| blob | a550546a70832dc727dd117ef1c4bbc3f0ef5e90 |
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>
60 /*
61 * Used in xfs_itruncate(). This is the maximum number of extents
62 * freed from a file in a single transaction.
63 */
64 #define XFS_ITRUNC_MAX_EXTENTS 2
71 #ifdef DEBUG
72 /*
73 * Make sure that the extents in the given memory buffer
74 * are valid.
75 */
76 STATIC void
80 xfs_exntfmt_t fmt)
81 {
82 xfs_bmbt_irec_t irec;
83 xfs_bmbt_rec_host_t rec;
93 }
94 }
96 #define xfs_validate_extents(ifp, nrecs, fmt)
99 /*
100 * Check that none of the inode's in the buffer have a next
101 * unlinked field of 0.
102 */
103 #if defined(DEBUG)
104 void
108 {
120 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
121 bp);
123 }
124 }
125 }
126 #endif
128 /*
129 * This routine is called to map an inode number within a file
130 * system to the buffer containing the on-disk version of the
131 * inode. It returns a pointer to the buffer containing the
132 * on-disk inode in the bpp parameter, and in the dip parameter
133 * it returns a pointer to the on-disk inode within that buffer.
134 *
135 * If a non-zero error is returned, then the contents of bpp and
136 * dipp are undefined.
137 *
138 * Use xfs_imap() to determine the size and location of the
139 * buffer to read from disk.
140 */
141 STATIC int
145 xfs_ino_t ino,
149 {
151 xfs_imap_t imap;
156 /*
157 * Call the space management code to find the location of the
158 * inode on disk.
159 */
164 "xfs_inotobp: xfs_imap() returned an "
167 }
169 /*
170 * If the inode number maps to a block outside the bounds of the
171 * file system then return NULL rather than calling read_buf
172 * and panicing when we get an error from the driver.
173 */
177 "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
182 }
184 /*
185 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
186 * default to just a read_buf() call.
187 */
193 "xfs_inotobp: xfs_trans_read_buf() returned an "
196 }
198 di_ok =
202 XFS_RANDOM_ITOBP_INOTOBP))) {
206 "xfs_inotobp: XFS_TEST_ERROR() returned an "
209 }
213 /*
214 * Set *dipp to point to the on-disk inode in the buffer.
215 */
220 }
223 /*
224 * This routine is called to map an inode to the buffer containing
225 * the on-disk version of the inode. It returns a pointer to the
226 * buffer containing the on-disk inode in the bpp parameter, and in
227 * the dip parameter it returns a pointer to the on-disk inode within
228 * that buffer.
229 *
230 * If a non-zero error is returned, then the contents of bpp and
231 * dipp are undefined.
232 *
233 * If the inode is new and has not yet been initialized, use xfs_imap()
234 * to determine the size and location of the buffer to read from disk.
235 * If the inode has already been mapped to its buffer and read in once,
236 * then use the mapping information stored in the inode rather than
237 * calling xfs_imap(). This allows us to avoid the overhead of looking
238 * at the inode btree for small block file systems (see xfs_dilocate()).
239 * We can tell whether the inode has been mapped in before by comparing
240 * its disk block address to 0. Only uninitialized inodes will have
241 * 0 for the disk block address.
242 */
243 int
250 xfs_daddr_t bno,
251 uint imap_flags)
252 {
253 xfs_imap_t imap;
260 /*
261 * Call the space management code to find the location of the
262 * inode on disk.
263 */
269 /*
270 * If the inode number maps to a block outside the bounds
271 * of the file system then return NULL rather than calling
272 * read_buf and panicing when we get an error from the
273 * driver.
274 */
277 #ifdef DEBUG
279 "(imap.im_blkno (0x%llx) "
280 "+ imap.im_len (0x%llx)) > "
281 " XFS_FSB_TO_BB(mp, "
288 }
290 /*
291 * Fill in the fields in the inode that will be used to
292 * map the inode to its buffer from now on.
293 */
298 /*
299 * We've already mapped the inode once, so just use the
300 * mapping that we saved the first time.
301 */
305 }
308 /*
309 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
310 * default to just a read_buf() call.
311 */
315 #ifdef DEBUG
317 "xfs_trans_read_buf() returned error %d, "
323 }
325 /*
326 * Validate the magic number and version of every inode in the buffer
327 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
328 * No validation is done here in userspace (xfs_repair).
329 */
330 #if !defined(__KERNEL__)
332 #elif defined(DEBUG)
336 #endif
347 XFS_ERRTAG_ITOBP_INOTOBP,
348 XFS_RANDOM_ITOBP_INOTOBP))) {
352 }
353 #ifdef DEBUG
355 "Device %s - bad inode magic/vsn "
360 #endif
365 }
366 }
370 /*
371 * Mark the buffer as an inode buffer now that it looks good
372 */
375 /*
376 * Set *dipp to point to the on-disk inode in the buffer.
377 */
381 }
383 /*
384 * Move inode type and inode format specific information from the
385 * on-disk inode to the in-core inode. For fifos, devs, and sockets
386 * this means set if_rdev to the proper value. For files, directories,
387 * and symlinks this means to bring in the in-line data or extent
388 * pointers. For a file in B-tree format, only the root is immediately
389 * brought in-core. The rest will be in-lined in if_extents when it
390 * is first referenced (see xfs_iread_extents()).
391 */
392 STATIC int
396 {
400 xfs_fsize_t di_size;
418 }
428 }
439 }
450 /*
451 * no local regular files yet
452 */
455 "corrupt inode %Lu "
459 XFS_ERRLEVEL_LOW,
462 }
467 "corrupt inode %Lu "
472 XFS_ERRLEVEL_LOW,
475 }
490 }
496 }
499 }
521 }
526 }
528 }
530 /*
531 * The file is in-lined in the on-disk inode.
532 * If it fits into if_inline_data, then copy
533 * it there, otherwise allocate a buffer for it
534 * and copy the data there. Either way, set
535 * if_data to point at the data.
536 * If we allocate a buffer for the data, make
537 * sure that its size is a multiple of 4 and
538 * record the real size in i_real_bytes.
539 */
540 STATIC int
546 {
550 /*
551 * If the size is unreasonable, then something
552 * is wrong and we just bail out rather than crash in
553 * kmem_alloc() or memcpy() below.
554 */
557 "corrupt inode %Lu "
564 }
574 }
582 }
584 /*
585 * The file consists of a set of extents all
586 * of which fit into the on-disk inode.
587 * If there are few enough extents to fit into
588 * the if_inline_ext, then copy them there.
589 * Otherwise allocate a buffer for them and copy
590 * them into it. Either way, set if_extents
591 * to point at the extents.
592 */
593 STATIC int
598 {
609 /*
610 * If the number of extents is unreasonable, then something
611 * is wrong and we just bail out rather than crash in
612 * kmem_alloc() or memcpy() below.
613 */
621 }
628 else
639 }
646 XFS_ERRLEVEL_LOW,
649 }
650 }
653 }
655 /*
656 * The file has too many extents to fit into
657 * the inode, so they are in B-tree format.
658 * Allocate a buffer for the root of the B-tree
659 * and copy the root into it. The i_extents
660 * field will remain NULL until all of the
661 * extents are read in (when they are needed).
662 */
663 STATIC int
668 {
671 /* REFERENCED */
680 /*
681 * blow out if -- fork has less extents than can fit in
682 * fork (fork shouldn't be a btree format), root btree
683 * block has more records than can fit into the fork,
684 * or the number of extents is greater than the number of
685 * blocks.
686 */
697 }
702 /*
703 * Copy and convert from the on-disk structure
704 * to the in-memory structure.
705 */
712 }
714 void
718 {
747 }
749 void
753 {
782 }
784 STATIC uint
786 __uint16_t di_flags)
787 {
819 }
822 }
824 uint
827 {
832 }
834 uint
837 {
842 }
844 /*
845 * Given a mount structure and an inode number, return a pointer
846 * to a newly allocated in-core inode corresponding to the given
847 * inode number.
848 *
849 * Initialize the inode's attributes and extent pointers if it
850 * already has them (it will not if the inode has no links).
851 */
852 int
856 xfs_ino_t ino,
858 xfs_daddr_t bno,
859 uint imap_flags)
860 {
874 /*
875 * Get pointer's to the on-disk inode and the buffer containing it.
876 * If the inode number refers to a block outside the file system
877 * then xfs_itobp() will return NULL. In this case we should
878 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
879 * know that this is a new incore inode.
880 */
885 }
887 /*
888 * Initialize inode's trace buffers.
889 * Do this before xfs_iformat in case it adds entries.
890 */
891 #ifdef XFS_INODE_TRACE
893 #endif
894 #ifdef XFS_BMAP_TRACE
896 #endif
897 #ifdef XFS_BMBT_TRACE
899 #endif
900 #ifdef XFS_RW_TRACE
902 #endif
903 #ifdef XFS_ILOCK_TRACE
905 #endif
906 #ifdef XFS_DIR2_TRACE
908 #endif
910 /*
911 * If we got something that isn't an inode it means someone
912 * (nfs or dmi) has a stale handle.
913 */
917 #ifdef DEBUG
919 "dip->di_core.di_magic (0x%x) != "
922 XFS_DINODE_MAGIC);
925 }
927 /*
928 * If the on-disk inode is already linked to a directory
929 * entry, copy all of the inode into the in-core inode.
930 * xfs_iformat() handles copying in the inode format
931 * specific information.
932 * Otherwise, just get the truly permanent information.
933 */
940 #ifdef DEBUG
943 error);
946 }
952 /*
953 * Make sure to pull in the mode here as well in
954 * case the inode is released without being used.
955 * This ensures that xfs_inactive() will see that
956 * the inode is already free and not try to mess
957 * with the uninitialized part of it.
958 */
960 /*
961 * Initialize the per-fork minima and maxima for a new
962 * inode here. xfs_iformat will do it for old inodes.
963 */
966 }
970 /*
971 * The inode format changed when we moved the link count and
972 * made it 32 bits long. If this is an old format inode,
973 * convert it in memory to look like a new one. If it gets
974 * flushed to disk we will convert back before flushing or
975 * logging it. We zero out the new projid field and the old link
976 * count field. We'll handle clearing the pad field (the remains
977 * of the old uuid field) when we actually convert the inode to
978 * the new format. We don't change the version number so that we
979 * can distinguish this from a real new format inode.
980 */
985 }
990 /*
991 * Mark the buffer containing the inode as something to keep
992 * around for a while. This helps to keep recently accessed
993 * meta-data in-core longer.
994 */
997 /*
998 * Use xfs_trans_brelse() to release the buffer containing the
999 * on-disk inode, because it was acquired with xfs_trans_read_buf()
1000 * in xfs_itobp() above. If tp is NULL, this is just a normal
1001 * brelse(). If we're within a transaction, then xfs_trans_brelse()
1002 * will only release the buffer if it is not dirty within the
1003 * transaction. It will be OK to release the buffer in this case,
1004 * because inodes on disk are never destroyed and we will be
1005 * locking the new in-core inode before putting it in the hash
1006 * table where other processes can find it. Thus we don't have
1007 * to worry about the inode being changed just because we released
1008 * the buffer.
1009 */
1013 }
1015 /*
1016 * Read in extents from a btree-format inode.
1017 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1018 */
1019 int
1024 {
1027 xfs_extnum_t nextents;
1034 }
1039 /*
1040 * We know that the size is valid (it's checked in iformat_btree)
1041 */
1051 }
1054 }
1056 /*
1057 * Allocate an inode on disk and return a copy of its in-core version.
1058 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1059 * appropriately within the inode. The uid and gid for the inode are
1060 * set according to the contents of the given cred structure.
1061 *
1062 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1063 * has a free inode available, call xfs_iget()
1064 * to obtain the in-core version of the allocated inode. Finally,
1065 * fill in the inode and log its initial contents. In this case,
1066 * ialloc_context would be set to NULL and call_again set to false.
1067 *
1068 * If xfs_dialloc() does not have an available inode,
1069 * it will replenish its supply by doing an allocation. Since we can
1070 * only do one allocation within a transaction without deadlocks, we
1071 * must commit the current transaction before returning the inode itself.
1072 * In this case, therefore, we will set call_again to true and return.
1073 * The caller should then commit the current transaction, start a new
1074 * transaction, and call xfs_ialloc() again to actually get the inode.
1075 *
1076 * To ensure that some other process does not grab the inode that
1077 * was allocated during the first call to xfs_ialloc(), this routine
1078 * also returns the [locked] bp pointing to the head of the freelist
1079 * as ialloc_context. The caller should hold this buffer across
1080 * the commit and pass it back into this routine on the second call.
1081 *
1082 * If we are allocating quota inodes, we do not have a parent inode
1083 * to attach to or associate with (i.e. pip == NULL) because they
1084 * are not linked into the directory structure - they are attached
1085 * directly to the superblock - and so have no parent.
1086 */
1087 int
1091 mode_t mode,
1092 xfs_nlink_t nlink,
1093 xfs_dev_t rdev,
1095 xfs_prid_t prid,
1100 {
1101 xfs_ino_t ino;
1104 uint flags;
1107 /*
1108 * Call the space management code to pick
1109 * the on-disk inode to be allocated.
1110 */
1115 }
1119 }
1122 /*
1123 * Get the in-core inode with the lock held exclusively.
1124 * This is because we're setting fields here we need
1125 * to prevent others from looking at until we're done.
1126 */
1131 }
1144 /*
1145 * If the superblock version is up to where we support new format
1146 * inodes and this is currently an old format inode, then change
1147 * the inode version number now. This way we only do the conversion
1148 * here rather than here and in the flush/logging code.
1149 */
1153 /*
1154 * We've already zeroed the old link count, the projid field,
1155 * and the pad field.
1156 */
1157 }
1159 /*
1160 * Project ids won't be stored on disk if we are using a version 1 inode.
1161 */
1169 }
1170 }
1172 /*
1173 * If the group ID of the new file does not match the effective group
1174 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1175 * (and only if the irix_sgid_inherit compatibility variable is set).
1176 */
1181 }
1188 /*
1189 * di_gen will have been taken care of in xfs_iread.
1190 */
1213 }
1214 /* fall through */
1225 }
1232 }
1233 }
1235 xfs_inherit_noatime)
1238 xfs_inherit_nodump)
1241 xfs_inherit_sync)
1244 xfs_inherit_nosymlinks)
1249 xfs_inherit_nodefrag)
1254 }
1255 /* FALLTHROUGH */
1264 }
1265 /*
1266 * Attribute fork settings for new inode.
1267 */
1271 /*
1272 * Log the new values stuffed into the inode.
1273 */
1276 /* now that we have an i_mode we can setup inode ops and unlock */
1281 }
1283 /*
1284 * Check to make sure that there are no blocks allocated to the
1285 * file beyond the size of the file. We don't check this for
1286 * files with fixed size extents or real time extents, but we
1287 * at least do it for regular files.
1288 */
1289 #ifdef DEBUG
1290 void
1294 xfs_fsize_t isize)
1295 {
1296 xfs_fileoff_t map_first;
1311 /*
1312 * The filesystem could be shutting down, so bmapi may return
1313 * an error.
1314 */
1318 map_first),
1324 }
1327 /*
1328 * Calculate the last possible buffered byte in a file. This must
1329 * include data that was buffered beyond the EOF by the write code.
1330 * This also needs to deal with overflowing the xfs_fsize_t type
1331 * which can happen for sizes near the limit.
1332 *
1333 * We also need to take into account any blocks beyond the EOF. It
1334 * may be the case that they were buffered by a write which failed.
1335 * In that case the pages will still be in memory, but the inode size
1336 * will never have been updated.
1337 */
1338 xfs_fsize_t
1341 {
1343 xfs_fsize_t last_byte;
1344 xfs_fileoff_t last_block;
1345 xfs_fileoff_t size_last_block;
1351 /*
1352 * Only check for blocks beyond the EOF if the extents have
1353 * been read in. This eliminates the need for the inode lock,
1354 * and it also saves us from looking when it really isn't
1355 * necessary.
1356 */
1359 XFS_DATA_FORK);
1362 }
1365 }
1372 }
1376 }
1378 }
1380 #if defined(XFS_RW_TRACE)
1381 STATIC void
1386 xfs_fsize_t new_size,
1387 xfs_off_t toss_start,
1388 xfs_off_t toss_finish)
1389 {
1392 }
1411 }
1412 #else
1413 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1414 #endif
1416 /*
1417 * Start the truncation of the file to new_size. The new size
1418 * must be smaller than the current size. This routine will
1419 * clear the buffer and page caches of file data in the removed
1420 * range, and xfs_itruncate_finish() will remove the underlying
1421 * disk blocks.
1422 *
1423 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1424 * must NOT have the inode lock held at all. This is because we're
1425 * calling into the buffer/page cache code and we can't hold the
1426 * inode lock when we do so.
1427 *
1428 * We need to wait for any direct I/Os in flight to complete before we
1429 * proceed with the truncate. This is needed to prevent the extents
1430 * being read or written by the direct I/Os from being removed while the
1431 * I/O is in flight as there is no other method of synchronising
1432 * direct I/O with the truncate operation. Also, because we hold
1433 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1434 * started until the truncate completes and drops the lock. Essentially,
1435 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1436 * between direct I/Os and the truncate operation.
1437 *
1438 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1439 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1440 * in the case that the caller is locking things out of order and
1441 * may not be able to call xfs_itruncate_finish() with the inode lock
1442 * held without dropping the I/O lock. If the caller must drop the
1443 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1444 * must be called again with all the same restrictions as the initial
1445 * call.
1446 */
1447 int
1450 uint flags,
1451 xfs_fsize_t new_size)
1452 {
1453 xfs_fsize_t last_byte;
1454 xfs_off_t toss_start;
1467 /* wait for the completion of any pending DIOs */
1471 /*
1472 * Call toss_pages or flushinval_pages to get rid of pages
1473 * overlapping the region being removed. We have to use
1474 * the less efficient flushinval_pages in the case that the
1475 * caller may not be able to finish the truncate without
1476 * dropping the inode's I/O lock. Make sure
1477 * to catch any pages brought in by buffers overlapping
1478 * the EOF by searching out beyond the isize by our
1479 * block size. We round new_size up to a block boundary
1480 * so that we don't toss things on the same block as
1481 * new_size but before it.
1482 *
1483 * Before calling toss_page or flushinval_pages, make sure to
1484 * call remapf() over the same region if the file is mapped.
1485 * This frees up mapped file references to the pages in the
1486 * given range and for the flushinval_pages case it ensures
1487 * that we get the latest mapped changes flushed out.
1488 */
1492 /*
1493 * The place to start tossing is beyond our maximum
1494 * file size, so there is no way that the data extended
1495 * out there.
1496 */
1498 }
1501 last_byte);
1509 }
1510 }
1512 #ifdef DEBUG
1515 }
1516 #endif
1518 }
1520 /*
1521 * Shrink the file to the given new_size. The new
1522 * size must be smaller than the current size.
1523 * This will free up the underlying blocks
1524 * in the removed range after a call to xfs_itruncate_start()
1525 * or xfs_atruncate_start().
1526 *
1527 * The transaction passed to this routine must have made
1528 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1529 * This routine may commit the given transaction and
1530 * start new ones, so make sure everything involved in
1531 * the transaction is tidy before calling here.
1532 * Some transaction will be returned to the caller to be
1533 * committed. The incoming transaction must already include
1534 * the inode, and both inode locks must be held exclusively.
1535 * The inode must also be "held" within the transaction. On
1536 * return the inode will be "held" within the returned transaction.
1537 * This routine does NOT require any disk space to be reserved
1538 * for it within the transaction.
1539 *
1540 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1541 * and it indicates the fork which is to be truncated. For the
1542 * attribute fork we only support truncation to size 0.
1543 *
1544 * We use the sync parameter to indicate whether or not the first
1545 * transaction we perform might have to be synchronous. For the attr fork,
1546 * it needs to be so if the unlink of the inode is not yet known to be
1547 * permanent in the log. This keeps us from freeing and reusing the
1548 * blocks of the attribute fork before the unlink of the inode becomes
1549 * permanent.
1550 *
1551 * For the data fork, we normally have to run synchronously if we're
1552 * being called out of the inactive path or we're being called
1553 * out of the create path where we're truncating an existing file.
1554 * Either way, the truncate needs to be sync so blocks don't reappear
1555 * in the file with altered data in case of a crash. wsync filesystems
1556 * can run the first case async because anything that shrinks the inode
1557 * has to run sync so by the time we're called here from inactive, the
1558 * inode size is permanently set to 0.
1559 *
1560 * Calls from the truncate path always need to be sync unless we're
1561 * in a wsync filesystem and the file has already been unlinked.
1562 *
1563 * The caller is responsible for correctly setting the sync parameter.
1564 * It gets too hard for us to guess here which path we're being called
1565 * out of just based on inode state.
1566 */
1567 int
1571 xfs_fsize_t new_size,
1574 {
1575 xfs_fsblock_t first_block;
1576 xfs_fileoff_t first_unmap_block;
1577 xfs_fileoff_t last_block;
1583 xfs_bmap_free_t free_list;
1600 /*
1601 * We only support truncating the entire attribute fork.
1602 */
1605 }
1608 /*
1609 * The first thing we do is set the size to new_size permanently
1610 * on disk. This way we don't have to worry about anyone ever
1611 * being able to look at the data being freed even in the face
1612 * of a crash. What we're getting around here is the case where
1613 * we free a block, it is allocated to another file, it is written
1614 * to, and then we crash. If the new data gets written to the
1615 * file but the log buffers containing the free and reallocation
1616 * don't, then we'd end up with garbage in the blocks being freed.
1617 * As long as we make the new_size permanent before actually
1618 * freeing any blocks it doesn't matter if they get writtten to.
1619 *
1620 * The callers must signal into us whether or not the size
1621 * setting here must be synchronous. There are a few cases
1622 * where it doesn't have to be synchronous. Those cases
1623 * occur if the file is unlinked and we know the unlink is
1624 * permanent or if the blocks being truncated are guaranteed
1625 * to be beyond the inode eof (regardless of the link count)
1626 * and the eof value is permanent. Both of these cases occur
1627 * only on wsync-mounted filesystems. In those cases, we're
1628 * guaranteed that no user will ever see the data in the blocks
1629 * that are being truncated so the truncate can run async.
1630 * In the free beyond eof case, the file may wind up with
1631 * more blocks allocated to it than it needs if we crash
1632 * and that won't get fixed until the next time the file
1633 * is re-opened and closed but that's ok as that shouldn't
1634 * be too many blocks.
1635 *
1636 * However, we can't just make all wsync xactions run async
1637 * because there's one call out of the create path that needs
1638 * to run sync where it's truncating an existing file to size
1639 * 0 whose size is > 0.
1640 *
1641 * It's probably possible to come up with a test in this
1642 * routine that would correctly distinguish all the above
1643 * cases from the values of the function parameters and the
1644 * inode state but for sanity's sake, I've decided to let the
1645 * layers above just tell us. It's simpler to correctly figure
1646 * out in the layer above exactly under what conditions we
1647 * can run async and I think it's easier for others read and
1648 * follow the logic in case something has to be changed.
1649 * cscope is your friend -- rcc.
1650 *
1651 * The attribute fork is much simpler.
1652 *
1653 * For the attribute fork we allow the caller to tell us whether
1654 * the unlink of the inode that led to this call is yet permanent
1655 * in the on disk log. If it is not and we will be freeing extents
1656 * in this inode then we make the first transaction synchronous
1657 * to make sure that the unlink is permanent by the time we free
1658 * the blocks.
1659 */
1662 /*
1663 * If we are not changing the file size then do
1664 * not update the on-disk file size - we may be
1665 * called from xfs_inactive_free_eofblocks(). If we
1666 * update the on-disk file size and then the system
1667 * crashes before the contents of the file are
1668 * flushed to disk then the files may be full of
1669 * holes (ie NULL files bug).
1670 */
1675 }
1676 }
1681 }
1687 /*
1688 * Since it is possible for space to become allocated beyond
1689 * the end of the file (in a crash where the space is allocated
1690 * but the inode size is not yet updated), simply remove any
1691 * blocks which show up between the new EOF and the maximum
1692 * possible file size. If the first block to be removed is
1693 * beyond the maximum file size (ie it is the same as last_block),
1694 * then there is nothing to do.
1695 */
1703 }
1705 /*
1706 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1707 * will tell us whether it freed the entire range or
1708 * not. If this is a synchronous mount (wsync),
1709 * then we can tell bunmapi to keep all the
1710 * transactions asynchronous since the unlink
1711 * transaction that made this inode inactive has
1712 * already hit the disk. There's no danger of
1713 * the freed blocks being reused, there being a
1714 * crash, and the reused blocks suddenly reappearing
1715 * in this file with garbage in them once recovery
1716 * runs.
1717 */
1723 XFS_ITRUNC_MAX_EXTENTS,
1727 /*
1728 * If the bunmapi call encounters an error,
1729 * return to the caller where the transaction
1730 * can be properly aborted. We just need to
1731 * make sure we're not holding any resources
1732 * that we were not when we came in.
1733 */
1736 }
1738 /*
1739 * Duplicate the transaction that has the permanent
1740 * reservation and commit the old transaction.
1741 */
1745 /*
1746 * If the bmap finish call encounters an error,
1747 * return to the caller where the transaction
1748 * can be properly aborted. We just need to
1749 * make sure we're not holding any resources
1750 * that we were not when we came in.
1751 *
1752 * Aborting from this point might lose some
1753 * blocks in the file system, but oh well.
1754 */
1757 /*
1758 * If the passed in transaction committed
1759 * in xfs_bmap_finish(), then we want to
1760 * add the inode to this one before returning.
1761 * This keeps things simple for the higher
1762 * level code, because it always knows that
1763 * the inode is locked and held in the
1764 * transaction that returns to it whether
1765 * errors occur or not. We don't mark the
1766 * inode dirty so that this transaction can
1767 * be easily aborted if possible.
1768 */
1772 }
1774 }
1777 /*
1778 * The first xact was committed,
1779 * so add the inode to the new one.
1780 * Mark it dirty so it will be logged
1781 * and moved forward in the log as
1782 * part of every commit.
1783 */
1788 }
1793 XFS_TRANS_PERM_LOG_RES,
1794 XFS_ITRUNCATE_LOG_COUNT);
1795 /*
1796 * Add the inode being truncated to the next chained
1797 * transaction.
1798 */
1803 }
1804 /*
1805 * Only update the size in the case of the data fork, but
1806 * always re-log the inode so that our permanent transaction
1807 * can keep on rolling it forward in the log.
1808 */
1811 /*
1812 * If we are not changing the file size then do
1813 * not update the on-disk file size - we may be
1814 * called from xfs_inactive_free_eofblocks(). If we
1815 * update the on-disk file size and then the system
1816 * crashes before the contents of the file are
1817 * flushed to disk then the files may be full of
1818 * holes (ie NULL files bug).
1819 */
1823 }
1824 }
1834 }
1837 /*
1838 * xfs_igrow_start
1839 *
1840 * Do the first part of growing a file: zero any data in the last
1841 * block that is beyond the old EOF. We need to do this before
1842 * the inode is joined to the transaction to modify the i_size.
1843 * That way we can drop the inode lock and call into the buffer
1844 * cache to get the buffer mapping the EOF.
1845 */
1846 int
1849 xfs_fsize_t new_size,
1851 {
1856 /*
1857 * Zero any pages that may have been created by
1858 * xfs_write_file() beyond the end of the file
1859 * and any blocks between the old and new file sizes.
1860 */
1862 }
1864 /*
1865 * xfs_igrow_finish
1866 *
1867 * This routine is called to extend the size of a file.
1868 * The inode must have both the iolock and the ilock locked
1869 * for update and it must be a part of the current transaction.
1870 * The xfs_igrow_start() function must have been called previously.
1871 * If the change_flag is not zero, the inode change timestamp will
1872 * be updated.
1873 */
1874 void
1878 xfs_fsize_t new_size,
1880 {
1886 /*
1887 * Update the file size. Update the inode change timestamp
1888 * if change_flag set.
1889 */
1896 }
1899 /*
1900 * This is called when the inode's link count goes to 0.
1901 * We place the on-disk inode on a list in the AGI. It
1902 * will be pulled from this list when the inode is freed.
1903 */
1904 int
1908 {
1914 xfs_agnumber_t agno;
1915 xfs_daddr_t agdaddr;
1916 xfs_agino_t agino;
1931 /*
1932 * Get the agi buffer first. It ensures lock ordering
1933 * on the list.
1934 */
1940 /*
1941 * Validate the magic number of the agi block.
1942 */
1944 agi_ok =
1948 XFS_RANDOM_IUNLINK))) {
1952 }
1953 /*
1954 * Get the index into the agi hash table for the
1955 * list this inode will go on.
1956 */
1964 /*
1965 * There is already another inode in the bucket we need
1966 * to add ourselves to. Add us at the front of the list.
1967 * Here we put the head pointer into our next pointer,
1968 * and then we fall through to point the head at us.
1969 */
1975 /* both on-disk, don't endian flip twice */
1983 }
1985 /*
1986 * Point the bucket head pointer at the inode being inserted.
1987 */
1995 }
1997 /*
1998 * Pull the on-disk inode from the AGI unlinked list.
1999 */
2000 STATIC int
2004 {
2005 xfs_ino_t next_ino;
2011 xfs_agnumber_t agno;
2012 xfs_daddr_t agdaddr;
2013 xfs_agino_t agino;
2014 xfs_agino_t next_agino;
2022 /*
2023 * First pull the on-disk inode from the AGI unlinked list.
2024 */
2030 /*
2031 * Get the agi buffer first. It ensures lock ordering
2032 * on the list.
2033 */
2041 }
2042 /*
2043 * Validate the magic number of the agi block.
2044 */
2046 agi_ok =
2050 XFS_RANDOM_IUNLINK_REMOVE))) {
2058 }
2059 /*
2060 * Get the index into the agi hash table for the
2061 * list this inode will go on.
2062 */
2070 /*
2071 * We're at the head of the list. Get the inode's
2072 * on-disk buffer to see if there is anyone after us
2073 * on the list. Only modify our next pointer if it
2074 * is not already NULLAGINO. This saves us the overhead
2075 * of dealing with the buffer when there is no need to
2076 * change it.
2077 */
2084 }
2097 }
2098 /*
2099 * Point the bucket head pointer at the next inode.
2100 */
2109 /*
2110 * We need to search the list for the inode being freed.
2111 */
2115 /*
2116 * If the last inode wasn't the one pointing to
2117 * us, then release its buffer since we're not
2118 * going to do anything with it.
2119 */
2122 }
2131 }
2135 }
2136 /*
2137 * Now last_ibp points to the buffer previous to us on
2138 * the unlinked list. Pull us from the list.
2139 */
2146 }
2160 }
2161 /*
2162 * Point the previous inode on the list to the next inode.
2163 */
2171 }
2173 }
2176 {
2180 }
2182 STATIC void
2186 xfs_ino_t inum)
2187 {
2193 xfs_daddr_t blkno;
2209 }
2218 /*
2219 * Look for each inode in memory and attempt to lock it,
2220 * we can be racing with flush and tail pushing here.
2221 * any inode we get the locks on, add to an array of
2222 * inode items to process later.
2223 *
2224 * The get the buffer lock, we could beat a flush
2225 * or tail pushing thread to the lock here, in which
2226 * case they will go looking for the inode buffer
2227 * and fail, we need some other form of interlock
2228 * here.
2229 */
2236 /* Inode not in memory or we found it already,
2237 * nothing to do
2238 */
2242 }
2247 }
2249 /* If we can get the locks then add it to the
2250 * list, otherwise by the time we get the bp lock
2251 * below it will already be attached to the
2252 * inode buffer.
2253 */
2255 /* This inode will already be locked - by us, lets
2256 * keep it that way.
2257 */
2266 }
2267 }
2270 }
2281 }
2284 }
2285 }
2287 }
2291 XFS_BUF_LOCK);
2304 pre_flushed++;
2305 }
2307 }
2318 }
2332 }
2333 }
2338 }
2342 }
2344 /*
2345 * This is called to return an inode to the inode free list.
2346 * The inode should already be truncated to 0 length and have
2347 * no pages associated with it. This routine also assumes that
2348 * the inode is already a part of the transaction.
2349 *
2350 * The on-disk copy of the inode will have been added to the list
2351 * of unlinked inodes in the AGI. We need to remove the inode from
2352 * that list atomically with respect to freeing it here.
2353 */
2354 int
2359 {
2362 xfs_ino_t first_ino;
2375 /*
2376 * Pull the on-disk inode from the AGI unlinked list.
2377 */
2381 }
2386 }
2395 /*
2396 * Bump the generation count so no one will be confused
2397 * by reincarnations of this inode.
2398 */
2407 /*
2408 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2409 * from picking up this inode when it is reclaimed (its incore state
2410 * initialzed but not flushed to disk yet). The in-core di_mode is
2411 * already cleared and a corresponding transaction logged.
2412 * The hack here just synchronizes the in-core to on-disk
2413 * di_mode value in advance before the actual inode sync to disk.
2414 * This is OK because the inode is already unlinked and would never
2415 * change its di_mode again for this inode generation.
2416 * This is a temporary hack that would require a proper fix
2417 * in the future.
2418 */
2423 }
2426 }
2428 /*
2429 * Reallocate the space for if_broot based on the number of records
2430 * being added or deleted as indicated in rec_diff. Move the records
2431 * and pointers in if_broot to fit the new size. When shrinking this
2432 * will eliminate holes between the records and pointers created by
2433 * the caller. When growing this will create holes to be filled in
2434 * by the caller.
2435 *
2436 * The caller must not request to add more records than would fit in
2437 * the on-disk inode root. If the if_broot is currently NULL, then
2438 * if we adding records one will be allocated. The caller must also
2439 * not request that the number of records go below zero, although
2440 * it can go to zero.
2441 *
2442 * ip -- the inode whose if_broot area is changing
2443 * ext_diff -- the change in the number of records, positive or negative,
2444 * requested for the if_broot array.
2445 */
2446 void
2451 {
2460 /*
2461 * Handle the degenerate case quietly.
2462 */
2465 }
2469 /*
2470 * If there wasn't any memory allocated before, just
2471 * allocate it now and get out.
2472 */
2476 KM_SLEEP);
2479 }
2481 /*
2482 * If there is already an existing if_broot, then we need
2483 * to realloc() it and shift the pointers to their new
2484 * location. The records don't change location because
2485 * they are kept butted up against the btree block header.
2486 */
2492 new_size,
2494 KM_SLEEP);
2504 }
2506 /*
2507 * rec_diff is less than 0. In this case, we are shrinking the
2508 * if_broot buffer. It must already exist. If we go to zero
2509 * records, just get rid of the root and clear the status bit.
2510 */
2517 else
2521 /*
2522 * First copy over the btree block header.
2523 */
2528 }
2530 /*
2531 * Only copy the records and pointers if there are any.
2532 */
2534 /*
2535 * First copy the records.
2536 */
2543 /*
2544 * Then copy the pointers.
2545 */
2551 }
2558 }
2561 /*
2562 * This is called when the amount of space needed for if_data
2563 * is increased or decreased. The change in size is indicated by
2564 * the number of bytes that need to be added or deleted in the
2565 * byte_diff parameter.
2566 *
2567 * If the amount of space needed has decreased below the size of the
2568 * inline buffer, then switch to using the inline buffer. Otherwise,
2569 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2570 * to what is needed.
2571 *
2572 * ip -- the inode whose if_data area is changing
2573 * byte_diff -- the change in the number of bytes, positive or negative,
2574 * requested for the if_data array.
2575 */
2576 void
2581 {
2588 }
2597 }
2601 /*
2602 * If the valid extents/data can fit in if_inline_ext/data,
2603 * copy them from the malloc'd vector and free it.
2604 */
2610 new_size);
2613 }
2616 /*
2617 * Stuck with malloc/realloc.
2618 * For inline data, the underlying buffer must be
2619 * a multiple of 4 bytes in size so that it can be
2620 * logged and stay on word boundaries. We enforce
2621 * that here.
2622 */
2628 /*
2629 * Only do the realloc if the underlying size
2630 * is really changing.
2631 */
2635 real_size,
2637 KM_SLEEP);
2638 }
2644 }
2645 }
2649 }
2654 /*
2655 * Map inode to disk block and offset.
2656 *
2657 * mp -- the mount point structure for the current file system
2658 * tp -- the current transaction
2659 * ino -- the inode number of the inode to be located
2660 * imap -- this structure is filled in with the information necessary
2661 * to retrieve the given inode from disk
2662 * flags -- flags to pass to xfs_dilocate indicating whether or not
2663 * lookups in the inode btree were OK or not
2664 */
2665 int
2669 xfs_ino_t ino,
2671 uint flags)
2672 {
2673 xfs_fsblock_t fsbno;
2683 }
2690 }
2692 void
2696 {
2703 }
2705 /*
2706 * If the format is local, then we can't have an extents
2707 * array so just look for an inline data array. If we're
2708 * not local then we may or may not have an extents list,
2709 * so check and free it up if we do.
2710 */
2718 }
2725 }
2732 }
2733 }
2735 /*
2736 * This is called free all the memory associated with an inode.
2737 * It must free the inode itself and any buffers allocated for
2738 * if_extents/if_data and if_broot. It must also free the lock
2739 * associated with the inode.
2740 */
2741 void
2744 {
2751 }
2758 #ifdef XFS_INODE_TRACE
2760 #endif
2761 #ifdef XFS_BMAP_TRACE
2763 #endif
2764 #ifdef XFS_BMBT_TRACE
2766 #endif
2767 #ifdef XFS_RW_TRACE
2769 #endif
2770 #ifdef XFS_ILOCK_TRACE
2772 #endif
2773 #ifdef XFS_DIR2_TRACE
2775 #endif
2777 /*
2778 * Only if we are shutting down the fs will we see an
2779 * inode still in the AIL. If it is there, we should remove
2780 * it to prevent a use-after-free from occurring.
2781 */
2791 else
2793 }
2795 }
2797 }
2800 /*
2801 * Increment the pin count of the given buffer.
2802 * This value is protected by ipinlock spinlock in the mount structure.
2803 */
2804 void
2807 {
2811 }
2813 /*
2814 * Decrement the pin count of the given inode, and wake up
2815 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2816 * inode must have been previously pinned with a call to xfs_ipin().
2817 */
2818 void
2821 {
2826 }
2828 /*
2829 * This is called to wait for the given inode to be unpinned.
2830 * It will sleep until this happens. The caller must have the
2831 * inode locked in at least shared mode so that the buffer cannot
2832 * be subsequently pinned once someone is waiting for it to be
2833 * unpinned.
2834 */
2835 STATIC void
2838 {
2840 xfs_lsn_t lsn;
2846 }
2853 }
2855 /*
2856 * Give the log a push so we don't wait here too long.
2857 */
2861 }
2864 /*
2865 * xfs_iextents_copy()
2866 *
2867 * This is called to copy the REAL extents (as opposed to the delayed
2868 * allocation extents) from the inode into the given buffer. It
2869 * returns the number of bytes copied into the buffer.
2870 *
2871 * If there are no delayed allocation extents, then we can just
2872 * memcpy() the extents into the buffer. Otherwise, we need to
2873 * examine each extent in turn and skip those which are delayed.
2874 */
2875 int
2880 {
2885 xfs_fsblock_t start_block;
2895 /*
2896 * There are some delayed allocation extents in the
2897 * inode, so copy the extents one at a time and skip
2898 * the delayed ones. There must be at least one
2899 * non-delayed extent.
2900 */
2906 /*
2907 * It's a delayed allocation extent, so skip it.
2908 */
2910 }
2912 /* Translate to on disk format */
2915 dp++;
2916 copied++;
2917 }
2922 }
2924 /*
2925 * Each of the following cases stores data into the same region
2926 * of the on-disk inode, so only one of them can be valid at
2927 * any given time. While it is possible to have conflicting formats
2928 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2929 * in EXTENTS format, this can only happen when the fork has
2930 * changed formats after being modified but before being flushed.
2931 * In these cases, the format always takes precedence, because the
2932 * format indicates the current state of the fork.
2933 */
2934 /*ARGSUSED*/
2935 STATIC int
2942 {
2946 #ifdef XFS_TRANS_DEBUG
2948 #endif
2959 /*
2960 * This can happen if we gave up in iformat in an error path,
2961 * for the attribute fork.
2962 */
2966 }
2976 }
2990 whichfork);
2991 }
3000 XFS_BROOT_SIZE_ADJ));
3004 }
3011 }
3019 }
3025 }
3028 }
3030 /*
3031 * xfs_iflush() will write a modified inode's changes out to the
3032 * inode's on disk home. The caller must have the inode lock held
3033 * in at least shared mode and the inode flush semaphore must be
3034 * held as well. The inode lock will still be held upon return from
3035 * the call and the caller is free to unlock it.
3036 * The inode flush lock will be unlocked when the inode reaches the disk.
3037 * The flags indicate how the inode's buffer should be written out.
3038 */
3039 int
3042 uint flags)
3043 {
3049 /* REFERENCED */
3066 /*
3067 * If the inode isn't dirty, then just release the inode
3068 * flush lock and do nothing.
3069 */
3076 }
3078 /*
3079 * We can't flush the inode until it is unpinned, so
3080 * wait for it. We know noone new can pin it, because
3081 * we are holding the inode lock shared and you need
3082 * to hold it exclusively to pin the inode.
3083 */
3086 /*
3087 * This may have been unpinned because the filesystem is shutting
3088 * down forcibly. If that's the case we must not write this inode
3089 * to disk, because the log record didn't make it to disk!
3090 */
3097 }
3099 /*
3100 * Get the buffer containing the on-disk inode.
3101 */
3106 }
3108 /*
3109 * Decide how buffer will be flushed out. This is done before
3110 * the call to xfs_iflush_int because this field is zeroed by it.
3111 */
3113 /*
3114 * Flush out the inode buffer according to the directions
3115 * of the caller. In the cases where the caller has given
3116 * us a choice choose the non-delwri case. This is because
3117 * the inode is in the AIL and we need to get it out soon.
3118 */
3135 }
3153 }
3154 }
3156 /*
3157 * First flush out the inode that xfs_iflush was called with.
3158 */
3162 }
3164 /*
3165 * inode clustering:
3166 * see if other inodes can be gathered into this write
3167 */
3176 /*
3177 * Do an un-protected check to see if the inode is dirty and
3178 * is a candidate for flushing. These checks will be repeated
3179 * later after the appropriate locks are acquired.
3180 */
3187 }
3189 /*
3190 * Try to get locks. If any are unavailable,
3191 * then this inode cannot be flushed and is skipped.
3192 */
3194 /* get inode locks (just i_lock) */
3196 /* get inode flush lock */
3198 /* check if pinned */
3200 /* arriving here means that
3201 * this inode can be flushed.
3202 * first re-check that it's
3203 * dirty
3204 */
3209 clcount++;
3213 XFS_ILOCK_SHARED);
3215 }
3218 }
3221 }
3222 }
3224 }
3225 }
3231 }
3233 /*
3234 * If the buffer is pinned then push on the log so we won't
3235 * get stuck waiting in the write for too long.
3236 */
3239 }
3247 }
3250 corrupt_out:
3254 /*
3255 * Unlocks the flush lock
3256 */
3259 cluster_corrupt_out:
3260 /* Corruption detected in the clustering loop. Invalidate the
3261 * inode buffer and shut down the filesystem.
3262 */
3265 /*
3266 * Clean up the buffer. If it was B_DELWRI, just release it --
3267 * brelse can handle it with no problems. If not, shut down the
3268 * filesystem before releasing the buffer.
3269 */
3272 }
3277 /*
3278 * Just like incore_relse: if we have b_iodone functions,
3279 * mark the buffer as an error and call them. Otherwise
3280 * mark it as stale and brelse.
3281 */
3292 }
3293 }
3296 /*
3297 * Unlocks the flush lock
3298 */
3300 }
3303 STATIC int
3307 {
3311 #ifdef XFS_TRANS_DEBUG
3313 #endif
3324 /*
3325 * If the inode isn't dirty, then just release the inode
3326 * flush lock and do nothing.
3327 */
3332 }
3334 /* set *dip = inode's place in the buffer */
3337 /*
3338 * Clear i_update_core before copying out the data.
3339 * This is for coordination with our timestamp updates
3340 * that don't hold the inode lock. They will always
3341 * update the timestamps BEFORE setting i_update_core,
3342 * so if we clear i_update_core after they set it we
3343 * are guaranteed to see their updates to the timestamps.
3344 * I believe that this depends on strongly ordered memory
3345 * semantics, but we have that. We use the SYNCHRONIZE
3346 * macro to make sure that the compiler does not reorder
3347 * the i_update_core access below the data copy below.
3348 */
3352 /*
3353 * Make sure to get the latest atime from the Linux inode.
3354 */
3363 }
3370 }
3380 }
3391 }
3392 }
3395 XFS_RANDOM_IFLUSH_5)) {
3401 ip);
3403 }
3410 }
3411 /*
3412 * bump the flush iteration count, used to detect flushes which
3413 * postdate a log record during recovery.
3414 */
3418 /*
3419 * Copy the dirty parts of the inode into the on-disk
3420 * inode. We always copy out the core of the inode,
3421 * because if the inode is dirty at all the core must
3422 * be.
3423 */
3426 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3430 /*
3431 * If this is really an old format inode and the superblock version
3432 * has not been updated to support only new format inodes, then
3433 * convert back to the old inode format. If the superblock version
3434 * has been updated, then make the conversion permanent.
3435 */
3440 /*
3441 * Convert it back.
3442 */
3446 /*
3447 * The superblock version has already been bumped,
3448 * so just make the conversion to the new inode
3449 * format permanent.
3450 */
3459 }
3460 }
3464 }
3467 /*
3468 * The only error from xfs_iflush_fork is on the data fork.
3469 */
3471 }
3474 /*
3475 * We've recorded everything logged in the inode, so we'd
3476 * like to clear the ilf_fields bits so we don't log and
3477 * flush things unnecessarily. However, we can't stop
3478 * logging all this information until the data we've copied
3479 * into the disk buffer is written to disk. If we did we might
3480 * overwrite the copy of the inode in the log with all the
3481 * data after re-logging only part of it, and in the face of
3482 * a crash we wouldn't have all the data we need to recover.
3483 *
3484 * What we do is move the bits to the ili_last_fields field.
3485 * When logging the inode, these bits are moved back to the
3486 * ilf_fields field. In the xfs_iflush_done() routine we
3487 * clear ili_last_fields, since we know that the information
3488 * those bits represent is permanently on disk. As long as
3489 * the flush completes before the inode is logged again, then
3490 * both ilf_fields and ili_last_fields will be cleared.
3491 *
3492 * We can play with the ilf_fields bits here, because the inode
3493 * lock must be held exclusively in order to set bits there
3494 * and the flush lock protects the ili_last_fields bits.
3495 * Set ili_logged so the flush done
3496 * routine can tell whether or not to look in the AIL.
3497 * Also, store the current LSN of the inode so that we can tell
3498 * whether the item has moved in the AIL from xfs_iflush_done().
3499 * In order to read the lsn we need the AIL lock, because
3500 * it is a 64 bit value that cannot be read atomically.
3501 */
3512 /*
3513 * Attach the function xfs_iflush_done to the inode's
3514 * buffer. This will remove the inode from the AIL
3515 * and unlock the inode's flush lock when the inode is
3516 * completely written to disk.
3517 */
3524 /*
3525 * We're flushing an inode which is not in the AIL and has
3526 * not been logged but has i_update_core set. For this
3527 * case we can use a B_DELWRI flush and immediately drop
3528 * the inode flush lock because we can avoid the whole
3529 * AIL state thing. It's OK to drop the flush lock now,
3530 * because we've already locked the buffer and to do anything
3531 * you really need both.
3532 */
3537 }
3539 }
3543 corrupt_out:
3545 }
3548 /*
3549 * Flush all inactive inodes in mp.
3550 */
3551 void
3554 {
3558 again:
3565 /* Make sure we skip markers inserted by sync */
3569 }
3576 }
3582 out:
3584 }
3586 #ifdef XFS_ILOCK_TRACE
3589 void
3591 {
3600 }
3601 #endif
3603 /*
3604 * Return a pointer to the extent record at file index idx.
3605 */
3606 xfs_bmbt_rec_host_t *
3610 {
3625 }
3626 }
3628 /*
3629 * Insert new item(s) into the extent records for incore inode
3630 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3631 */
3632 void
3638 {
3645 }
3647 /*
3648 * This is called when the amount of space required for incore file
3649 * extents needs to be increased. The ext_diff parameter stores the
3650 * number of new extents being added and the idx parameter contains
3651 * the extent index where the new extents will be added. If the new
3652 * extents are being appended, then we just need to (re)allocate and
3653 * initialize the space. Otherwise, if the new extents are being
3654 * inserted into the middle of the existing entries, a bit more work
3655 * is required to make room for the new extents to be inserted. The
3656 * caller is responsible for filling in the new extent entries upon
3657 * return.
3658 */
3659 void
3664 {
3673 /*
3674 * If the new number of extents (nextents + ext_diff)
3675 * fits inside the inode, then continue to use the inline
3676 * extent buffer.
3677 */
3684 }
3688 }
3689 /*
3690 * Otherwise use a linear (direct) extent list.
3691 * If the extents are currently inside the inode,
3692 * xfs_iext_realloc_direct will switch us from
3693 * inline to direct extent allocation mode.
3694 */
3702 }
3703 }
3704 /* Indirection array */
3717 }
3718 /* Extents fit in target extent page */
3726 }
3729 }
3730 /* Insert a new extent page */
3734 }
3735 /*
3736 * If extent(s) are being appended to the last page in
3737 * the indirection array and the new extent(s) don't fit
3738 * in the page, then erp is NULL and erp_idx is set to
3739 * the next index needed in the indirection array.
3740 */
3749 erp_idx++;
3750 }
3751 }
3752 }
3753 }
3755 }
3757 /*
3758 * This is called when incore extents are being added to the indirection
3759 * array and the new extents do not fit in the target extent list. The
3760 * erp_idx parameter contains the irec index for the target extent list
3761 * in the indirection array, and the idx parameter contains the extent
3762 * index within the list. The number of extents being added is stored
3763 * in the count parameter.
3764 *
3765 * |-------| |-------|
3766 * | | | | idx - number of extents before idx
3767 * | idx | | count |
3768 * | | | | count - number of extents being inserted at idx
3769 * |-------| |-------|
3770 * | count | | nex2 | nex2 - number of extents after idx + count
3771 * |-------| |-------|
3772 */
3773 void
3779 {
3793 /*
3794 * Save second part of target extent list
3795 * (all extents past */
3803 }
3805 /*
3806 * Add the new extents to the end of the target
3807 * list, then allocate new irec record(s) and
3808 * extent buffer(s) as needed to store the rest
3809 * of the new extents.
3810 */
3817 }
3819 erp_idx++;
3825 }
3827 /* Add nex2 extents back to indirection array */
3829 xfs_extnum_t ext_avail;
3835 /*
3836 * If nex2 extents fit in the current page, append
3837 * nex2_ep after the new extents.
3838 */
3841 }
3842 /*
3843 * Otherwise, check if space is available in the
3844 * next page.
3845 */
3849 erp_idx++;
3850 erp++;
3851 /* Create a hole for nex2 extents */
3854 }
3855 /*
3856 * Final choice, create a new extent page for
3857 * nex2 extents.
3858 */
3860 erp_idx++;
3862 }
3867 }
3868 }
3870 /*
3871 * This is called when the amount of space required for incore file
3872 * extents needs to be decreased. The ext_diff parameter stores the
3873 * number of extents to be removed and the idx parameter contains
3874 * the extent index where the extents will be removed from.
3875 *
3876 * If the amount of space needed has decreased below the linear
3877 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3878 * extent array. Otherwise, use kmem_realloc() to adjust the
3879 * size to what is needed.
3880 */
3881 void
3886 {
3902 }
3904 }
3906 /*
3907 * This removes ext_diff extents from the inline buffer, beginning
3908 * at extent index idx.
3909 */
3910 void
3915 {
3934 }
3935 }
3937 /*
3938 * This removes ext_diff extents from a linear (direct) extent list,
3939 * beginning at extent index idx. If the extents are being removed
3940 * from the end of the list (ie. truncate) then we just need to re-
3941 * allocate the list to remove the extra space. Otherwise, if the
3942 * extents are being removed from the middle of the existing extent
3943 * entries, then we first need to move the extent records beginning
3944 * at idx + ext_diff up in the list to overwrite the records being
3945 * removed, then remove the extra space via kmem_realloc.
3946 */
3947 void
3952 {
3964 }
3965 /* Move extents up in the list (if needed) */
3971 }
3974 /*
3975 * Reallocate the direct extent list. If the extents
3976 * will fit inside the inode then xfs_iext_realloc_direct
3977 * will switch from direct to inline extent allocation
3978 * mode for us.
3979 */
3982 }
3984 /*
3985 * This is called when incore extents are being removed from the
3986 * indirection array and the extents being removed span multiple extent
3987 * buffers. The idx parameter contains the file extent index where we
3988 * want to begin removing extents, and the count parameter contains
3989 * how many extents need to be removed.
3990 *
3991 * |-------| |-------|
3992 * | nex1 | | | nex1 - number of extents before idx
3993 * |-------| | count |
3994 * | | | | count - number of extents being removed at idx
3995 * | count | |-------|
3996 * | | | nex2 | nex2 - number of extents after idx + count
3997 * |-------| |-------|
3998 */
3999 void
4004 {
4023 /*
4024 * Check for deletion of entire list;
4025 * xfs_iext_irec_remove() updates extent offsets.
4026 */
4033 XFS_IEXT_BUFSZ);
4039 }
4040 }
4041 /* Move extents up (if needed) */
4046 }
4047 /* Zero out rest of page */
4050 /* Update remaining counters */
4055 erp_idx++;
4056 erp++;
4057 }
4060 }
4062 /*
4063 * Create, destroy, or resize a linear (direct) block of extents.
4064 */
4065 void
4069 {
4078 /* Free extent records */
4081 }
4082 /* Resize direct extent list and zero any new bytes */
4084 /* Check if extents will fit inside the inode */
4090 }
4093 }
4097 rnew_size,
4099 KM_SLEEP);
4100 }
4105 }
4106 }
4107 /*
4108 * Switch from the inline extent buffer to a direct
4109 * extent list. Be sure to include the inline extent
4110 * bytes in new_size.
4111 */
4116 }
4118 }
4121 }
4123 /*
4124 * Switch from linear (direct) extent records to inline buffer.
4125 */
4126 void
4130 {
4133 /*
4134 * The inline buffer was zeroed when we switched
4135 * from inline to direct extent allocation mode,
4136 * so we don't need to clear it here.
4137 */
4143 }
4145 /*
4146 * Switch from inline buffer to linear (direct) extent records.
4147 * new_size should already be rounded up to the next power of 2
4148 * by the caller (when appropriate), so use new_size as it is.
4149 * However, since new_size may be rounded up, we can't update
4150 * if_bytes here. It is the caller's responsibility to update
4151 * if_bytes upon return.
4152 */
4153 void
4157 {
4165 }
4167 }
4169 /*
4170 * Resize an extent indirection array to new_size bytes.
4171 */
4172 void
4176 {
4191 }
4192 }
4194 /*
4195 * Switch from indirection array to linear (direct) extent allocations.
4196 */
4197 void
4200 {
4220 }
4221 }
4223 /*
4224 * Free incore file extents.
4225 */
4226 void
4229 {
4237 }
4244 }
4248 }
4250 /*
4251 * Return a pointer to the extent record for file system block bno.
4252 */
4258 {
4273 }
4276 /* Find target extent list */
4284 }
4285 /* Binary search extent records */
4296 /* Convert back to file-based extent index */
4299 }
4302 }
4303 }
4304 /* Convert back to file-based extent index */
4307 }
4313 }
4314 }
4317 }
4319 /*
4320 * Return a pointer to the indirection array entry containing the
4321 * extent record for filesystem block bno. Store the index of the
4322 * target irec in *erp_idxp.
4323 */
4329 {
4353 }
4354 }
4357 }
4359 /*
4360 * Return a pointer to the indirection array entry containing the
4361 * extent record at file extent index *idxp. Store the index of the
4362 * target irec in *erp_idxp and store the page index of the target
4363 * extent record in *idxp.
4364 */
4365 xfs_ext_irec_t *
4371 {
4388 /* Binary search extent irec's */
4404 erp_idx++;
4410 }
4411 }
4415 }
4417 /*
4418 * Allocate and initialize an indirection array once the space needed
4419 * for incore extents increases above XFS_IEXT_BUFSZ.
4420 */
4421 void
4424 {
4441 }
4452 }
4454 /*
4455 * Allocate and initialize a new entry in the indirection array.
4456 */
4457 xfs_ext_irec_t *
4461 {
4469 /* Resize indirection array */
4472 /*
4473 * Move records down in the array so the
4474 * new page can use erp_idx.
4475 */
4479 }
4482 /* Initialize new extent record */
4491 }
4493 /*
4494 * Remove a record from the indirection array.
4495 */
4496 void
4500 {
4512 }
4513 /* Compact extent records */
4517 }
4518 /*
4519 * Manually free the last extent record from the indirection
4520 * array. A call to xfs_iext_realloc_indirect() with a size
4521 * of zero would result in a call to xfs_iext_destroy() which
4522 * would in turn call this function again, creating a nasty
4523 * infinite loop.
4524 */
4531 }
4533 }
4535 /*
4536 * This is called to clean up large amounts of unused memory allocated
4537 * by the indirection array. Before compacting anything though, verify
4538 * that the indirection array is still needed and switch back to the
4539 * linear extent list (or even the inline buffer) if possible. The
4540 * compaction policy is as follows:
4541 *
4542 * Full Compaction: Extents fit into a single page (or inline buffer)
4543 * Full Compaction: Extents occupy less than 10% of allocated space
4544 * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
4545 * No Compaction: Extents occupy at least 50% of allocated space
4546 */
4547 void
4550 {
4569 }
4570 }
4572 /*
4573 * Combine extents from neighboring extent pages.
4574 */
4575 void
4578 {
4594 /*
4595 * Free page before removing extent record
4596 * so er_extoffs don't get modified in
4597 * xfs_iext_irec_remove.
4598 */
4604 erp_idx++;
4605 }
4606 }
4607 }
4609 /*
4610 * Fully compact the extent records managed by the indirection array.
4611 */
4612 void
4615 {
4635 /* Remove next page */
4637 /*
4638 * Free page before removing extent record
4639 * so er_extoffs don't get modified in
4640 * xfs_iext_irec_remove.
4641 */
4648 /* Update next page */
4650 /* Move rest of page up to become next new page */
4657 }
4659 erp_idx++;
4662 else
4664 }
4668 }
4669 }
4671 /*
4672 * This is called to update the er_extoff field in the indirection
4673 * array when extents have been added or removed from one of the
4674 * extent lists. erp_idx contains the irec index to begin updating
4675 * at and ext_diff contains the number of extents that were added
4676 * or removed.
4677 */
4678 void
4683 {
4691 }
4692 }