| blob | ca12acb90394193571f4b09f9febba47ac091ea7 |
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;
1050 uint flags;
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 }
1090 /*
1091 * If the superblock version is up to where we support new format
1092 * inodes and this is currently an old format inode, then change
1093 * the inode version number now. This way we only do the conversion
1094 * here rather than here and in the flush/logging code.
1095 */
1099 /*
1100 * We've already zeroed the old link count, the projid field,
1101 * and the pad field.
1102 */
1103 }
1105 /*
1106 * Project ids won't be stored on disk if we are using a version 1 inode.
1107 */
1115 }
1116 }
1118 /*
1119 * If the group ID of the new file does not match the effective group
1120 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1121 * (and only if the irix_sgid_inherit compatibility variable is set).
1122 */
1127 }
1134 /*
1135 * di_gen will have been taken care of in xfs_iread.
1136 */
1159 }
1160 /* fall through */
1171 }
1178 }
1179 }
1181 xfs_inherit_noatime)
1184 xfs_inherit_nodump)
1187 xfs_inherit_sync)
1190 xfs_inherit_nosymlinks)
1195 xfs_inherit_nodefrag)
1200 }
1201 /* FALLTHROUGH */
1210 }
1211 /*
1212 * Attribute fork settings for new inode.
1213 */
1217 /*
1218 * Log the new values stuffed into the inode.
1219 */
1222 /* now that we have an i_mode we can setup inode ops and unlock */
1227 }
1229 /*
1230 * Check to make sure that there are no blocks allocated to the
1231 * file beyond the size of the file. We don't check this for
1232 * files with fixed size extents or real time extents, but we
1233 * at least do it for regular files.
1234 */
1235 #ifdef DEBUG
1236 void
1240 xfs_fsize_t isize)
1241 {
1242 xfs_fileoff_t map_first;
1257 /*
1258 * The filesystem could be shutting down, so bmapi may return
1259 * an error.
1260 */
1264 map_first),
1270 }
1273 /*
1274 * Calculate the last possible buffered byte in a file. This must
1275 * include data that was buffered beyond the EOF by the write code.
1276 * This also needs to deal with overflowing the xfs_fsize_t type
1277 * which can happen for sizes near the limit.
1278 *
1279 * We also need to take into account any blocks beyond the EOF. It
1280 * may be the case that they were buffered by a write which failed.
1281 * In that case the pages will still be in memory, but the inode size
1282 * will never have been updated.
1283 */
1284 xfs_fsize_t
1287 {
1289 xfs_fsize_t last_byte;
1290 xfs_fileoff_t last_block;
1291 xfs_fileoff_t size_last_block;
1297 /*
1298 * Only check for blocks beyond the EOF if the extents have
1299 * been read in. This eliminates the need for the inode lock,
1300 * and it also saves us from looking when it really isn't
1301 * necessary.
1302 */
1305 XFS_DATA_FORK);
1308 }
1311 }
1318 }
1322 }
1324 }
1326 #if defined(XFS_RW_TRACE)
1327 STATIC void
1332 xfs_fsize_t new_size,
1333 xfs_off_t toss_start,
1334 xfs_off_t toss_finish)
1335 {
1338 }
1357 }
1358 #else
1359 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1360 #endif
1362 /*
1363 * Start the truncation of the file to new_size. The new size
1364 * must be smaller than the current size. This routine will
1365 * clear the buffer and page caches of file data in the removed
1366 * range, and xfs_itruncate_finish() will remove the underlying
1367 * disk blocks.
1368 *
1369 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1370 * must NOT have the inode lock held at all. This is because we're
1371 * calling into the buffer/page cache code and we can't hold the
1372 * inode lock when we do so.
1373 *
1374 * We need to wait for any direct I/Os in flight to complete before we
1375 * proceed with the truncate. This is needed to prevent the extents
1376 * being read or written by the direct I/Os from being removed while the
1377 * I/O is in flight as there is no other method of synchronising
1378 * direct I/O with the truncate operation. Also, because we hold
1379 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1380 * started until the truncate completes and drops the lock. Essentially,
1381 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1382 * between direct I/Os and the truncate operation.
1383 *
1384 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1385 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1386 * in the case that the caller is locking things out of order and
1387 * may not be able to call xfs_itruncate_finish() with the inode lock
1388 * held without dropping the I/O lock. If the caller must drop the
1389 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1390 * must be called again with all the same restrictions as the initial
1391 * call.
1392 */
1393 int
1396 uint flags,
1397 xfs_fsize_t new_size)
1398 {
1399 xfs_fsize_t last_byte;
1400 xfs_off_t toss_start;
1413 /* wait for the completion of any pending DIOs */
1417 /*
1418 * Call toss_pages or flushinval_pages to get rid of pages
1419 * overlapping the region being removed. We have to use
1420 * the less efficient flushinval_pages in the case that the
1421 * caller may not be able to finish the truncate without
1422 * dropping the inode's I/O lock. Make sure
1423 * to catch any pages brought in by buffers overlapping
1424 * the EOF by searching out beyond the isize by our
1425 * block size. We round new_size up to a block boundary
1426 * so that we don't toss things on the same block as
1427 * new_size but before it.
1428 *
1429 * Before calling toss_page or flushinval_pages, make sure to
1430 * call remapf() over the same region if the file is mapped.
1431 * This frees up mapped file references to the pages in the
1432 * given range and for the flushinval_pages case it ensures
1433 * that we get the latest mapped changes flushed out.
1434 */
1438 /*
1439 * The place to start tossing is beyond our maximum
1440 * file size, so there is no way that the data extended
1441 * out there.
1442 */
1444 }
1447 last_byte);
1455 }
1456 }
1458 #ifdef DEBUG
1461 }
1462 #endif
1464 }
1466 /*
1467 * Shrink the file to the given new_size. The new size must be smaller than
1468 * the current size. This will free up the underlying blocks in the removed
1469 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
1470 *
1471 * The transaction passed to this routine must have made a permanent log
1472 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1473 * given transaction and start new ones, so make sure everything involved in
1474 * the transaction is tidy before calling here. Some transaction will be
1475 * returned to the caller to be committed. The incoming transaction must
1476 * already include the inode, and both inode locks must be held exclusively.
1477 * The inode must also be "held" within the transaction. On return the inode
1478 * will be "held" within the returned transaction. This routine does NOT
1479 * require any disk space to be reserved for it within the transaction.
1480 *
1481 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1482 * indicates the fork which is to be truncated. For the attribute fork we only
1483 * support truncation to size 0.
1484 *
1485 * We use the sync parameter to indicate whether or not the first transaction
1486 * we perform might have to be synchronous. For the attr fork, it needs to be
1487 * so if the unlink of the inode is not yet known to be permanent in the log.
1488 * This keeps us from freeing and reusing the blocks of the attribute fork
1489 * before the unlink of the inode becomes permanent.
1490 *
1491 * For the data fork, we normally have to run synchronously if we're being
1492 * called out of the inactive path or we're being called out of the create path
1493 * where we're truncating an existing file. Either way, the truncate needs to
1494 * be sync so blocks don't reappear in the file with altered data in case of a
1495 * crash. wsync filesystems can run the first case async because anything that
1496 * shrinks the inode has to run sync so by the time we're called here from
1497 * inactive, the inode size is permanently set to 0.
1498 *
1499 * Calls from the truncate path always need to be sync unless we're in a wsync
1500 * filesystem and the file has already been unlinked.
1501 *
1502 * The caller is responsible for correctly setting the sync parameter. It gets
1503 * too hard for us to guess here which path we're being called out of just
1504 * based on inode state.
1505 *
1506 * If we get an error, we must return with the inode locked and linked into the
1507 * current transaction. This keeps things simple for the higher level code,
1508 * because it always knows that the inode is locked and held in the transaction
1509 * that returns to it whether errors occur or not. We don't mark the inode
1510 * dirty on error so that transactions can be easily aborted if possible.
1511 */
1512 int
1516 xfs_fsize_t new_size,
1519 {
1520 xfs_fsblock_t first_block;
1521 xfs_fileoff_t first_unmap_block;
1522 xfs_fileoff_t last_block;
1528 xfs_bmap_free_t free_list;
1545 /*
1546 * We only support truncating the entire attribute fork.
1547 */
1550 }
1553 /*
1554 * The first thing we do is set the size to new_size permanently
1555 * on disk. This way we don't have to worry about anyone ever
1556 * being able to look at the data being freed even in the face
1557 * of a crash. What we're getting around here is the case where
1558 * we free a block, it is allocated to another file, it is written
1559 * to, and then we crash. If the new data gets written to the
1560 * file but the log buffers containing the free and reallocation
1561 * don't, then we'd end up with garbage in the blocks being freed.
1562 * As long as we make the new_size permanent before actually
1563 * freeing any blocks it doesn't matter if they get writtten to.
1564 *
1565 * The callers must signal into us whether or not the size
1566 * setting here must be synchronous. There are a few cases
1567 * where it doesn't have to be synchronous. Those cases
1568 * occur if the file is unlinked and we know the unlink is
1569 * permanent or if the blocks being truncated are guaranteed
1570 * to be beyond the inode eof (regardless of the link count)
1571 * and the eof value is permanent. Both of these cases occur
1572 * only on wsync-mounted filesystems. In those cases, we're
1573 * guaranteed that no user will ever see the data in the blocks
1574 * that are being truncated so the truncate can run async.
1575 * In the free beyond eof case, the file may wind up with
1576 * more blocks allocated to it than it needs if we crash
1577 * and that won't get fixed until the next time the file
1578 * is re-opened and closed but that's ok as that shouldn't
1579 * be too many blocks.
1580 *
1581 * However, we can't just make all wsync xactions run async
1582 * because there's one call out of the create path that needs
1583 * to run sync where it's truncating an existing file to size
1584 * 0 whose size is > 0.
1585 *
1586 * It's probably possible to come up with a test in this
1587 * routine that would correctly distinguish all the above
1588 * cases from the values of the function parameters and the
1589 * inode state but for sanity's sake, I've decided to let the
1590 * layers above just tell us. It's simpler to correctly figure
1591 * out in the layer above exactly under what conditions we
1592 * can run async and I think it's easier for others read and
1593 * follow the logic in case something has to be changed.
1594 * cscope is your friend -- rcc.
1595 *
1596 * The attribute fork is much simpler.
1597 *
1598 * For the attribute fork we allow the caller to tell us whether
1599 * the unlink of the inode that led to this call is yet permanent
1600 * in the on disk log. If it is not and we will be freeing extents
1601 * in this inode then we make the first transaction synchronous
1602 * to make sure that the unlink is permanent by the time we free
1603 * the blocks.
1604 */
1607 /*
1608 * If we are not changing the file size then do
1609 * not update the on-disk file size - we may be
1610 * called from xfs_inactive_free_eofblocks(). If we
1611 * update the on-disk file size and then the system
1612 * crashes before the contents of the file are
1613 * flushed to disk then the files may be full of
1614 * holes (ie NULL files bug).
1615 */
1620 }
1621 }
1626 }
1632 /*
1633 * Since it is possible for space to become allocated beyond
1634 * the end of the file (in a crash where the space is allocated
1635 * but the inode size is not yet updated), simply remove any
1636 * blocks which show up between the new EOF and the maximum
1637 * possible file size. If the first block to be removed is
1638 * beyond the maximum file size (ie it is the same as last_block),
1639 * then there is nothing to do.
1640 */
1648 }
1650 /*
1651 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1652 * will tell us whether it freed the entire range or
1653 * not. If this is a synchronous mount (wsync),
1654 * then we can tell bunmapi to keep all the
1655 * transactions asynchronous since the unlink
1656 * transaction that made this inode inactive has
1657 * already hit the disk. There's no danger of
1658 * the freed blocks being reused, there being a
1659 * crash, and the reused blocks suddenly reappearing
1660 * in this file with garbage in them once recovery
1661 * runs.
1662 */
1668 XFS_ITRUNC_MAX_EXTENTS,
1672 /*
1673 * If the bunmapi call encounters an error,
1674 * return to the caller where the transaction
1675 * can be properly aborted. We just need to
1676 * make sure we're not holding any resources
1677 * that we were not when we came in.
1678 */
1681 }
1683 /*
1684 * Duplicate the transaction that has the permanent
1685 * reservation and commit the old transaction.
1686 */
1690 /* link the inode into the next xact in the chain */
1694 }
1697 /*
1698 * If the bmap finish call encounters an error, return
1699 * to the caller where the transaction can be properly
1700 * aborted. We just need to make sure we're not
1701 * holding any resources that we were not when we came
1702 * in.
1703 *
1704 * Aborting from this point might lose some blocks in
1705 * the file system, but oh well.
1706 */
1709 }
1712 /*
1713 * Mark the inode dirty so it will be logged and
1714 * moved forward in the log as part of every commit.
1715 */
1717 }
1723 /* link the inode into the next transaction in the chain */
1730 XFS_TRANS_PERM_LOG_RES,
1731 XFS_ITRUNCATE_LOG_COUNT);
1734 }
1735 /*
1736 * Only update the size in the case of the data fork, but
1737 * always re-log the inode so that our permanent transaction
1738 * can keep on rolling it forward in the log.
1739 */
1742 /*
1743 * If we are not changing the file size then do
1744 * not update the on-disk file size - we may be
1745 * called from xfs_inactive_free_eofblocks(). If we
1746 * update the on-disk file size and then the system
1747 * crashes before the contents of the file are
1748 * flushed to disk then the files may be full of
1749 * holes (ie NULL files bug).
1750 */
1754 }
1755 }
1765 }
1768 /*
1769 * xfs_igrow_start
1770 *
1771 * Do the first part of growing a file: zero any data in the last
1772 * block that is beyond the old EOF. We need to do this before
1773 * the inode is joined to the transaction to modify the i_size.
1774 * That way we can drop the inode lock and call into the buffer
1775 * cache to get the buffer mapping the EOF.
1776 */
1777 int
1780 xfs_fsize_t new_size,
1782 {
1787 /*
1788 * Zero any pages that may have been created by
1789 * xfs_write_file() beyond the end of the file
1790 * and any blocks between the old and new file sizes.
1791 */
1793 }
1795 /*
1796 * xfs_igrow_finish
1797 *
1798 * This routine is called to extend the size of a file.
1799 * The inode must have both the iolock and the ilock locked
1800 * for update and it must be a part of the current transaction.
1801 * The xfs_igrow_start() function must have been called previously.
1802 * If the change_flag is not zero, the inode change timestamp will
1803 * be updated.
1804 */
1805 void
1809 xfs_fsize_t new_size,
1811 {
1817 /*
1818 * Update the file size. Update the inode change timestamp
1819 * if change_flag set.
1820 */
1827 }
1830 /*
1831 * This is called when the inode's link count goes to 0.
1832 * We place the on-disk inode on a list in the AGI. It
1833 * will be pulled from this list when the inode is freed.
1834 */
1835 int
1839 {
1845 xfs_agnumber_t agno;
1846 xfs_daddr_t agdaddr;
1847 xfs_agino_t agino;
1862 /*
1863 * Get the agi buffer first. It ensures lock ordering
1864 * on the list.
1865 */
1871 /*
1872 * Validate the magic number of the agi block.
1873 */
1875 agi_ok =
1879 XFS_RANDOM_IUNLINK))) {
1883 }
1884 /*
1885 * Get the index into the agi hash table for the
1886 * list this inode will go on.
1887 */
1895 /*
1896 * There is already another inode in the bucket we need
1897 * to add ourselves to. Add us at the front of the list.
1898 * Here we put the head pointer into our next pointer,
1899 * and then we fall through to point the head at us.
1900 */
1906 /* both on-disk, don't endian flip twice */
1914 }
1916 /*
1917 * Point the bucket head pointer at the inode being inserted.
1918 */
1926 }
1928 /*
1929 * Pull the on-disk inode from the AGI unlinked list.
1930 */
1931 STATIC int
1935 {
1936 xfs_ino_t next_ino;
1942 xfs_agnumber_t agno;
1943 xfs_daddr_t agdaddr;
1944 xfs_agino_t agino;
1945 xfs_agino_t next_agino;
1953 /*
1954 * First pull the on-disk inode from the AGI unlinked list.
1955 */
1961 /*
1962 * Get the agi buffer first. It ensures lock ordering
1963 * on the list.
1964 */
1972 }
1973 /*
1974 * Validate the magic number of the agi block.
1975 */
1977 agi_ok =
1981 XFS_RANDOM_IUNLINK_REMOVE))) {
1989 }
1990 /*
1991 * Get the index into the agi hash table for the
1992 * list this inode will go on.
1993 */
2001 /*
2002 * We're at the head of the list. Get the inode's
2003 * on-disk buffer to see if there is anyone after us
2004 * on the list. Only modify our next pointer if it
2005 * is not already NULLAGINO. This saves us the overhead
2006 * of dealing with the buffer when there is no need to
2007 * change it.
2008 */
2015 }
2028 }
2029 /*
2030 * Point the bucket head pointer at the next inode.
2031 */
2040 /*
2041 * We need to search the list for the inode being freed.
2042 */
2046 /*
2047 * If the last inode wasn't the one pointing to
2048 * us, then release its buffer since we're not
2049 * going to do anything with it.
2050 */
2053 }
2062 }
2066 }
2067 /*
2068 * Now last_ibp points to the buffer previous to us on
2069 * the unlinked list. Pull us from the list.
2070 */
2077 }
2091 }
2092 /*
2093 * Point the previous inode on the list to the next inode.
2094 */
2102 }
2104 }
2106 STATIC void
2110 xfs_ino_t inum)
2111 {
2117 xfs_daddr_t blkno;
2133 }
2142 /*
2143 * Look for each inode in memory and attempt to lock it,
2144 * we can be racing with flush and tail pushing here.
2145 * any inode we get the locks on, add to an array of
2146 * inode items to process later.
2147 *
2148 * The get the buffer lock, we could beat a flush
2149 * or tail pushing thread to the lock here, in which
2150 * case they will go looking for the inode buffer
2151 * and fail, we need some other form of interlock
2152 * here.
2153 */
2160 /* Inode not in memory or we found it already,
2161 * nothing to do
2162 */
2166 }
2171 }
2173 /* If we can get the locks then add it to the
2174 * list, otherwise by the time we get the bp lock
2175 * below it will already be attached to the
2176 * inode buffer.
2177 */
2179 /* This inode will already be locked - by us, lets
2180 * keep it that way.
2181 */
2190 }
2191 }
2194 }
2205 }
2208 }
2209 }
2211 }
2215 XFS_BUF_LOCK);
2228 pre_flushed++;
2229 }
2231 }
2242 }
2256 }
2257 }
2262 }
2266 }
2268 /*
2269 * This is called to return an inode to the inode free list.
2270 * The inode should already be truncated to 0 length and have
2271 * no pages associated with it. This routine also assumes that
2272 * the inode is already a part of the transaction.
2273 *
2274 * The on-disk copy of the inode will have been added to the list
2275 * of unlinked inodes in the AGI. We need to remove the inode from
2276 * that list atomically with respect to freeing it here.
2277 */
2278 int
2283 {
2286 xfs_ino_t first_ino;
2299 /*
2300 * Pull the on-disk inode from the AGI unlinked list.
2301 */
2305 }
2310 }
2319 /*
2320 * Bump the generation count so no one will be confused
2321 * by reincarnations of this inode.
2322 */
2331 /*
2332 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2333 * from picking up this inode when it is reclaimed (its incore state
2334 * initialzed but not flushed to disk yet). The in-core di_mode is
2335 * already cleared and a corresponding transaction logged.
2336 * The hack here just synchronizes the in-core to on-disk
2337 * di_mode value in advance before the actual inode sync to disk.
2338 * This is OK because the inode is already unlinked and would never
2339 * change its di_mode again for this inode generation.
2340 * This is a temporary hack that would require a proper fix
2341 * in the future.
2342 */
2347 }
2350 }
2352 /*
2353 * Reallocate the space for if_broot based on the number of records
2354 * being added or deleted as indicated in rec_diff. Move the records
2355 * and pointers in if_broot to fit the new size. When shrinking this
2356 * will eliminate holes between the records and pointers created by
2357 * the caller. When growing this will create holes to be filled in
2358 * by the caller.
2359 *
2360 * The caller must not request to add more records than would fit in
2361 * the on-disk inode root. If the if_broot is currently NULL, then
2362 * if we adding records one will be allocated. The caller must also
2363 * not request that the number of records go below zero, although
2364 * it can go to zero.
2365 *
2366 * ip -- the inode whose if_broot area is changing
2367 * ext_diff -- the change in the number of records, positive or negative,
2368 * requested for the if_broot array.
2369 */
2370 void
2375 {
2384 /*
2385 * Handle the degenerate case quietly.
2386 */
2389 }
2393 /*
2394 * If there wasn't any memory allocated before, just
2395 * allocate it now and get out.
2396 */
2400 KM_SLEEP);
2403 }
2405 /*
2406 * If there is already an existing if_broot, then we need
2407 * to realloc() it and shift the pointers to their new
2408 * location. The records don't change location because
2409 * they are kept butted up against the btree block header.
2410 */
2416 new_size,
2418 KM_SLEEP);
2428 }
2430 /*
2431 * rec_diff is less than 0. In this case, we are shrinking the
2432 * if_broot buffer. It must already exist. If we go to zero
2433 * records, just get rid of the root and clear the status bit.
2434 */
2441 else
2445 /*
2446 * First copy over the btree block header.
2447 */
2452 }
2454 /*
2455 * Only copy the records and pointers if there are any.
2456 */
2458 /*
2459 * First copy the records.
2460 */
2467 /*
2468 * Then copy the pointers.
2469 */
2475 }
2482 }
2485 /*
2486 * This is called when the amount of space needed for if_data
2487 * is increased or decreased. The change in size is indicated by
2488 * the number of bytes that need to be added or deleted in the
2489 * byte_diff parameter.
2490 *
2491 * If the amount of space needed has decreased below the size of the
2492 * inline buffer, then switch to using the inline buffer. Otherwise,
2493 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2494 * to what is needed.
2495 *
2496 * ip -- the inode whose if_data area is changing
2497 * byte_diff -- the change in the number of bytes, positive or negative,
2498 * requested for the if_data array.
2499 */
2500 void
2505 {
2512 }
2521 }
2525 /*
2526 * If the valid extents/data can fit in if_inline_ext/data,
2527 * copy them from the malloc'd vector and free it.
2528 */
2534 new_size);
2537 }
2540 /*
2541 * Stuck with malloc/realloc.
2542 * For inline data, the underlying buffer must be
2543 * a multiple of 4 bytes in size so that it can be
2544 * logged and stay on word boundaries. We enforce
2545 * that here.
2546 */
2552 /*
2553 * Only do the realloc if the underlying size
2554 * is really changing.
2555 */
2559 real_size,
2561 KM_SLEEP);
2562 }
2568 }
2569 }
2573 }
2578 /*
2579 * Map inode to disk block and offset.
2580 *
2581 * mp -- the mount point structure for the current file system
2582 * tp -- the current transaction
2583 * ino -- the inode number of the inode to be located
2584 * imap -- this structure is filled in with the information necessary
2585 * to retrieve the given inode from disk
2586 * flags -- flags to pass to xfs_dilocate indicating whether or not
2587 * lookups in the inode btree were OK or not
2588 */
2589 int
2593 xfs_ino_t ino,
2595 uint flags)
2596 {
2597 xfs_fsblock_t fsbno;
2614 /*
2615 * If the inode number maps to a block outside the bounds
2616 * of the file system then return NULL rather than calling
2617 * read_buf and panicing when we get an error from the
2618 * driver.
2619 */
2623 "(imap->im_blkno (0x%llx) + imap->im_len (0x%llx)) > "
2629 }
2631 }
2633 void
2637 {
2644 }
2646 /*
2647 * If the format is local, then we can't have an extents
2648 * array so just look for an inline data array. If we're
2649 * not local then we may or may not have an extents list,
2650 * so check and free it up if we do.
2651 */
2659 }
2666 }
2673 }
2674 }
2676 /*
2677 * This is called free all the memory associated with an inode.
2678 * It must free the inode itself and any buffers allocated for
2679 * if_extents/if_data and if_broot. It must also free the lock
2680 * associated with the inode.
2681 */
2682 void
2685 {
2692 }
2699 #ifdef XFS_INODE_TRACE
2701 #endif
2702 #ifdef XFS_BMAP_TRACE
2704 #endif
2705 #ifdef XFS_BMBT_TRACE
2707 #endif
2708 #ifdef XFS_RW_TRACE
2710 #endif
2711 #ifdef XFS_ILOCK_TRACE
2713 #endif
2714 #ifdef XFS_DIR2_TRACE
2716 #endif
2718 /*
2719 * Only if we are shutting down the fs will we see an
2720 * inode still in the AIL. If it is there, we should remove
2721 * it to prevent a use-after-free from occurring.
2722 */
2732 else
2734 }
2736 }
2738 }
2741 /*
2742 * Increment the pin count of the given buffer.
2743 * This value is protected by ipinlock spinlock in the mount structure.
2744 */
2745 void
2748 {
2752 }
2754 /*
2755 * Decrement the pin count of the given inode, and wake up
2756 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2757 * inode must have been previously pinned with a call to xfs_ipin().
2758 */
2759 void
2762 {
2767 }
2769 /*
2770 * This is called to unpin an inode. It can be directed to wait or to return
2771 * immediately without waiting for the inode to be unpinned. The caller must
2772 * have the inode locked in at least shared mode so that the buffer cannot be
2773 * subsequently pinned once someone is waiting for it to be unpinned.
2774 */
2775 STATIC void
2779 {
2786 /* Give the log a push to start the unpinning I/O */
2791 }
2796 {
2798 }
2803 {
2805 }
2808 /*
2809 * xfs_iextents_copy()
2810 *
2811 * This is called to copy the REAL extents (as opposed to the delayed
2812 * allocation extents) from the inode into the given buffer. It
2813 * returns the number of bytes copied into the buffer.
2814 *
2815 * If there are no delayed allocation extents, then we can just
2816 * memcpy() the extents into the buffer. Otherwise, we need to
2817 * examine each extent in turn and skip those which are delayed.
2818 */
2819 int
2824 {
2829 xfs_fsblock_t start_block;
2839 /*
2840 * There are some delayed allocation extents in the
2841 * inode, so copy the extents one at a time and skip
2842 * the delayed ones. There must be at least one
2843 * non-delayed extent.
2844 */
2850 /*
2851 * It's a delayed allocation extent, so skip it.
2852 */
2854 }
2856 /* Translate to on disk format */
2859 dp++;
2860 copied++;
2861 }
2866 }
2868 /*
2869 * Each of the following cases stores data into the same region
2870 * of the on-disk inode, so only one of them can be valid at
2871 * any given time. While it is possible to have conflicting formats
2872 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2873 * in EXTENTS format, this can only happen when the fork has
2874 * changed formats after being modified but before being flushed.
2875 * In these cases, the format always takes precedence, because the
2876 * format indicates the current state of the fork.
2877 */
2878 /*ARGSUSED*/
2879 STATIC void
2886 {
2890 #ifdef XFS_TRANS_DEBUG
2892 #endif
2903 /*
2904 * This can happen if we gave up in iformat in an error path,
2905 * for the attribute fork.
2906 */
2910 }
2920 }
2934 whichfork);
2935 }
2944 XFS_BROOT_SIZE_ADJ));
2948 }
2955 }
2963 }
2969 }
2970 }
2972 STATIC int
2976 {
2999 /* really need a gang lookup range call here */
3001 first_index,
3010 /* if the inode lies outside this cluster, we're done. */
3013 /*
3014 * Do an un-protected check to see if the inode is dirty and
3015 * is a candidate for flushing. These checks will be repeated
3016 * later after the appropriate locks are acquired.
3017 */
3021 /*
3022 * Try to get locks. If any are unavailable or it is pinned,
3023 * then this inode cannot be flushed and is skipped.
3024 */
3031 }
3036 }
3038 /*
3039 * arriving here means that this inode can be flushed. First
3040 * re-check that it's dirty before flushing.
3041 */
3048 }
3049 clcount++;
3052 }
3054 }
3059 }
3061 out_free:
3067 cluster_corrupt_out:
3068 /*
3069 * Corruption detected in the clustering loop. Invalidate the
3070 * inode buffer and shut down the filesystem.
3071 */
3073 /*
3074 * Clean up the buffer. If it was B_DELWRI, just release it --
3075 * brelse can handle it with no problems. If not, shut down the
3076 * filesystem before releasing the buffer.
3077 */
3085 /*
3086 * Just like incore_relse: if we have b_iodone functions,
3087 * mark the buffer as an error and call them. Otherwise
3088 * mark it as stale and brelse.
3089 */
3100 }
3101 }
3103 /*
3104 * Unlocks the flush lock
3105 */
3109 }
3111 /*
3112 * xfs_iflush() will write a modified inode's changes out to the
3113 * inode's on disk home. The caller must have the inode lock held
3114 * in at least shared mode and the inode flush semaphore must be
3115 * held as well. The inode lock will still be held upon return from
3116 * the call and the caller is free to unlock it.
3117 * The inode flush lock will be unlocked when the inode reaches the disk.
3118 * The flags indicate how the inode's buffer should be written out.
3119 */
3120 int
3123 uint flags)
3124 {
3143 /*
3144 * If the inode isn't dirty, then just release the inode
3145 * flush lock and do nothing.
3146 */
3152 }
3154 /*
3155 * We can't flush the inode until it is unpinned, so wait for it if we
3156 * are allowed to block. We know noone new can pin it, because we are
3157 * holding the inode lock shared and you need to hold it exclusively to
3158 * pin the inode.
3159 *
3160 * If we are not allowed to block, force the log out asynchronously so
3161 * that when we come back the inode will be unpinned. If other inodes
3162 * in the same cluster are dirty, they will probably write the inode
3163 * out for us if they occur after the log force completes.
3164 */
3169 }
3172 /*
3173 * This may have been unpinned because the filesystem is shutting
3174 * down forcibly. If that's the case we must not write this inode
3175 * to disk, because the log record didn't make it to disk!
3176 */
3183 }
3185 /*
3186 * Decide how buffer will be flushed out. This is done before
3187 * the call to xfs_iflush_int because this field is zeroed by it.
3188 */
3190 /*
3191 * Flush out the inode buffer according to the directions
3192 * of the caller. In the cases where the caller has given
3193 * us a choice choose the non-delwri case. This is because
3194 * the inode is in the AIL and we need to get it out soon.
3195 */
3213 }
3232 }
3233 }
3235 /*
3236 * Get the buffer containing the on-disk inode.
3237 */
3243 }
3245 /*
3246 * First flush out the inode that xfs_iflush was called with.
3247 */
3252 /*
3253 * If the buffer is pinned then push on the log now so we won't
3254 * get stuck waiting in the write for too long.
3255 */
3259 /*
3260 * inode clustering:
3261 * see if other inodes can be gathered into this write
3262 */
3273 }
3276 corrupt_out:
3279 cluster_corrupt_out:
3280 /*
3281 * Unlocks the flush lock
3282 */
3285 }
3288 STATIC int
3292 {
3296 #ifdef XFS_TRANS_DEBUG
3298 #endif
3309 /*
3310 * If the inode isn't dirty, then just release the inode
3311 * flush lock and do nothing.
3312 */
3316 }
3318 /* set *dip = inode's place in the buffer */
3321 /*
3322 * Clear i_update_core before copying out the data.
3323 * This is for coordination with our timestamp updates
3324 * that don't hold the inode lock. They will always
3325 * update the timestamps BEFORE setting i_update_core,
3326 * so if we clear i_update_core after they set it we
3327 * are guaranteed to see their updates to the timestamps.
3328 * I believe that this depends on strongly ordered memory
3329 * semantics, but we have that. We use the SYNCHRONIZE
3330 * macro to make sure that the compiler does not reorder
3331 * the i_update_core access below the data copy below.
3332 */
3336 /*
3337 * Make sure to get the latest atime from the Linux inode.
3338 */
3347 }
3354 }
3364 }
3375 }
3376 }
3379 XFS_RANDOM_IFLUSH_5)) {
3385 ip);
3387 }
3394 }
3395 /*
3396 * bump the flush iteration count, used to detect flushes which
3397 * postdate a log record during recovery.
3398 */
3402 /*
3403 * Copy the dirty parts of the inode into the on-disk
3404 * inode. We always copy out the core of the inode,
3405 * because if the inode is dirty at all the core must
3406 * be.
3407 */
3410 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3414 /*
3415 * If this is really an old format inode and the superblock version
3416 * has not been updated to support only new format inodes, then
3417 * convert back to the old inode format. If the superblock version
3418 * has been updated, then make the conversion permanent.
3419 */
3424 /*
3425 * Convert it back.
3426 */
3430 /*
3431 * The superblock version has already been bumped,
3432 * so just make the conversion to the new inode
3433 * format permanent.
3434 */
3443 }
3444 }
3451 /*
3452 * We've recorded everything logged in the inode, so we'd
3453 * like to clear the ilf_fields bits so we don't log and
3454 * flush things unnecessarily. However, we can't stop
3455 * logging all this information until the data we've copied
3456 * into the disk buffer is written to disk. If we did we might
3457 * overwrite the copy of the inode in the log with all the
3458 * data after re-logging only part of it, and in the face of
3459 * a crash we wouldn't have all the data we need to recover.
3460 *
3461 * What we do is move the bits to the ili_last_fields field.
3462 * When logging the inode, these bits are moved back to the
3463 * ilf_fields field. In the xfs_iflush_done() routine we
3464 * clear ili_last_fields, since we know that the information
3465 * those bits represent is permanently on disk. As long as
3466 * the flush completes before the inode is logged again, then
3467 * both ilf_fields and ili_last_fields will be cleared.
3468 *
3469 * We can play with the ilf_fields bits here, because the inode
3470 * lock must be held exclusively in order to set bits there
3471 * and the flush lock protects the ili_last_fields bits.
3472 * Set ili_logged so the flush done
3473 * routine can tell whether or not to look in the AIL.
3474 * Also, store the current LSN of the inode so that we can tell
3475 * whether the item has moved in the AIL from xfs_iflush_done().
3476 * In order to read the lsn we need the AIL lock, because
3477 * it is a 64 bit value that cannot be read atomically.
3478 */
3489 /*
3490 * Attach the function xfs_iflush_done to the inode's
3491 * buffer. This will remove the inode from the AIL
3492 * and unlock the inode's flush lock when the inode is
3493 * completely written to disk.
3494 */
3501 /*
3502 * We're flushing an inode which is not in the AIL and has
3503 * not been logged but has i_update_core set. For this
3504 * case we can use a B_DELWRI flush and immediately drop
3505 * the inode flush lock because we can avoid the whole
3506 * AIL state thing. It's OK to drop the flush lock now,
3507 * because we've already locked the buffer and to do anything
3508 * you really need both.
3509 */
3514 }
3516 }
3520 corrupt_out:
3522 }
3525 /*
3526 * Flush all inactive inodes in mp.
3527 */
3528 void
3531 {
3535 again:
3542 /* Make sure we skip markers inserted by sync */
3546 }
3553 }
3559 out:
3561 }
3563 #ifdef XFS_ILOCK_TRACE
3566 void
3568 {
3577 }
3578 #endif
3580 /*
3581 * Return a pointer to the extent record at file index idx.
3582 */
3583 xfs_bmbt_rec_host_t *
3587 {
3602 }
3603 }
3605 /*
3606 * Insert new item(s) into the extent records for incore inode
3607 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3608 */
3609 void
3615 {
3622 }
3624 /*
3625 * This is called when the amount of space required for incore file
3626 * extents needs to be increased. The ext_diff parameter stores the
3627 * number of new extents being added and the idx parameter contains
3628 * the extent index where the new extents will be added. If the new
3629 * extents are being appended, then we just need to (re)allocate and
3630 * initialize the space. Otherwise, if the new extents are being
3631 * inserted into the middle of the existing entries, a bit more work
3632 * is required to make room for the new extents to be inserted. The
3633 * caller is responsible for filling in the new extent entries upon
3634 * return.
3635 */
3636 void
3641 {
3650 /*
3651 * If the new number of extents (nextents + ext_diff)
3652 * fits inside the inode, then continue to use the inline
3653 * extent buffer.
3654 */
3661 }
3665 }
3666 /*
3667 * Otherwise use a linear (direct) extent list.
3668 * If the extents are currently inside the inode,
3669 * xfs_iext_realloc_direct will switch us from
3670 * inline to direct extent allocation mode.
3671 */
3679 }
3680 }
3681 /* Indirection array */
3694 }
3695 /* Extents fit in target extent page */
3703 }
3706 }
3707 /* Insert a new extent page */
3711 }
3712 /*
3713 * If extent(s) are being appended to the last page in
3714 * the indirection array and the new extent(s) don't fit
3715 * in the page, then erp is NULL and erp_idx is set to
3716 * the next index needed in the indirection array.
3717 */
3726 erp_idx++;
3727 }
3728 }
3729 }
3730 }
3732 }
3734 /*
3735 * This is called when incore extents are being added to the indirection
3736 * array and the new extents do not fit in the target extent list. The
3737 * erp_idx parameter contains the irec index for the target extent list
3738 * in the indirection array, and the idx parameter contains the extent
3739 * index within the list. The number of extents being added is stored
3740 * in the count parameter.
3741 *
3742 * |-------| |-------|
3743 * | | | | idx - number of extents before idx
3744 * | idx | | count |
3745 * | | | | count - number of extents being inserted at idx
3746 * |-------| |-------|
3747 * | count | | nex2 | nex2 - number of extents after idx + count
3748 * |-------| |-------|
3749 */
3750 void
3756 {
3770 /*
3771 * Save second part of target extent list
3772 * (all extents past */
3780 }
3782 /*
3783 * Add the new extents to the end of the target
3784 * list, then allocate new irec record(s) and
3785 * extent buffer(s) as needed to store the rest
3786 * of the new extents.
3787 */
3794 }
3796 erp_idx++;
3802 }
3804 /* Add nex2 extents back to indirection array */
3806 xfs_extnum_t ext_avail;
3812 /*
3813 * If nex2 extents fit in the current page, append
3814 * nex2_ep after the new extents.
3815 */
3818 }
3819 /*
3820 * Otherwise, check if space is available in the
3821 * next page.
3822 */
3826 erp_idx++;
3827 erp++;
3828 /* Create a hole for nex2 extents */
3831 }
3832 /*
3833 * Final choice, create a new extent page for
3834 * nex2 extents.
3835 */
3837 erp_idx++;
3839 }
3844 }
3845 }
3847 /*
3848 * This is called when the amount of space required for incore file
3849 * extents needs to be decreased. The ext_diff parameter stores the
3850 * number of extents to be removed and the idx parameter contains
3851 * the extent index where the extents will be removed from.
3852 *
3853 * If the amount of space needed has decreased below the linear
3854 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3855 * extent array. Otherwise, use kmem_realloc() to adjust the
3856 * size to what is needed.
3857 */
3858 void
3863 {
3879 }
3881 }
3883 /*
3884 * This removes ext_diff extents from the inline buffer, beginning
3885 * at extent index idx.
3886 */
3887 void
3892 {
3911 }
3912 }
3914 /*
3915 * This removes ext_diff extents from a linear (direct) extent list,
3916 * beginning at extent index idx. If the extents are being removed
3917 * from the end of the list (ie. truncate) then we just need to re-
3918 * allocate the list to remove the extra space. Otherwise, if the
3919 * extents are being removed from the middle of the existing extent
3920 * entries, then we first need to move the extent records beginning
3921 * at idx + ext_diff up in the list to overwrite the records being
3922 * removed, then remove the extra space via kmem_realloc.
3923 */
3924 void
3929 {
3941 }
3942 /* Move extents up in the list (if needed) */
3948 }
3951 /*
3952 * Reallocate the direct extent list. If the extents
3953 * will fit inside the inode then xfs_iext_realloc_direct
3954 * will switch from direct to inline extent allocation
3955 * mode for us.
3956 */
3959 }
3961 /*
3962 * This is called when incore extents are being removed from the
3963 * indirection array and the extents being removed span multiple extent
3964 * buffers. The idx parameter contains the file extent index where we
3965 * want to begin removing extents, and the count parameter contains
3966 * how many extents need to be removed.
3967 *
3968 * |-------| |-------|
3969 * | nex1 | | | nex1 - number of extents before idx
3970 * |-------| | count |
3971 * | | | | count - number of extents being removed at idx
3972 * | count | |-------|
3973 * | | | nex2 | nex2 - number of extents after idx + count
3974 * |-------| |-------|
3975 */
3976 void
3981 {
4000 /*
4001 * Check for deletion of entire list;
4002 * xfs_iext_irec_remove() updates extent offsets.
4003 */
4010 XFS_IEXT_BUFSZ);
4016 }
4017 }
4018 /* Move extents up (if needed) */
4023 }
4024 /* Zero out rest of page */
4027 /* Update remaining counters */
4032 erp_idx++;
4033 erp++;
4034 }
4037 }
4039 /*
4040 * Create, destroy, or resize a linear (direct) block of extents.
4041 */
4042 void
4046 {
4055 /* Free extent records */
4058 }
4059 /* Resize direct extent list and zero any new bytes */
4061 /* Check if extents will fit inside the inode */
4067 }
4070 }
4074 rnew_size,
4076 KM_SLEEP);
4077 }
4082 }
4083 }
4084 /*
4085 * Switch from the inline extent buffer to a direct
4086 * extent list. Be sure to include the inline extent
4087 * bytes in new_size.
4088 */
4093 }
4095 }
4098 }
4100 /*
4101 * Switch from linear (direct) extent records to inline buffer.
4102 */
4103 void
4107 {
4110 /*
4111 * The inline buffer was zeroed when we switched
4112 * from inline to direct extent allocation mode,
4113 * so we don't need to clear it here.
4114 */
4120 }
4122 /*
4123 * Switch from inline buffer to linear (direct) extent records.
4124 * new_size should already be rounded up to the next power of 2
4125 * by the caller (when appropriate), so use new_size as it is.
4126 * However, since new_size may be rounded up, we can't update
4127 * if_bytes here. It is the caller's responsibility to update
4128 * if_bytes upon return.
4129 */
4130 void
4134 {
4142 }
4144 }
4146 /*
4147 * Resize an extent indirection array to new_size bytes.
4148 */
4149 void
4153 {
4168 }
4169 }
4171 /*
4172 * Switch from indirection array to linear (direct) extent allocations.
4173 */
4174 void
4177 {
4197 }
4198 }
4200 /*
4201 * Free incore file extents.
4202 */
4203 void
4206 {
4214 }
4221 }
4225 }
4227 /*
4228 * Return a pointer to the extent record for file system block bno.
4229 */
4235 {
4250 }
4253 /* Find target extent list */
4261 }
4262 /* Binary search extent records */
4273 /* Convert back to file-based extent index */
4276 }
4279 }
4280 }
4281 /* Convert back to file-based extent index */
4284 }
4290 }
4291 }
4294 }
4296 /*
4297 * Return a pointer to the indirection array entry containing the
4298 * extent record for filesystem block bno. Store the index of the
4299 * target irec in *erp_idxp.
4300 */
4306 {
4330 }
4331 }
4334 }
4336 /*
4337 * Return a pointer to the indirection array entry containing the
4338 * extent record at file extent index *idxp. Store the index of the
4339 * target irec in *erp_idxp and store the page index of the target
4340 * extent record in *idxp.
4341 */
4342 xfs_ext_irec_t *
4348 {
4365 /* Binary search extent irec's */
4381 erp_idx++;
4387 }
4388 }
4392 }
4394 /*
4395 * Allocate and initialize an indirection array once the space needed
4396 * for incore extents increases above XFS_IEXT_BUFSZ.
4397 */
4398 void
4401 {
4418 }
4429 }
4431 /*
4432 * Allocate and initialize a new entry in the indirection array.
4433 */
4434 xfs_ext_irec_t *
4438 {
4446 /* Resize indirection array */
4449 /*
4450 * Move records down in the array so the
4451 * new page can use erp_idx.
4452 */
4456 }
4459 /* Initialize new extent record */
4468 }
4470 /*
4471 * Remove a record from the indirection array.
4472 */
4473 void
4477 {
4489 }
4490 /* Compact extent records */
4494 }
4495 /*
4496 * Manually free the last extent record from the indirection
4497 * array. A call to xfs_iext_realloc_indirect() with a size
4498 * of zero would result in a call to xfs_iext_destroy() which
4499 * would in turn call this function again, creating a nasty
4500 * infinite loop.
4501 */
4508 }
4510 }
4512 /*
4513 * This is called to clean up large amounts of unused memory allocated
4514 * by the indirection array. Before compacting anything though, verify
4515 * that the indirection array is still needed and switch back to the
4516 * linear extent list (or even the inline buffer) if possible. The
4517 * compaction policy is as follows:
4518 *
4519 * Full Compaction: Extents fit into a single page (or inline buffer)
4520 * Full Compaction: Extents occupy less than 10% of allocated space
4521 * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
4522 * No Compaction: Extents occupy at least 50% of allocated space
4523 */
4524 void
4527 {
4546 }
4547 }
4549 /*
4550 * Combine extents from neighboring extent pages.
4551 */
4552 void
4555 {
4571 /*
4572 * Free page before removing extent record
4573 * so er_extoffs don't get modified in
4574 * xfs_iext_irec_remove.
4575 */
4581 erp_idx++;
4582 }
4583 }
4584 }
4586 /*
4587 * Fully compact the extent records managed by the indirection array.
4588 */
4589 void
4592 {
4612 /* Remove next page */
4614 /*
4615 * Free page before removing extent record
4616 * so er_extoffs don't get modified in
4617 * xfs_iext_irec_remove.
4618 */
4625 /* Update next page */
4627 /* Move rest of page up to become next new page */
4634 }
4636 erp_idx++;
4639 else
4641 }
4645 }
4646 }
4648 /*
4649 * This is called to update the er_extoff field in the indirection
4650 * array when extents have been added or removed from one of the
4651 * extent lists. erp_idx contains the irec index to begin updating
4652 * at and ext_diff contains the number of extents that were added
4653 * or removed.
4654 */
4655 void
4660 {
4668 }
4669 }