 }
3349 XFS_RANDOM_IFLUSH_5)) {
3355 ip);
3357 }
3364 }
3365 /*
3366 * bump the flush iteration count, used to detect flushes which
3367 * postdate a log record during recovery.
3368 */
3372 /*
3373 * Copy the dirty parts of the inode into the on-disk
3374 * inode. We always copy out the core of the inode,
3375 * because if the inode is dirty at all the core must
3376 * be.
3377 */
3380 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3384 /*
3385 * If this is really an old format inode and the superblock version
3386 * has not been updated to support only new format inodes, then
3387 * convert back to the old inode format. If the superblock version
3388 * has been updated, then make the conversion permanent.
3389 */
3394 /*
3395 * Convert it back.
3396 */
3400 /*
3401 * The superblock version has already been bumped,
3402 * so just make the conversion to the new inode
3403 * format permanent.
3404 */
3413 }
3414 }
3418 }
3421 /*
3422 * The only error from xfs_iflush_fork is on the data fork.
3423 */
3425 }
3428 /*
3429 * We've recorded everything logged in the inode, so we'd
3430 * like to clear the ilf_fields bits so we don't log and
3431 * flush things unnecessarily. However, we can't stop
3432 * logging all this information until the data we've copied
3433 * into the disk buffer is written to disk. If we did we might
3434 * overwrite the copy of the inode in the log with all the
3435 * data after re-logging only part of it, and in the face of
3436 * a crash we wouldn't have all the data we need to recover.
3437 *
3438 * What we do is move the bits to the ili_last_fields field.
3439 * When logging the inode, these bits are moved back to the
3440 * ilf_fields field. In the xfs_iflush_done() routine we
3441 * clear ili_last_fields, since we know that the information
3442 * those bits represent is permanently on disk. As long as
3443 * the flush completes before the inode is logged again, then
3444 * both ilf_fields and ili_last_fields will be cleared.
3445 *
3446 * We can play with the ilf_fields bits here, because the inode
3447 * lock must be held exclusively in order to set bits there
3448 * and the flush lock protects the ili_last_fields bits.
3449 * Set ili_logged so the flush done
3450 * routine can tell whether or not to look in the AIL.
3451 * Also, store the current LSN of the inode so that we can tell
3452 * whether the item has moved in the AIL from xfs_iflush_done().
3453 * In order to read the lsn we need the AIL lock, because
3454 * it is a 64 bit value that cannot be read atomically.
3455 */
3466 /*
3467 * Attach the function xfs_iflush_done to the inode's
3468 * buffer. This will remove the inode from the AIL
3469 * and unlock the inode's flush lock when the inode is
3470 * completely written to disk.
3471 */
3478 /*
3479 * We're flushing an inode which is not in the AIL and has
3480 * not been logged but has i_update_core set. For this
3481 * case we can use a B_DELWRI flush and immediately drop
3482 * the inode flush lock because we can avoid the whole
3483 * AIL state thing. It's OK to drop the flush lock now,
3484 * because we've already locked the buffer and to do anything
3485 * you really need both.
3486 */
3491 }
3493 }
3497 corrupt_out:
3499 }
3502 /*
3503 * Flush all inactive inodes in mp.
3504 */
3505 void
3508 {
3512 again:
3519 /* Make sure we skip markers inserted by sync */
3523 }
3530 }
3536 out:
3538 }
3540 /*
3541 * xfs_iaccess: check accessibility of inode for mode.
3542 */
3543 int
3546 mode_t mode,
3548 {
3562 }
3564 /*
3565 * If there's an Access Control List it's used instead of
3566 * the mode bits.
3567 */
3575 }
3577 /*
3578 * If the DACs are ok we don't need any capability check.
3579 */
3582 /*
3583 * Read/write DACs are always overridable.
3584 * Executable DACs are overridable if at least one exec bit is set.
3585 */
3595 #ifdef NOISE
3599 }
3601 }
3603 /*
3604 * xfs_iroundup: round up argument to next power of two
3605 */
3606 uint
3608 uint v)
3609 {
3611 uint m;
3624 }
3627 }
3629 #ifdef XFS_ILOCK_TRACE
3632 void
3634 {
3643 }
3644 #endif
3646 /*
3647 * Return a pointer to the extent record at file index idx.
3648 */
3649 xfs_bmbt_rec_t *
3653 {
3668 }
3669 }
3671 /*
3672 * Insert new item(s) into the extent records for incore inode
3673 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3674 */
3675 void
3681 {
3690 }
3691 }
3693 /*
3694 * This is called when the amount of space required for incore file
3695 * extents needs to be increased. The ext_diff parameter stores the
3696 * number of new extents being added and the idx parameter contains
3697 * the extent index where the new extents will be added. If the new
3698 * extents are being appended, then we just need to (re)allocate and
3699 * initialize the space. Otherwise, if the new extents are being
3700 * inserted into the middle of the existing entries, a bit more work
3701 * is required to make room for the new extents to be inserted. The
3702 * caller is responsible for filling in the new extent entries upon
3703 * return.
3704 */
3705 void
3710 {
3719 /*
3720 * If the new number of extents (nextents + ext_diff)
3721 * fits inside the inode, then continue to use the inline
3722 * extent buffer.
3723 */
3730 }
3734 }
3735 /*
3736 * Otherwise use a linear (direct) extent list.
3737 * If the extents are currently inside the inode,
3738 * xfs_iext_realloc_direct will switch us from
3739 * inline to direct extent allocation mode.
3740 */
3748 }
3749 }
3750 /* Indirection array */
3763 }
3764 /* Extents fit in target extent page */
3772 }
3775 }
3776 /* Insert a new extent page */
3780 }
3781 /*
3782 * If extent(s) are being appended to the last page in
3783 * the indirection array and the new extent(s) don't fit
3784 * in the page, then erp is NULL and erp_idx is set to
3785 * the next index needed in the indirection array.
3786 */
3795 erp_idx++;
3796 }
3797 }
3798 }
3799 }
3801 }
3803 /*
3804 * This is called when incore extents are being added to the indirection
3805 * array and the new extents do not fit in the target extent list. The
3806 * erp_idx parameter contains the irec index for the target extent list
3807 * in the indirection array, and the idx parameter contains the extent
3808 * index within the list. The number of extents being added is stored
3809 * in the count parameter.
3810 *
3811 * |-------| |-------|
3812 * | | | | idx - number of extents before idx
3813 * | idx | | count |
3814 * | | | | count - number of extents being inserted at idx
3815 * |-------| |-------|
3816 * | count | | nex2 | nex2 - number of extents after idx + count
3817 * |-------| |-------|
3818 */
3819 void
3825 {
3839 /*
3840 * Save second part of target extent list
3841 * (all extents past */
3849 }
3851 /*
3852 * Add the new extents to the end of the target
3853 * list, then allocate new irec record(s) and
3854 * extent buffer(s) as needed to store the rest
3855 * of the new extents.
3856 */
3863 }
3865 erp_idx++;
3871 }
3873 /* Add nex2 extents back to indirection array */
3875 xfs_extnum_t ext_avail;
3881 /*
3882 * If nex2 extents fit in the current page, append
3883 * nex2_ep after the new extents.
3884 */
3887 }
3888 /*
3889 * Otherwise, check if space is available in the
3890 * next page.
3891 */
3895 erp_idx++;
3896 erp++;
3897 /* Create a hole for nex2 extents */
3900 }
3901 /*
3902 * Final choice, create a new extent page for
3903 * nex2 extents.
3904 */
3906 erp_idx++;
3908 }
3913 }
3914 }
3916 /*
3917 * This is called when the amount of space required for incore file
3918 * extents needs to be decreased. The ext_diff parameter stores the
3919 * number of extents to be removed and the idx parameter contains
3920 * the extent index where the extents will be removed from.
3921 *
3922 * If the amount of space needed has decreased below the linear
3923 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3924 * extent array. Otherwise, use kmem_realloc() to adjust the
3925 * size to what is needed.
3926 */
3927 void
3932 {
3948 }
3950 }
3952 /*
3953 * This removes ext_diff extents from the inline buffer, beginning
3954 * at extent index idx.
3955 */
3956 void
3961 {
3980 }
3981 }
3983 /*
3984 * This removes ext_diff extents from a linear (direct) extent list,
3985 * beginning at extent index idx. If the extents are being removed
3986 * from the end of the list (ie. truncate) then we just need to re-
3987 * allocate the list to remove the extra space. Otherwise, if the
3988 * extents are being removed from the middle of the existing extent
3989 * entries, then we first need to move the extent records beginning
3990 * at idx + ext_diff up in the list to overwrite the records being
3991 * removed, then remove the extra space via kmem_realloc.
3992 */
3993 void
3998 {
4010 }
4011 /* Move extents up in the list (if needed) */
4017 }
4020 /*
4021 * Reallocate the direct extent list. If the extents
4022 * will fit inside the inode then xfs_iext_realloc_direct
4023 * will switch from direct to inline extent allocation
4024 * mode for us.
4025 */
4028 }
4030 /*
4031 * This is called when incore extents are being removed from the
4032 * indirection array and the extents being removed span multiple extent
4033 * buffers. The idx parameter contains the file extent index where we
4034 * want to begin removing extents, and the count parameter contains
4035 * how many extents need to be removed.
4036 *
4037 * |-------| |-------|
4038 * | nex1 | | | nex1 - number of extents before idx
4039 * |-------| | count |
4040 * | | | | count - number of extents being removed at idx
4041 * | count | |-------|
4042 * | | | nex2 | nex2 - number of extents after idx + count
4043 * |-------| |-------|
4044 */
4045 void
4050 {
4069 /*
4070 * Check for deletion of entire list;
4071 * xfs_iext_irec_remove() updates extent offsets.
4072 */
4079 XFS_IEXT_BUFSZ);
4085 }
4086 }
4087 /* Move extents up (if needed) */
4092 }
4093 /* Zero out rest of page */
4096 /* Update remaining counters */
4101 erp_idx++;
4102 erp++;
4103 }
4106 }
4108 /*
4109 * Create, destroy, or resize a linear (direct) block of extents.
4110 */
4111 void
4115 {
4124 /* Free extent records */
4127 }
4128 /* Resize direct extent list and zero any new bytes */
4130 /* Check if extents will fit inside the inode */
4136 }
4139 }
4143 rnew_size,
4145 KM_SLEEP);
4146 }
4151 }
4152 }
4153 /*
4154 * Switch from the inline extent buffer to a direct
4155 * extent list. Be sure to include the inline extent
4156 * bytes in new_size.
4157 */
4162 }
4164 }
4167 }
4169 /*
4170 * Switch from linear (direct) extent records to inline buffer.
4171 */
4172 void
4176 {
4179 /*
4180 * The inline buffer was zeroed when we switched
4181 * from inline to direct extent allocation mode,
4182 * so we don't need to clear it here.
4183 */
4189 }
4191 /*
4192 * Switch from inline buffer to linear (direct) extent records.
4193 * new_size should already be rounded up to the next power of 2
4194 * by the caller (when appropriate), so use new_size as it is.
4195 * However, since new_size may be rounded up, we can't update
4196 * if_bytes here. It is the caller's responsibility to update
4197 * if_bytes upon return.
4198 */
4199 void
4203 {
4212 }
4214 }
4216 /*
4217 * Resize an extent indirection array to new_size bytes.
4218 */
4219 void
4223 {
4238 }
4239 }
4241 /*
4242 * Switch from indirection array to linear (direct) extent allocations.
4243 */
4244 void
4247 {
4267 }
4268 }
4270 /*
4271 * Free incore file extents.
4272 */
4273 void
4276 {
4284 }
4291 }
4295 }
4297 /*
4298 * Return a pointer to the extent record for file system block bno.
4299 */
4305 {
4320 }
4323 /* Find target extent list */
4331 }
4332 /* Binary search extent records */
4343 /* Convert back to file-based extent index */
4346 }
4349 }
4350 }
4351 /* Convert back to file-based extent index */
4354 }
4360 }
4361 }
4364 }
4366 /*
4367 * Return a pointer to the indirection array entry containing the
4368 * extent record for filesystem block bno. Store the index of the
4369 * target irec in *erp_idxp.
4370 */
4376 {
4400 }
4401 }
4404 }
4406 /*
4407 * Return a pointer to the indirection array entry containing the
4408 * extent record at file extent index *idxp. Store the index of the
4409 * target irec in *erp_idxp and store the page index of the target
4410 * extent record in *idxp.
4411 */
4412 xfs_ext_irec_t *
4418 {
4435 /* Binary search extent irec's */
4451 erp_idx++;
4457 }
4458 }
4462 }
4464 /*
4465 * Allocate and initialize an indirection array once the space needed
4466 * for incore extents increases above XFS_IEXT_BUFSZ.
4467 */
4468 void
4471 {
4489 }
4500 }
4502 /*
4503 * Allocate and initialize a new entry in the indirection array.
4504 */
4505 xfs_ext_irec_t *
4509 {
4517 /* Resize indirection array */
4520 /*
4521 * Move records down in the array so the
4522 * new page can use erp_idx.
4523 */
4527 }
4530 /* Initialize new extent record */
4540 }
4542 /*
4543 * Remove a record from the indirection array.
4544 */
4545 void
4549 {
4561 }
4562 /* Compact extent records */
4566 }
4567 /*
4568 * Manually free the last extent record from the indirection
4569 * array. A call to xfs_iext_realloc_indirect() with a size
4570 * of zero would result in a call to xfs_iext_destroy() which
4571 * would in turn call this function again, creating a nasty
4572 * infinite loop.
4573 */
4580 }
4582 }
4584 /*
4585 * This is called to clean up large amounts of unused memory allocated
4586 * by the indirection array. Before compacting anything though, verify
4587 * that the indirection array is still needed and switch back to the
4588 * linear extent list (or even the inline buffer) if possible. The
4589 * compaction policy is as follows:
4590 *
4591 * Full Compaction: Extents fit into a single page (or inline buffer)
4592 * Full Compaction: Extents occupy less than 10% of allocated space
4593 * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
4594 * No Compaction: Extents occupy at least 50% of allocated space
4595 */
4596 void
4599 {
4618 }
4619 }
4621 /*
4622 * Combine extents from neighboring extent pages.
4623 */
4624 void
4627 {
4643 /*
4644 * Free page before removing extent record
4645 * so er_extoffs don't get modified in
4646 * xfs_iext_irec_remove.
4647 */
4653 erp_idx++;
4654 }
4655 }
4656 }
4658 /*
4659 * Fully compact the extent records managed by the indirection array.
4660 */
4661 void
4664 {
4684 /* Remove next page */
4686 /*
4687 * Free page before removing extent record
4688 * so er_extoffs don't get modified in
4689 * xfs_iext_irec_remove.
4690 */
4697 /* Update next page */
4699 /* Move rest of page up to become next new page */
4706 }
4708 erp_idx++;
4711 else
4713 }
4717 }
4718 }
4720 /*
4721 * This is called to update the er_extoff field in the indirection
4722 * array when extents have been added or removed from one of the
4723 * extent lists. erp_idx contains the irec index to begin updating
4724 * at and ext_diff contains the number of extents that were added
4725 * or removed.
4726 */
4727 void
4732 {
4740 }
4741 }