| blob | 3ca5d43b83456ccdf8ca4c61f13b363d13c9e76c |
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 */
57 /*
58 * Used in xfs_itruncate(). This is the maximum number of extents
59 * freed from a file in a single transaction.
60 */
61 #define XFS_ITRUNC_MAX_EXTENTS 2
69 #ifdef DEBUG
70 /*
71 * Make sure that the extents in the given memory buffer
72 * are valid.
73 */
74 STATIC void
79 xfs_exntfmt_t fmt)
80 {
82 xfs_bmbt_irec_t irec;
83 xfs_bmbt_rec_t rec;
92 else
96 }
97 }
99 #define xfs_validate_extents(ifp, nrecs, disk, fmt)
102 /*
103 * Check that none of the inode's in the buffer have a next
104 * unlinked field of 0.
105 */
106 #if defined(DEBUG)
107 void
111 {
123 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
124 bp);
126 }
127 }
128 }
129 #endif
131 /*
132 * This routine is called to map an inode number within a file
133 * system to the buffer containing the on-disk version of the
134 * inode. It returns a pointer to the buffer containing the
135 * on-disk inode in the bpp parameter, and in the dip parameter
136 * it returns a pointer to the on-disk inode within that buffer.
137 *
138 * If a non-zero error is returned, then the contents of bpp and
139 * dipp are undefined.
140 *
141 * Use xfs_imap() to determine the size and location of the
142 * buffer to read from disk.
143 */
144 STATIC int
148 xfs_ino_t ino,
152 {
154 xfs_imap_t imap;
159 /*
160 * Call the space management code to find the location of the
161 * inode on disk.
162 */
167 "xfs_inotobp: xfs_imap() returned an "
170 }
172 /*
173 * If the inode number maps to a block outside the bounds of the
174 * file system then return NULL rather than calling read_buf
175 * and panicing when we get an error from the driver.
176 */
180 "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
185 }
187 /*
188 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
189 * default to just a read_buf() call.
190 */
196 "xfs_inotobp: xfs_trans_read_buf() returned an "
199 }
201 di_ok =
205 XFS_RANDOM_ITOBP_INOTOBP))) {
209 "xfs_inotobp: XFS_TEST_ERROR() returned an "
212 }
216 /*
217 * Set *dipp to point to the on-disk inode in the buffer.
218 */
223 }
226 /*
227 * This routine is called to map an inode to the buffer containing
228 * the on-disk version of the inode. It returns a pointer to the
229 * buffer containing the on-disk inode in the bpp parameter, and in
230 * the dip parameter it returns a pointer to the on-disk inode within
231 * that buffer.
232 *
233 * If a non-zero error is returned, then the contents of bpp and
234 * dipp are undefined.
235 *
236 * If the inode is new and has not yet been initialized, use xfs_imap()
237 * to determine the size and location of the buffer to read from disk.
238 * If the inode has already been mapped to its buffer and read in once,
239 * then use the mapping information stored in the inode rather than
240 * calling xfs_imap(). This allows us to avoid the overhead of looking
241 * at the inode btree for small block file systems (see xfs_dilocate()).
242 * We can tell whether the inode has been mapped in before by comparing
243 * its disk block address to 0. Only uninitialized inodes will have
244 * 0 for the disk block address.
245 */
246 int
253 xfs_daddr_t bno,
254 uint imap_flags)
255 {
256 xfs_imap_t imap;
263 /*
264 * Call the space management code to find the location of the
265 * inode on disk.
266 */
272 /*
273 * If the inode number maps to a block outside the bounds
274 * of the file system then return NULL rather than calling
275 * read_buf and panicing when we get an error from the
276 * driver.
277 */
280 #ifdef DEBUG
282 "(imap.im_blkno (0x%llx) "
283 "+ imap.im_len (0x%llx)) > "
284 " XFS_FSB_TO_BB(mp, "
291 }
293 /*
294 * Fill in the fields in the inode that will be used to
295 * map the inode to its buffer from now on.
296 */
301 /*
302 * We've already mapped the inode once, so just use the
303 * mapping that we saved the first time.
304 */
308 }
311 /*
312 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
313 * default to just a read_buf() call.
314 */
318 #ifdef DEBUG
320 "xfs_trans_read_buf() returned error %d, "
326 }
328 /*
329 * Validate the magic number and version of every inode in the buffer
330 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
331 * No validation is done here in userspace (xfs_repair).
332 */
333 #if !defined(__KERNEL__)
335 #elif defined(DEBUG)
339 #endif
350 XFS_ERRTAG_ITOBP_INOTOBP,
351 XFS_RANDOM_ITOBP_INOTOBP))) {
355 }
356 #ifdef DEBUG
358 "Device %s - bad inode magic/vsn "
363 #endif
368 }
369 }
373 /*
374 * Mark the buffer as an inode buffer now that it looks good
375 */
378 /*
379 * Set *dipp to point to the on-disk inode in the buffer.
380 */
384 }
386 /*
387 * Move inode type and inode format specific information from the
388 * on-disk inode to the in-core inode. For fifos, devs, and sockets
389 * this means set if_rdev to the proper value. For files, directories,
390 * and symlinks this means to bring in the in-line data or extent
391 * pointers. For a file in B-tree format, only the root is immediately
392 * brought in-core. The rest will be in-lined in if_extents when it
393 * is first referenced (see xfs_iread_extents()).
394 */
395 STATIC int
399 {
403 xfs_fsize_t di_size;
422 }
424 if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) {
432 }
443 }
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 uint imap_flags)
859 {
872 /*
873 * Get pointer's to the on-disk inode and the buffer containing it.
874 * If the inode number refers to a block outside the file system
875 * then xfs_itobp() will return NULL. In this case we should
876 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
877 * know that this is a new incore inode.
878 */
883 }
885 /*
886 * Initialize inode's trace buffers.
887 * Do this before xfs_iformat in case it adds entries.
888 */
889 #ifdef XFS_BMAP_TRACE
891 #endif
892 #ifdef XFS_BMBT_TRACE
894 #endif
895 #ifdef XFS_RW_TRACE
897 #endif
898 #ifdef XFS_ILOCK_TRACE
900 #endif
901 #ifdef XFS_DIR2_TRACE
903 #endif
905 /*
906 * If we got something that isn't an inode it means someone
907 * (nfs or dmi) has a stale handle.
908 */
912 #ifdef DEBUG
914 "dip->di_core.di_magic (0x%x) != "
917 XFS_DINODE_MAGIC);
920 }
922 /*
923 * If the on-disk inode is already linked to a directory
924 * entry, copy all of the inode into the in-core inode.
925 * xfs_iformat() handles copying in the inode format
926 * specific information.
927 * Otherwise, just get the truly permanent information.
928 */
936 #ifdef DEBUG
939 error);
942 }
948 /*
949 * Make sure to pull in the mode here as well in
950 * case the inode is released without being used.
951 * This ensures that xfs_inactive() will see that
952 * the inode is already free and not try to mess
953 * with the uninitialized part of it.
954 */
956 /*
957 * Initialize the per-fork minima and maxima for a new
958 * inode here. xfs_iformat will do it for old inodes.
959 */
962 }
966 /*
967 * The inode format changed when we moved the link count and
968 * made it 32 bits long. If this is an old format inode,
969 * convert it in memory to look like a new one. If it gets
970 * flushed to disk we will convert back before flushing or
971 * logging it. We zero out the new projid field and the old link
972 * count field. We'll handle clearing the pad field (the remains
973 * of the old uuid field) when we actually convert the inode to
974 * the new format. We don't change the version number so that we
975 * can distinguish this from a real new format inode.
976 */
981 }
986 /*
987 * Mark the buffer containing the inode as something to keep
988 * around for a while. This helps to keep recently accessed
989 * meta-data in-core longer.
990 */
993 /*
994 * Use xfs_trans_brelse() to release the buffer containing the
995 * on-disk inode, because it was acquired with xfs_trans_read_buf()
996 * in xfs_itobp() above. If tp is NULL, this is just a normal
997 * brelse(). If we're within a transaction, then xfs_trans_brelse()
998 * will only release the buffer if it is not dirty within the
999 * transaction. It will be OK to release the buffer in this case,
1000 * because inodes on disk are never destroyed and we will be
1001 * locking the new in-core inode before putting it in the hash
1002 * table where other processes can find it. Thus we don't have
1003 * to worry about the inode being changed just because we released
1004 * the buffer.
1005 */
1009 }
1011 /*
1012 * Read in extents from a btree-format inode.
1013 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1014 */
1015 int
1020 {
1023 xfs_extnum_t nextents;
1030 }
1035 /*
1036 * We know that the size is valid (it's checked in iformat_btree)
1037 */
1047 }
1050 }
1052 /*
1053 * Allocate an inode on disk and return a copy of its in-core version.
1054 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1055 * appropriately within the inode. The uid and gid for the inode are
1056 * set according to the contents of the given cred structure.
1057 *
1058 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1059 * has a free inode available, call xfs_iget()
1060 * to obtain the in-core version of the allocated inode. Finally,
1061 * fill in the inode and log its initial contents. In this case,
1062 * ialloc_context would be set to NULL and call_again set to false.
1063 *
1064 * If xfs_dialloc() does not have an available inode,
1065 * it will replenish its supply by doing an allocation. Since we can
1066 * only do one allocation within a transaction without deadlocks, we
1067 * must commit the current transaction before returning the inode itself.
1068 * In this case, therefore, we will set call_again to true and return.
1069 * The caller should then commit the current transaction, start a new
1070 * transaction, and call xfs_ialloc() again to actually get the inode.
1071 *
1072 * To ensure that some other process does not grab the inode that
1073 * was allocated during the first call to xfs_ialloc(), this routine
1074 * also returns the [locked] bp pointing to the head of the freelist
1075 * as ialloc_context. The caller should hold this buffer across
1076 * the commit and pass it back into this routine on the second call.
1077 */
1078 int
1082 mode_t mode,
1083 xfs_nlink_t nlink,
1084 xfs_dev_t rdev,
1086 xfs_prid_t prid,
1091 {
1092 xfs_ino_t ino;
1095 uint flags;
1098 /*
1099 * Call the space management code to pick
1100 * the on-disk inode to be allocated.
1101 */
1106 }
1110 }
1113 /*
1114 * Get the in-core inode with the lock held exclusively.
1115 * This is because we're setting fields here we need
1116 * to prevent others from looking at until we're done.
1117 */
1122 }
1135 /*
1136 * If the superblock version is up to where we support new format
1137 * inodes and this is currently an old format inode, then change
1138 * the inode version number now. This way we only do the conversion
1139 * here rather than here and in the flush/logging code.
1140 */
1144 /*
1145 * We've already zeroed the old link count, the projid field,
1146 * and the pad field.
1147 */
1148 }
1150 /*
1151 * Project ids won't be stored on disk if we are using a version 1 inode.
1152 */
1160 }
1161 }
1163 /*
1164 * If the group ID of the new file does not match the effective group
1165 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1166 * (and only if the irix_sgid_inherit compatibility variable is set).
1167 */
1172 }
1179 /*
1180 * di_gen will have been taken care of in xfs_iread.
1181 */
1208 }
1213 }
1217 }
1218 }
1220 xfs_inherit_noatime)
1223 xfs_inherit_nodump)
1226 xfs_inherit_sync)
1229 xfs_inherit_nosymlinks)
1234 xfs_inherit_nodefrag)
1237 }
1238 /* FALLTHROUGH */
1247 }
1248 /*
1249 * Attribute fork settings for new inode.
1250 */
1254 /*
1255 * Log the new values stuffed into the inode.
1256 */
1259 /* now that we have an i_mode we can setup inode ops and unlock */
1264 }
1266 /*
1267 * Check to make sure that there are no blocks allocated to the
1268 * file beyond the size of the file. We don't check this for
1269 * files with fixed size extents or real time extents, but we
1270 * at least do it for regular files.
1271 */
1272 #ifdef DEBUG
1273 void
1277 xfs_fsize_t isize)
1278 {
1279 xfs_fileoff_t map_first;
1291 /*
1292 * The filesystem could be shutting down, so bmapi may return
1293 * an error.
1294 */
1298 map_first),
1304 }
1307 /*
1308 * Calculate the last possible buffered byte in a file. This must
1309 * include data that was buffered beyond the EOF by the write code.
1310 * This also needs to deal with overflowing the xfs_fsize_t type
1311 * which can happen for sizes near the limit.
1312 *
1313 * We also need to take into account any blocks beyond the EOF. It
1314 * may be the case that they were buffered by a write which failed.
1315 * In that case the pages will still be in memory, but the inode size
1316 * will never have been updated.
1317 */
1318 xfs_fsize_t
1321 {
1323 xfs_fsize_t last_byte;
1324 xfs_fileoff_t last_block;
1325 xfs_fileoff_t size_last_block;
1331 /*
1332 * Only check for blocks beyond the EOF if the extents have
1333 * been read in. This eliminates the need for the inode lock,
1334 * and it also saves us from looking when it really isn't
1335 * necessary.
1336 */
1339 XFS_DATA_FORK);
1342 }
1345 }
1352 }
1356 }
1358 }
1360 #if defined(XFS_RW_TRACE)
1361 STATIC void
1366 xfs_fsize_t new_size,
1367 xfs_off_t toss_start,
1368 xfs_off_t toss_finish)
1369 {
1372 }
1391 }
1392 #else
1393 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1394 #endif
1396 /*
1397 * Start the truncation of the file to new_size. The new size
1398 * must be smaller than the current size. This routine will
1399 * clear the buffer and page caches of file data in the removed
1400 * range, and xfs_itruncate_finish() will remove the underlying
1401 * disk blocks.
1402 *
1403 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1404 * must NOT have the inode lock held at all. This is because we're
1405 * calling into the buffer/page cache code and we can't hold the
1406 * inode lock when we do so.
1407 *
1408 * We need to wait for any direct I/Os in flight to complete before we
1409 * proceed with the truncate. This is needed to prevent the extents
1410 * being read or written by the direct I/Os from being removed while the
1411 * I/O is in flight as there is no other method of synchronising
1412 * direct I/O with the truncate operation. Also, because we hold
1413 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1414 * started until the truncate completes and drops the lock. Essentially,
1415 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1416 * between direct I/Os and the truncate operation.
1417 *
1418 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1419 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1420 * in the case that the caller is locking things out of order and
1421 * may not be able to call xfs_itruncate_finish() with the inode lock
1422 * held without dropping the I/O lock. If the caller must drop the
1423 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1424 * must be called again with all the same restrictions as the initial
1425 * call.
1426 */
1427 int
1430 uint flags,
1431 xfs_fsize_t new_size)
1432 {
1433 xfs_fsize_t last_byte;
1434 xfs_off_t toss_start;
1449 /*
1450 * Call toss_pages or flushinval_pages to get rid of pages
1451 * overlapping the region being removed. We have to use
1452 * the less efficient flushinval_pages in the case that the
1453 * caller may not be able to finish the truncate without
1454 * dropping the inode's I/O lock. Make sure
1455 * to catch any pages brought in by buffers overlapping
1456 * the EOF by searching out beyond the isize by our
1457 * block size. We round new_size up to a block boundary
1458 * so that we don't toss things on the same block as
1459 * new_size but before it.
1460 *
1461 * Before calling toss_page or flushinval_pages, make sure to
1462 * call remapf() over the same region if the file is mapped.
1463 * This frees up mapped file references to the pages in the
1464 * given range and for the flushinval_pages case it ensures
1465 * that we get the latest mapped changes flushed out.
1466 */
1470 /*
1471 * The place to start tossing is beyond our maximum
1472 * file size, so there is no way that the data extended
1473 * out there.
1474 */
1476 }
1479 last_byte);
1485 }
1486 }
1488 #ifdef DEBUG
1491 }
1492 #endif
1494 }
1496 /*
1497 * Shrink the file to the given new_size. The new
1498 * size must be smaller than the current size.
1499 * This will free up the underlying blocks
1500 * in the removed range after a call to xfs_itruncate_start()
1501 * or xfs_atruncate_start().
1502 *
1503 * The transaction passed to this routine must have made
1504 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1505 * This routine may commit the given transaction and
1506 * start new ones, so make sure everything involved in
1507 * the transaction is tidy before calling here.
1508 * Some transaction will be returned to the caller to be
1509 * committed. The incoming transaction must already include
1510 * the inode, and both inode locks must be held exclusively.
1511 * The inode must also be "held" within the transaction. On
1512 * return the inode will be "held" within the returned transaction.
1513 * This routine does NOT require any disk space to be reserved
1514 * for it within the transaction.
1515 *
1516 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1517 * and it indicates the fork which is to be truncated. For the
1518 * attribute fork we only support truncation to size 0.
1519 *
1520 * We use the sync parameter to indicate whether or not the first
1521 * transaction we perform might have to be synchronous. For the attr fork,
1522 * it needs to be so if the unlink of the inode is not yet known to be
1523 * permanent in the log. This keeps us from freeing and reusing the
1524 * blocks of the attribute fork before the unlink of the inode becomes
1525 * permanent.
1526 *
1527 * For the data fork, we normally have to run synchronously if we're
1528 * being called out of the inactive path or we're being called
1529 * out of the create path where we're truncating an existing file.
1530 * Either way, the truncate needs to be sync so blocks don't reappear
1531 * in the file with altered data in case of a crash. wsync filesystems
1532 * can run the first case async because anything that shrinks the inode
1533 * has to run sync so by the time we're called here from inactive, the
1534 * inode size is permanently set to 0.
1535 *
1536 * Calls from the truncate path always need to be sync unless we're
1537 * in a wsync filesystem and the file has already been unlinked.
1538 *
1539 * The caller is responsible for correctly setting the sync parameter.
1540 * It gets too hard for us to guess here which path we're being called
1541 * out of just based on inode state.
1542 */
1543 int
1547 xfs_fsize_t new_size,
1550 {
1551 xfs_fsblock_t first_block;
1552 xfs_fileoff_t first_unmap_block;
1553 xfs_fileoff_t last_block;
1559 xfs_bmap_free_t free_list;
1576 /*
1577 * We only support truncating the entire attribute fork.
1578 */
1581 }
1584 /*
1585 * The first thing we do is set the size to new_size permanently
1586 * on disk. This way we don't have to worry about anyone ever
1587 * being able to look at the data being freed even in the face
1588 * of a crash. What we're getting around here is the case where
1589 * we free a block, it is allocated to another file, it is written
1590 * to, and then we crash. If the new data gets written to the
1591 * file but the log buffers containing the free and reallocation
1592 * don't, then we'd end up with garbage in the blocks being freed.
1593 * As long as we make the new_size permanent before actually
1594 * freeing any blocks it doesn't matter if they get writtten to.
1595 *
1596 * The callers must signal into us whether or not the size
1597 * setting here must be synchronous. There are a few cases
1598 * where it doesn't have to be synchronous. Those cases
1599 * occur if the file is unlinked and we know the unlink is
1600 * permanent or if the blocks being truncated are guaranteed
1601 * to be beyond the inode eof (regardless of the link count)
1602 * and the eof value is permanent. Both of these cases occur
1603 * only on wsync-mounted filesystems. In those cases, we're
1604 * guaranteed that no user will ever see the data in the blocks
1605 * that are being truncated so the truncate can run async.
1606 * In the free beyond eof case, the file may wind up with
1607 * more blocks allocated to it than it needs if we crash
1608 * and that won't get fixed until the next time the file
1609 * is re-opened and closed but that's ok as that shouldn't
1610 * be too many blocks.
1611 *
1612 * However, we can't just make all wsync xactions run async
1613 * because there's one call out of the create path that needs
1614 * to run sync where it's truncating an existing file to size
1615 * 0 whose size is > 0.
1616 *
1617 * It's probably possible to come up with a test in this
1618 * routine that would correctly distinguish all the above
1619 * cases from the values of the function parameters and the
1620 * inode state but for sanity's sake, I've decided to let the
1621 * layers above just tell us. It's simpler to correctly figure
1622 * out in the layer above exactly under what conditions we
1623 * can run async and I think it's easier for others read and
1624 * follow the logic in case something has to be changed.
1625 * cscope is your friend -- rcc.
1626 *
1627 * The attribute fork is much simpler.
1628 *
1629 * For the attribute fork we allow the caller to tell us whether
1630 * the unlink of the inode that led to this call is yet permanent
1631 * in the on disk log. If it is not and we will be freeing extents
1632 * in this inode then we make the first transaction synchronous
1633 * to make sure that the unlink is permanent by the time we free
1634 * the blocks.
1635 */
1638 /*
1639 * If we are not changing the file size then do
1640 * not update the on-disk file size - we may be
1641 * called from xfs_inactive_free_eofblocks(). If we
1642 * update the on-disk file size and then the system
1643 * crashes before the contents of the file are
1644 * flushed to disk then the files may be full of
1645 * holes (ie NULL files bug).
1646 */
1651 }
1652 }
1657 }
1663 /*
1664 * Since it is possible for space to become allocated beyond
1665 * the end of the file (in a crash where the space is allocated
1666 * but the inode size is not yet updated), simply remove any
1667 * blocks which show up between the new EOF and the maximum
1668 * possible file size. If the first block to be removed is
1669 * beyond the maximum file size (ie it is the same as last_block),
1670 * then there is nothing to do.
1671 */
1679 }
1681 /*
1682 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1683 * will tell us whether it freed the entire range or
1684 * not. If this is a synchronous mount (wsync),
1685 * then we can tell bunmapi to keep all the
1686 * transactions asynchronous since the unlink
1687 * transaction that made this inode inactive has
1688 * already hit the disk. There's no danger of
1689 * the freed blocks being reused, there being a
1690 * crash, and the reused blocks suddenly reappearing
1691 * in this file with garbage in them once recovery
1692 * runs.
1693 */
1699 XFS_ITRUNC_MAX_EXTENTS,
1703 /*
1704 * If the bunmapi call encounters an error,
1705 * return to the caller where the transaction
1706 * can be properly aborted. We just need to
1707 * make sure we're not holding any resources
1708 * that we were not when we came in.
1709 */
1712 }
1714 /*
1715 * Duplicate the transaction that has the permanent
1716 * reservation and commit the old transaction.
1717 */
1721 /*
1722 * If the bmap finish call encounters an error,
1723 * return to the caller where the transaction
1724 * can be properly aborted. We just need to
1725 * make sure we're not holding any resources
1726 * that we were not when we came in.
1727 *
1728 * Aborting from this point might lose some
1729 * blocks in the file system, but oh well.
1730 */
1733 /*
1734 * If the passed in transaction committed
1735 * in xfs_bmap_finish(), then we want to
1736 * add the inode to this one before returning.
1737 * This keeps things simple for the higher
1738 * level code, because it always knows that
1739 * the inode is locked and held in the
1740 * transaction that returns to it whether
1741 * errors occur or not. We don't mark the
1742 * inode dirty so that this transaction can
1743 * be easily aborted if possible.
1744 */
1748 }
1750 }
1753 /*
1754 * The first xact was committed,
1755 * so add the inode to the new one.
1756 * Mark it dirty so it will be logged
1757 * and moved forward in the log as
1758 * part of every commit.
1759 */
1764 }
1769 XFS_TRANS_PERM_LOG_RES,
1770 XFS_ITRUNCATE_LOG_COUNT);
1771 /*
1772 * Add the inode being truncated to the next chained
1773 * transaction.
1774 */
1779 }
1780 /*
1781 * Only update the size in the case of the data fork, but
1782 * always re-log the inode so that our permanent transaction
1783 * can keep on rolling it forward in the log.
1784 */
1787 /*
1788 * If we are not changing the file size then do
1789 * not update the on-disk file size - we may be
1790 * called from xfs_inactive_free_eofblocks(). If we
1791 * update the on-disk file size and then the system
1792 * crashes before the contents of the file are
1793 * flushed to disk then the files may be full of
1794 * holes (ie NULL files bug).
1795 */
1799 }
1800 }
1810 }
1813 /*
1814 * xfs_igrow_start
1815 *
1816 * Do the first part of growing a file: zero any data in the last
1817 * block that is beyond the old EOF. We need to do this before
1818 * the inode is joined to the transaction to modify the i_size.
1819 * That way we can drop the inode lock and call into the buffer
1820 * cache to get the buffer mapping the EOF.
1821 */
1822 int
1825 xfs_fsize_t new_size,
1827 {
1834 /*
1835 * Zero any pages that may have been created by
1836 * xfs_write_file() beyond the end of the file
1837 * and any blocks between the old and new file sizes.
1838 */
1842 }
1844 /*
1845 * xfs_igrow_finish
1846 *
1847 * This routine is called to extend the size of a file.
1848 * The inode must have both the iolock and the ilock locked
1849 * for update and it must be a part of the current transaction.
1850 * The xfs_igrow_start() function must have been called previously.
1851 * If the change_flag is not zero, the inode change timestamp will
1852 * be updated.
1853 */
1854 void
1858 xfs_fsize_t new_size,
1860 {
1866 /*
1867 * Update the file size. Update the inode change timestamp
1868 * if change_flag set.
1869 */
1876 }
1879 /*
1880 * This is called when the inode's link count goes to 0.
1881 * We place the on-disk inode on a list in the AGI. It
1882 * will be pulled from this list when the inode is freed.
1883 */
1884 int
1888 {
1894 xfs_agnumber_t agno;
1895 xfs_daddr_t agdaddr;
1896 xfs_agino_t agino;
1911 /*
1912 * Get the agi buffer first. It ensures lock ordering
1913 * on the list.
1914 */
1919 }
1920 /*
1921 * Validate the magic number of the agi block.
1922 */
1924 agi_ok =
1928 XFS_RANDOM_IUNLINK))) {
1932 }
1933 /*
1934 * Get the index into the agi hash table for the
1935 * list this inode will go on.
1936 */
1944 /*
1945 * There is already another inode in the bucket we need
1946 * to add ourselves to. Add us at the front of the list.
1947 * Here we put the head pointer into our next pointer,
1948 * and then we fall through to point the head at us.
1949 */
1953 }
1956 /* both on-disk, don't endian flip twice */
1964 }
1966 /*
1967 * Point the bucket head pointer at the inode being inserted.
1968 */
1976 }
1978 /*
1979 * Pull the on-disk inode from the AGI unlinked list.
1980 */
1981 STATIC int
1985 {
1986 xfs_ino_t next_ino;
1992 xfs_agnumber_t agno;
1993 xfs_daddr_t agdaddr;
1994 xfs_agino_t agino;
1995 xfs_agino_t next_agino;
2003 /*
2004 * First pull the on-disk inode from the AGI unlinked list.
2005 */
2011 /*
2012 * Get the agi buffer first. It ensures lock ordering
2013 * on the list.
2014 */
2022 }
2023 /*
2024 * Validate the magic number of the agi block.
2025 */
2027 agi_ok =
2031 XFS_RANDOM_IUNLINK_REMOVE))) {
2039 }
2040 /*
2041 * Get the index into the agi hash table for the
2042 * list this inode will go on.
2043 */
2051 /*
2052 * We're at the head of the list. Get the inode's
2053 * on-disk buffer to see if there is anyone after us
2054 * on the list. Only modify our next pointer if it
2055 * is not already NULLAGINO. This saves us the overhead
2056 * of dealing with the buffer when there is no need to
2057 * change it.
2058 */
2065 }
2078 }
2079 /*
2080 * Point the bucket head pointer at the next inode.
2081 */
2090 /*
2091 * We need to search the list for the inode being freed.
2092 */
2096 /*
2097 * If the last inode wasn't the one pointing to
2098 * us, then release its buffer since we're not
2099 * going to do anything with it.
2100 */
2103 }
2112 }
2116 }
2117 /*
2118 * Now last_ibp points to the buffer previous to us on
2119 * the unlinked list. Pull us from the list.
2120 */
2127 }
2141 }
2142 /*
2143 * Point the previous inode on the list to the next inode.
2144 */
2152 }
2154 }
2157 {
2161 }
2163 STATIC void
2167 xfs_ino_t inum)
2168 {
2174 xfs_daddr_t blkno;
2191 }
2200 /*
2201 * Look for each inode in memory and attempt to lock it,
2202 * we can be racing with flush and tail pushing here.
2203 * any inode we get the locks on, add to an array of
2204 * inode items to process later.
2205 *
2206 * The get the buffer lock, we could beat a flush
2207 * or tail pushing thread to the lock here, in which
2208 * case they will go looking for the inode buffer
2209 * and fail, we need some other form of interlock
2210 * here.
2211 */
2219 }
2221 /* Inode not in memory or we found it already,
2222 * nothing to do
2223 */
2227 }
2232 }
2234 /* If we can get the locks then add it to the
2235 * list, otherwise by the time we get the bp lock
2236 * below it will already be attached to the
2237 * inode buffer.
2238 */
2240 /* This inode will already be locked - by us, lets
2241 * keep it that way.
2242 */
2251 }
2252 }
2255 }
2266 }
2269 }
2270 }
2273 }
2277 XFS_BUF_LOCK);
2290 pre_flushed++;
2291 }
2293 }
2304 }
2318 }
2319 }
2324 }
2327 }
2329 /*
2330 * This is called to return an inode to the inode free list.
2331 * The inode should already be truncated to 0 length and have
2332 * no pages associated with it. This routine also assumes that
2333 * the inode is already a part of the transaction.
2334 *
2335 * The on-disk copy of the inode will have been added to the list
2336 * of unlinked inodes in the AGI. We need to remove the inode from
2337 * that list atomically with respect to freeing it here.
2338 */
2339 int
2344 {
2347 xfs_ino_t first_ino;
2358 /*
2359 * Pull the on-disk inode from the AGI unlinked list.
2360 */
2364 }
2369 }
2378 /*
2379 * Bump the generation count so no one will be confused
2380 * by reincarnations of this inode.
2381 */
2387 }
2390 }
2392 /*
2393 * Reallocate the space for if_broot based on the number of records
2394 * being added or deleted as indicated in rec_diff. Move the records
2395 * and pointers in if_broot to fit the new size. When shrinking this
2396 * will eliminate holes between the records and pointers created by
2397 * the caller. When growing this will create holes to be filled in
2398 * by the caller.
2399 *
2400 * The caller must not request to add more records than would fit in
2401 * the on-disk inode root. If the if_broot is currently NULL, then
2402 * if we adding records one will be allocated. The caller must also
2403 * not request that the number of records go below zero, although
2404 * it can go to zero.
2405 *
2406 * ip -- the inode whose if_broot area is changing
2407 * ext_diff -- the change in the number of records, positive or negative,
2408 * requested for the if_broot array.
2409 */
2410 void
2415 {
2424 /*
2425 * Handle the degenerate case quietly.
2426 */
2429 }
2433 /*
2434 * If there wasn't any memory allocated before, just
2435 * allocate it now and get out.
2436 */
2440 KM_SLEEP);
2443 }
2445 /*
2446 * If there is already an existing if_broot, then we need
2447 * to realloc() it and shift the pointers to their new
2448 * location. The records don't change location because
2449 * they are kept butted up against the btree block header.
2450 */
2456 new_size,
2458 KM_SLEEP);
2468 }
2470 /*
2471 * rec_diff is less than 0. In this case, we are shrinking the
2472 * if_broot buffer. It must already exist. If we go to zero
2473 * records, just get rid of the root and clear the status bit.
2474 */
2481 else
2485 /*
2486 * First copy over the btree block header.
2487 */
2492 }
2494 /*
2495 * Only copy the records and pointers if there are any.
2496 */
2498 /*
2499 * First copy the records.
2500 */
2507 /*
2508 * Then copy the pointers.
2509 */
2515 }
2522 }
2525 /*
2526 * This is called when the amount of space needed for if_data
2527 * is increased or decreased. The change in size is indicated by
2528 * the number of bytes that need to be added or deleted in the
2529 * byte_diff parameter.
2530 *
2531 * If the amount of space needed has decreased below the size of the
2532 * inline buffer, then switch to using the inline buffer. Otherwise,
2533 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2534 * to what is needed.
2535 *
2536 * ip -- the inode whose if_data area is changing
2537 * byte_diff -- the change in the number of bytes, positive or negative,
2538 * requested for the if_data array.
2539 */
2540 void
2545 {
2552 }
2561 }
2565 /*
2566 * If the valid extents/data can fit in if_inline_ext/data,
2567 * copy them from the malloc'd vector and free it.
2568 */
2574 new_size);
2577 }
2580 /*
2581 * Stuck with malloc/realloc.
2582 * For inline data, the underlying buffer must be
2583 * a multiple of 4 bytes in size so that it can be
2584 * logged and stay on word boundaries. We enforce
2585 * that here.
2586 */
2592 /*
2593 * Only do the realloc if the underlying size
2594 * is really changing.
2595 */
2599 real_size,
2601 KM_SLEEP);
2602 }
2608 }
2609 }
2613 }
2618 /*
2619 * Map inode to disk block and offset.
2620 *
2621 * mp -- the mount point structure for the current file system
2622 * tp -- the current transaction
2623 * ino -- the inode number of the inode to be located
2624 * imap -- this structure is filled in with the information necessary
2625 * to retrieve the given inode from disk
2626 * flags -- flags to pass to xfs_dilocate indicating whether or not
2627 * lookups in the inode btree were OK or not
2628 */
2629 int
2633 xfs_ino_t ino,
2635 uint flags)
2636 {
2637 xfs_fsblock_t fsbno;
2647 }
2654 }
2656 void
2660 {
2667 }
2669 /*
2670 * If the format is local, then we can't have an extents
2671 * array so just look for an inline data array. If we're
2672 * not local then we may or may not have an extents list,
2673 * so check and free it up if we do.
2674 */
2682 }
2689 }
2696 }
2697 }
2699 /*
2700 * This is called free all the memory associated with an inode.
2701 * It must free the inode itself and any buffers allocated for
2702 * if_extents/if_data and if_broot. It must also free the lock
2703 * associated with the inode.
2704 */
2705 void
2708 {
2716 }
2722 #ifdef XFS_BMAP_TRACE
2724 #endif
2725 #ifdef XFS_BMBT_TRACE
2727 #endif
2728 #ifdef XFS_RW_TRACE
2730 #endif
2731 #ifdef XFS_ILOCK_TRACE
2733 #endif
2734 #ifdef XFS_DIR2_TRACE
2736 #endif
2738 /*
2739 * Only if we are shutting down the fs will we see an
2740 * inode still in the AIL. If it is there, we should remove
2741 * it to prevent a use-after-free from occurring.
2742 */
2753 else
2755 }
2757 }
2759 }
2762 /*
2763 * Increment the pin count of the given buffer.
2764 * This value is protected by ipinlock spinlock in the mount structure.
2765 */
2766 void
2769 {
2773 }
2775 /*
2776 * Decrement the pin count of the given inode, and wake up
2777 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2778 * inode must have been previously pinned with a call to xfs_ipin().
2779 */
2780 void
2783 {
2788 /*
2789 * If the inode is currently being reclaimed, the link between
2790 * the bhv_vnode and the xfs_inode will be broken after the
2791 * XFS_IRECLAIM* flag is set. Hence, if these flags are not
2792 * set, then we can move forward and mark the linux inode dirty
2793 * knowing that it is still valid as it won't freed until after
2794 * the bhv_vnode<->xfs_inode link is broken in xfs_reclaim. The
2795 * i_flags_lock is used to synchronise the setting of the
2796 * XFS_IRECLAIM* flags and the breaking of the link, and so we
2797 * can execute atomically w.r.t to reclaim by holding this lock
2798 * here.
2799 *
2800 * However, we still need to issue the unpin wakeup call as the
2801 * inode reclaim may be blocked waiting for the inode to become
2802 * unpinned.
2803 */
2813 /* make sync come back and flush this inode */
2816 }
2819 }
2820 }
2822 /*
2823 * This is called to wait for the given inode to be unpinned.
2824 * It will sleep until this happens. The caller must have the
2825 * inode locked in at least shared mode so that the buffer cannot
2826 * be subsequently pinned once someone is waiting for it to be
2827 * unpinned.
2828 */
2829 STATIC void
2832 {
2834 xfs_lsn_t lsn;
2840 }
2847 }
2849 /*
2850 * Give the log a push so we don't wait here too long.
2851 */
2855 }
2858 /*
2859 * xfs_iextents_copy()
2860 *
2861 * This is called to copy the REAL extents (as opposed to the delayed
2862 * allocation extents) from the inode into the given buffer. It
2863 * returns the number of bytes copied into the buffer.
2864 *
2865 * If there are no delayed allocation extents, then we can just
2866 * memcpy() the extents into the buffer. Otherwise, we need to
2867 * examine each extent in turn and skip those which are delayed.
2868 */
2869 int
2874 {
2878 #ifdef XFS_BMAP_TRACE
2880 #endif
2884 xfs_fsblock_t start_block;
2894 /*
2895 * There are some delayed allocation extents in the
2896 * inode, so copy the extents one at a time and skip
2897 * the delayed ones. There must be at least one
2898 * non-delayed extent.
2899 */
2906 /*
2907 * It's a delayed allocation extent, so skip it.
2908 */
2910 }
2912 /* Translate to on disk format */
2917 dest_ep++;
2918 copied++;
2919 }
2924 }
2926 /*
2927 * Each of the following cases stores data into the same region
2928 * of the on-disk inode, so only one of them can be valid at
2929 * any given time. While it is possible to have conflicting formats
2930 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2931 * in EXTENTS format, this can only happen when the fork has
2932 * changed formats after being modified but before being flushed.
2933 * In these cases, the format always takes precedence, because the
2934 * format indicates the current state of the fork.
2935 */
2936 /*ARGSUSED*/
2937 STATIC int
2944 {
2948 #ifdef XFS_TRANS_DEBUG
2950 #endif
2961 /*
2962 * This can happen if we gave up in iformat in an error path,
2963 * for the attribute fork.
2964 */
2968 }
2978 }
2992 whichfork);
2993 }
3002 XFS_BROOT_SIZE_ADJ));
3006 }
3013 }
3021 }
3027 }
3030 }
3032 /*
3033 * xfs_iflush() will write a modified inode's changes out to the
3034 * inode's on disk home. The caller must have the inode lock held
3035 * in at least shared mode and the inode flush semaphore must be
3036 * held as well. The inode lock will still be held upon return from
3037 * the call and the caller is free to unlock it.
3038 * The inode flush lock will be unlocked when the inode reaches the disk.
3039 * The flags indicate how the inode's buffer should be written out.
3040 */
3041 int
3044 uint flags)
3045 {
3051 /* REFERENCED */
3069 /*
3070 * If the inode isn't dirty, then just release the inode
3071 * flush lock and do nothing.
3072 */
3079 }
3081 /*
3082 * We can't flush the inode until it is unpinned, so
3083 * wait for it. We know noone new can pin it, because
3084 * we are holding the inode lock shared and you need
3085 * to hold it exclusively to pin the inode.
3086 */
3089 /*
3090 * This may have been unpinned because the filesystem is shutting
3091 * down forcibly. If that's the case we must not write this inode
3092 * to disk, because the log record didn't make it to disk!
3093 */
3100 }
3102 /*
3103 * Get the buffer containing the on-disk inode.
3104 */
3109 }
3111 /*
3112 * Decide how buffer will be flushed out. This is done before
3113 * the call to xfs_iflush_int because this field is zeroed by it.
3114 */
3116 /*
3117 * Flush out the inode buffer according to the directions
3118 * of the caller. In the cases where the caller has given
3119 * us a choice choose the non-delwri case. This is because
3120 * the inode is in the AIL and we need to get it out soon.
3121 */
3138 }
3156 }
3157 }
3159 /*
3160 * First flush out the inode that xfs_iflush was called with.
3161 */
3165 }
3167 /*
3168 * inode clustering:
3169 * see if other inodes can be gathered into this write
3170 */
3179 /*
3180 * Do an un-protected check to see if the inode is dirty and
3181 * is a candidate for flushing. These checks will be repeated
3182 * later after the appropriate locks are acquired.
3183 */
3190 }
3192 /*
3193 * Try to get locks. If any are unavailable,
3194 * then this inode cannot be flushed and is skipped.
3195 */
3197 /* get inode locks (just i_lock) */
3199 /* get inode flush lock */
3201 /* check if pinned */
3203 /* arriving here means that
3204 * this inode can be flushed.
3205 * first re-check that it's
3206 * dirty
3207 */
3212 clcount++;
3216 XFS_ILOCK_SHARED);
3218 }
3221 }
3224 }
3225 }
3227 }
3228 }
3234 }
3236 /*
3237 * If the buffer is pinned then push on the log so we won't
3238 * get stuck waiting in the write for too long.
3239 */
3242 }
3250 }
3253 corrupt_out:
3257 /*
3258 * Unlocks the flush lock
3259 */
3262 cluster_corrupt_out:
3263 /* Corruption detected in the clustering loop. Invalidate the
3264 * inode buffer and shut down the filesystem.
3265 */
3268 /*
3269 * Clean up the buffer. If it was B_DELWRI, just release it --
3270 * brelse can handle it with no problems. If not, shut down the
3271 * filesystem before releasing the buffer.
3272 */
3275 }
3280 /*
3281 * Just like incore_relse: if we have b_iodone functions,
3282 * mark the buffer as an error and call them. Otherwise
3283 * mark it as stale and brelse.
3284 */
3295 }
3296 }
3299 /*
3300 * Unlocks the flush lock
3301 */
3303 }
3306 STATIC int
3310 {
3314 #ifdef XFS_TRANS_DEBUG
3316 #endif
3328 /*
3329 * If the inode isn't dirty, then just release the inode
3330 * flush lock and do nothing.
3331 */
3336 }
3338 /* set *dip = inode's place in the buffer */
3341 /*
3342 * Clear i_update_core before copying out the data.
3343 * This is for coordination with our timestamp updates
3344 * that don't hold the inode lock. They will always
3345 * update the timestamps BEFORE setting i_update_core,
3346 * so if we clear i_update_core after they set it we
3347 * are guaranteed to see their updates to the timestamps.
3348 * I believe that this depends on strongly ordered memory
3349 * semantics, but we have that. We use the SYNCHRONIZE
3350 * macro to make sure that the compiler does not reorder
3351 * the i_update_core access below the data copy below.
3352 */
3356 /*
3357 * Make sure to get the latest atime from the Linux inode.
3358 */
3367 }
3374 }
3384 }
3395 }
3396 }
3399 XFS_RANDOM_IFLUSH_5)) {
3405 ip);
3407 }
3414 }
3415 /*
3416 * bump the flush iteration count, used to detect flushes which
3417 * postdate a log record during recovery.
3418 */
3422 /*
3423 * Copy the dirty parts of the inode into the on-disk
3424 * inode. We always copy out the core of the inode,
3425 * because if the inode is dirty at all the core must
3426 * be.
3427 */
3430 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3434 /*
3435 * If this is really an old format inode and the superblock version
3436 * has not been updated to support only new format inodes, then
3437 * convert back to the old inode format. If the superblock version
3438 * has been updated, then make the conversion permanent.
3439 */
3444 /*
3445 * Convert it back.
3446 */
3450 /*
3451 * The superblock version has already been bumped,
3452 * so just make the conversion to the new inode
3453 * format permanent.
3454 */
3463 }
3464 }
3468 }
3471 /*
3472 * The only error from xfs_iflush_fork is on the data fork.
3473 */
3475 }
3478 /*
3479 * We've recorded everything logged in the inode, so we'd
3480 * like to clear the ilf_fields bits so we don't log and
3481 * flush things unnecessarily. However, we can't stop
3482 * logging all this information until the data we've copied
3483 * into the disk buffer is written to disk. If we did we might
3484 * overwrite the copy of the inode in the log with all the
3485 * data after re-logging only part of it, and in the face of
3486 * a crash we wouldn't have all the data we need to recover.
3487 *
3488 * What we do is move the bits to the ili_last_fields field.
3489 * When logging the inode, these bits are moved back to the
3490 * ilf_fields field. In the xfs_iflush_done() routine we
3491 * clear ili_last_fields, since we know that the information
3492 * those bits represent is permanently on disk. As long as
3493 * the flush completes before the inode is logged again, then
3494 * both ilf_fields and ili_last_fields will be cleared.
3495 *
3496 * We can play with the ilf_fields bits here, because the inode
3497 * lock must be held exclusively in order to set bits there
3498 * and the flush lock protects the ili_last_fields bits.
3499 * Set ili_logged so the flush done
3500 * routine can tell whether or not to look in the AIL.
3501 * Also, store the current LSN of the inode so that we can tell
3502 * whether the item has moved in the AIL from xfs_iflush_done().
3503 * In order to read the lsn we need the AIL lock, because
3504 * it is a 64 bit value that cannot be read atomically.
3505 */
3516 /*
3517 * Attach the function xfs_iflush_done to the inode's
3518 * buffer. This will remove the inode from the AIL
3519 * and unlock the inode's flush lock when the inode is
3520 * completely written to disk.
3521 */
3528 /*
3529 * We're flushing an inode which is not in the AIL and has
3530 * not been logged but has i_update_core set. For this
3531 * case we can use a B_DELWRI flush and immediately drop
3532 * the inode flush lock because we can avoid the whole
3533 * AIL state thing. It's OK to drop the flush lock now,
3534 * because we've already locked the buffer and to do anything
3535 * you really need both.
3536 */
3541 }
3543 }
3547 corrupt_out:
3549 }
3552 /*
3553 * Flush all inactive inodes in mp.
3554 */
3555 void
3558 {
3562 again:
3569 /* Make sure we skip markers inserted by sync */
3573 }
3580 }
3586 out:
3588 }
3590 /*
3591 * xfs_iaccess: check accessibility of inode for mode.
3592 */
3593 int
3596 mode_t mode,
3598 {
3612 }
3614 /*
3615 * If there's an Access Control List it's used instead of
3616 * the mode bits.
3617 */
3625 }
3627 /*
3628 * If the DACs are ok we don't need any capability check.
3629 */
3632 /*
3633 * Read/write DACs are always overridable.
3634 * Executable DACs are overridable if at least one exec bit is set.
3635 */
3645 #ifdef NOISE
3649 }
3651 }
3653 /*
3654 * xfs_iroundup: round up argument to next power of two
3655 */
3656 uint
3658 uint v)
3659 {
3661 uint m;
3674 }
3677 }
3679 #ifdef XFS_ILOCK_TRACE
3682 void
3684 {
3693 }
3694 #endif
3696 /*
3697 * Return a pointer to the extent record at file index idx.
3698 */
3699 xfs_bmbt_rec_t *
3703 {
3718 }
3719 }
3721 /*
3722 * Insert new item(s) into the extent records for incore inode
3723 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3724 */
3725 void
3731 {
3740 }
3741 }
3743 /*
3744 * This is called when the amount of space required for incore file
3745 * extents needs to be increased. The ext_diff parameter stores the
3746 * number of new extents being added and the idx parameter contains
3747 * the extent index where the new extents will be added. If the new
3748 * extents are being appended, then we just need to (re)allocate and
3749 * initialize the space. Otherwise, if the new extents are being
3750 * inserted into the middle of the existing entries, a bit more work
3751 * is required to make room for the new extents to be inserted. The
3752 * caller is responsible for filling in the new extent entries upon
3753 * return.
3754 */
3755 void
3760 {
3769 /*
3770 * If the new number of extents (nextents + ext_diff)
3771 * fits inside the inode, then continue to use the inline
3772 * extent buffer.
3773 */
3780 }
3784 }
3785 /*
3786 * Otherwise use a linear (direct) extent list.
3787 * If the extents are currently inside the inode,
3788 * xfs_iext_realloc_direct will switch us from
3789 * inline to direct extent allocation mode.
3790 */
3798 }
3799 }
3800 /* Indirection array */
3813 }
3814 /* Extents fit in target extent page */
3822 }
3825 }
3826 /* Insert a new extent page */
3830 }
3831 /*
3832 * If extent(s) are being appended to the last page in
3833 * the indirection array and the new extent(s) don't fit
3834 * in the page, then erp is NULL and erp_idx is set to
3835 * the next index needed in the indirection array.
3836 */
3845 erp_idx++;
3846 }
3847 }
3848 }
3849 }
3851 }
3853 /*
3854 * This is called when incore extents are being added to the indirection
3855 * array and the new extents do not fit in the target extent list. The
3856 * erp_idx parameter contains the irec index for the target extent list
3857 * in the indirection array, and the idx parameter contains the extent
3858 * index within the list. The number of extents being added is stored
3859 * in the count parameter.
3860 *
3861 * |-------| |-------|
3862 * | | | | idx - number of extents before idx
3863 * | idx | | count |
3864 * | | | | count - number of extents being inserted at idx
3865 * |-------| |-------|
3866 * | count | | nex2 | nex2 - number of extents after idx + count
3867 * |-------| |-------|
3868 */
3869 void
3875 {
3889 /*
3890 * Save second part of target extent list
3891 * (all extents past */
3899 }
3901 /*
3902 * Add the new extents to the end of the target
3903 * list, then allocate new irec record(s) and
3904 * extent buffer(s) as needed to store the rest
3905 * of the new extents.
3906 */
3913 }
3915 erp_idx++;
3921 }
3923 /* Add nex2 extents back to indirection array */
3925 xfs_extnum_t ext_avail;
3931 /*
3932 * If nex2 extents fit in the current page, append
3933 * nex2_ep after the new extents.
3934 */
3937 }
3938 /*
3939 * Otherwise, check if space is available in the
3940 * next page.
3941 */
3945 erp_idx++;
3946 erp++;
3947 /* Create a hole for nex2 extents */
3950 }
3951 /*
3952 * Final choice, create a new extent page for
3953 * nex2 extents.
3954 */
3956 erp_idx++;
3958 }
3963 }
3964 }
3966 /*
3967 * This is called when the amount of space required for incore file
3968 * extents needs to be decreased. The ext_diff parameter stores the
3969 * number of extents to be removed and the idx parameter contains
3970 * the extent index where the extents will be removed from.
3971 *
3972 * If the amount of space needed has decreased below the linear
3973 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3974 * extent array. Otherwise, use kmem_realloc() to adjust the
3975 * size to what is needed.
3976 */
3977 void
3982 {
3998 }
4000 }
4002 /*
4003 * This removes ext_diff extents from the inline buffer, beginning
4004 * at extent index idx.
4005 */
4006 void
4011 {
4030 }
4031 }
4033 /*
4034 * This removes ext_diff extents from a linear (direct) extent list,
4035 * beginning at extent index idx. If the extents are being removed
4036 * from the end of the list (ie. truncate) then we just need to re-
4037 * allocate the list to remove the extra space. Otherwise, if the
4038 * extents are being removed from the middle of the existing extent
4039 * entries, then we first need to move the extent records beginning
4040 * at idx + ext_diff up in the list to overwrite the records being
4041 * removed, then remove the extra space via kmem_realloc.
4042 */
4043 void
4048 {
4060 }
4061 /* Move extents up in the list (if needed) */
4067 }
4070 /*
4071 * Reallocate the direct extent list. If the extents
4072 * will fit inside the inode then xfs_iext_realloc_direct
4073 * will switch from direct to inline extent allocation
4074 * mode for us.
4075 */
4078 }
4080 /*
4081 * This is called when incore extents are being removed from the
4082 * indirection array and the extents being removed span multiple extent
4083 * buffers. The idx parameter contains the file extent index where we
4084 * want to begin removing extents, and the count parameter contains
4085 * how many extents need to be removed.
4086 *
4087 * |-------| |-------|
4088 * | nex1 | | | nex1 - number of extents before idx
4089 * |-------| | count |
4090 * | | | | count - number of extents being removed at idx
4091 * | count | |-------|
4092 * | | | nex2 | nex2 - number of extents after idx + count
4093 * |-------| |-------|
4094 */
4095 void
4100 {
4119 /*
4120 * Check for deletion of entire list;
4121 * xfs_iext_irec_remove() updates extent offsets.
4122 */
4129 XFS_IEXT_BUFSZ);
4135 }
4136 }
4137 /* Move extents up (if needed) */
4142 }
4143 /* Zero out rest of page */
4146 /* Update remaining counters */
4151 erp_idx++;
4152 erp++;
4153 }
4156 }
4158 /*
4159 * Create, destroy, or resize a linear (direct) block of extents.
4160 */
4161 void
4165 {
4174 /* Free extent records */
4177 }
4178 /* Resize direct extent list and zero any new bytes */
4180 /* Check if extents will fit inside the inode */
4186 }
4189 }
4193 rnew_size,
4195 KM_SLEEP);
4196 }
4201 }
4202 }
4203 /*
4204 * Switch from the inline extent buffer to a direct
4205 * extent list. Be sure to include the inline extent
4206 * bytes in new_size.
4207 */
4212 }
4214 }
4217 }
4219 /*
4220 * Switch from linear (direct) extent records to inline buffer.
4221 */
4222 void
4226 {
4229 /*
4230 * The inline buffer was zeroed when we switched
4231 * from inline to direct extent allocation mode,
4232 * so we don't need to clear it here.
4233 */
4239 }
4241 /*
4242 * Switch from inline buffer to linear (direct) extent records.
4243 * new_size should already be rounded up to the next power of 2
4244 * by the caller (when appropriate), so use new_size as it is.
4245 * However, since new_size may be rounded up, we can't update
4246 * if_bytes here. It is the caller's responsibility to update
4247 * if_bytes upon return.
4248 */
4249 void
4253 {
4262 }
4264 }
4266 /*
4267 * Resize an extent indirection array to new_size bytes.
4268 */
4269 void
4273 {
4288 }
4289 }
4291 /*
4292 * Switch from indirection array to linear (direct) extent allocations.
4293 */
4294 void
4297 {
4317 }
4318 }
4320 /*
4321 * Free incore file extents.
4322 */
4323 void
4326 {
4334 }
4341 }
4345 }
4347 /*
4348 * Return a pointer to the extent record for file system block bno.
4349 */
4355 {
4370 }
4373 /* Find target extent list */
4381 }
4382 /* Binary search extent records */
4393 /* Convert back to file-based extent index */
4396 }
4399 }
4400 }
4401 /* Convert back to file-based extent index */
4404 }
4410 }
4411 }
4414 }
4416 /*
4417 * Return a pointer to the indirection array entry containing the
4418 * extent record for filesystem block bno. Store the index of the
4419 * target irec in *erp_idxp.
4420 */
4426 {
4450 }
4451 }
4454 }
4456 /*
4457 * Return a pointer to the indirection array entry containing the
4458 * extent record at file extent index *idxp. Store the index of the
4459 * target irec in *erp_idxp and store the page index of the target
4460 * extent record in *idxp.
4461 */
4462 xfs_ext_irec_t *
4468 {
4485 /* Binary search extent irec's */
4501 erp_idx++;
4507 }
4508 }
4512 }
4514 /*
4515 * Allocate and initialize an indirection array once the space needed
4516 * for incore extents increases above XFS_IEXT_BUFSZ.
4517 */
4518 void
4521 {
4539 }
4550 }
4552 /*
4553 * Allocate and initialize a new entry in the indirection array.
4554 */
4555 xfs_ext_irec_t *
4559 {
4567 /* Resize indirection array */
4570 /*
4571 * Move records down in the array so the
4572 * new page can use erp_idx.
4573 */
4577 }
4580 /* Initialize new extent record */
4590 }
4592 /*
4593 * Remove a record from the indirection array.
4594 */
4595 void
4599 {
4611 }
4612 /* Compact extent records */
4616 }
4617 /*
4618 * Manually free the last extent record from the indirection
4619 * array. A call to xfs_iext_realloc_indirect() with a size
4620 * of zero would result in a call to xfs_iext_destroy() which
4621 * would in turn call this function again, creating a nasty
4622 * infinite loop.
4623 */
4630 }
4632 }
4634 /*
4635 * This is called to clean up large amounts of unused memory allocated
4636 * by the indirection array. Before compacting anything though, verify
4637 * that the indirection array is still needed and switch back to the
4638 * linear extent list (or even the inline buffer) if possible. The
4639 * compaction policy is as follows:
4640 *
4641 * Full Compaction: Extents fit into a single page (or inline buffer)
4642 * Full Compaction: Extents occupy less than 10% of allocated space
4643 * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
4644 * No Compaction: Extents occupy at least 50% of allocated space
4645 */
4646 void
4649 {
4668 }
4669 }
4671 /*
4672 * Combine extents from neighboring extent pages.
4673 */
4674 void
4677 {
4693 /*
4694 * Free page before removing extent record
4695 * so er_extoffs don't get modified in
4696 * xfs_iext_irec_remove.
4697 */
4703 erp_idx++;
4704 }
4705 }
4706 }
4708 /*
4709 * Fully compact the extent records managed by the indirection array.
4710 */
4711 void
4714 {
4734 /* Remove next page */
4736 /*
4737 * Free page before removing extent record
4738 * so er_extoffs don't get modified in
4739 * xfs_iext_irec_remove.
4740 */
4747 /* Update next page */
4749 /* Move rest of page up to become next new page */
4756 }
4758 erp_idx++;
4761 else
4763 }
4767 }
4768 }
4770 /*
4771 * This is called to update the er_extoff field in the indirection
4772 * array when extents have been added or removed from one of the
4773 * extent lists. erp_idx contains the irec index to begin updating
4774 * at and ext_diff contains the number of extents that were added
4775 * or removed.
4776 */
4777 void
4782 {
4790 }
4791 }