| blob | a391b955df0176969727b3768d08e6162216e96c |
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 */
18 #include <linux/log2.h>
59 /*
60 * Used in xfs_itruncate(). This is the maximum number of extents
61 * freed from a file in a single transaction.
62 */
63 #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
79 xfs_exntfmt_t fmt)
80 {
81 xfs_bmbt_irec_t irec;
82 xfs_bmbt_rec_host_t rec;
92 }
93 }
95 #define xfs_validate_extents(ifp, nrecs, fmt)
98 /*
99 * Check that none of the inode's in the buffer have a next
100 * unlinked field of 0.
101 */
102 #if defined(DEBUG)
103 void
107 {
119 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
120 bp);
122 }
123 }
124 }
125 #endif
127 /*
128 * Find the buffer associated with the given inode map
129 * We do basic validation checks on the buffer once it has been
130 * retrieved from disk.
131 */
132 STATIC int
138 uint buf_flags,
139 uint imap_flags)
140 {
151 "xfs_imap_to_bp: xfs_trans_read_buf()returned "
156 }
158 }
160 /*
161 * Validate the magic number and version of every inode in the buffer
162 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
163 */
164 #ifdef DEBUG
168 #endif
179 XFS_ERRTAG_ITOBP_INOTOBP,
180 XFS_RANDOM_ITOBP_INOTOBP))) {
184 }
187 #ifdef DEBUG
189 "Device %s - bad inode magic/vsn "
194 #endif
197 }
198 }
202 /*
203 * Mark the buffer as an inode buffer now that it looks good
204 */
209 }
211 /*
212 * This routine is called to map an inode number within a file
213 * system to the buffer containing the on-disk version of the
214 * inode. It returns a pointer to the buffer containing the
215 * on-disk inode in the bpp parameter, and in the dip parameter
216 * it returns a pointer to the on-disk inode within that buffer.
217 *
218 * If a non-zero error is returned, then the contents of bpp and
219 * dipp are undefined.
220 *
221 * Use xfs_imap() to determine the size and location of the
222 * buffer to read from disk.
223 */
224 STATIC int
228 xfs_ino_t ino,
232 {
233 xfs_imap_t imap;
250 }
253 /*
254 * This routine is called to map an inode to the buffer containing
255 * the on-disk version of the inode. It returns a pointer to the
256 * buffer containing the on-disk inode in the bpp parameter, and in
257 * the dip parameter it returns a pointer to the on-disk inode within
258 * that buffer.
259 *
260 * If a non-zero error is returned, then the contents of bpp and
261 * dipp are undefined.
262 *
263 * If the inode is new and has not yet been initialized, use xfs_imap()
264 * to determine the size and location of the buffer to read from disk.
265 * If the inode has already been mapped to its buffer and read in once,
266 * then use the mapping information stored in the inode rather than
267 * calling xfs_imap(). This allows us to avoid the overhead of looking
268 * at the inode btree for small block file systems (see xfs_dilocate()).
269 * We can tell whether the inode has been mapped in before by comparing
270 * its disk block address to 0. Only uninitialized inodes will have
271 * 0 for the disk block address.
272 */
273 int
280 xfs_daddr_t bno,
281 uint imap_flags,
282 uint buf_flags)
283 {
284 xfs_imap_t imap;
295 /*
296 * Fill in the fields in the inode that will be used to
297 * map the inode to its buffer from now on.
298 */
303 /*
304 * We've already mapped the inode once, so just use the
305 * mapping that we saved the first time.
306 */
310 }
322 }
327 }
329 /*
330 * Move inode type and inode format specific information from the
331 * on-disk inode to the in-core inode. For fifos, devs, and sockets
332 * this means set if_rdev to the proper value. For files, directories,
333 * and symlinks this means to bring in the in-line data or extent
334 * pointers. For a file in B-tree format, only the root is immediately
335 * brought in-core. The rest will be in-lined in if_extents when it
336 * is first referenced (see xfs_iread_extents()).
337 */
338 STATIC int
342 {
346 xfs_fsize_t di_size;
364 }
374 }
385 }
396 /*
397 * no local regular files yet
398 */
401 "corrupt inode %Lu "
405 XFS_ERRLEVEL_LOW,
408 }
413 "corrupt inode %Lu "
418 XFS_ERRLEVEL_LOW,
421 }
436 }
442 }
445 }
467 }
472 }
474 }
476 /*
477 * The file is in-lined in the on-disk inode.
478 * If it fits into if_inline_data, then copy
479 * it there, otherwise allocate a buffer for it
480 * and copy the data there. Either way, set
481 * if_data to point at the data.
482 * If we allocate a buffer for the data, make
483 * sure that its size is a multiple of 4 and
484 * record the real size in i_real_bytes.
485 */
486 STATIC int
492 {
496 /*
497 * If the size is unreasonable, then something
498 * is wrong and we just bail out rather than crash in
499 * kmem_alloc() or memcpy() below.
500 */
503 "corrupt inode %Lu "
510 }
520 }
528 }
530 /*
531 * The file consists of a set of extents all
532 * of which fit into the on-disk inode.
533 * If there are few enough extents to fit into
534 * the if_inline_ext, then copy them there.
535 * Otherwise allocate a buffer for them and copy
536 * them into it. Either way, set if_extents
537 * to point at the extents.
538 */
539 STATIC int
544 {
555 /*
556 * If the number of extents is unreasonable, then something
557 * is wrong and we just bail out rather than crash in
558 * kmem_alloc() or memcpy() below.
559 */
567 }
574 else
585 }
592 XFS_ERRLEVEL_LOW,
595 }
596 }
599 }
601 /*
602 * The file has too many extents to fit into
603 * the inode, so they are in B-tree format.
604 * Allocate a buffer for the root of the B-tree
605 * and copy the root into it. The i_extents
606 * field will remain NULL until all of the
607 * extents are read in (when they are needed).
608 */
609 STATIC int
614 {
617 /* REFERENCED */
626 /*
627 * blow out if -- fork has less extents than can fit in
628 * fork (fork shouldn't be a btree format), root btree
629 * block has more records than can fit into the fork,
630 * or the number of extents is greater than the number of
631 * blocks.
632 */
643 }
648 /*
649 * Copy and convert from the on-disk structure
650 * to the in-memory structure.
651 */
658 }
660 void
664 {
693 }
695 void
699 {
728 }
730 STATIC uint
732 __uint16_t di_flags)
733 {
765 }
768 }
770 uint
773 {
778 }
780 uint
783 {
788 }
790 /*
791 * Given a mount structure and an inode number, return a pointer
792 * to a newly allocated in-core inode corresponding to the given
793 * inode number.
794 *
795 * Initialize the inode's attributes and extent pointers if it
796 * already has them (it will not if the inode has no links).
797 */
798 int
802 xfs_ino_t ino,
804 xfs_daddr_t bno,
805 uint imap_flags)
806 {
820 /*
821 * Get pointer's to the on-disk inode and the buffer containing it.
822 * If the inode number refers to a block outside the file system
823 * then xfs_itobp() will return NULL. In this case we should
824 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
825 * know that this is a new incore inode.
826 */
831 }
833 /*
834 * Initialize inode's trace buffers.
835 * Do this before xfs_iformat in case it adds entries.
836 */
837 #ifdef XFS_INODE_TRACE
839 #endif
840 #ifdef XFS_BMAP_TRACE
842 #endif
843 #ifdef XFS_BMBT_TRACE
845 #endif
846 #ifdef XFS_RW_TRACE
848 #endif
849 #ifdef XFS_ILOCK_TRACE
851 #endif
852 #ifdef XFS_DIR2_TRACE
854 #endif
856 /*
857 * If we got something that isn't an inode it means someone
858 * (nfs or dmi) has a stale handle.
859 */
863 #ifdef DEBUG
865 "dip->di_core.di_magic (0x%x) != "
868 XFS_DINODE_MAGIC);
871 }
873 /*
874 * If the on-disk inode is already linked to a directory
875 * entry, copy all of the inode into the in-core inode.
876 * xfs_iformat() handles copying in the inode format
877 * specific information.
878 * Otherwise, just get the truly permanent information.
879 */
886 #ifdef DEBUG
889 error);
892 }
898 /*
899 * Make sure to pull in the mode here as well in
900 * case the inode is released without being used.
901 * This ensures that xfs_inactive() will see that
902 * the inode is already free and not try to mess
903 * with the uninitialized part of it.
904 */
906 /*
907 * Initialize the per-fork minima and maxima for a new
908 * inode here. xfs_iformat will do it for old inodes.
909 */
912 }
916 /*
917 * The inode format changed when we moved the link count and
918 * made it 32 bits long. If this is an old format inode,
919 * convert it in memory to look like a new one. If it gets
920 * flushed to disk we will convert back before flushing or
921 * logging it. We zero out the new projid field and the old link
922 * count field. We'll handle clearing the pad field (the remains
923 * of the old uuid field) when we actually convert the inode to
924 * the new format. We don't change the version number so that we
925 * can distinguish this from a real new format inode.
926 */
931 }
936 /*
937 * Mark the buffer containing the inode as something to keep
938 * around for a while. This helps to keep recently accessed
939 * meta-data in-core longer.
940 */
943 /*
944 * Use xfs_trans_brelse() to release the buffer containing the
945 * on-disk inode, because it was acquired with xfs_trans_read_buf()
946 * in xfs_itobp() above. If tp is NULL, this is just a normal
947 * brelse(). If we're within a transaction, then xfs_trans_brelse()
948 * will only release the buffer if it is not dirty within the
949 * transaction. It will be OK to release the buffer in this case,
950 * because inodes on disk are never destroyed and we will be
951 * locking the new in-core inode before putting it in the hash
952 * table where other processes can find it. Thus we don't have
953 * to worry about the inode being changed just because we released
954 * the buffer.
955 */
959 }
961 /*
962 * Read in extents from a btree-format inode.
963 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
964 */
965 int
970 {
973 xfs_extnum_t nextents;
980 }
985 /*
986 * We know that the size is valid (it's checked in iformat_btree)
987 */
997 }
1000 }
1002 /*
1003 * Allocate an inode on disk and return a copy of its in-core version.
1004 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1005 * appropriately within the inode. The uid and gid for the inode are
1006 * set according to the contents of the given cred structure.
1007 *
1008 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1009 * has a free inode available, call xfs_iget()
1010 * to obtain the in-core version of the allocated inode. Finally,
1011 * fill in the inode and log its initial contents. In this case,
1012 * ialloc_context would be set to NULL and call_again set to false.
1013 *
1014 * If xfs_dialloc() does not have an available inode,
1015 * it will replenish its supply by doing an allocation. Since we can
1016 * only do one allocation within a transaction without deadlocks, we
1017 * must commit the current transaction before returning the inode itself.
1018 * In this case, therefore, we will set call_again to true and return.
1019 * The caller should then commit the current transaction, start a new
1020 * transaction, and call xfs_ialloc() again to actually get the inode.
1021 *
1022 * To ensure that some other process does not grab the inode that
1023 * was allocated during the first call to xfs_ialloc(), this routine
1024 * also returns the [locked] bp pointing to the head of the freelist
1025 * as ialloc_context. The caller should hold this buffer across
1026 * the commit and pass it back into this routine on the second call.
1027 *
1028 * If we are allocating quota inodes, we do not have a parent inode
1029 * to attach to or associate with (i.e. pip == NULL) because they
1030 * are not linked into the directory structure - they are attached
1031 * directly to the superblock - and so have no parent.
1032 */
1033 int
1037 mode_t mode,
1038 xfs_nlink_t nlink,
1039 xfs_dev_t rdev,
1041 xfs_prid_t prid,
1046 {
1047 xfs_ino_t ino;
1049 uint flags;
1051 timespec_t tv;
1053 /*
1054 * Call the space management code to pick
1055 * the on-disk inode to be allocated.
1056 */
1061 }
1065 }
1068 /*
1069 * Get the in-core inode with the lock held exclusively.
1070 * This is because we're setting fields here we need
1071 * to prevent others from looking at until we're done.
1072 */
1077 }
1089 /*
1090 * If the superblock version is up to where we support new format
1091 * inodes and this is currently an old format inode, then change
1092 * the inode version number now. This way we only do the conversion
1093 * here rather than here and in the flush/logging code.
1094 */
1098 /*
1099 * We've already zeroed the old link count, the projid field,
1100 * and the pad field.
1101 */
1102 }
1104 /*
1105 * Project ids won't be stored on disk if we are using a version 1 inode.
1106 */
1114 }
1115 }
1117 /*
1118 * If the group ID of the new file does not match the effective group
1119 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1120 * (and only if the irix_sgid_inherit compatibility variable is set).
1121 */
1126 }
1139 /*
1140 * di_gen will have been taken care of in xfs_iread.
1141 */
1164 }
1165 /* fall through */
1176 }
1183 }
1184 }
1186 xfs_inherit_noatime)
1189 xfs_inherit_nodump)
1192 xfs_inherit_sync)
1195 xfs_inherit_nosymlinks)
1200 xfs_inherit_nodefrag)
1205 }
1206 /* FALLTHROUGH */
1215 }
1216 /*
1217 * Attribute fork settings for new inode.
1218 */
1222 /*
1223 * Log the new values stuffed into the inode.
1224 */
1227 /* now that we have an i_mode we can setup inode ops and unlock */
1232 }
1234 /*
1235 * Check to make sure that there are no blocks allocated to the
1236 * file beyond the size of the file. We don't check this for
1237 * files with fixed size extents or real time extents, but we
1238 * at least do it for regular files.
1239 */
1240 #ifdef DEBUG
1241 void
1245 xfs_fsize_t isize)
1246 {
1247 xfs_fileoff_t map_first;
1262 /*
1263 * The filesystem could be shutting down, so bmapi may return
1264 * an error.
1265 */
1269 map_first),
1275 }
1278 /*
1279 * Calculate the last possible buffered byte in a file. This must
1280 * include data that was buffered beyond the EOF by the write code.
1281 * This also needs to deal with overflowing the xfs_fsize_t type
1282 * which can happen for sizes near the limit.
1283 *
1284 * We also need to take into account any blocks beyond the EOF. It
1285 * may be the case that they were buffered by a write which failed.
1286 * In that case the pages will still be in memory, but the inode size
1287 * will never have been updated.
1288 */
1289 xfs_fsize_t
1292 {
1294 xfs_fsize_t last_byte;
1295 xfs_fileoff_t last_block;
1296 xfs_fileoff_t size_last_block;
1302 /*
1303 * Only check for blocks beyond the EOF if the extents have
1304 * been read in. This eliminates the need for the inode lock,
1305 * and it also saves us from looking when it really isn't
1306 * necessary.
1307 */
1310 XFS_DATA_FORK);
1313 }
1316 }
1323 }
1327 }
1329 }
1331 #if defined(XFS_RW_TRACE)
1332 STATIC void
1337 xfs_fsize_t new_size,
1338 xfs_off_t toss_start,
1339 xfs_off_t toss_finish)
1340 {
1343 }
1362 }
1363 #else
1364 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1365 #endif
1367 /*
1368 * Start the truncation of the file to new_size. The new size
1369 * must be smaller than the current size. This routine will
1370 * clear the buffer and page caches of file data in the removed
1371 * range, and xfs_itruncate_finish() will remove the underlying
1372 * disk blocks.
1373 *
1374 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1375 * must NOT have the inode lock held at all. This is because we're
1376 * calling into the buffer/page cache code and we can't hold the
1377 * inode lock when we do so.
1378 *
1379 * We need to wait for any direct I/Os in flight to complete before we
1380 * proceed with the truncate. This is needed to prevent the extents
1381 * being read or written by the direct I/Os from being removed while the
1382 * I/O is in flight as there is no other method of synchronising
1383 * direct I/O with the truncate operation. Also, because we hold
1384 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1385 * started until the truncate completes and drops the lock. Essentially,
1386 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1387 * between direct I/Os and the truncate operation.
1388 *
1389 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1390 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1391 * in the case that the caller is locking things out of order and
1392 * may not be able to call xfs_itruncate_finish() with the inode lock
1393 * held without dropping the I/O lock. If the caller must drop the
1394 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1395 * must be called again with all the same restrictions as the initial
1396 * call.
1397 */
1398 int
1401 uint flags,
1402 xfs_fsize_t new_size)
1403 {
1404 xfs_fsize_t last_byte;
1405 xfs_off_t toss_start;
1416 /* wait for the completion of any pending DIOs */
1420 /*
1421 * Call toss_pages or flushinval_pages to get rid of pages
1422 * overlapping the region being removed. We have to use
1423 * the less efficient flushinval_pages in the case that the
1424 * caller may not be able to finish the truncate without
1425 * dropping the inode's I/O lock. Make sure
1426 * to catch any pages brought in by buffers overlapping
1427 * the EOF by searching out beyond the isize by our
1428 * block size. We round new_size up to a block boundary
1429 * so that we don't toss things on the same block as
1430 * new_size but before it.
1431 *
1432 * Before calling toss_page or flushinval_pages, make sure to
1433 * call remapf() over the same region if the file is mapped.
1434 * This frees up mapped file references to the pages in the
1435 * given range and for the flushinval_pages case it ensures
1436 * that we get the latest mapped changes flushed out.
1437 */
1441 /*
1442 * The place to start tossing is beyond our maximum
1443 * file size, so there is no way that the data extended
1444 * out there.
1445 */
1447 }
1450 last_byte);
1458 }
1459 }
1461 #ifdef DEBUG
1464 }
1465 #endif
1467 }
1469 /*
1470 * Shrink the file to the given new_size. The new size must be smaller than
1471 * the current size. This will free up the underlying blocks in the removed
1472 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
1473 *
1474 * The transaction passed to this routine must have made a permanent log
1475 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1476 * given transaction and start new ones, so make sure everything involved in
1477 * the transaction is tidy before calling here. Some transaction will be
1478 * returned to the caller to be committed. The incoming transaction must
1479 * already include the inode, and both inode locks must be held exclusively.
1480 * The inode must also be "held" within the transaction. On return the inode
1481 * will be "held" within the returned transaction. This routine does NOT
1482 * require any disk space to be reserved for it within the transaction.
1483 *
1484 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1485 * indicates the fork which is to be truncated. For the attribute fork we only
1486 * support truncation to size 0.
1487 *
1488 * We use the sync parameter to indicate whether or not the first transaction
1489 * we perform might have to be synchronous. For the attr fork, it needs to be
1490 * so if the unlink of the inode is not yet known to be permanent in the log.
1491 * This keeps us from freeing and reusing the blocks of the attribute fork
1492 * before the unlink of the inode becomes permanent.
1493 *
1494 * For the data fork, we normally have to run synchronously if we're being
1495 * called out of the inactive path or we're being called out of the create path
1496 * where we're truncating an existing file. Either way, the truncate needs to
1497 * be sync so blocks don't reappear in the file with altered data in case of a
1498 * crash. wsync filesystems can run the first case async because anything that
1499 * shrinks the inode has to run sync so by the time we're called here from
1500 * inactive, the inode size is permanently set to 0.
1501 *
1502 * Calls from the truncate path always need to be sync unless we're in a wsync
1503 * filesystem and the file has already been unlinked.
1504 *
1505 * The caller is responsible for correctly setting the sync parameter. It gets
1506 * too hard for us to guess here which path we're being called out of just
1507 * based on inode state.
1508 *
1509 * If we get an error, we must return with the inode locked and linked into the
1510 * current transaction. This keeps things simple for the higher level code,
1511 * because it always knows that the inode is locked and held in the transaction
1512 * that returns to it whether errors occur or not. We don't mark the inode
1513 * dirty on error so that transactions can be easily aborted if possible.
1514 */
1515 int
1519 xfs_fsize_t new_size,
1522 {
1523 xfs_fsblock_t first_block;
1524 xfs_fileoff_t first_unmap_block;
1525 xfs_fileoff_t last_block;
1531 xfs_bmap_free_t free_list;
1547 /*
1548 * We only support truncating the entire attribute fork.
1549 */
1552 }
1555 /*
1556 * The first thing we do is set the size to new_size permanently
1557 * on disk. This way we don't have to worry about anyone ever
1558 * being able to look at the data being freed even in the face
1559 * of a crash. What we're getting around here is the case where
1560 * we free a block, it is allocated to another file, it is written
1561 * to, and then we crash. If the new data gets written to the
1562 * file but the log buffers containing the free and reallocation
1563 * don't, then we'd end up with garbage in the blocks being freed.
1564 * As long as we make the new_size permanent before actually
1565 * freeing any blocks it doesn't matter if they get writtten to.
1566 *
1567 * The callers must signal into us whether or not the size
1568 * setting here must be synchronous. There are a few cases
1569 * where it doesn't have to be synchronous. Those cases
1570 * occur if the file is unlinked and we know the unlink is
1571 * permanent or if the blocks being truncated are guaranteed
1572 * to be beyond the inode eof (regardless of the link count)
1573 * and the eof value is permanent. Both of these cases occur
1574 * only on wsync-mounted filesystems. In those cases, we're
1575 * guaranteed that no user will ever see the data in the blocks
1576 * that are being truncated so the truncate can run async.
1577 * In the free beyond eof case, the file may wind up with
1578 * more blocks allocated to it than it needs if we crash
1579 * and that won't get fixed until the next time the file
1580 * is re-opened and closed but that's ok as that shouldn't
1581 * be too many blocks.
1582 *
1583 * However, we can't just make all wsync xactions run async
1584 * because there's one call out of the create path that needs
1585 * to run sync where it's truncating an existing file to size
1586 * 0 whose size is > 0.
1587 *
1588 * It's probably possible to come up with a test in this
1589 * routine that would correctly distinguish all the above
1590 * cases from the values of the function parameters and the
1591 * inode state but for sanity's sake, I've decided to let the
1592 * layers above just tell us. It's simpler to correctly figure
1593 * out in the layer above exactly under what conditions we
1594 * can run async and I think it's easier for others read and
1595 * follow the logic in case something has to be changed.
1596 * cscope is your friend -- rcc.
1597 *
1598 * The attribute fork is much simpler.
1599 *
1600 * For the attribute fork we allow the caller to tell us whether
1601 * the unlink of the inode that led to this call is yet permanent
1602 * in the on disk log. If it is not and we will be freeing extents
1603 * in this inode then we make the first transaction synchronous
1604 * to make sure that the unlink is permanent by the time we free
1605 * the blocks.
1606 */
1609 /*
1610 * If we are not changing the file size then do
1611 * not update the on-disk file size - we may be
1612 * called from xfs_inactive_free_eofblocks(). If we
1613 * update the on-disk file size and then the system
1614 * crashes before the contents of the file are
1615 * flushed to disk then the files may be full of
1616 * holes (ie NULL files bug).
1617 */
1622 }
1623 }
1628 }
1634 /*
1635 * Since it is possible for space to become allocated beyond
1636 * the end of the file (in a crash where the space is allocated
1637 * but the inode size is not yet updated), simply remove any
1638 * blocks which show up between the new EOF and the maximum
1639 * possible file size. If the first block to be removed is
1640 * beyond the maximum file size (ie it is the same as last_block),
1641 * then there is nothing to do.
1642 */
1650 }
1652 /*
1653 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1654 * will tell us whether it freed the entire range or
1655 * not. If this is a synchronous mount (wsync),
1656 * then we can tell bunmapi to keep all the
1657 * transactions asynchronous since the unlink
1658 * transaction that made this inode inactive has
1659 * already hit the disk. There's no danger of
1660 * the freed blocks being reused, there being a
1661 * crash, and the reused blocks suddenly reappearing
1662 * in this file with garbage in them once recovery
1663 * runs.
1664 */
1670 XFS_ITRUNC_MAX_EXTENTS,
1674 /*
1675 * If the bunmapi call encounters an error,
1676 * return to the caller where the transaction
1677 * can be properly aborted. We just need to
1678 * make sure we're not holding any resources
1679 * that we were not when we came in.
1680 */
1683 }
1685 /*
1686 * Duplicate the transaction that has the permanent
1687 * reservation and commit the old transaction.
1688 */
1692 /* link the inode into the next xact in the chain */
1696 }
1699 /*
1700 * If the bmap finish call encounters an error, return
1701 * to the caller where the transaction can be properly
1702 * aborted. We just need to make sure we're not
1703 * holding any resources that we were not when we came
1704 * in.
1705 *
1706 * Aborting from this point might lose some blocks in
1707 * the file system, but oh well.
1708 */
1711 }
1714 /*
1715 * Mark the inode dirty so it will be logged and
1716 * moved forward in the log as part of every commit.
1717 */
1719 }
1725 /* link the inode into the next transaction in the chain */
1732 XFS_TRANS_PERM_LOG_RES,
1733 XFS_ITRUNCATE_LOG_COUNT);
1736 }
1737 /*
1738 * Only update the size in the case of the data fork, but
1739 * always re-log the inode so that our permanent transaction
1740 * can keep on rolling it forward in the log.
1741 */
1744 /*
1745 * If we are not changing the file size then do
1746 * not update the on-disk file size - we may be
1747 * called from xfs_inactive_free_eofblocks(). If we
1748 * update the on-disk file size and then the system
1749 * crashes before the contents of the file are
1750 * flushed to disk then the files may be full of
1751 * holes (ie NULL files bug).
1752 */
1756 }
1757 }
1767 }
1769 /*
1770 * This is called when the inode's link count goes to 0.
1771 * We place the on-disk inode on a list in the AGI. It
1772 * will be pulled from this list when the inode is freed.
1773 */
1774 int
1778 {
1784 xfs_agnumber_t agno;
1785 xfs_daddr_t agdaddr;
1786 xfs_agino_t agino;
1801 /*
1802 * Get the agi buffer first. It ensures lock ordering
1803 * on the list.
1804 */
1810 /*
1811 * Validate the magic number of the agi block.
1812 */
1814 agi_ok =
1818 XFS_RANDOM_IUNLINK))) {
1822 }
1823 /*
1824 * Get the index into the agi hash table for the
1825 * list this inode will go on.
1826 */
1834 /*
1835 * There is already another inode in the bucket we need
1836 * to add ourselves to. Add us at the front of the list.
1837 * Here we put the head pointer into our next pointer,
1838 * and then we fall through to point the head at us.
1839 */
1845 /* both on-disk, don't endian flip twice */
1853 }
1855 /*
1856 * Point the bucket head pointer at the inode being inserted.
1857 */
1865 }
1867 /*
1868 * Pull the on-disk inode from the AGI unlinked list.
1869 */
1870 STATIC int
1874 {
1875 xfs_ino_t next_ino;
1881 xfs_agnumber_t agno;
1882 xfs_daddr_t agdaddr;
1883 xfs_agino_t agino;
1884 xfs_agino_t next_agino;
1892 /*
1893 * First pull the on-disk inode from the AGI unlinked list.
1894 */
1900 /*
1901 * Get the agi buffer first. It ensures lock ordering
1902 * on the list.
1903 */
1911 }
1912 /*
1913 * Validate the magic number of the agi block.
1914 */
1916 agi_ok =
1920 XFS_RANDOM_IUNLINK_REMOVE))) {
1928 }
1929 /*
1930 * Get the index into the agi hash table for the
1931 * list this inode will go on.
1932 */
1940 /*
1941 * We're at the head of the list. Get the inode's
1942 * on-disk buffer to see if there is anyone after us
1943 * on the list. Only modify our next pointer if it
1944 * is not already NULLAGINO. This saves us the overhead
1945 * of dealing with the buffer when there is no need to
1946 * change it.
1947 */
1954 }
1967 }
1968 /*
1969 * Point the bucket head pointer at the next inode.
1970 */
1979 /*
1980 * We need to search the list for the inode being freed.
1981 */
1985 /*
1986 * If the last inode wasn't the one pointing to
1987 * us, then release its buffer since we're not
1988 * going to do anything with it.
1989 */
1992 }
2001 }
2005 }
2006 /*
2007 * Now last_ibp points to the buffer previous to us on
2008 * the unlinked list. Pull us from the list.
2009 */
2016 }
2030 }
2031 /*
2032 * Point the previous inode on the list to the next inode.
2033 */
2041 }
2043 }
2045 STATIC void
2049 xfs_ino_t inum)
2050 {
2056 xfs_daddr_t blkno;
2072 }
2081 /*
2082 * Look for each inode in memory and attempt to lock it,
2083 * we can be racing with flush and tail pushing here.
2084 * any inode we get the locks on, add to an array of
2085 * inode items to process later.
2086 *
2087 * The get the buffer lock, we could beat a flush
2088 * or tail pushing thread to the lock here, in which
2089 * case they will go looking for the inode buffer
2090 * and fail, we need some other form of interlock
2091 * here.
2092 */
2099 /* Inode not in memory or we found it already,
2100 * nothing to do
2101 */
2105 }
2110 }
2112 /* If we can get the locks then add it to the
2113 * list, otherwise by the time we get the bp lock
2114 * below it will already be attached to the
2115 * inode buffer.
2116 */
2118 /* This inode will already be locked - by us, lets
2119 * keep it that way.
2120 */
2129 }
2130 }
2133 }
2144 }
2147 }
2148 }
2150 }
2154 XFS_BUF_LOCK);
2167 pre_flushed++;
2168 }
2170 }
2181 }
2195 }
2196 }
2201 }
2205 }
2207 /*
2208 * This is called to return an inode to the inode free list.
2209 * The inode should already be truncated to 0 length and have
2210 * no pages associated with it. This routine also assumes that
2211 * the inode is already a part of the transaction.
2212 *
2213 * The on-disk copy of the inode will have been added to the list
2214 * of unlinked inodes in the AGI. We need to remove the inode from
2215 * that list atomically with respect to freeing it here.
2216 */
2217 int
2222 {
2225 xfs_ino_t first_ino;
2238 /*
2239 * Pull the on-disk inode from the AGI unlinked list.
2240 */
2244 }
2249 }
2258 /*
2259 * Bump the generation count so no one will be confused
2260 * by reincarnations of this inode.
2261 */
2270 /*
2271 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2272 * from picking up this inode when it is reclaimed (its incore state
2273 * initialzed but not flushed to disk yet). The in-core di_mode is
2274 * already cleared and a corresponding transaction logged.
2275 * The hack here just synchronizes the in-core to on-disk
2276 * di_mode value in advance before the actual inode sync to disk.
2277 * This is OK because the inode is already unlinked and would never
2278 * change its di_mode again for this inode generation.
2279 * This is a temporary hack that would require a proper fix
2280 * in the future.
2281 */
2286 }
2289 }
2291 /*
2292 * Reallocate the space for if_broot based on the number of records
2293 * being added or deleted as indicated in rec_diff. Move the records
2294 * and pointers in if_broot to fit the new size. When shrinking this
2295 * will eliminate holes between the records and pointers created by
2296 * the caller. When growing this will create holes to be filled in
2297 * by the caller.
2298 *
2299 * The caller must not request to add more records than would fit in
2300 * the on-disk inode root. If the if_broot is currently NULL, then
2301 * if we adding records one will be allocated. The caller must also
2302 * not request that the number of records go below zero, although
2303 * it can go to zero.
2304 *
2305 * ip -- the inode whose if_broot area is changing
2306 * ext_diff -- the change in the number of records, positive or negative,
2307 * requested for the if_broot array.
2308 */
2309 void
2314 {
2323 /*
2324 * Handle the degenerate case quietly.
2325 */
2328 }
2332 /*
2333 * If there wasn't any memory allocated before, just
2334 * allocate it now and get out.
2335 */
2339 KM_SLEEP);
2342 }
2344 /*
2345 * If there is already an existing if_broot, then we need
2346 * to realloc() it and shift the pointers to their new
2347 * location. The records don't change location because
2348 * they are kept butted up against the btree block header.
2349 */
2355 new_size,
2357 KM_SLEEP);
2367 }
2369 /*
2370 * rec_diff is less than 0. In this case, we are shrinking the
2371 * if_broot buffer. It must already exist. If we go to zero
2372 * records, just get rid of the root and clear the status bit.
2373 */
2380 else
2384 /*
2385 * First copy over the btree block header.
2386 */
2391 }
2393 /*
2394 * Only copy the records and pointers if there are any.
2395 */
2397 /*
2398 * First copy the records.
2399 */
2406 /*
2407 * Then copy the pointers.
2408 */
2414 }
2421 }
2424 /*
2425 * This is called when the amount of space needed for if_data
2426 * is increased or decreased. The change in size is indicated by
2427 * the number of bytes that need to be added or deleted in the
2428 * byte_diff parameter.
2429 *
2430 * If the amount of space needed has decreased below the size of the
2431 * inline buffer, then switch to using the inline buffer. Otherwise,
2432 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2433 * to what is needed.
2434 *
2435 * ip -- the inode whose if_data area is changing
2436 * byte_diff -- the change in the number of bytes, positive or negative,
2437 * requested for the if_data array.
2438 */
2439 void
2444 {
2451 }
2460 }
2464 /*
2465 * If the valid extents/data can fit in if_inline_ext/data,
2466 * copy them from the malloc'd vector and free it.
2467 */
2473 new_size);
2476 }
2479 /*
2480 * Stuck with malloc/realloc.
2481 * For inline data, the underlying buffer must be
2482 * a multiple of 4 bytes in size so that it can be
2483 * logged and stay on word boundaries. We enforce
2484 * that here.
2485 */
2491 /*
2492 * Only do the realloc if the underlying size
2493 * is really changing.
2494 */
2498 real_size,
2500 KM_SLEEP);
2501 }
2507 }
2508 }
2512 }
2517 /*
2518 * Map inode to disk block and offset.
2519 *
2520 * mp -- the mount point structure for the current file system
2521 * tp -- the current transaction
2522 * ino -- the inode number of the inode to be located
2523 * imap -- this structure is filled in with the information necessary
2524 * to retrieve the given inode from disk
2525 * flags -- flags to pass to xfs_dilocate indicating whether or not
2526 * lookups in the inode btree were OK or not
2527 */
2528 int
2532 xfs_ino_t ino,
2534 uint flags)
2535 {
2536 xfs_fsblock_t fsbno;
2553 /*
2554 * If the inode number maps to a block outside the bounds
2555 * of the file system then return NULL rather than calling
2556 * read_buf and panicing when we get an error from the
2557 * driver.
2558 */
2562 "(imap->im_blkno (0x%llx) + imap->im_len (0x%llx)) > "
2568 }
2570 }
2572 void
2576 {
2583 }
2585 /*
2586 * If the format is local, then we can't have an extents
2587 * array so just look for an inline data array. If we're
2588 * not local then we may or may not have an extents list,
2589 * so check and free it up if we do.
2590 */
2598 }
2605 }
2612 }
2613 }
2615 /*
2616 * This is called free all the memory associated with an inode.
2617 * It must free the inode itself and any buffers allocated for
2618 * if_extents/if_data and if_broot. It must also free the lock
2619 * associated with the inode.
2620 */
2621 void
2624 {
2631 }
2637 #ifdef XFS_INODE_TRACE
2639 #endif
2640 #ifdef XFS_BMAP_TRACE
2642 #endif
2643 #ifdef XFS_BMBT_TRACE
2645 #endif
2646 #ifdef XFS_RW_TRACE
2648 #endif
2649 #ifdef XFS_ILOCK_TRACE
2651 #endif
2652 #ifdef XFS_DIR2_TRACE
2654 #endif
2656 /*
2657 * Only if we are shutting down the fs will we see an
2658 * inode still in the AIL. If it is there, we should remove
2659 * it to prevent a use-after-free from occurring.
2660 */
2670 else
2672 }
2674 }
2676 }
2679 /*
2680 * Increment the pin count of the given buffer.
2681 * This value is protected by ipinlock spinlock in the mount structure.
2682 */
2683 void
2686 {
2690 }
2692 /*
2693 * Decrement the pin count of the given inode, and wake up
2694 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2695 * inode must have been previously pinned with a call to xfs_ipin().
2696 */
2697 void
2700 {
2705 }
2707 /*
2708 * This is called to unpin an inode. It can be directed to wait or to return
2709 * immediately without waiting for the inode to be unpinned. The caller must
2710 * have the inode locked in at least shared mode so that the buffer cannot be
2711 * subsequently pinned once someone is waiting for it to be unpinned.
2712 */
2713 STATIC void
2717 {
2724 /* Give the log a push to start the unpinning I/O */
2729 }
2734 {
2736 }
2741 {
2743 }
2746 /*
2747 * xfs_iextents_copy()
2748 *
2749 * This is called to copy the REAL extents (as opposed to the delayed
2750 * allocation extents) from the inode into the given buffer. It
2751 * returns the number of bytes copied into the buffer.
2752 *
2753 * If there are no delayed allocation extents, then we can just
2754 * memcpy() the extents into the buffer. Otherwise, we need to
2755 * examine each extent in turn and skip those which are delayed.
2756 */
2757 int
2762 {
2767 xfs_fsblock_t start_block;
2777 /*
2778 * There are some delayed allocation extents in the
2779 * inode, so copy the extents one at a time and skip
2780 * the delayed ones. There must be at least one
2781 * non-delayed extent.
2782 */
2788 /*
2789 * It's a delayed allocation extent, so skip it.
2790 */
2792 }
2794 /* Translate to on disk format */
2797 dp++;
2798 copied++;
2799 }
2804 }
2806 /*
2807 * Each of the following cases stores data into the same region
2808 * of the on-disk inode, so only one of them can be valid at
2809 * any given time. While it is possible to have conflicting formats
2810 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2811 * in EXTENTS format, this can only happen when the fork has
2812 * changed formats after being modified but before being flushed.
2813 * In these cases, the format always takes precedence, because the
2814 * format indicates the current state of the fork.
2815 */
2816 /*ARGSUSED*/
2817 STATIC void
2824 {
2828 #ifdef XFS_TRANS_DEBUG
2830 #endif
2841 /*
2842 * This can happen if we gave up in iformat in an error path,
2843 * for the attribute fork.
2844 */
2848 }
2858 }
2872 whichfork);
2873 }
2882 XFS_BROOT_SIZE_ADJ));
2886 }
2893 }
2901 }
2907 }
2908 }
2910 STATIC int
2914 {
2939 /* really need a gang lookup range call here */
2949 /* if the inode lies outside this cluster, we're done. */
2952 /*
2953 * Do an un-protected check to see if the inode is dirty and
2954 * is a candidate for flushing. These checks will be repeated
2955 * later after the appropriate locks are acquired.
2956 */
2960 /*
2961 * Try to get locks. If any are unavailable or it is pinned,
2962 * then this inode cannot be flushed and is skipped.
2963 */
2970 }
2975 }
2977 /*
2978 * arriving here means that this inode can be flushed. First
2979 * re-check that it's dirty before flushing.
2980 */
2987 }
2988 clcount++;
2991 }
2993 }
2998 }
3000 out_free:
3006 cluster_corrupt_out:
3007 /*
3008 * Corruption detected in the clustering loop. Invalidate the
3009 * inode buffer and shut down the filesystem.
3010 */
3012 /*
3013 * Clean up the buffer. If it was B_DELWRI, just release it --
3014 * brelse can handle it with no problems. If not, shut down the
3015 * filesystem before releasing the buffer.
3016 */
3024 /*
3025 * Just like incore_relse: if we have b_iodone functions,
3026 * mark the buffer as an error and call them. Otherwise
3027 * mark it as stale and brelse.
3028 */
3039 }
3040 }
3042 /*
3043 * Unlocks the flush lock
3044 */
3048 }
3050 /*
3051 * xfs_iflush() will write a modified inode's changes out to the
3052 * inode's on disk home. The caller must have the inode lock held
3053 * in at least shared mode and the inode flush completion must be
3054 * active as well. The inode lock will still be held upon return from
3055 * the call and the caller is free to unlock it.
3056 * The inode flush will be completed when the inode reaches the disk.
3057 * The flags indicate how the inode's buffer should be written out.
3058 */
3059 int
3062 uint flags)
3063 {
3082 /*
3083 * If the inode isn't dirty, then just release the inode
3084 * flush lock and do nothing.
3085 */
3089 }
3091 /*
3092 * We can't flush the inode until it is unpinned, so wait for it if we
3093 * are allowed to block. We know noone new can pin it, because we are
3094 * holding the inode lock shared and you need to hold it exclusively to
3095 * pin the inode.
3096 *
3097 * If we are not allowed to block, force the log out asynchronously so
3098 * that when we come back the inode will be unpinned. If other inodes
3099 * in the same cluster are dirty, they will probably write the inode
3100 * out for us if they occur after the log force completes.
3101 */
3106 }
3109 /*
3110 * This may have been unpinned because the filesystem is shutting
3111 * down forcibly. If that's the case we must not write this inode
3112 * to disk, because the log record didn't make it to disk!
3113 */
3120 }
3122 /*
3123 * Decide how buffer will be flushed out. This is done before
3124 * the call to xfs_iflush_int because this field is zeroed by it.
3125 */
3127 /*
3128 * Flush out the inode buffer according to the directions
3129 * of the caller. In the cases where the caller has given
3130 * us a choice choose the non-delwri case. This is because
3131 * the inode is in the AIL and we need to get it out soon.
3132 */
3150 }
3169 }
3170 }
3172 /*
3173 * Get the buffer containing the on-disk inode.
3174 */
3180 }
3182 /*
3183 * First flush out the inode that xfs_iflush was called with.
3184 */
3189 /*
3190 * If the buffer is pinned then push on the log now so we won't
3191 * get stuck waiting in the write for too long.
3192 */
3196 /*
3197 * inode clustering:
3198 * see if other inodes can be gathered into this write
3199 */
3210 }
3213 corrupt_out:
3216 cluster_corrupt_out:
3217 /*
3218 * Unlocks the flush lock
3219 */
3222 }
3225 STATIC int
3229 {
3233 #ifdef XFS_TRANS_DEBUG
3235 #endif
3246 /*
3247 * If the inode isn't dirty, then just release the inode
3248 * flush lock and do nothing.
3249 */
3253 }
3255 /* set *dip = inode's place in the buffer */
3258 /*
3259 * Clear i_update_core before copying out the data.
3260 * This is for coordination with our timestamp updates
3261 * that don't hold the inode lock. They will always
3262 * update the timestamps BEFORE setting i_update_core,
3263 * so if we clear i_update_core after they set it we
3264 * are guaranteed to see their updates to the timestamps.
3265 * I believe that this depends on strongly ordered memory
3266 * semantics, but we have that. We use the SYNCHRONIZE
3267 * macro to make sure that the compiler does not reorder
3268 * the i_update_core access below the data copy below.
3269 */
3273 /*
3274 * Make sure to get the latest atime from the Linux inode.
3275 */
3284 }
3291 }
3301 }
3312 }
3313 }
3316 XFS_RANDOM_IFLUSH_5)) {
3322 ip);
3324 }
3331 }
3332 /*
3333 * bump the flush iteration count, used to detect flushes which
3334 * postdate a log record during recovery.
3335 */
3339 /*
3340 * Copy the dirty parts of the inode into the on-disk
3341 * inode. We always copy out the core of the inode,
3342 * because if the inode is dirty at all the core must
3343 * be.
3344 */
3347 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3351 /*
3352 * If this is really an old format inode and the superblock version
3353 * has not been updated to support only new format inodes, then
3354 * convert back to the old inode format. If the superblock version
3355 * has been updated, then make the conversion permanent.
3356 */
3361 /*
3362 * Convert it back.
3363 */
3367 /*
3368 * The superblock version has already been bumped,
3369 * so just make the conversion to the new inode
3370 * format permanent.
3371 */
3380 }
3381 }
3388 /*
3389 * We've recorded everything logged in the inode, so we'd
3390 * like to clear the ilf_fields bits so we don't log and
3391 * flush things unnecessarily. However, we can't stop
3392 * logging all this information until the data we've copied
3393 * into the disk buffer is written to disk. If we did we might
3394 * overwrite the copy of the inode in the log with all the
3395 * data after re-logging only part of it, and in the face of
3396 * a crash we wouldn't have all the data we need to recover.
3397 *
3398 * What we do is move the bits to the ili_last_fields field.
3399 * When logging the inode, these bits are moved back to the
3400 * ilf_fields field. In the xfs_iflush_done() routine we
3401 * clear ili_last_fields, since we know that the information
3402 * those bits represent is permanently on disk. As long as
3403 * the flush completes before the inode is logged again, then
3404 * both ilf_fields and ili_last_fields will be cleared.
3405 *
3406 * We can play with the ilf_fields bits here, because the inode
3407 * lock must be held exclusively in order to set bits there
3408 * and the flush lock protects the ili_last_fields bits.
3409 * Set ili_logged so the flush done
3410 * routine can tell whether or not to look in the AIL.
3411 * Also, store the current LSN of the inode so that we can tell
3412 * whether the item has moved in the AIL from xfs_iflush_done().
3413 * In order to read the lsn we need the AIL lock, because
3414 * it is a 64 bit value that cannot be read atomically.
3415 */
3426 /*
3427 * Attach the function xfs_iflush_done to the inode's
3428 * buffer. This will remove the inode from the AIL
3429 * and unlock the inode's flush lock when the inode is
3430 * completely written to disk.
3431 */
3438 /*
3439 * We're flushing an inode which is not in the AIL and has
3440 * not been logged but has i_update_core set. For this
3441 * case we can use a B_DELWRI flush and immediately drop
3442 * the inode flush lock because we can avoid the whole
3443 * AIL state thing. It's OK to drop the flush lock now,
3444 * because we've already locked the buffer and to do anything
3445 * you really need both.
3446 */
3451 }
3453 }
3457 corrupt_out:
3459 }
3462 /*
3463 * Flush all inactive inodes in mp.
3464 */
3465 void
3468 {
3471 again:
3478 /* Make sure we skip markers inserted by sync */
3482 }
3488 }
3494 out:
3496 }
3498 #ifdef XFS_ILOCK_TRACE
3501 void
3503 {
3512 }
3513 #endif
3515 /*
3516 * Return a pointer to the extent record at file index idx.
3517 */
3518 xfs_bmbt_rec_host_t *
3522 {
3537 }
3538 }
3540 /*
3541 * Insert new item(s) into the extent records for incore inode
3542 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3543 */
3544 void
3550 {
3557 }
3559 /*
3560 * This is called when the amount of space required for incore file
3561 * extents needs to be increased. The ext_diff parameter stores the
3562 * number of new extents being added and the idx parameter contains
3563 * the extent index where the new extents will be added. If the new
3564 * extents are being appended, then we just need to (re)allocate and
3565 * initialize the space. Otherwise, if the new extents are being
3566 * inserted into the middle of the existing entries, a bit more work
3567 * is required to make room for the new extents to be inserted. The
3568 * caller is responsible for filling in the new extent entries upon
3569 * return.
3570 */
3571 void
3576 {
3585 /*
3586 * If the new number of extents (nextents + ext_diff)
3587 * fits inside the inode, then continue to use the inline
3588 * extent buffer.
3589 */
3596 }
3600 }
3601 /*
3602 * Otherwise use a linear (direct) extent list.
3603 * If the extents are currently inside the inode,
3604 * xfs_iext_realloc_direct will switch us from
3605 * inline to direct extent allocation mode.
3606 */
3614 }
3615 }
3616 /* Indirection array */
3629 }
3630 /* Extents fit in target extent page */
3638 }
3641 }
3642 /* Insert a new extent page */
3646 }
3647 /*
3648 * If extent(s) are being appended to the last page in
3649 * the indirection array and the new extent(s) don't fit
3650 * in the page, then erp is NULL and erp_idx is set to
3651 * the next index needed in the indirection array.
3652 */
3661 erp_idx++;
3662 }
3663 }
3664 }
3665 }
3667 }
3669 /*
3670 * This is called when incore extents are being added to the indirection
3671 * array and the new extents do not fit in the target extent list. The
3672 * erp_idx parameter contains the irec index for the target extent list
3673 * in the indirection array, and the idx parameter contains the extent
3674 * index within the list. The number of extents being added is stored
3675 * in the count parameter.
3676 *
3677 * |-------| |-------|
3678 * | | | | idx - number of extents before idx
3679 * | idx | | count |
3680 * | | | | count - number of extents being inserted at idx
3681 * |-------| |-------|
3682 * | count | | nex2 | nex2 - number of extents after idx + count
3683 * |-------| |-------|
3684 */
3685 void
3691 {
3705 /*
3706 * Save second part of target extent list
3707 * (all extents past */
3715 }
3717 /*
3718 * Add the new extents to the end of the target
3719 * list, then allocate new irec record(s) and
3720 * extent buffer(s) as needed to store the rest
3721 * of the new extents.
3722 */
3729 }
3731 erp_idx++;
3737 }
3739 /* Add nex2 extents back to indirection array */
3741 xfs_extnum_t ext_avail;
3747 /*
3748 * If nex2 extents fit in the current page, append
3749 * nex2_ep after the new extents.
3750 */
3753 }
3754 /*
3755 * Otherwise, check if space is available in the
3756 * next page.
3757 */
3761 erp_idx++;
3762 erp++;
3763 /* Create a hole for nex2 extents */
3766 }
3767 /*
3768 * Final choice, create a new extent page for
3769 * nex2 extents.
3770 */
3772 erp_idx++;
3774 }
3779 }
3780 }
3782 /*
3783 * This is called when the amount of space required for incore file
3784 * extents needs to be decreased. The ext_diff parameter stores the
3785 * number of extents to be removed and the idx parameter contains
3786 * the extent index where the extents will be removed from.
3787 *
3788 * If the amount of space needed has decreased below the linear
3789 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3790 * extent array. Otherwise, use kmem_realloc() to adjust the
3791 * size to what is needed.
3792 */
3793 void
3798 {
3814 }
3816 }
3818 /*
3819 * This removes ext_diff extents from the inline buffer, beginning
3820 * at extent index idx.
3821 */
3822 void
3827 {
3846 }
3847 }
3849 /*
3850 * This removes ext_diff extents from a linear (direct) extent list,
3851 * beginning at extent index idx. If the extents are being removed
3852 * from the end of the list (ie. truncate) then we just need to re-
3853 * allocate the list to remove the extra space. Otherwise, if the
3854 * extents are being removed from the middle of the existing extent
3855 * entries, then we first need to move the extent records beginning
3856 * at idx + ext_diff up in the list to overwrite the records being
3857 * removed, then remove the extra space via kmem_realloc.
3858 */
3859 void
3864 {
3876 }
3877 /* Move extents up in the list (if needed) */
3883 }
3886 /*
3887 * Reallocate the direct extent list. If the extents
3888 * will fit inside the inode then xfs_iext_realloc_direct
3889 * will switch from direct to inline extent allocation
3890 * mode for us.
3891 */
3894 }
3896 /*
3897 * This is called when incore extents are being removed from the
3898 * indirection array and the extents being removed span multiple extent
3899 * buffers. The idx parameter contains the file extent index where we
3900 * want to begin removing extents, and the count parameter contains
3901 * how many extents need to be removed.
3902 *
3903 * |-------| |-------|
3904 * | nex1 | | | nex1 - number of extents before idx
3905 * |-------| | count |
3906 * | | | | count - number of extents being removed at idx
3907 * | count | |-------|
3908 * | | | nex2 | nex2 - number of extents after idx + count
3909 * |-------| |-------|
3910 */
3911 void
3916 {
3935 /*
3936 * Check for deletion of entire list;
3937 * xfs_iext_irec_remove() updates extent offsets.
3938 */
3945 XFS_IEXT_BUFSZ);
3951 }
3952 }
3953 /* Move extents up (if needed) */
3958 }
3959 /* Zero out rest of page */
3962 /* Update remaining counters */
3967 erp_idx++;
3968 erp++;
3969 }
3972 }
3974 /*
3975 * Create, destroy, or resize a linear (direct) block of extents.
3976 */
3977 void
3981 {
3990 /* Free extent records */
3993 }
3994 /* Resize direct extent list and zero any new bytes */
3996 /* Check if extents will fit inside the inode */
4002 }
4005 }
4009 rnew_size,
4011 }
4016 }
4017 }
4018 /*
4019 * Switch from the inline extent buffer to a direct
4020 * extent list. Be sure to include the inline extent
4021 * bytes in new_size.
4022 */
4027 }
4029 }
4032 }
4034 /*
4035 * Switch from linear (direct) extent records to inline buffer.
4036 */
4037 void
4041 {
4044 /*
4045 * The inline buffer was zeroed when we switched
4046 * from inline to direct extent allocation mode,
4047 * so we don't need to clear it here.
4048 */
4054 }
4056 /*
4057 * Switch from inline buffer to linear (direct) extent records.
4058 * new_size should already be rounded up to the next power of 2
4059 * by the caller (when appropriate), so use new_size as it is.
4060 * However, since new_size may be rounded up, we can't update
4061 * if_bytes here. It is the caller's responsibility to update
4062 * if_bytes upon return.
4063 */
4064 void
4068 {
4076 }
4078 }
4080 /*
4081 * Resize an extent indirection array to new_size bytes.
4082 */
4083 void
4087 {
4102 }
4103 }
4105 /*
4106 * Switch from indirection array to linear (direct) extent allocations.
4107 */
4108 void
4111 {
4131 }
4132 }
4134 /*
4135 * Free incore file extents.
4136 */
4137 void
4140 {
4148 }
4155 }
4159 }
4161 /*
4162 * Return a pointer to the extent record for file system block bno.
4163 */
4169 {
4184 }
4187 /* Find target extent list */
4195 }
4196 /* Binary search extent records */
4207 /* Convert back to file-based extent index */
4210 }
4213 }
4214 }
4215 /* Convert back to file-based extent index */
4218 }
4224 }
4225 }
4228 }
4230 /*
4231 * Return a pointer to the indirection array entry containing the
4232 * extent record for filesystem block bno. Store the index of the
4233 * target irec in *erp_idxp.
4234 */
4240 {
4264 }
4265 }
4268 }
4270 /*
4271 * Return a pointer to the indirection array entry containing the
4272 * extent record at file extent index *idxp. Store the index of the
4273 * target irec in *erp_idxp and store the page index of the target
4274 * extent record in *idxp.
4275 */
4276 xfs_ext_irec_t *
4282 {
4299 /* Binary search extent irec's */
4315 erp_idx++;
4321 }
4322 }
4326 }
4328 /*
4329 * Allocate and initialize an indirection array once the space needed
4330 * for incore extents increases above XFS_IEXT_BUFSZ.
4331 */
4332 void
4335 {
4351 }
4362 }
4364 /*
4365 * Allocate and initialize a new entry in the indirection array.
4366 */
4367 xfs_ext_irec_t *
4371 {
4379 /* Resize indirection array */
4382 /*
4383 * Move records down in the array so the
4384 * new page can use erp_idx.
4385 */
4389 }
4392 /* Initialize new extent record */
4401 }
4403 /*
4404 * Remove a record from the indirection array.
4405 */
4406 void
4410 {
4422 }
4423 /* Compact extent records */
4427 }
4428 /*
4429 * Manually free the last extent record from the indirection
4430 * array. A call to xfs_iext_realloc_indirect() with a size
4431 * of zero would result in a call to xfs_iext_destroy() which
4432 * would in turn call this function again, creating a nasty
4433 * infinite loop.
4434 */
4440 }
4442 }
4444 /*
4445 * This is called to clean up large amounts of unused memory allocated
4446 * by the indirection array. Before compacting anything though, verify
4447 * that the indirection array is still needed and switch back to the
4448 * linear extent list (or even the inline buffer) if possible. The
4449 * compaction policy is as follows:
4450 *
4451 * Full Compaction: Extents fit into a single page (or inline buffer)
4452 * Partial Compaction: Extents occupy less than 50% of allocated space
4453 * No Compaction: Extents occupy at least 50% of allocated space
4454 */
4455 void
4458 {
4475 }
4476 }
4478 /*
4479 * Combine extents from neighboring extent pages.
4480 */
4481 void
4484 {
4500 /*
4501 * Free page before removing extent record
4502 * so er_extoffs don't get modified in
4503 * xfs_iext_irec_remove.
4504 */
4510 erp_idx++;
4511 }
4512 }
4513 }
4515 /*
4516 * This is called to update the er_extoff field in the indirection
4517 * array when extents have been added or removed from one of the
4518 * extent lists. erp_idx contains the irec index to begin updating
4519 * at and ext_diff contains the number of extents that were added
4520 * or removed.
4521 */
4522 void
4527 {
4535 }
4536 }