| blob | c27d7d495aa0f498ade09aa6a5db5a67a8fa01ce |
1 /*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
58 /*
59 * Used in xfs_itruncate(). This is the maximum number of extents
60 * freed from a file in a single transaction.
61 */
62 #define XFS_ITRUNC_MAX_EXTENTS 2
70 #ifdef DEBUG
71 /*
72 * Make sure that the extents in the given memory buffer
73 * are valid.
74 */
75 STATIC void
80 xfs_exntfmt_t fmt)
81 {
83 xfs_bmbt_irec_t irec;
84 xfs_bmbt_rec_t rec;
93 else
97 }
98 }
100 #define xfs_validate_extents(ifp, nrecs, disk, fmt)
103 /*
104 * Check that none of the inode's in the buffer have a next
105 * unlinked field of 0.
106 */
107 #if defined(DEBUG)
108 void
112 {
124 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
125 bp);
127 }
128 }
129 }
130 #endif
132 /*
133 * This routine is called to map an inode number within a file
134 * system to the buffer containing the on-disk version of the
135 * inode. It returns a pointer to the buffer containing the
136 * on-disk inode in the bpp parameter, and in the dip parameter
137 * it returns a pointer to the on-disk inode within that buffer.
138 *
139 * If a non-zero error is returned, then the contents of bpp and
140 * dipp are undefined.
141 *
142 * Use xfs_imap() to determine the size and location of the
143 * buffer to read from disk.
144 */
145 STATIC int
149 xfs_ino_t ino,
153 {
155 xfs_imap_t imap;
160 /*
161 * Call the space management code to find the location of the
162 * inode on disk.
163 */
168 "xfs_inotobp: xfs_imap() returned an "
171 }
173 /*
174 * If the inode number maps to a block outside the bounds of the
175 * file system then return NULL rather than calling read_buf
176 * and panicing when we get an error from the driver.
177 */
181 "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
186 }
188 /*
189 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
190 * default to just a read_buf() call.
191 */
197 "xfs_inotobp: xfs_trans_read_buf() returned an "
200 }
202 di_ok =
206 XFS_RANDOM_ITOBP_INOTOBP))) {
210 "xfs_inotobp: XFS_TEST_ERROR() returned an "
213 }
217 /*
218 * Set *dipp to point to the on-disk inode in the buffer.
219 */
224 }
227 /*
228 * This routine is called to map an inode to the buffer containing
229 * the on-disk version of the inode. It returns a pointer to the
230 * buffer containing the on-disk inode in the bpp parameter, and in
231 * the dip parameter it returns a pointer to the on-disk inode within
232 * that buffer.
233 *
234 * If a non-zero error is returned, then the contents of bpp and
235 * dipp are undefined.
236 *
237 * If the inode is new and has not yet been initialized, use xfs_imap()
238 * to determine the size and location of the buffer to read from disk.
239 * If the inode has already been mapped to its buffer and read in once,
240 * then use the mapping information stored in the inode rather than
241 * calling xfs_imap(). This allows us to avoid the overhead of looking
242 * at the inode btree for small block file systems (see xfs_dilocate()).
243 * We can tell whether the inode has been mapped in before by comparing
244 * its disk block address to 0. Only uninitialized inodes will have
245 * 0 for the disk block address.
246 */
247 int
254 xfs_daddr_t bno,
255 uint imap_flags)
256 {
257 xfs_imap_t imap;
264 /*
265 * Call the space management code to find the location of the
266 * inode on disk.
267 */
273 /*
274 * If the inode number maps to a block outside the bounds
275 * of the file system then return NULL rather than calling
276 * read_buf and panicing when we get an error from the
277 * driver.
278 */
281 #ifdef DEBUG
283 "(imap.im_blkno (0x%llx) "
284 "+ imap.im_len (0x%llx)) > "
285 " XFS_FSB_TO_BB(mp, "
292 }
294 /*
295 * Fill in the fields in the inode that will be used to
296 * map the inode to its buffer from now on.
297 */
302 /*
303 * We've already mapped the inode once, so just use the
304 * mapping that we saved the first time.
305 */
309 }
312 /*
313 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
314 * default to just a read_buf() call.
315 */
319 #ifdef DEBUG
321 "xfs_trans_read_buf() returned error %d, "
327 }
329 /*
330 * Validate the magic number and version of every inode in the buffer
331 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
332 * No validation is done here in userspace (xfs_repair).
333 */
334 #if !defined(__KERNEL__)
336 #elif defined(DEBUG)
340 #endif
351 XFS_ERRTAG_ITOBP_INOTOBP,
352 XFS_RANDOM_ITOBP_INOTOBP))) {
356 }
357 #ifdef DEBUG
359 "Device %s - bad inode magic/vsn "
364 #endif
369 }
370 }
374 /*
375 * Mark the buffer as an inode buffer now that it looks good
376 */
379 /*
380 * Set *dipp to point to the on-disk inode in the buffer.
381 */
385 }
387 /*
388 * Move inode type and inode format specific information from the
389 * on-disk inode to the in-core inode. For fifos, devs, and sockets
390 * this means set if_rdev to the proper value. For files, directories,
391 * and symlinks this means to bring in the in-line data or extent
392 * pointers. For a file in B-tree format, only the root is immediately
393 * brought in-core. The rest will be in-lined in if_extents when it
394 * is first referenced (see xfs_iread_extents()).
395 */
396 STATIC int
400 {
404 xfs_fsize_t di_size;
423 }
425 if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) {
433 }
444 }
454 /*
455 * no local regular files yet
456 */
459 "corrupt inode %Lu "
463 XFS_ERRLEVEL_LOW,
466 }
471 "corrupt inode %Lu "
476 XFS_ERRLEVEL_LOW,
479 }
494 }
500 }
503 }
525 }
530 }
532 }
534 /*
535 * The file is in-lined in the on-disk inode.
536 * If it fits into if_inline_data, then copy
537 * it there, otherwise allocate a buffer for it
538 * and copy the data there. Either way, set
539 * if_data to point at the data.
540 * If we allocate a buffer for the data, make
541 * sure that its size is a multiple of 4 and
542 * record the real size in i_real_bytes.
543 */
544 STATIC int
550 {
554 /*
555 * If the size is unreasonable, then something
556 * is wrong and we just bail out rather than crash in
557 * kmem_alloc() or memcpy() below.
558 */
561 "corrupt inode %Lu "
568 }
578 }
586 }
588 /*
589 * The file consists of a set of extents all
590 * of which fit into the on-disk inode.
591 * If there are few enough extents to fit into
592 * the if_inline_ext, then copy them there.
593 * Otherwise allocate a buffer for them and copy
594 * them into it. Either way, set if_extents
595 * to point at the extents.
596 */
597 STATIC int
602 {
613 /*
614 * If the number of extents is unreasonable, then something
615 * is wrong and we just bail out rather than crash in
616 * kmem_alloc() or memcpy() below.
617 */
625 }
632 else
642 ARCH_CONVERT);
644 ARCH_CONVERT);
645 }
647 whichfork);
653 XFS_ERRLEVEL_LOW,
656 }
657 }
660 }
662 /*
663 * The file has too many extents to fit into
664 * the inode, so they are in B-tree format.
665 * Allocate a buffer for the root of the B-tree
666 * and copy the root into it. The i_extents
667 * field will remain NULL until all of the
668 * extents are read in (when they are needed).
669 */
670 STATIC int
675 {
678 /* REFERENCED */
687 /*
688 * blow out if -- fork has less extents than can fit in
689 * fork (fork shouldn't be a btree format), root btree
690 * block has more records than can fit into the fork,
691 * or the number of extents is greater than the number of
692 * blocks.
693 */
704 }
709 /*
710 * Copy and convert from the on-disk structure
711 * to the in-memory structure.
712 */
719 }
721 /*
722 * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
723 * and native format
724 *
725 * buf = on-disk representation
726 * dip = native representation
727 * dir = direction - +ve -> disk to native
728 * -ve -> native to disk
729 */
730 void
732 xfs_caddr_t buf,
735 {
758 }
785 }
787 STATIC uint
789 __uint16_t di_flags)
790 {
820 }
823 }
825 uint
828 {
833 }
835 uint
838 {
841 }
843 /*
844 * Given a mount structure and an inode number, return a pointer
845 * to a newly allocated in-core inode corresponding to the given
846 * inode number.
847 *
848 * Initialize the inode's attributes and extent pointers if it
849 * already has them (it will not if the inode has no links).
850 */
851 int
855 xfs_ino_t ino,
857 xfs_daddr_t bno,
858 uint imap_flags)
859 {
872 /*
873 * Get pointer's to the on-disk inode and the buffer containing it.
874 * If the inode number refers to a block outside the file system
875 * then xfs_itobp() will return NULL. In this case we should
876 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
877 * know that this is a new incore inode.
878 */
883 }
885 /*
886 * Initialize inode's trace buffers.
887 * Do this before xfs_iformat in case it adds entries.
888 */
889 #ifdef XFS_BMAP_TRACE
891 #endif
892 #ifdef XFS_BMBT_TRACE
894 #endif
895 #ifdef XFS_RW_TRACE
897 #endif
898 #ifdef XFS_ILOCK_TRACE
900 #endif
901 #ifdef XFS_DIR2_TRACE
903 #endif
905 /*
906 * If we got something that isn't an inode it means someone
907 * (nfs or dmi) has a stale handle.
908 */
912 #ifdef DEBUG
914 "dip->di_core.di_magic (0x%x) != "
917 XFS_DINODE_MAGIC);
920 }
922 /*
923 * If the on-disk inode is already linked to a directory
924 * entry, copy all of the inode into the in-core inode.
925 * xfs_iformat() handles copying in the inode format
926 * specific information.
927 * Otherwise, just get the truly permanent information.
928 */
936 #ifdef DEBUG
939 error);
942 }
948 /*
949 * Make sure to pull in the mode here as well in
950 * case the inode is released without being used.
951 * This ensures that xfs_inactive() will see that
952 * the inode is already free and not try to mess
953 * with the uninitialized part of it.
954 */
956 /*
957 * Initialize the per-fork minima and maxima for a new
958 * inode here. xfs_iformat will do it for old inodes.
959 */
962 }
966 /*
967 * The inode format changed when we moved the link count and
968 * made it 32 bits long. If this is an old format inode,
969 * convert it in memory to look like a new one. If it gets
970 * flushed to disk we will convert back before flushing or
971 * logging it. We zero out the new projid field and the old link
972 * count field. We'll handle clearing the pad field (the remains
973 * of the old uuid field) when we actually convert the inode to
974 * the new format. We don't change the version number so that we
975 * can distinguish this from a real new format inode.
976 */
981 }
985 /*
986 * Mark the buffer containing the inode as something to keep
987 * around for a while. This helps to keep recently accessed
988 * meta-data in-core longer.
989 */
992 /*
993 * Use xfs_trans_brelse() to release the buffer containing the
994 * on-disk inode, because it was acquired with xfs_trans_read_buf()
995 * in xfs_itobp() above. If tp is NULL, this is just a normal
996 * brelse(). If we're within a transaction, then xfs_trans_brelse()
997 * will only release the buffer if it is not dirty within the
998 * transaction. It will be OK to release the buffer in this case,
999 * because inodes on disk are never destroyed and we will be
1000 * locking the new in-core inode before putting it in the hash
1001 * table where other processes can find it. Thus we don't have
1002 * to worry about the inode being changed just because we released
1003 * the buffer.
1004 */
1008 }
1010 /*
1011 * Read in extents from a btree-format inode.
1012 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1013 */
1014 int
1019 {
1022 xfs_extnum_t nextents;
1029 }
1034 /*
1035 * We know that the size is valid (it's checked in iformat_btree)
1036 */
1046 }
1049 }
1051 /*
1052 * Allocate an inode on disk and return a copy of its in-core version.
1053 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1054 * appropriately within the inode. The uid and gid for the inode are
1055 * set according to the contents of the given cred structure.
1056 *
1057 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1058 * has a free inode available, call xfs_iget()
1059 * to obtain the in-core version of the allocated inode. Finally,
1060 * fill in the inode and log its initial contents. In this case,
1061 * ialloc_context would be set to NULL and call_again set to false.
1062 *
1063 * If xfs_dialloc() does not have an available inode,
1064 * it will replenish its supply by doing an allocation. Since we can
1065 * only do one allocation within a transaction without deadlocks, we
1066 * must commit the current transaction before returning the inode itself.
1067 * In this case, therefore, we will set call_again to true and return.
1068 * The caller should then commit the current transaction, start a new
1069 * transaction, and call xfs_ialloc() again to actually get the inode.
1070 *
1071 * To ensure that some other process does not grab the inode that
1072 * was allocated during the first call to xfs_ialloc(), this routine
1073 * also returns the [locked] bp pointing to the head of the freelist
1074 * as ialloc_context. The caller should hold this buffer across
1075 * the commit and pass it back into this routine on the second call.
1076 */
1077 int
1081 mode_t mode,
1082 xfs_nlink_t nlink,
1083 xfs_dev_t rdev,
1085 xfs_prid_t prid,
1090 {
1091 xfs_ino_t ino;
1094 uint flags;
1097 /*
1098 * Call the space management code to pick
1099 * the on-disk inode to be allocated.
1100 */
1105 }
1109 }
1112 /*
1113 * Get the in-core inode with the lock held exclusively.
1114 * This is because we're setting fields here we need
1115 * to prevent others from looking at until we're done.
1116 */
1121 }
1134 /*
1135 * If the superblock version is up to where we support new format
1136 * inodes and this is currently an old format inode, then change
1137 * the inode version number now. This way we only do the conversion
1138 * here rather than here and in the flush/logging code.
1139 */
1143 /*
1144 * We've already zeroed the old link count, the projid field,
1145 * and the pad field.
1146 */
1147 }
1149 /*
1150 * Project ids won't be stored on disk if we are using a version 1 inode.
1151 */
1159 }
1160 }
1162 /*
1163 * If the group ID of the new file does not match the effective group
1164 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1165 * (and only if the irix_sgid_inherit compatibility variable is set).
1166 */
1171 }
1177 /*
1178 * di_gen will have been taken care of in xfs_iread.
1179 */
1206 }
1211 }
1215 }
1216 }
1218 xfs_inherit_noatime)
1221 xfs_inherit_nodump)
1224 xfs_inherit_sync)
1227 xfs_inherit_nosymlinks)
1232 xfs_inherit_nodefrag)
1235 }
1236 /* FALLTHROUGH */
1245 }
1246 /*
1247 * Attribute fork settings for new inode.
1248 */
1252 /*
1253 * Log the new values stuffed into the inode.
1254 */
1257 /* now that we have an i_mode we can setup inode ops and unlock */
1262 }
1264 /*
1265 * Check to make sure that there are no blocks allocated to the
1266 * file beyond the size of the file. We don't check this for
1267 * files with fixed size extents or real time extents, but we
1268 * at least do it for regular files.
1269 */
1270 #ifdef DEBUG
1271 void
1275 xfs_fsize_t isize)
1276 {
1277 xfs_fileoff_t map_first;
1289 /*
1290 * The filesystem could be shutting down, so bmapi may return
1291 * an error.
1292 */
1296 map_first),
1302 }
1305 /*
1306 * Calculate the last possible buffered byte in a file. This must
1307 * include data that was buffered beyond the EOF by the write code.
1308 * This also needs to deal with overflowing the xfs_fsize_t type
1309 * which can happen for sizes near the limit.
1310 *
1311 * We also need to take into account any blocks beyond the EOF. It
1312 * may be the case that they were buffered by a write which failed.
1313 * In that case the pages will still be in memory, but the inode size
1314 * will never have been updated.
1315 */
1316 xfs_fsize_t
1319 {
1321 xfs_fsize_t last_byte;
1322 xfs_fileoff_t last_block;
1323 xfs_fileoff_t size_last_block;
1329 /*
1330 * Only check for blocks beyond the EOF if the extents have
1331 * been read in. This eliminates the need for the inode lock,
1332 * and it also saves us from looking when it really isn't
1333 * necessary.
1334 */
1337 XFS_DATA_FORK);
1340 }
1343 }
1350 }
1354 }
1356 }
1358 #if defined(XFS_RW_TRACE)
1359 STATIC void
1364 xfs_fsize_t new_size,
1365 xfs_off_t toss_start,
1366 xfs_off_t toss_finish)
1367 {
1370 }
1389 }
1390 #else
1391 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1392 #endif
1394 /*
1395 * Start the truncation of the file to new_size. The new size
1396 * must be smaller than the current size. This routine will
1397 * clear the buffer and page caches of file data in the removed
1398 * range, and xfs_itruncate_finish() will remove the underlying
1399 * disk blocks.
1400 *
1401 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1402 * must NOT have the inode lock held at all. This is because we're
1403 * calling into the buffer/page cache code and we can't hold the
1404 * inode lock when we do so.
1405 *
1406 * We need to wait for any direct I/Os in flight to complete before we
1407 * proceed with the truncate. This is needed to prevent the extents
1408 * being read or written by the direct I/Os from being removed while the
1409 * I/O is in flight as there is no other method of synchronising
1410 * direct I/O with the truncate operation. Also, because we hold
1411 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1412 * started until the truncate completes and drops the lock. Essentially,
1413 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1414 * between direct I/Os and the truncate operation.
1415 *
1416 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1417 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1418 * in the case that the caller is locking things out of order and
1419 * may not be able to call xfs_itruncate_finish() with the inode lock
1420 * held without dropping the I/O lock. If the caller must drop the
1421 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1422 * must be called again with all the same restrictions as the initial
1423 * call.
1424 */
1425 void
1428 uint flags,
1429 xfs_fsize_t new_size)
1430 {
1431 xfs_fsize_t last_byte;
1432 xfs_off_t toss_start;
1446 /*
1447 * Call toss_pages or flushinval_pages to get rid of pages
1448 * overlapping the region being removed. We have to use
1449 * the less efficient flushinval_pages in the case that the
1450 * caller may not be able to finish the truncate without
1451 * dropping the inode's I/O lock. Make sure
1452 * to catch any pages brought in by buffers overlapping
1453 * the EOF by searching out beyond the isize by our
1454 * block size. We round new_size up to a block boundary
1455 * so that we don't toss things on the same block as
1456 * new_size but before it.
1457 *
1458 * Before calling toss_page or flushinval_pages, make sure to
1459 * call remapf() over the same region if the file is mapped.
1460 * This frees up mapped file references to the pages in the
1461 * given range and for the flushinval_pages case it ensures
1462 * that we get the latest mapped changes flushed out.
1463 */
1467 /*
1468 * The place to start tossing is beyond our maximum
1469 * file size, so there is no way that the data extended
1470 * out there.
1471 */
1473 }
1476 last_byte);
1482 }
1483 }
1485 #ifdef DEBUG
1488 }
1489 #endif
1490 }
1492 /*
1493 * Shrink the file to the given new_size. The new
1494 * size must be smaller than the current size.
1495 * This will free up the underlying blocks
1496 * in the removed range after a call to xfs_itruncate_start()
1497 * or xfs_atruncate_start().
1498 *
1499 * The transaction passed to this routine must have made
1500 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1501 * This routine may commit the given transaction and
1502 * start new ones, so make sure everything involved in
1503 * the transaction is tidy before calling here.
1504 * Some transaction will be returned to the caller to be
1505 * committed. The incoming transaction must already include
1506 * the inode, and both inode locks must be held exclusively.
1507 * The inode must also be "held" within the transaction. On
1508 * return the inode will be "held" within the returned transaction.
1509 * This routine does NOT require any disk space to be reserved
1510 * for it within the transaction.
1511 *
1512 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1513 * and it indicates the fork which is to be truncated. For the
1514 * attribute fork we only support truncation to size 0.
1515 *
1516 * We use the sync parameter to indicate whether or not the first
1517 * transaction we perform might have to be synchronous. For the attr fork,
1518 * it needs to be so if the unlink of the inode is not yet known to be
1519 * permanent in the log. This keeps us from freeing and reusing the
1520 * blocks of the attribute fork before the unlink of the inode becomes
1521 * permanent.
1522 *
1523 * For the data fork, we normally have to run synchronously if we're
1524 * being called out of the inactive path or we're being called
1525 * out of the create path where we're truncating an existing file.
1526 * Either way, the truncate needs to be sync so blocks don't reappear
1527 * in the file with altered data in case of a crash. wsync filesystems
1528 * can run the first case async because anything that shrinks the inode
1529 * has to run sync so by the time we're called here from inactive, the
1530 * inode size is permanently set to 0.
1531 *
1532 * Calls from the truncate path always need to be sync unless we're
1533 * in a wsync filesystem and the file has already been unlinked.
1534 *
1535 * The caller is responsible for correctly setting the sync parameter.
1536 * It gets too hard for us to guess here which path we're being called
1537 * out of just based on inode state.
1538 */
1539 int
1543 xfs_fsize_t new_size,
1546 {
1547 xfs_fsblock_t first_block;
1548 xfs_fileoff_t first_unmap_block;
1549 xfs_fileoff_t last_block;
1555 xfs_bmap_free_t free_list;
1572 /*
1573 * We only support truncating the entire attribute fork.
1574 */
1577 }
1580 /*
1581 * The first thing we do is set the size to new_size permanently
1582 * on disk. This way we don't have to worry about anyone ever
1583 * being able to look at the data being freed even in the face
1584 * of a crash. What we're getting around here is the case where
1585 * we free a block, it is allocated to another file, it is written
1586 * to, and then we crash. If the new data gets written to the
1587 * file but the log buffers containing the free and reallocation
1588 * don't, then we'd end up with garbage in the blocks being freed.
1589 * As long as we make the new_size permanent before actually
1590 * freeing any blocks it doesn't matter if they get writtten to.
1591 *
1592 * The callers must signal into us whether or not the size
1593 * setting here must be synchronous. There are a few cases
1594 * where it doesn't have to be synchronous. Those cases
1595 * occur if the file is unlinked and we know the unlink is
1596 * permanent or if the blocks being truncated are guaranteed
1597 * to be beyond the inode eof (regardless of the link count)
1598 * and the eof value is permanent. Both of these cases occur
1599 * only on wsync-mounted filesystems. In those cases, we're
1600 * guaranteed that no user will ever see the data in the blocks
1601 * that are being truncated so the truncate can run async.
1602 * In the free beyond eof case, the file may wind up with
1603 * more blocks allocated to it than it needs if we crash
1604 * and that won't get fixed until the next time the file
1605 * is re-opened and closed but that's ok as that shouldn't
1606 * be too many blocks.
1607 *
1608 * However, we can't just make all wsync xactions run async
1609 * because there's one call out of the create path that needs
1610 * to run sync where it's truncating an existing file to size
1611 * 0 whose size is > 0.
1612 *
1613 * It's probably possible to come up with a test in this
1614 * routine that would correctly distinguish all the above
1615 * cases from the values of the function parameters and the
1616 * inode state but for sanity's sake, I've decided to let the
1617 * layers above just tell us. It's simpler to correctly figure
1618 * out in the layer above exactly under what conditions we
1619 * can run async and I think it's easier for others read and
1620 * follow the logic in case something has to be changed.
1621 * cscope is your friend -- rcc.
1622 *
1623 * The attribute fork is much simpler.
1624 *
1625 * For the attribute fork we allow the caller to tell us whether
1626 * the unlink of the inode that led to this call is yet permanent
1627 * in the on disk log. If it is not and we will be freeing extents
1628 * in this inode then we make the first transaction synchronous
1629 * to make sure that the unlink is permanent by the time we free
1630 * the blocks.
1631 */
1636 }
1641 }
1647 /*
1648 * Since it is possible for space to become allocated beyond
1649 * the end of the file (in a crash where the space is allocated
1650 * but the inode size is not yet updated), simply remove any
1651 * blocks which show up between the new EOF and the maximum
1652 * possible file size. If the first block to be removed is
1653 * beyond the maximum file size (ie it is the same as last_block),
1654 * then there is nothing to do.
1655 */
1663 }
1665 /*
1666 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1667 * will tell us whether it freed the entire range or
1668 * not. If this is a synchronous mount (wsync),
1669 * then we can tell bunmapi to keep all the
1670 * transactions asynchronous since the unlink
1671 * transaction that made this inode inactive has
1672 * already hit the disk. There's no danger of
1673 * the freed blocks being reused, there being a
1674 * crash, and the reused blocks suddenly reappearing
1675 * in this file with garbage in them once recovery
1676 * runs.
1677 */
1683 XFS_ITRUNC_MAX_EXTENTS,
1687 /*
1688 * If the bunmapi call encounters an error,
1689 * return to the caller where the transaction
1690 * can be properly aborted. We just need to
1691 * make sure we're not holding any resources
1692 * that we were not when we came in.
1693 */
1696 }
1698 /*
1699 * Duplicate the transaction that has the permanent
1700 * reservation and commit the old transaction.
1701 */
1706 /*
1707 * If the bmap finish call encounters an error,
1708 * return to the caller where the transaction
1709 * can be properly aborted. We just need to
1710 * make sure we're not holding any resources
1711 * that we were not when we came in.
1712 *
1713 * Aborting from this point might lose some
1714 * blocks in the file system, but oh well.
1715 */
1718 /*
1719 * If the passed in transaction committed
1720 * in xfs_bmap_finish(), then we want to
1721 * add the inode to this one before returning.
1722 * This keeps things simple for the higher
1723 * level code, because it always knows that
1724 * the inode is locked and held in the
1725 * transaction that returns to it whether
1726 * errors occur or not. We don't mark the
1727 * inode dirty so that this transaction can
1728 * be easily aborted if possible.
1729 */
1733 }
1735 }
1738 /*
1739 * The first xact was committed,
1740 * so add the inode to the new one.
1741 * Mark it dirty so it will be logged
1742 * and moved forward in the log as
1743 * part of every commit.
1744 */
1749 }
1754 XFS_TRANS_PERM_LOG_RES,
1755 XFS_ITRUNCATE_LOG_COUNT);
1756 /*
1757 * Add the inode being truncated to the next chained
1758 * transaction.
1759 */
1764 }
1765 /*
1766 * Only update the size in the case of the data fork, but
1767 * always re-log the inode so that our permanent transaction
1768 * can keep on rolling it forward in the log.
1769 */
1773 }
1783 }
1786 /*
1787 * xfs_igrow_start
1788 *
1789 * Do the first part of growing a file: zero any data in the last
1790 * block that is beyond the old EOF. We need to do this before
1791 * the inode is joined to the transaction to modify the i_size.
1792 * That way we can drop the inode lock and call into the buffer
1793 * cache to get the buffer mapping the EOF.
1794 */
1795 int
1798 xfs_fsize_t new_size,
1800 {
1807 /*
1808 * Zero any pages that may have been created by
1809 * xfs_write_file() beyond the end of the file
1810 * and any blocks between the old and new file sizes.
1811 */
1815 }
1817 /*
1818 * xfs_igrow_finish
1819 *
1820 * This routine is called to extend the size of a file.
1821 * The inode must have both the iolock and the ilock locked
1822 * for update and it must be a part of the current transaction.
1823 * The xfs_igrow_start() function must have been called previously.
1824 * If the change_flag is not zero, the inode change timestamp will
1825 * be updated.
1826 */
1827 void
1831 xfs_fsize_t new_size,
1833 {
1839 /*
1840 * Update the file size. Update the inode change timestamp
1841 * if change_flag set.
1842 */
1848 }
1851 /*
1852 * This is called when the inode's link count goes to 0.
1853 * We place the on-disk inode on a list in the AGI. It
1854 * will be pulled from this list when the inode is freed.
1855 */
1856 int
1860 {
1866 xfs_agnumber_t agno;
1867 xfs_daddr_t agdaddr;
1868 xfs_agino_t agino;
1883 /*
1884 * Get the agi buffer first. It ensures lock ordering
1885 * on the list.
1886 */
1891 }
1892 /*
1893 * Validate the magic number of the agi block.
1894 */
1896 agi_ok =
1900 XFS_RANDOM_IUNLINK))) {
1904 }
1905 /*
1906 * Get the index into the agi hash table for the
1907 * list this inode will go on.
1908 */
1916 /*
1917 * There is already another inode in the bucket we need
1918 * to add ourselves to. Add us at the front of the list.
1919 * Here we put the head pointer into our next pointer,
1920 * and then we fall through to point the head at us.
1921 */
1925 }
1928 /* both on-disk, don't endian flip twice */
1936 }
1938 /*
1939 * Point the bucket head pointer at the inode being inserted.
1940 */
1948 }
1950 /*
1951 * Pull the on-disk inode from the AGI unlinked list.
1952 */
1953 STATIC int
1957 {
1958 xfs_ino_t next_ino;
1964 xfs_agnumber_t agno;
1965 xfs_daddr_t agdaddr;
1966 xfs_agino_t agino;
1967 xfs_agino_t next_agino;
1975 /*
1976 * First pull the on-disk inode from the AGI unlinked list.
1977 */
1983 /*
1984 * Get the agi buffer first. It ensures lock ordering
1985 * on the list.
1986 */
1994 }
1995 /*
1996 * Validate the magic number of the agi block.
1997 */
1999 agi_ok =
2003 XFS_RANDOM_IUNLINK_REMOVE))) {
2011 }
2012 /*
2013 * Get the index into the agi hash table for the
2014 * list this inode will go on.
2015 */
2023 /*
2024 * We're at the head of the list. Get the inode's
2025 * on-disk buffer to see if there is anyone after us
2026 * on the list. Only modify our next pointer if it
2027 * is not already NULLAGINO. This saves us the overhead
2028 * of dealing with the buffer when there is no need to
2029 * change it.
2030 */
2037 }
2050 }
2051 /*
2052 * Point the bucket head pointer at the next inode.
2053 */
2062 /*
2063 * We need to search the list for the inode being freed.
2064 */
2068 /*
2069 * If the last inode wasn't the one pointing to
2070 * us, then release its buffer since we're not
2071 * going to do anything with it.
2072 */
2075 }
2084 }
2088 }
2089 /*
2090 * Now last_ibp points to the buffer previous to us on
2091 * the unlinked list. Pull us from the list.
2092 */
2099 }
2113 }
2114 /*
2115 * Point the previous inode on the list to the next inode.
2116 */
2124 }
2126 }
2129 {
2133 }
2135 STATIC void
2139 xfs_ino_t inum)
2140 {
2146 xfs_daddr_t blkno;
2163 }
2172 /*
2173 * Look for each inode in memory and attempt to lock it,
2174 * we can be racing with flush and tail pushing here.
2175 * any inode we get the locks on, add to an array of
2176 * inode items to process later.
2177 *
2178 * The get the buffer lock, we could beat a flush
2179 * or tail pushing thread to the lock here, in which
2180 * case they will go looking for the inode buffer
2181 * and fail, we need some other form of interlock
2182 * here.
2183 */
2191 }
2193 /* Inode not in memory or we found it already,
2194 * nothing to do
2195 */
2199 }
2204 }
2206 /* If we can get the locks then add it to the
2207 * list, otherwise by the time we get the bp lock
2208 * below it will already be attached to the
2209 * inode buffer.
2210 */
2212 /* This inode will already be locked - by us, lets
2213 * keep it that way.
2214 */
2226 }
2227 }
2230 }
2243 }
2246 }
2247 }
2250 }
2254 XFS_BUF_LOCK);
2269 pre_flushed++;
2270 }
2272 }
2283 }
2297 }
2298 }
2303 }
2306 }
2308 /*
2309 * This is called to return an inode to the inode free list.
2310 * The inode should already be truncated to 0 length and have
2311 * no pages associated with it. This routine also assumes that
2312 * the inode is already a part of the transaction.
2313 *
2314 * The on-disk copy of the inode will have been added to the list
2315 * of unlinked inodes in the AGI. We need to remove the inode from
2316 * that list atomically with respect to freeing it here.
2317 */
2318 int
2323 {
2326 xfs_ino_t first_ino;
2337 /*
2338 * Pull the on-disk inode from the AGI unlinked list.
2339 */
2343 }
2348 }
2357 /*
2358 * Bump the generation count so no one will be confused
2359 * by reincarnations of this inode.
2360 */
2366 }
2369 }
2371 /*
2372 * Reallocate the space for if_broot based on the number of records
2373 * being added or deleted as indicated in rec_diff. Move the records
2374 * and pointers in if_broot to fit the new size. When shrinking this
2375 * will eliminate holes between the records and pointers created by
2376 * the caller. When growing this will create holes to be filled in
2377 * by the caller.
2378 *
2379 * The caller must not request to add more records than would fit in
2380 * the on-disk inode root. If the if_broot is currently NULL, then
2381 * if we adding records one will be allocated. The caller must also
2382 * not request that the number of records go below zero, although
2383 * it can go to zero.
2384 *
2385 * ip -- the inode whose if_broot area is changing
2386 * ext_diff -- the change in the number of records, positive or negative,
2387 * requested for the if_broot array.
2388 */
2389 void
2394 {
2403 /*
2404 * Handle the degenerate case quietly.
2405 */
2408 }
2412 /*
2413 * If there wasn't any memory allocated before, just
2414 * allocate it now and get out.
2415 */
2419 KM_SLEEP);
2422 }
2424 /*
2425 * If there is already an existing if_broot, then we need
2426 * to realloc() it and shift the pointers to their new
2427 * location. The records don't change location because
2428 * they are kept butted up against the btree block header.
2429 */
2435 new_size,
2437 KM_SLEEP);
2447 }
2449 /*
2450 * rec_diff is less than 0. In this case, we are shrinking the
2451 * if_broot buffer. It must already exist. If we go to zero
2452 * records, just get rid of the root and clear the status bit.
2453 */
2460 else
2464 /*
2465 * First copy over the btree block header.
2466 */
2471 }
2473 /*
2474 * Only copy the records and pointers if there are any.
2475 */
2477 /*
2478 * First copy the records.
2479 */
2486 /*
2487 * Then copy the pointers.
2488 */
2494 }
2501 }
2504 /*
2505 * This is called when the amount of space needed for if_data
2506 * is increased or decreased. The change in size is indicated by
2507 * the number of bytes that need to be added or deleted in the
2508 * byte_diff parameter.
2509 *
2510 * If the amount of space needed has decreased below the size of the
2511 * inline buffer, then switch to using the inline buffer. Otherwise,
2512 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2513 * to what is needed.
2514 *
2515 * ip -- the inode whose if_data area is changing
2516 * byte_diff -- the change in the number of bytes, positive or negative,
2517 * requested for the if_data array.
2518 */
2519 void
2524 {
2531 }
2540 }
2544 /*
2545 * If the valid extents/data can fit in if_inline_ext/data,
2546 * copy them from the malloc'd vector and free it.
2547 */
2553 new_size);
2556 }
2559 /*
2560 * Stuck with malloc/realloc.
2561 * For inline data, the underlying buffer must be
2562 * a multiple of 4 bytes in size so that it can be
2563 * logged and stay on word boundaries. We enforce
2564 * that here.
2565 */
2571 /*
2572 * Only do the realloc if the underlying size
2573 * is really changing.
2574 */
2578 real_size,
2580 KM_SLEEP);
2581 }
2587 }
2588 }
2592 }
2597 /*
2598 * Map inode to disk block and offset.
2599 *
2600 * mp -- the mount point structure for the current file system
2601 * tp -- the current transaction
2602 * ino -- the inode number of the inode to be located
2603 * imap -- this structure is filled in with the information necessary
2604 * to retrieve the given inode from disk
2605 * flags -- flags to pass to xfs_dilocate indicating whether or not
2606 * lookups in the inode btree were OK or not
2607 */
2608 int
2612 xfs_ino_t ino,
2614 uint flags)
2615 {
2616 xfs_fsblock_t fsbno;
2626 }
2633 }
2635 void
2639 {
2646 }
2648 /*
2649 * If the format is local, then we can't have an extents
2650 * array so just look for an inline data array. If we're
2651 * not local then we may or may not have an extents list,
2652 * so check and free it up if we do.
2653 */
2661 }
2668 }
2675 }
2676 }
2678 /*
2679 * This is called free all the memory associated with an inode.
2680 * It must free the inode itself and any buffers allocated for
2681 * if_extents/if_data and if_broot. It must also free the lock
2682 * associated with the inode.
2683 */
2684 void
2687 {
2695 }
2701 #ifdef XFS_BMAP_TRACE
2703 #endif
2704 #ifdef XFS_BMBT_TRACE
2706 #endif
2707 #ifdef XFS_RW_TRACE
2709 #endif
2710 #ifdef XFS_ILOCK_TRACE
2712 #endif
2713 #ifdef XFS_DIR2_TRACE
2715 #endif
2717 /* XXXdpd should be able to assert this but shutdown
2718 * is leaving the AIL behind. */
2722 }
2724 }
2727 /*
2728 * Increment the pin count of the given buffer.
2729 * This value is protected by ipinlock spinlock in the mount structure.
2730 */
2731 void
2734 {
2738 }
2740 /*
2741 * Decrement the pin count of the given inode, and wake up
2742 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2743 * inode must have been previously pinned with a call to xfs_ipin().
2744 */
2745 void
2748 {
2752 /*
2753 * If the inode is currently being reclaimed, the
2754 * linux inode _and_ the xfs vnode may have been
2755 * freed so we cannot reference either of them safely.
2756 * Hence we should not try to do anything to them
2757 * if the xfs inode is currently in the reclaim
2758 * path.
2759 *
2760 * However, we still need to issue the unpin wakeup
2761 * call as the inode reclaim may be blocked waiting for
2762 * the inode to become unpinned.
2763 */
2770 /* make sync come back and flush this inode */
2781 }
2782 }
2787 }
2788 }
2790 /*
2791 * This is called to wait for the given inode to be unpinned.
2792 * It will sleep until this happens. The caller must have the
2793 * inode locked in at least shared mode so that the buffer cannot
2794 * be subsequently pinned once someone is waiting for it to be
2795 * unpinned.
2796 */
2797 STATIC void
2800 {
2802 xfs_lsn_t lsn;
2808 }
2815 }
2817 /*
2818 * Give the log a push so we don't wait here too long.
2819 */
2823 }
2826 /*
2827 * xfs_iextents_copy()
2828 *
2829 * This is called to copy the REAL extents (as opposed to the delayed
2830 * allocation extents) from the inode into the given buffer. It
2831 * returns the number of bytes copied into the buffer.
2832 *
2833 * If there are no delayed allocation extents, then we can just
2834 * memcpy() the extents into the buffer. Otherwise, we need to
2835 * examine each extent in turn and skip those which are delayed.
2836 */
2837 int
2842 {
2846 #ifdef XFS_BMAP_TRACE
2848 #endif
2852 xfs_fsblock_t start_block;
2862 /*
2863 * There are some delayed allocation extents in the
2864 * inode, so copy the extents one at a time and skip
2865 * the delayed ones. There must be at least one
2866 * non-delayed extent.
2867 */
2874 /*
2875 * It's a delayed allocation extent, so skip it.
2876 */
2878 }
2880 /* Translate to on disk format */
2885 dest_ep++;
2886 copied++;
2887 }
2892 }
2894 /*
2895 * Each of the following cases stores data into the same region
2896 * of the on-disk inode, so only one of them can be valid at
2897 * any given time. While it is possible to have conflicting formats
2898 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2899 * in EXTENTS format, this can only happen when the fork has
2900 * changed formats after being modified but before being flushed.
2901 * In these cases, the format always takes precedence, because the
2902 * format indicates the current state of the fork.
2903 */
2904 /*ARGSUSED*/
2905 STATIC int
2912 {
2916 #ifdef XFS_TRANS_DEBUG
2918 #endif
2929 /*
2930 * This can happen if we gave up in iformat in an error path,
2931 * for the attribute fork.
2932 */
2936 }
2946 }
2960 whichfork);
2961 }
2970 XFS_BROOT_SIZE_ADJ));
2974 }
2981 }
2989 }
2995 }
2998 }
3000 /*
3001 * xfs_iflush() will write a modified inode's changes out to the
3002 * inode's on disk home. The caller must have the inode lock held
3003 * in at least shared mode and the inode flush semaphore must be
3004 * held as well. The inode lock will still be held upon return from
3005 * the call and the caller is free to unlock it.
3006 * The inode flush lock will be unlocked when the inode reaches the disk.
3007 * The flags indicate how the inode's buffer should be written out.
3008 */
3009 int
3012 uint flags)
3013 {
3019 /* REFERENCED */
3037 /*
3038 * If the inode isn't dirty, then just release the inode
3039 * flush lock and do nothing.
3040 */
3047 }
3049 /*
3050 * We can't flush the inode until it is unpinned, so
3051 * wait for it. We know noone new can pin it, because
3052 * we are holding the inode lock shared and you need
3053 * to hold it exclusively to pin the inode.
3054 */
3057 /*
3058 * This may have been unpinned because the filesystem is shutting
3059 * down forcibly. If that's the case we must not write this inode
3060 * to disk, because the log record didn't make it to disk!
3061 */
3068 }
3070 /*
3071 * Get the buffer containing the on-disk inode.
3072 */
3077 }
3079 /*
3080 * Decide how buffer will be flushed out. This is done before
3081 * the call to xfs_iflush_int because this field is zeroed by it.
3082 */
3084 /*
3085 * Flush out the inode buffer according to the directions
3086 * of the caller. In the cases where the caller has given
3087 * us a choice choose the non-delwri case. This is because
3088 * the inode is in the AIL and we need to get it out soon.
3089 */
3106 }
3124 }
3125 }
3127 /*
3128 * First flush out the inode that xfs_iflush was called with.
3129 */
3133 }
3135 /*
3136 * inode clustering:
3137 * see if other inodes can be gathered into this write
3138 */
3147 /*
3148 * Do an un-protected check to see if the inode is dirty and
3149 * is a candidate for flushing. These checks will be repeated
3150 * later after the appropriate locks are acquired.
3151 */
3158 }
3160 /*
3161 * Try to get locks. If any are unavailable,
3162 * then this inode cannot be flushed and is skipped.
3163 */
3165 /* get inode locks (just i_lock) */
3167 /* get inode flush lock */
3169 /* check if pinned */
3171 /* arriving here means that
3172 * this inode can be flushed.
3173 * first re-check that it's
3174 * dirty
3175 */
3180 clcount++;
3184 XFS_ILOCK_SHARED);
3186 }
3189 }
3192 }
3193 }
3195 }
3196 }
3202 }
3204 /*
3205 * If the buffer is pinned then push on the log so we won't
3206 * get stuck waiting in the write for too long.
3207 */
3210 }
3218 }
3221 corrupt_out:
3225 /*
3226 * Unlocks the flush lock
3227 */
3230 cluster_corrupt_out:
3231 /* Corruption detected in the clustering loop. Invalidate the
3232 * inode buffer and shut down the filesystem.
3233 */
3236 /*
3237 * Clean up the buffer. If it was B_DELWRI, just release it --
3238 * brelse can handle it with no problems. If not, shut down the
3239 * filesystem before releasing the buffer.
3240 */
3243 }
3248 /*
3249 * Just like incore_relse: if we have b_iodone functions,
3250 * mark the buffer as an error and call them. Otherwise
3251 * mark it as stale and brelse.
3252 */
3263 }
3264 }
3267 /*
3268 * Unlocks the flush lock
3269 */
3271 }
3274 STATIC int
3278 {
3282 #ifdef XFS_TRANS_DEBUG
3284 #endif
3296 /*
3297 * If the inode isn't dirty, then just release the inode
3298 * flush lock and do nothing.
3299 */
3304 }
3306 /* set *dip = inode's place in the buffer */
3309 /*
3310 * Clear i_update_core before copying out the data.
3311 * This is for coordination with our timestamp updates
3312 * that don't hold the inode lock. They will always
3313 * update the timestamps BEFORE setting i_update_core,
3314 * so if we clear i_update_core after they set it we
3315 * are guaranteed to see their updates to the timestamps.
3316 * I believe that this depends on strongly ordered memory
3317 * semantics, but we have that. We use the SYNCHRONIZE
3318 * macro to make sure that the compiler does not reorder
3319 * the i_update_core access below the data copy below.
3320 */
3324 /*
3325 * Make sure to get the latest atime from the Linux inode.
3326 */
3335 }
3342 }
3352 }
3363 }
3364 }
3367 XFS_RANDOM_IFLUSH_5)) {
3373 ip);
3375 }
3382 }
3383 /*
3384 * bump the flush iteration count, used to detect flushes which
3385 * postdate a log record during recovery.
3386 */
3390 /*
3391 * Copy the dirty parts of the inode into the on-disk
3392 * inode. We always copy out the core of the inode,
3393 * because if the inode is dirty at all the core must
3394 * be.
3395 */
3398 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3402 /*
3403 * If this is really an old format inode and the superblock version
3404 * has not been updated to support only new format inodes, then
3405 * convert back to the old inode format. If the superblock version
3406 * has been updated, then make the conversion permanent.
3407 */
3412 /*
3413 * Convert it back.
3414 */
3418 /*
3419 * The superblock version has already been bumped,
3420 * so just make the conversion to the new inode
3421 * format permanent.
3422 */
3431 }
3432 }
3436 }
3439 /*
3440 * The only error from xfs_iflush_fork is on the data fork.
3441 */
3443 }
3446 /*
3447 * We've recorded everything logged in the inode, so we'd
3448 * like to clear the ilf_fields bits so we don't log and
3449 * flush things unnecessarily. However, we can't stop
3450 * logging all this information until the data we've copied
3451 * into the disk buffer is written to disk. If we did we might
3452 * overwrite the copy of the inode in the log with all the
3453 * data after re-logging only part of it, and in the face of
3454 * a crash we wouldn't have all the data we need to recover.
3455 *
3456 * What we do is move the bits to the ili_last_fields field.
3457 * When logging the inode, these bits are moved back to the
3458 * ilf_fields field. In the xfs_iflush_done() routine we
3459 * clear ili_last_fields, since we know that the information
3460 * those bits represent is permanently on disk. As long as
3461 * the flush completes before the inode is logged again, then
3462 * both ilf_fields and ili_last_fields will be cleared.
3463 *
3464 * We can play with the ilf_fields bits here, because the inode
3465 * lock must be held exclusively in order to set bits there
3466 * and the flush lock protects the ili_last_fields bits.
3467 * Set ili_logged so the flush done
3468 * routine can tell whether or not to look in the AIL.
3469 * Also, store the current LSN of the inode so that we can tell
3470 * whether the item has moved in the AIL from xfs_iflush_done().
3471 * In order to read the lsn we need the AIL lock, because
3472 * it is a 64 bit value that cannot be read atomically.
3473 */
3484 /*
3485 * Attach the function xfs_iflush_done to the inode's
3486 * buffer. This will remove the inode from the AIL
3487 * and unlock the inode's flush lock when the inode is
3488 * completely written to disk.
3489 */
3496 /*
3497 * We're flushing an inode which is not in the AIL and has
3498 * not been logged but has i_update_core set. For this
3499 * case we can use a B_DELWRI flush and immediately drop
3500 * the inode flush lock because we can avoid the whole
3501 * AIL state thing. It's OK to drop the flush lock now,
3502 * because we've already locked the buffer and to do anything
3503 * you really need both.
3504 */
3509 }
3511 }
3515 corrupt_out:
3517 }
3520 /*
3521 * Flush all inactive inodes in mp.
3522 */
3523 void
3526 {
3530 again:
3537 /* Make sure we skip markers inserted by sync */
3541 }
3548 }
3554 out:
3556 }
3558 /*
3559 * xfs_iaccess: check accessibility of inode for mode.
3560 */
3561 int
3564 mode_t mode,
3566 {
3580 }
3582 /*
3583 * If there's an Access Control List it's used instead of
3584 * the mode bits.
3585 */
3593 }
3595 /*
3596 * If the DACs are ok we don't need any capability check.
3597 */
3600 /*
3601 * Read/write DACs are always overridable.
3602 * Executable DACs are overridable if at least one exec bit is set.
3603 */
3613 #ifdef NOISE
3617 }
3619 }
3621 /*
3622 * xfs_iroundup: round up argument to next power of two
3623 */
3624 uint
3626 uint v)
3627 {
3629 uint m;
3642 }
3645 }
3647 #ifdef XFS_ILOCK_TRACE
3650 void
3652 {
3661 }
3662 #endif
3664 /*
3665 * Return a pointer to the extent record at file index idx.
3666 */
3667 xfs_bmbt_rec_t *
3671 {
3686 }
3687 }
3689 /*
3690 * Insert new item(s) into the extent records for incore inode
3691 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3692 */
3693 void
3699 {
3708 }
3709 }
3711 /*
3712 * This is called when the amount of space required for incore file
3713 * extents needs to be increased. The ext_diff parameter stores the
3714 * number of new extents being added and the idx parameter contains
3715 * the extent index where the new extents will be added. If the new
3716 * extents are being appended, then we just need to (re)allocate and
3717 * initialize the space. Otherwise, if the new extents are being
3718 * inserted into the middle of the existing entries, a bit more work
3719 * is required to make room for the new extents to be inserted. The
3720 * caller is responsible for filling in the new extent entries upon
3721 * return.
3722 */
3723 void
3728 {
3737 /*
3738 * If the new number of extents (nextents + ext_diff)
3739 * fits inside the inode, then continue to use the inline
3740 * extent buffer.
3741 */
3748 }
3752 }
3753 /*
3754 * Otherwise use a linear (direct) extent list.
3755 * If the extents are currently inside the inode,
3756 * xfs_iext_realloc_direct will switch us from
3757 * inline to direct extent allocation mode.
3758 */
3766 }
3767 }
3768 /* Indirection array */
3781 }
3782 /* Extents fit in target extent page */
3790 }
3793 }
3794 /* Insert a new extent page */
3798 }
3799 /*
3800 * If extent(s) are being appended to the last page in
3801 * the indirection array and the new extent(s) don't fit
3802 * in the page, then erp is NULL and erp_idx is set to
3803 * the next index needed in the indirection array.
3804 */
3813 erp_idx++;
3814 }
3815 }
3816 }
3817 }
3819 }
3821 /*
3822 * This is called when incore extents are being added to the indirection
3823 * array and the new extents do not fit in the target extent list. The
3824 * erp_idx parameter contains the irec index for the target extent list
3825 * in the indirection array, and the idx parameter contains the extent
3826 * index within the list. The number of extents being added is stored
3827 * in the count parameter.
3828 *
3829 * |-------| |-------|
3830 * | | | | idx - number of extents before idx
3831 * | idx | | count |
3832 * | | | | count - number of extents being inserted at idx
3833 * |-------| |-------|
3834 * | count | | nex2 | nex2 - number of extents after idx + count
3835 * |-------| |-------|
3836 */
3837 void
3843 {
3857 /*
3858 * Save second part of target extent list
3859 * (all extents past */
3867 }
3869 /*
3870 * Add the new extents to the end of the target
3871 * list, then allocate new irec record(s) and
3872 * extent buffer(s) as needed to store the rest
3873 * of the new extents.
3874 */
3881 }
3883 erp_idx++;
3889 }
3891 /* Add nex2 extents back to indirection array */
3893 xfs_extnum_t ext_avail;
3899 /*
3900 * If nex2 extents fit in the current page, append
3901 * nex2_ep after the new extents.
3902 */
3905 }
3906 /*
3907 * Otherwise, check if space is available in the
3908 * next page.
3909 */
3913 erp_idx++;
3914 erp++;
3915 /* Create a hole for nex2 extents */
3918 }
3919 /*
3920 * Final choice, create a new extent page for
3921 * nex2 extents.
3922 */
3924 erp_idx++;
3926 }
3931 }
3932 }
3934 /*
3935 * This is called when the amount of space required for incore file
3936 * extents needs to be decreased. The ext_diff parameter stores the
3937 * number of extents to be removed and the idx parameter contains
3938 * the extent index where the extents will be removed from.
3939 *
3940 * If the amount of space needed has decreased below the linear
3941 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3942 * extent array. Otherwise, use kmem_realloc() to adjust the
3943 * size to what is needed.
3944 */
3945 void
3950 {
3966 }
3968 }
3970 /*
3971 * This removes ext_diff extents from the inline buffer, beginning
3972 * at extent index idx.
3973 */
3974 void
3979 {
3998 }
3999 }
4001 /*
4002 * This removes ext_diff extents from a linear (direct) extent list,
4003 * beginning at extent index idx. If the extents are being removed
4004 * from the end of the list (ie. truncate) then we just need to re-
4005 * allocate the list to remove the extra space. Otherwise, if the
4006 * extents are being removed from the middle of the existing extent
4007 * entries, then we first need to move the extent records beginning
4008 * at idx + ext_diff up in the list to overwrite the records being
4009 * removed, then remove the extra space via kmem_realloc.
4010 */
4011 void
4016 {
4028 }
4029 /* Move extents up in the list (if needed) */
4035 }
4038 /*
4039 * Reallocate the direct extent list. If the extents
4040 * will fit inside the inode then xfs_iext_realloc_direct
4041 * will switch from direct to inline extent allocation
4042 * mode for us.
4043 */
4046 }
4048 /*
4049 * This is called when incore extents are being removed from the
4050 * indirection array and the extents being removed span multiple extent
4051 * buffers. The idx parameter contains the file extent index where we
4052 * want to begin removing extents, and the count parameter contains
4053 * how many extents need to be removed.
4054 *
4055 * |-------| |-------|
4056 * | nex1 | | | nex1 - number of extents before idx
4057 * |-------| | count |
4058 * | | | | count - number of extents being removed at idx
4059 * | count | |-------|
4060 * | | | nex2 | nex2 - number of extents after idx + count
4061 * |-------| |-------|
4062 */
4063 void
4068 {
4087 /*
4088 * Check for deletion of entire list;
4089 * xfs_iext_irec_remove() updates extent offsets.
4090 */
4097 XFS_IEXT_BUFSZ);
4103 }
4104 }
4105 /* Move extents up (if needed) */
4110 }
4111 /* Zero out rest of page */
4114 /* Update remaining counters */
4119 erp_idx++;
4120 erp++;
4121 }
4124 }
4126 /*
4127 * Create, destroy, or resize a linear (direct) block of extents.
4128 */
4129 void
4133 {
4142 /* Free extent records */
4145 }
4146 /* Resize direct extent list and zero any new bytes */
4148 /* Check if extents will fit inside the inode */
4154 }
4157 }
4161 rnew_size,
4163 KM_SLEEP);
4164 }
4169 }
4170 }
4171 /*
4172 * Switch from the inline extent buffer to a direct
4173 * extent list. Be sure to include the inline extent
4174 * bytes in new_size.
4175 */
4180 }
4182 }
4185 }
4187 /*
4188 * Switch from linear (direct) extent records to inline buffer.
4189 */
4190 void
4194 {
4197 /*
4198 * The inline buffer was zeroed when we switched
4199 * from inline to direct extent allocation mode,
4200 * so we don't need to clear it here.
4201 */
4207 }
4209 /*
4210 * Switch from inline buffer to linear (direct) extent records.
4211 * new_size should already be rounded up to the next power of 2
4212 * by the caller (when appropriate), so use new_size as it is.
4213 * However, since new_size may be rounded up, we can't update
4214 * if_bytes here. It is the caller's responsibility to update
4215 * if_bytes upon return.
4216 */
4217 void
4221 {
4230 }
4232 }
4234 /*
4235 * Resize an extent indirection array to new_size bytes.
4236 */
4237 void
4241 {
4256 }
4257 }
4259 /*
4260 * Switch from indirection array to linear (direct) extent allocations.
4261 */
4262 void
4265 {
4285 }
4286 }
4288 /*
4289 * Free incore file extents.
4290 */
4291 void
4294 {
4302 }
4309 }
4313 }
4315 /*
4316 * Return a pointer to the extent record for file system block bno.
4317 */
4323 {
4338 }
4341 /* Find target extent list */
4349 }
4350 /* Binary search extent records */
4361 /* Convert back to file-based extent index */
4364 }
4367 }
4368 }
4369 /* Convert back to file-based extent index */
4372 }
4378 }
4379 }
4382 }
4384 /*
4385 * Return a pointer to the indirection array entry containing the
4386 * extent record for filesystem block bno. Store the index of the
4387 * target irec in *erp_idxp.
4388 */
4394 {
4418 }
4419 }
4422 }
4424 /*
4425 * Return a pointer to the indirection array entry containing the
4426 * extent record at file extent index *idxp. Store the index of the
4427 * target irec in *erp_idxp and store the page index of the target
4428 * extent record in *idxp.
4429 */
4430 xfs_ext_irec_t *
4436 {
4453 /* Binary search extent irec's */
4469 erp_idx++;
4475 }
4476 }
4480 }
4482 /*
4483 * Allocate and initialize an indirection array once the space needed
4484 * for incore extents increases above XFS_IEXT_BUFSZ.
4485 */
4486 void
4489 {
4507 }
4518 }
4520 /*
4521 * Allocate and initialize a new entry in the indirection array.
4522 */
4523 xfs_ext_irec_t *
4527 {
4535 /* Resize indirection array */
4538 /*
4539 * Move records down in the array so the
4540 * new page can use erp_idx.
4541 */
4545 }
4548 /* Initialize new extent record */
4558 }
4560 /*
4561 * Remove a record from the indirection array.
4562 */
4563 void
4567 {
4579 }
4580 /* Compact extent records */
4584 }
4585 /*
4586 * Manually free the last extent record from the indirection
4587 * array. A call to xfs_iext_realloc_indirect() with a size
4588 * of zero would result in a call to xfs_iext_destroy() which
4589 * would in turn call this function again, creating a nasty
4590 * infinite loop.
4591 */
4598 }
4600 }
4602 /*
4603 * This is called to clean up large amounts of unused memory allocated
4604 * by the indirection array. Before compacting anything though, verify
4605 * that the indirection array is still needed and switch back to the
4606 * linear extent list (or even the inline buffer) if possible. The
4607 * compaction policy is as follows:
4608 *
4609 * Full Compaction: Extents fit into a single page (or inline buffer)
4610 * Full Compaction: Extents occupy less than 10% of allocated space
4611 * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
4612 * No Compaction: Extents occupy at least 50% of allocated space
4613 */
4614 void
4617 {
4636 }
4637 }
4639 /*
4640 * Combine extents from neighboring extent pages.
4641 */
4642 void
4645 {
4661 /*
4662 * Free page before removing extent record
4663 * so er_extoffs don't get modified in
4664 * xfs_iext_irec_remove.
4665 */
4671 erp_idx++;
4672 }
4673 }
4674 }
4676 /*
4677 * Fully compact the extent records managed by the indirection array.
4678 */
4679 void
4682 {
4702 /* Remove next page */
4704 /*
4705 * Free page before removing extent record
4706 * so er_extoffs don't get modified in
4707 * xfs_iext_irec_remove.
4708 */
4715 /* Update next page */
4717 /* Move rest of page up to become next new page */
4724 }
4726 erp_idx++;
4729 else
4731 }
4735 }
4736 }
4738 /*
4739 * This is called to update the er_extoff field in the indirection
4740 * array when extents have been added or removed from one of the
4741 * extent lists. erp_idx contains the irec index to begin updating
4742 * at and ext_diff contains the number of extents that were added
4743 * or removed.
4744 */
4745 void
4750 {
4758 }
4759 }