| blob | bc9e7c8918091690b407c3d08dc1293c18c0e2fe |
1 /*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
57 /*
58 * Used in xfs_itruncate(). This is the maximum number of extents
59 * freed from a file in a single transaction.
60 */
61 #define XFS_ITRUNC_MAX_EXTENTS 2
68 #ifdef DEBUG
69 /*
70 * Make sure that the extents in the given memory buffer
71 * are valid.
72 */
73 STATIC void
77 xfs_exntfmt_t fmt)
78 {
79 xfs_bmbt_irec_t irec;
80 xfs_bmbt_rec_host_t rec;
90 }
91 }
93 #define xfs_validate_extents(ifp, nrecs, fmt)
96 /*
97 * Check that none of the inode's in the buffer have a next
98 * unlinked field of 0.
99 */
100 #if defined(DEBUG)
101 void
105 {
117 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
118 bp);
120 }
121 }
122 }
123 #endif
125 /*
126 * This routine is called to map an inode number within a file
127 * system to the buffer containing the on-disk version of the
128 * inode. It returns a pointer to the buffer containing the
129 * on-disk inode in the bpp parameter, and in the dip parameter
130 * it returns a pointer to the on-disk inode within that buffer.
131 *
132 * If a non-zero error is returned, then the contents of bpp and
133 * dipp are undefined.
134 *
135 * Use xfs_imap() to determine the size and location of the
136 * buffer to read from disk.
137 */
138 STATIC int
142 xfs_ino_t ino,
146 {
148 xfs_imap_t imap;
153 /*
154 * Call the space management code to find the location of the
155 * inode on disk.
156 */
161 "xfs_inotobp: xfs_imap() returned an "
164 }
166 /*
167 * If the inode number maps to a block outside the bounds of the
168 * file system then return NULL rather than calling read_buf
169 * and panicing when we get an error from the driver.
170 */
174 "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
179 }
181 /*
182 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
183 * default to just a read_buf() call.
184 */
190 "xfs_inotobp: xfs_trans_read_buf() returned an "
193 }
195 di_ok =
199 XFS_RANDOM_ITOBP_INOTOBP))) {
203 "xfs_inotobp: XFS_TEST_ERROR() returned an "
206 }
210 /*
211 * Set *dipp to point to the on-disk inode in the buffer.
212 */
217 }
220 /*
221 * This routine is called to map an inode to the buffer containing
222 * the on-disk version of the inode. It returns a pointer to the
223 * buffer containing the on-disk inode in the bpp parameter, and in
224 * the dip parameter it returns a pointer to the on-disk inode within
225 * that buffer.
226 *
227 * If a non-zero error is returned, then the contents of bpp and
228 * dipp are undefined.
229 *
230 * If the inode is new and has not yet been initialized, use xfs_imap()
231 * to determine the size and location of the buffer to read from disk.
232 * If the inode has already been mapped to its buffer and read in once,
233 * then use the mapping information stored in the inode rather than
234 * calling xfs_imap(). This allows us to avoid the overhead of looking
235 * at the inode btree for small block file systems (see xfs_dilocate()).
236 * We can tell whether the inode has been mapped in before by comparing
237 * its disk block address to 0. Only uninitialized inodes will have
238 * 0 for the disk block address.
239 */
240 int
247 xfs_daddr_t bno,
248 uint imap_flags)
249 {
250 xfs_imap_t imap;
257 /*
258 * Call the space management code to find the location of the
259 * inode on disk.
260 */
266 /*
267 * If the inode number maps to a block outside the bounds
268 * of the file system then return NULL rather than calling
269 * read_buf and panicing when we get an error from the
270 * driver.
271 */
274 #ifdef DEBUG
276 "(imap.im_blkno (0x%llx) "
277 "+ imap.im_len (0x%llx)) > "
278 " XFS_FSB_TO_BB(mp, "
285 }
287 /*
288 * Fill in the fields in the inode that will be used to
289 * map the inode to its buffer from now on.
290 */
295 /*
296 * We've already mapped the inode once, so just use the
297 * mapping that we saved the first time.
298 */
302 }
305 /*
306 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
307 * default to just a read_buf() call.
308 */
312 #ifdef DEBUG
314 "xfs_trans_read_buf() returned error %d, "
320 }
322 /*
323 * Validate the magic number and version of every inode in the buffer
324 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
325 * No validation is done here in userspace (xfs_repair).
326 */
327 #if !defined(__KERNEL__)
329 #elif defined(DEBUG)
333 #endif
344 XFS_ERRTAG_ITOBP_INOTOBP,
345 XFS_RANDOM_ITOBP_INOTOBP))) {
349 }
350 #ifdef DEBUG
352 "Device %s - bad inode magic/vsn "
357 #endif
362 }
363 }
367 /*
368 * Mark the buffer as an inode buffer now that it looks good
369 */
372 /*
373 * Set *dipp to point to the on-disk inode in the buffer.
374 */
378 }
380 /*
381 * Move inode type and inode format specific information from the
382 * on-disk inode to the in-core inode. For fifos, devs, and sockets
383 * this means set if_rdev to the proper value. For files, directories,
384 * and symlinks this means to bring in the in-line data or extent
385 * pointers. For a file in B-tree format, only the root is immediately
386 * brought in-core. The rest will be in-lined in if_extents when it
387 * is first referenced (see xfs_iread_extents()).
388 */
389 STATIC int
393 {
397 xfs_fsize_t di_size;
416 }
418 if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) {
426 }
437 }
448 /*
449 * no local regular files yet
450 */
453 "corrupt inode %Lu "
457 XFS_ERRLEVEL_LOW,
460 }
465 "corrupt inode %Lu "
470 XFS_ERRLEVEL_LOW,
473 }
488 }
494 }
497 }
519 }
524 }
526 }
528 /*
529 * The file is in-lined in the on-disk inode.
530 * If it fits into if_inline_data, then copy
531 * it there, otherwise allocate a buffer for it
532 * and copy the data there. Either way, set
533 * if_data to point at the data.
534 * If we allocate a buffer for the data, make
535 * sure that its size is a multiple of 4 and
536 * record the real size in i_real_bytes.
537 */
538 STATIC int
544 {
548 /*
549 * If the size is unreasonable, then something
550 * is wrong and we just bail out rather than crash in
551 * kmem_alloc() or memcpy() below.
552 */
555 "corrupt inode %Lu "
562 }
572 }
580 }
582 /*
583 * The file consists of a set of extents all
584 * of which fit into the on-disk inode.
585 * If there are few enough extents to fit into
586 * the if_inline_ext, then copy them there.
587 * Otherwise allocate a buffer for them and copy
588 * them into it. Either way, set if_extents
589 * to point at the extents.
590 */
591 STATIC int
596 {
607 /*
608 * If the number of extents is unreasonable, then something
609 * is wrong and we just bail out rather than crash in
610 * kmem_alloc() or memcpy() below.
611 */
619 }
626 else
637 }
644 XFS_ERRLEVEL_LOW,
647 }
648 }
651 }
653 /*
654 * The file has too many extents to fit into
655 * the inode, so they are in B-tree format.
656 * Allocate a buffer for the root of the B-tree
657 * and copy the root into it. The i_extents
658 * field will remain NULL until all of the
659 * extents are read in (when they are needed).
660 */
661 STATIC int
666 {
669 /* REFERENCED */
678 /*
679 * blow out if -- fork has less extents than can fit in
680 * fork (fork shouldn't be a btree format), root btree
681 * block has more records than can fit into the fork,
682 * or the number of extents is greater than the number of
683 * blocks.
684 */
695 }
700 /*
701 * Copy and convert from the on-disk structure
702 * to the in-memory structure.
703 */
710 }
712 /*
713 * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
714 * and native format
715 *
716 * buf = on-disk representation
717 * dip = native representation
718 * dir = direction - +ve -> disk to native
719 * -ve -> native to disk
720 */
721 void
723 xfs_caddr_t buf,
726 {
749 }
776 }
778 STATIC uint
780 __uint16_t di_flags)
781 {
813 }
816 }
818 uint
821 {
826 }
828 uint
831 {
834 }
836 /*
837 * Given a mount structure and an inode number, return a pointer
838 * to a newly allocated in-core inode corresponding to the given
839 * inode number.
840 *
841 * Initialize the inode's attributes and extent pointers if it
842 * already has them (it will not if the inode has no links).
843 */
844 int
848 xfs_ino_t ino,
850 xfs_daddr_t bno,
851 uint imap_flags)
852 {
865 /*
866 * Get pointer's to the on-disk inode and the buffer containing it.
867 * If the inode number refers to a block outside the file system
868 * then xfs_itobp() will return NULL. In this case we should
869 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
870 * know that this is a new incore inode.
871 */
876 }
878 /*
879 * Initialize inode's trace buffers.
880 * Do this before xfs_iformat in case it adds entries.
881 */
882 #ifdef XFS_BMAP_TRACE
884 #endif
885 #ifdef XFS_BMBT_TRACE
887 #endif
888 #ifdef XFS_RW_TRACE
890 #endif
891 #ifdef XFS_ILOCK_TRACE
893 #endif
894 #ifdef XFS_DIR2_TRACE
896 #endif
898 /*
899 * If we got something that isn't an inode it means someone
900 * (nfs or dmi) has a stale handle.
901 */
905 #ifdef DEBUG
907 "dip->di_core.di_magic (0x%x) != "
910 XFS_DINODE_MAGIC);
913 }
915 /*
916 * If the on-disk inode is already linked to a directory
917 * entry, copy all of the inode into the in-core inode.
918 * xfs_iformat() handles copying in the inode format
919 * specific information.
920 * Otherwise, just get the truly permanent information.
921 */
929 #ifdef DEBUG
932 error);
935 }
941 /*
942 * Make sure to pull in the mode here as well in
943 * case the inode is released without being used.
944 * This ensures that xfs_inactive() will see that
945 * the inode is already free and not try to mess
946 * with the uninitialized part of it.
947 */
949 /*
950 * Initialize the per-fork minima and maxima for a new
951 * inode here. xfs_iformat will do it for old inodes.
952 */
955 }
959 /*
960 * The inode format changed when we moved the link count and
961 * made it 32 bits long. If this is an old format inode,
962 * convert it in memory to look like a new one. If it gets
963 * flushed to disk we will convert back before flushing or
964 * logging it. We zero out the new projid field and the old link
965 * count field. We'll handle clearing the pad field (the remains
966 * of the old uuid field) when we actually convert the inode to
967 * the new format. We don't change the version number so that we
968 * can distinguish this from a real new format inode.
969 */
974 }
979 /*
980 * Mark the buffer containing the inode as something to keep
981 * around for a while. This helps to keep recently accessed
982 * meta-data in-core longer.
983 */
986 /*
987 * Use xfs_trans_brelse() to release the buffer containing the
988 * on-disk inode, because it was acquired with xfs_trans_read_buf()
989 * in xfs_itobp() above. If tp is NULL, this is just a normal
990 * brelse(). If we're within a transaction, then xfs_trans_brelse()
991 * will only release the buffer if it is not dirty within the
992 * transaction. It will be OK to release the buffer in this case,
993 * because inodes on disk are never destroyed and we will be
994 * locking the new in-core inode before putting it in the hash
995 * table where other processes can find it. Thus we don't have
996 * to worry about the inode being changed just because we released
997 * the buffer.
998 */
1002 }
1004 /*
1005 * Read in extents from a btree-format inode.
1006 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1007 */
1008 int
1013 {
1016 xfs_extnum_t nextents;
1023 }
1028 /*
1029 * We know that the size is valid (it's checked in iformat_btree)
1030 */
1040 }
1043 }
1045 /*
1046 * Allocate an inode on disk and return a copy of its in-core version.
1047 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1048 * appropriately within the inode. The uid and gid for the inode are
1049 * set according to the contents of the given cred structure.
1050 *
1051 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1052 * has a free inode available, call xfs_iget()
1053 * to obtain the in-core version of the allocated inode. Finally,
1054 * fill in the inode and log its initial contents. In this case,
1055 * ialloc_context would be set to NULL and call_again set to false.
1056 *
1057 * If xfs_dialloc() does not have an available inode,
1058 * it will replenish its supply by doing an allocation. Since we can
1059 * only do one allocation within a transaction without deadlocks, we
1060 * must commit the current transaction before returning the inode itself.
1061 * In this case, therefore, we will set call_again to true and return.
1062 * The caller should then commit the current transaction, start a new
1063 * transaction, and call xfs_ialloc() again to actually get the inode.
1064 *
1065 * To ensure that some other process does not grab the inode that
1066 * was allocated during the first call to xfs_ialloc(), this routine
1067 * also returns the [locked] bp pointing to the head of the freelist
1068 * as ialloc_context. The caller should hold this buffer across
1069 * the commit and pass it back into this routine on the second call.
1070 *
1071 * If we are allocating quota inodes, we do not have a parent inode
1072 * to attach to or associate with (i.e. pip == NULL) because they
1073 * are not linked into the directory structure - they are attached
1074 * directly to the superblock - and so have no parent.
1075 */
1076 int
1080 mode_t mode,
1081 xfs_nlink_t nlink,
1082 xfs_dev_t rdev,
1084 xfs_prid_t prid,
1089 {
1090 xfs_ino_t ino;
1093 uint flags;
1096 /*
1097 * Call the space management code to pick
1098 * the on-disk inode to be allocated.
1099 */
1104 }
1108 }
1111 /*
1112 * Get the in-core inode with the lock held exclusively.
1113 * This is because we're setting fields here we need
1114 * to prevent others from looking at until we're done.
1115 */
1120 }
1133 /*
1134 * If the superblock version is up to where we support new format
1135 * inodes and this is currently an old format inode, then change
1136 * the inode version number now. This way we only do the conversion
1137 * here rather than here and in the flush/logging code.
1138 */
1142 /*
1143 * We've already zeroed the old link count, the projid field,
1144 * and the pad field.
1145 */
1146 }
1148 /*
1149 * Project ids won't be stored on disk if we are using a version 1 inode.
1150 */
1158 }
1159 }
1161 /*
1162 * If the group ID of the new file does not match the effective group
1163 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1164 * (and only if the irix_sgid_inherit compatibility variable is set).
1165 */
1170 }
1177 /*
1178 * di_gen will have been taken care of in xfs_iread.
1179 */
1202 }
1203 /* fall through */
1214 }
1219 }
1223 }
1224 }
1226 xfs_inherit_noatime)
1229 xfs_inherit_nodump)
1232 xfs_inherit_sync)
1235 xfs_inherit_nosymlinks)
1240 xfs_inherit_nodefrag)
1245 }
1246 /* FALLTHROUGH */
1255 }
1256 /*
1257 * Attribute fork settings for new inode.
1258 */
1262 /*
1263 * Log the new values stuffed into the inode.
1264 */
1267 /* now that we have an i_mode we can setup inode ops and unlock */
1272 }
1274 /*
1275 * Check to make sure that there are no blocks allocated to the
1276 * file beyond the size of the file. We don't check this for
1277 * files with fixed size extents or real time extents, but we
1278 * at least do it for regular files.
1279 */
1280 #ifdef DEBUG
1281 void
1285 xfs_fsize_t isize)
1286 {
1287 xfs_fileoff_t map_first;
1299 /*
1300 * The filesystem could be shutting down, so bmapi may return
1301 * an error.
1302 */
1306 map_first),
1312 }
1315 /*
1316 * Calculate the last possible buffered byte in a file. This must
1317 * include data that was buffered beyond the EOF by the write code.
1318 * This also needs to deal with overflowing the xfs_fsize_t type
1319 * which can happen for sizes near the limit.
1320 *
1321 * We also need to take into account any blocks beyond the EOF. It
1322 * may be the case that they were buffered by a write which failed.
1323 * In that case the pages will still be in memory, but the inode size
1324 * will never have been updated.
1325 */
1326 xfs_fsize_t
1329 {
1331 xfs_fsize_t last_byte;
1332 xfs_fileoff_t last_block;
1333 xfs_fileoff_t size_last_block;
1339 /*
1340 * Only check for blocks beyond the EOF if the extents have
1341 * been read in. This eliminates the need for the inode lock,
1342 * and it also saves us from looking when it really isn't
1343 * necessary.
1344 */
1347 XFS_DATA_FORK);
1350 }
1353 }
1360 }
1364 }
1366 }
1368 #if defined(XFS_RW_TRACE)
1369 STATIC void
1374 xfs_fsize_t new_size,
1375 xfs_off_t toss_start,
1376 xfs_off_t toss_finish)
1377 {
1380 }
1399 }
1400 #else
1401 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1402 #endif
1404 /*
1405 * Start the truncation of the file to new_size. The new size
1406 * must be smaller than the current size. This routine will
1407 * clear the buffer and page caches of file data in the removed
1408 * range, and xfs_itruncate_finish() will remove the underlying
1409 * disk blocks.
1410 *
1411 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1412 * must NOT have the inode lock held at all. This is because we're
1413 * calling into the buffer/page cache code and we can't hold the
1414 * inode lock when we do so.
1415 *
1416 * We need to wait for any direct I/Os in flight to complete before we
1417 * proceed with the truncate. This is needed to prevent the extents
1418 * being read or written by the direct I/Os from being removed while the
1419 * I/O is in flight as there is no other method of synchronising
1420 * direct I/O with the truncate operation. Also, because we hold
1421 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1422 * started until the truncate completes and drops the lock. Essentially,
1423 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1424 * between direct I/Os and the truncate operation.
1425 *
1426 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1427 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1428 * in the case that the caller is locking things out of order and
1429 * may not be able to call xfs_itruncate_finish() with the inode lock
1430 * held without dropping the I/O lock. If the caller must drop the
1431 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1432 * must be called again with all the same restrictions as the initial
1433 * call.
1434 */
1435 int
1438 uint flags,
1439 xfs_fsize_t new_size)
1440 {
1441 xfs_fsize_t last_byte;
1442 xfs_off_t toss_start;
1457 /*
1458 * Call toss_pages or flushinval_pages to get rid of pages
1459 * overlapping the region being removed. We have to use
1460 * the less efficient flushinval_pages in the case that the
1461 * caller may not be able to finish the truncate without
1462 * dropping the inode's I/O lock. Make sure
1463 * to catch any pages brought in by buffers overlapping
1464 * the EOF by searching out beyond the isize by our
1465 * block size. We round new_size up to a block boundary
1466 * so that we don't toss things on the same block as
1467 * new_size but before it.
1468 *
1469 * Before calling toss_page or flushinval_pages, make sure to
1470 * call remapf() over the same region if the file is mapped.
1471 * This frees up mapped file references to the pages in the
1472 * given range and for the flushinval_pages case it ensures
1473 * that we get the latest mapped changes flushed out.
1474 */
1478 /*
1479 * The place to start tossing is beyond our maximum
1480 * file size, so there is no way that the data extended
1481 * out there.
1482 */
1484 }
1487 last_byte);
1493 }
1494 }
1496 #ifdef DEBUG
1499 }
1500 #endif
1502 }
1504 /*
1505 * Shrink the file to the given new_size. The new
1506 * size must be smaller than the current size.
1507 * This will free up the underlying blocks
1508 * in the removed range after a call to xfs_itruncate_start()
1509 * or xfs_atruncate_start().
1510 *
1511 * The transaction passed to this routine must have made
1512 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1513 * This routine may commit the given transaction and
1514 * start new ones, so make sure everything involved in
1515 * the transaction is tidy before calling here.
1516 * Some transaction will be returned to the caller to be
1517 * committed. The incoming transaction must already include
1518 * the inode, and both inode locks must be held exclusively.
1519 * The inode must also be "held" within the transaction. On
1520 * return the inode will be "held" within the returned transaction.
1521 * This routine does NOT require any disk space to be reserved
1522 * for it within the transaction.
1523 *
1524 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1525 * and it indicates the fork which is to be truncated. For the
1526 * attribute fork we only support truncation to size 0.
1527 *
1528 * We use the sync parameter to indicate whether or not the first
1529 * transaction we perform might have to be synchronous. For the attr fork,
1530 * it needs to be so if the unlink of the inode is not yet known to be
1531 * permanent in the log. This keeps us from freeing and reusing the
1532 * blocks of the attribute fork before the unlink of the inode becomes
1533 * permanent.
1534 *
1535 * For the data fork, we normally have to run synchronously if we're
1536 * being called out of the inactive path or we're being called
1537 * out of the create path where we're truncating an existing file.
1538 * Either way, the truncate needs to be sync so blocks don't reappear
1539 * in the file with altered data in case of a crash. wsync filesystems
1540 * can run the first case async because anything that shrinks the inode
1541 * has to run sync so by the time we're called here from inactive, the
1542 * inode size is permanently set to 0.
1543 *
1544 * Calls from the truncate path always need to be sync unless we're
1545 * in a wsync filesystem and the file has already been unlinked.
1546 *
1547 * The caller is responsible for correctly setting the sync parameter.
1548 * It gets too hard for us to guess here which path we're being called
1549 * out of just based on inode state.
1550 */
1551 int
1555 xfs_fsize_t new_size,
1558 {
1559 xfs_fsblock_t first_block;
1560 xfs_fileoff_t first_unmap_block;
1561 xfs_fileoff_t last_block;
1567 xfs_bmap_free_t free_list;
1584 /*
1585 * We only support truncating the entire attribute fork.
1586 */
1589 }
1592 /*
1593 * The first thing we do is set the size to new_size permanently
1594 * on disk. This way we don't have to worry about anyone ever
1595 * being able to look at the data being freed even in the face
1596 * of a crash. What we're getting around here is the case where
1597 * we free a block, it is allocated to another file, it is written
1598 * to, and then we crash. If the new data gets written to the
1599 * file but the log buffers containing the free and reallocation
1600 * don't, then we'd end up with garbage in the blocks being freed.
1601 * As long as we make the new_size permanent before actually
1602 * freeing any blocks it doesn't matter if they get writtten to.
1603 *
1604 * The callers must signal into us whether or not the size
1605 * setting here must be synchronous. There are a few cases
1606 * where it doesn't have to be synchronous. Those cases
1607 * occur if the file is unlinked and we know the unlink is
1608 * permanent or if the blocks being truncated are guaranteed
1609 * to be beyond the inode eof (regardless of the link count)
1610 * and the eof value is permanent. Both of these cases occur
1611 * only on wsync-mounted filesystems. In those cases, we're
1612 * guaranteed that no user will ever see the data in the blocks
1613 * that are being truncated so the truncate can run async.
1614 * In the free beyond eof case, the file may wind up with
1615 * more blocks allocated to it than it needs if we crash
1616 * and that won't get fixed until the next time the file
1617 * is re-opened and closed but that's ok as that shouldn't
1618 * be too many blocks.
1619 *
1620 * However, we can't just make all wsync xactions run async
1621 * because there's one call out of the create path that needs
1622 * to run sync where it's truncating an existing file to size
1623 * 0 whose size is > 0.
1624 *
1625 * It's probably possible to come up with a test in this
1626 * routine that would correctly distinguish all the above
1627 * cases from the values of the function parameters and the
1628 * inode state but for sanity's sake, I've decided to let the
1629 * layers above just tell us. It's simpler to correctly figure
1630 * out in the layer above exactly under what conditions we
1631 * can run async and I think it's easier for others read and
1632 * follow the logic in case something has to be changed.
1633 * cscope is your friend -- rcc.
1634 *
1635 * The attribute fork is much simpler.
1636 *
1637 * For the attribute fork we allow the caller to tell us whether
1638 * the unlink of the inode that led to this call is yet permanent
1639 * in the on disk log. If it is not and we will be freeing extents
1640 * in this inode then we make the first transaction synchronous
1641 * to make sure that the unlink is permanent by the time we free
1642 * the blocks.
1643 */
1646 /*
1647 * If we are not changing the file size then do
1648 * not update the on-disk file size - we may be
1649 * called from xfs_inactive_free_eofblocks(). If we
1650 * update the on-disk file size and then the system
1651 * crashes before the contents of the file are
1652 * flushed to disk then the files may be full of
1653 * holes (ie NULL files bug).
1654 */
1659 }
1660 }
1665 }
1671 /*
1672 * Since it is possible for space to become allocated beyond
1673 * the end of the file (in a crash where the space is allocated
1674 * but the inode size is not yet updated), simply remove any
1675 * blocks which show up between the new EOF and the maximum
1676 * possible file size. If the first block to be removed is
1677 * beyond the maximum file size (ie it is the same as last_block),
1678 * then there is nothing to do.
1679 */
1687 }
1689 /*
1690 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1691 * will tell us whether it freed the entire range or
1692 * not. If this is a synchronous mount (wsync),
1693 * then we can tell bunmapi to keep all the
1694 * transactions asynchronous since the unlink
1695 * transaction that made this inode inactive has
1696 * already hit the disk. There's no danger of
1697 * the freed blocks being reused, there being a
1698 * crash, and the reused blocks suddenly reappearing
1699 * in this file with garbage in them once recovery
1700 * runs.
1701 */
1707 XFS_ITRUNC_MAX_EXTENTS,
1711 /*
1712 * If the bunmapi call encounters an error,
1713 * return to the caller where the transaction
1714 * can be properly aborted. We just need to
1715 * make sure we're not holding any resources
1716 * that we were not when we came in.
1717 */
1720 }
1722 /*
1723 * Duplicate the transaction that has the permanent
1724 * reservation and commit the old transaction.
1725 */
1729 /*
1730 * If the bmap finish call encounters an error,
1731 * return to the caller where the transaction
1732 * can be properly aborted. We just need to
1733 * make sure we're not holding any resources
1734 * that we were not when we came in.
1735 *
1736 * Aborting from this point might lose some
1737 * blocks in the file system, but oh well.
1738 */
1741 /*
1742 * If the passed in transaction committed
1743 * in xfs_bmap_finish(), then we want to
1744 * add the inode to this one before returning.
1745 * This keeps things simple for the higher
1746 * level code, because it always knows that
1747 * the inode is locked and held in the
1748 * transaction that returns to it whether
1749 * errors occur or not. We don't mark the
1750 * inode dirty so that this transaction can
1751 * be easily aborted if possible.
1752 */
1756 }
1758 }
1761 /*
1762 * The first xact was committed,
1763 * so add the inode to the new one.
1764 * Mark it dirty so it will be logged
1765 * and moved forward in the log as
1766 * part of every commit.
1767 */
1772 }
1777 XFS_TRANS_PERM_LOG_RES,
1778 XFS_ITRUNCATE_LOG_COUNT);
1779 /*
1780 * Add the inode being truncated to the next chained
1781 * transaction.
1782 */
1787 }
1788 /*
1789 * Only update the size in the case of the data fork, but
1790 * always re-log the inode so that our permanent transaction
1791 * can keep on rolling it forward in the log.
1792 */
1795 /*
1796 * If we are not changing the file size then do
1797 * not update the on-disk file size - we may be
1798 * called from xfs_inactive_free_eofblocks(). If we
1799 * update the on-disk file size and then the system
1800 * crashes before the contents of the file are
1801 * flushed to disk then the files may be full of
1802 * holes (ie NULL files bug).
1803 */
1807 }
1808 }
1818 }
1821 /*
1822 * xfs_igrow_start
1823 *
1824 * Do the first part of growing a file: zero any data in the last
1825 * block that is beyond the old EOF. We need to do this before
1826 * the inode is joined to the transaction to modify the i_size.
1827 * That way we can drop the inode lock and call into the buffer
1828 * cache to get the buffer mapping the EOF.
1829 */
1830 int
1833 xfs_fsize_t new_size,
1835 {
1842 /*
1843 * Zero any pages that may have been created by
1844 * xfs_write_file() beyond the end of the file
1845 * and any blocks between the old and new file sizes.
1846 */
1850 }
1852 /*
1853 * xfs_igrow_finish
1854 *
1855 * This routine is called to extend the size of a file.
1856 * The inode must have both the iolock and the ilock locked
1857 * for update and it must be a part of the current transaction.
1858 * The xfs_igrow_start() function must have been called previously.
1859 * If the change_flag is not zero, the inode change timestamp will
1860 * be updated.
1861 */
1862 void
1866 xfs_fsize_t new_size,
1868 {
1874 /*
1875 * Update the file size. Update the inode change timestamp
1876 * if change_flag set.
1877 */
1884 }
1887 /*
1888 * This is called when the inode's link count goes to 0.
1889 * We place the on-disk inode on a list in the AGI. It
1890 * will be pulled from this list when the inode is freed.
1891 */
1892 int
1896 {
1902 xfs_agnumber_t agno;
1903 xfs_daddr_t agdaddr;
1904 xfs_agino_t agino;
1919 /*
1920 * Get the agi buffer first. It ensures lock ordering
1921 * on the list.
1922 */
1927 }
1928 /*
1929 * Validate the magic number of the agi block.
1930 */
1932 agi_ok =
1936 XFS_RANDOM_IUNLINK))) {
1940 }
1941 /*
1942 * Get the index into the agi hash table for the
1943 * list this inode will go on.
1944 */
1952 /*
1953 * There is already another inode in the bucket we need
1954 * to add ourselves to. Add us at the front of the list.
1955 * Here we put the head pointer into our next pointer,
1956 * and then we fall through to point the head at us.
1957 */
1961 }
1964 /* both on-disk, don't endian flip twice */
1972 }
1974 /*
1975 * Point the bucket head pointer at the inode being inserted.
1976 */
1984 }
1986 /*
1987 * Pull the on-disk inode from the AGI unlinked list.
1988 */
1989 STATIC int
1993 {
1994 xfs_ino_t next_ino;
2000 xfs_agnumber_t agno;
2001 xfs_daddr_t agdaddr;
2002 xfs_agino_t agino;
2003 xfs_agino_t next_agino;
2011 /*
2012 * First pull the on-disk inode from the AGI unlinked list.
2013 */
2019 /*
2020 * Get the agi buffer first. It ensures lock ordering
2021 * on the list.
2022 */
2030 }
2031 /*
2032 * Validate the magic number of the agi block.
2033 */
2035 agi_ok =
2039 XFS_RANDOM_IUNLINK_REMOVE))) {
2047 }
2048 /*
2049 * Get the index into the agi hash table for the
2050 * list this inode will go on.
2051 */
2059 /*
2060 * We're at the head of the list. Get the inode's
2061 * on-disk buffer to see if there is anyone after us
2062 * on the list. Only modify our next pointer if it
2063 * is not already NULLAGINO. This saves us the overhead
2064 * of dealing with the buffer when there is no need to
2065 * change it.
2066 */
2073 }
2086 }
2087 /*
2088 * Point the bucket head pointer at the next inode.
2089 */
2098 /*
2099 * We need to search the list for the inode being freed.
2100 */
2104 /*
2105 * If the last inode wasn't the one pointing to
2106 * us, then release its buffer since we're not
2107 * going to do anything with it.
2108 */
2111 }
2120 }
2124 }
2125 /*
2126 * Now last_ibp points to the buffer previous to us on
2127 * the unlinked list. Pull us from the list.
2128 */
2135 }
2149 }
2150 /*
2151 * Point the previous inode on the list to the next inode.
2152 */
2160 }
2162 }
2165 {
2169 }
2171 STATIC void
2175 xfs_ino_t inum)
2176 {
2182 xfs_daddr_t blkno;
2199 }
2208 /*
2209 * Look for each inode in memory and attempt to lock it,
2210 * we can be racing with flush and tail pushing here.
2211 * any inode we get the locks on, add to an array of
2212 * inode items to process later.
2213 *
2214 * The get the buffer lock, we could beat a flush
2215 * or tail pushing thread to the lock here, in which
2216 * case they will go looking for the inode buffer
2217 * and fail, we need some other form of interlock
2218 * here.
2219 */
2227 }
2229 /* Inode not in memory or we found it already,
2230 * nothing to do
2231 */
2235 }
2240 }
2242 /* If we can get the locks then add it to the
2243 * list, otherwise by the time we get the bp lock
2244 * below it will already be attached to the
2245 * inode buffer.
2246 */
2248 /* This inode will already be locked - by us, lets
2249 * keep it that way.
2250 */
2259 }
2260 }
2263 }
2274 }
2277 }
2278 }
2281 }
2285 XFS_BUF_LOCK);
2298 pre_flushed++;
2299 }
2301 }
2312 }
2326 }
2327 }
2332 }
2335 }
2337 /*
2338 * This is called to return an inode to the inode free list.
2339 * The inode should already be truncated to 0 length and have
2340 * no pages associated with it. This routine also assumes that
2341 * the inode is already a part of the transaction.
2342 *
2343 * The on-disk copy of the inode will have been added to the list
2344 * of unlinked inodes in the AGI. We need to remove the inode from
2345 * that list atomically with respect to freeing it here.
2346 */
2347 int
2352 {
2355 xfs_ino_t first_ino;
2366 /*
2367 * Pull the on-disk inode from the AGI unlinked list.
2368 */
2372 }
2377 }
2386 /*
2387 * Bump the generation count so no one will be confused
2388 * by reincarnations of this inode.
2389 */
2395 }
2398 }
2400 /*
2401 * Reallocate the space for if_broot based on the number of records
2402 * being added or deleted as indicated in rec_diff. Move the records
2403 * and pointers in if_broot to fit the new size. When shrinking this
2404 * will eliminate holes between the records and pointers created by
2405 * the caller. When growing this will create holes to be filled in
2406 * by the caller.
2407 *
2408 * The caller must not request to add more records than would fit in
2409 * the on-disk inode root. If the if_broot is currently NULL, then
2410 * if we adding records one will be allocated. The caller must also
2411 * not request that the number of records go below zero, although
2412 * it can go to zero.
2413 *
2414 * ip -- the inode whose if_broot area is changing
2415 * ext_diff -- the change in the number of records, positive or negative,
2416 * requested for the if_broot array.
2417 */
2418 void
2423 {
2432 /*
2433 * Handle the degenerate case quietly.
2434 */
2437 }
2441 /*
2442 * If there wasn't any memory allocated before, just
2443 * allocate it now and get out.
2444 */
2448 KM_SLEEP);
2451 }
2453 /*
2454 * If there is already an existing if_broot, then we need
2455 * to realloc() it and shift the pointers to their new
2456 * location. The records don't change location because
2457 * they are kept butted up against the btree block header.
2458 */
2464 new_size,
2466 KM_SLEEP);
2476 }
2478 /*
2479 * rec_diff is less than 0. In this case, we are shrinking the
2480 * if_broot buffer. It must already exist. If we go to zero
2481 * records, just get rid of the root and clear the status bit.
2482 */
2489 else
2493 /*
2494 * First copy over the btree block header.
2495 */
2500 }
2502 /*
2503 * Only copy the records and pointers if there are any.
2504 */
2506 /*
2507 * First copy the records.
2508 */
2515 /*
2516 * Then copy the pointers.
2517 */
2523 }
2530 }
2533 /*
2534 * This is called when the amount of space needed for if_data
2535 * is increased or decreased. The change in size is indicated by
2536 * the number of bytes that need to be added or deleted in the
2537 * byte_diff parameter.
2538 *
2539 * If the amount of space needed has decreased below the size of the
2540 * inline buffer, then switch to using the inline buffer. Otherwise,
2541 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2542 * to what is needed.
2543 *
2544 * ip -- the inode whose if_data area is changing
2545 * byte_diff -- the change in the number of bytes, positive or negative,
2546 * requested for the if_data array.
2547 */
2548 void
2553 {
2560 }
2569 }
2573 /*
2574 * If the valid extents/data can fit in if_inline_ext/data,
2575 * copy them from the malloc'd vector and free it.
2576 */
2582 new_size);
2585 }
2588 /*
2589 * Stuck with malloc/realloc.
2590 * For inline data, the underlying buffer must be
2591 * a multiple of 4 bytes in size so that it can be
2592 * logged and stay on word boundaries. We enforce
2593 * that here.
2594 */
2600 /*
2601 * Only do the realloc if the underlying size
2602 * is really changing.
2603 */
2607 real_size,
2609 KM_SLEEP);
2610 }
2616 }
2617 }
2621 }
2626 /*
2627 * Map inode to disk block and offset.
2628 *
2629 * mp -- the mount point structure for the current file system
2630 * tp -- the current transaction
2631 * ino -- the inode number of the inode to be located
2632 * imap -- this structure is filled in with the information necessary
2633 * to retrieve the given inode from disk
2634 * flags -- flags to pass to xfs_dilocate indicating whether or not
2635 * lookups in the inode btree were OK or not
2636 */
2637 int
2641 xfs_ino_t ino,
2643 uint flags)
2644 {
2645 xfs_fsblock_t fsbno;
2655 }
2662 }
2664 void
2668 {
2675 }
2677 /*
2678 * If the format is local, then we can't have an extents
2679 * array so just look for an inline data array. If we're
2680 * not local then we may or may not have an extents list,
2681 * so check and free it up if we do.
2682 */
2690 }
2697 }
2704 }
2705 }
2707 /*
2708 * This is called free all the memory associated with an inode.
2709 * It must free the inode itself and any buffers allocated for
2710 * if_extents/if_data and if_broot. It must also free the lock
2711 * associated with the inode.
2712 */
2713 void
2716 {
2724 }
2730 #ifdef XFS_BMAP_TRACE
2732 #endif
2733 #ifdef XFS_BMBT_TRACE
2735 #endif
2736 #ifdef XFS_RW_TRACE
2738 #endif
2739 #ifdef XFS_ILOCK_TRACE
2741 #endif
2742 #ifdef XFS_DIR2_TRACE
2744 #endif
2746 /*
2747 * Only if we are shutting down the fs will we see an
2748 * inode still in the AIL. If it is there, we should remove
2749 * it to prevent a use-after-free from occurring.
2750 */
2761 else
2763 }
2765 }
2767 }
2770 /*
2771 * Increment the pin count of the given buffer.
2772 * This value is protected by ipinlock spinlock in the mount structure.
2773 */
2774 void
2777 {
2781 }
2783 /*
2784 * Decrement the pin count of the given inode, and wake up
2785 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2786 * inode must have been previously pinned with a call to xfs_ipin().
2787 */
2788 void
2791 {
2796 /*
2797 * If the inode is currently being reclaimed, the link between
2798 * the bhv_vnode and the xfs_inode will be broken after the
2799 * XFS_IRECLAIM* flag is set. Hence, if these flags are not
2800 * set, then we can move forward and mark the linux inode dirty
2801 * knowing that it is still valid as it won't freed until after
2802 * the bhv_vnode<->xfs_inode link is broken in xfs_reclaim. The
2803 * i_flags_lock is used to synchronise the setting of the
2804 * XFS_IRECLAIM* flags and the breaking of the link, and so we
2805 * can execute atomically w.r.t to reclaim by holding this lock
2806 * here.
2807 *
2808 * However, we still need to issue the unpin wakeup call as the
2809 * inode reclaim may be blocked waiting for the inode to become
2810 * unpinned.
2811 */
2821 /* make sync come back and flush this inode */
2824 }
2827 }
2828 }
2830 /*
2831 * This is called to wait for the given inode to be unpinned.
2832 * It will sleep until this happens. The caller must have the
2833 * inode locked in at least shared mode so that the buffer cannot
2834 * be subsequently pinned once someone is waiting for it to be
2835 * unpinned.
2836 */
2837 STATIC void
2840 {
2842 xfs_lsn_t lsn;
2848 }
2855 }
2857 /*
2858 * Give the log a push so we don't wait here too long.
2859 */
2863 }
2866 /*
2867 * xfs_iextents_copy()
2868 *
2869 * This is called to copy the REAL extents (as opposed to the delayed
2870 * allocation extents) from the inode into the given buffer. It
2871 * returns the number of bytes copied into the buffer.
2872 *
2873 * If there are no delayed allocation extents, then we can just
2874 * memcpy() the extents into the buffer. Otherwise, we need to
2875 * examine each extent in turn and skip those which are delayed.
2876 */
2877 int
2882 {
2887 xfs_fsblock_t start_block;
2897 /*
2898 * There are some delayed allocation extents in the
2899 * inode, so copy the extents one at a time and skip
2900 * the delayed ones. There must be at least one
2901 * non-delayed extent.
2902 */
2908 /*
2909 * It's a delayed allocation extent, so skip it.
2910 */
2912 }
2914 /* Translate to on disk format */
2917 dp++;
2918 copied++;
2919 }
2924 }
2926 /*
2927 * Each of the following cases stores data into the same region
2928 * of the on-disk inode, so only one of them can be valid at
2929 * any given time. While it is possible to have conflicting formats
2930 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2931 * in EXTENTS format, this can only happen when the fork has
2932 * changed formats after being modified but before being flushed.
2933 * In these cases, the format always takes precedence, because the
2934 * format indicates the current state of the fork.
2935 */
2936 /*ARGSUSED*/
2937 STATIC int
2944 {
2948 #ifdef XFS_TRANS_DEBUG
2950 #endif
2961 /*
2962 * This can happen if we gave up in iformat in an error path,
2963 * for the attribute fork.
2964 */
2968 }
2978 }
2992 whichfork);
2993 }
3002 XFS_BROOT_SIZE_ADJ));
3006 }
3013 }
3021 }
3027 }
3030 }
3032 /*
3033 * xfs_iflush() will write a modified inode's changes out to the
3034 * inode's on disk home. The caller must have the inode lock held
3035 * in at least shared mode and the inode flush semaphore must be
3036 * held as well. The inode lock will still be held upon return from
3037 * the call and the caller is free to unlock it.
3038 * The inode flush lock will be unlocked when the inode reaches the disk.
3039 * The flags indicate how the inode's buffer should be written out.
3040 */
3041 int
3044 uint flags)
3045 {
3051 /* REFERENCED */
3069 /*
3070 * If the inode isn't dirty, then just release the inode
3071 * flush lock and do nothing.
3072 */
3079 }
3081 /*
3082 * We can't flush the inode until it is unpinned, so
3083 * wait for it. We know noone new can pin it, because
3084 * we are holding the inode lock shared and you need
3085 * to hold it exclusively to pin the inode.
3086 */
3089 /*
3090 * This may have been unpinned because the filesystem is shutting
3091 * down forcibly. If that's the case we must not write this inode
3092 * to disk, because the log record didn't make it to disk!
3093 */
3100 }
3102 /*
3103 * Get the buffer containing the on-disk inode.
3104 */
3109 }
3111 /*
3112 * Decide how buffer will be flushed out. This is done before
3113 * the call to xfs_iflush_int because this field is zeroed by it.
3114 */
3116 /*
3117 * Flush out the inode buffer according to the directions
3118 * of the caller. In the cases where the caller has given
3119 * us a choice choose the non-delwri case. This is because
3120 * the inode is in the AIL and we need to get it out soon.
3121 */
3138 }
3156 }
3157 }
3159 /*
3160 * First flush out the inode that xfs_iflush was called with.
3161 */
3165 }
3167 /*
3168 * inode clustering:
3169 * see if other inodes can be gathered into this write
3170 */
3179 /*
3180 * Do an un-protected check to see if the inode is dirty and
3181 * is a candidate for flushing. These checks will be repeated
3182 * later after the appropriate locks are acquired.
3183 */
3190 }
3192 /*
3193 * Try to get locks. If any are unavailable,
3194 * then this inode cannot be flushed and is skipped.
3195 */
3197 /* get inode locks (just i_lock) */
3199 /* get inode flush lock */
3201 /* check if pinned */
3203 /* arriving here means that
3204 * this inode can be flushed.
3205 * first re-check that it's
3206 * dirty
3207 */
3212 clcount++;
3216 XFS_ILOCK_SHARED);
3218 }
3221 }
3224 }
3225 }
3227 }
3228 }
3234 }
3236 /*
3237 * If the buffer is pinned then push on the log so we won't
3238 * get stuck waiting in the write for too long.
3239 */
3242 }
3250 }
3253 corrupt_out:
3257 /*
3258 * Unlocks the flush lock
3259 */
3262 cluster_corrupt_out:
3263 /* Corruption detected in the clustering loop. Invalidate the
3264 * inode buffer and shut down the filesystem.
3265 */
3268 /*
3269 * Clean up the buffer. If it was B_DELWRI, just release it --
3270 * brelse can handle it with no problems. If not, shut down the
3271 * filesystem before releasing the buffer.
3272 */
3275 }
3280 /*
3281 * Just like incore_relse: if we have b_iodone functions,
3282 * mark the buffer as an error and call them. Otherwise
3283 * mark it as stale and brelse.
3284 */
3295 }
3296 }
3299 /*
3300 * Unlocks the flush lock
3301 */
3303 }
3306 STATIC int
3310 {
3314 #ifdef XFS_TRANS_DEBUG
3316 #endif
3328 /*
3329 * If the inode isn't dirty, then just release the inode
3330 * flush lock and do nothing.
3331 */
3336 }
3338 /* set *dip = inode's place in the buffer */
3341 /*
3342 * Clear i_update_core before copying out the data.
3343 * This is for coordination with our timestamp updates
3344 * that don't hold the inode lock. They will always
3345 * update the timestamps BEFORE setting i_update_core,
3346 * so if we clear i_update_core after they set it we
3347 * are guaranteed to see their updates to the timestamps.
3348 * I believe that this depends on strongly ordered memory
3349 * semantics, but we have that. We use the SYNCHRONIZE
3350 * macro to make sure that the compiler does not reorder
3351 * the i_update_core access below the data copy below.
3352 */
3356 /*
3357 * Make sure to get the latest atime from the Linux inode.
3358 */
3367 }
3374 }
3384 }
3395 }
3396 }
3399 XFS_RANDOM_IFLUSH_5)) {
3405 ip);
3407 }
3414 }
3415 /*
3416 * bump the flush iteration count, used to detect flushes which
3417 * postdate a log record during recovery.
3418 */
3422 /*
3423 * Copy the dirty parts of the inode into the on-disk
3424 * inode. We always copy out the core of the inode,
3425 * because if the inode is dirty at all the core must
3426 * be.
3427 */
3430 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3434 /*
3435 * If this is really an old format inode and the superblock version
3436 * has not been updated to support only new format inodes, then
3437 * convert back to the old inode format. If the superblock version
3438 * has been updated, then make the conversion permanent.
3439 */
3444 /*
3445 * Convert it back.
3446 */
3450 /*
3451 * The superblock version has already been bumped,
3452 * so just make the conversion to the new inode
3453 * format permanent.
3454 */
3463 }
3464 }
3468 }
3471 /*
3472 * The only error from xfs_iflush_fork is on the data fork.
3473 */
3475 }
3478 /*
3479 * We've recorded everything logged in the inode, so we'd
3480 * like to clear the ilf_fields bits so we don't log and
3481 * flush things unnecessarily. However, we can't stop
3482 * logging all this information until the data we've copied
3483 * into the disk buffer is written to disk. If we did we might
3484 * overwrite the copy of the inode in the log with all the
3485 * data after re-logging only part of it, and in the face of
3486 * a crash we wouldn't have all the data we need to recover.
3487 *
3488 * What we do is move the bits to the ili_last_fields field.
3489 * When logging the inode, these bits are moved back to the
3490 * ilf_fields field. In the xfs_iflush_done() routine we
3491 * clear ili_last_fields, since we know that the information
3492 * those bits represent is permanently on disk. As long as
3493 * the flush completes before the inode is logged again, then
3494 * both ilf_fields and ili_last_fields will be cleared.
3495 *
3496 * We can play with the ilf_fields bits here, because the inode
3497 * lock must be held exclusively in order to set bits there
3498 * and the flush lock protects the ili_last_fields bits.
3499 * Set ili_logged so the flush done
3500 * routine can tell whether or not to look in the AIL.
3501 * Also, store the current LSN of the inode so that we can tell
3502 * whether the item has moved in the AIL from xfs_iflush_done().
3503 * In order to read the lsn we need the AIL lock, because
3504 * it is a 64 bit value that cannot be read atomically.
3505 */
3516 /*
3517 * Attach the function xfs_iflush_done to the inode's
3518 * buffer. This will remove the inode from the AIL
3519 * and unlock the inode's flush lock when the inode is
3520 * completely written to disk.
3521 */
3528 /*
3529 * We're flushing an inode which is not in the AIL and has
3530 * not been logged but has i_update_core set. For this
3531 * case we can use a B_DELWRI flush and immediately drop
3532 * the inode flush lock because we can avoid the whole
3533 * AIL state thing. It's OK to drop the flush lock now,
3534 * because we've already locked the buffer and to do anything
3535 * you really need both.
3536 */
3541 }
3543 }
3547 corrupt_out:
3549 }
3552 /*
3553 * Flush all inactive inodes in mp.
3554 */
3555 void
3558 {
3562 again:
3569 /* Make sure we skip markers inserted by sync */
3573 }
3580 }
3586 out:
3588 }
3590 /*
3591 * xfs_iaccess: check accessibility of inode for mode.
3592 */
3593 int
3596 mode_t mode,
3598 {
3612 }
3614 /*
3615 * If there's an Access Control List it's used instead of
3616 * the mode bits.
3617 */
3625 }
3627 /*
3628 * If the DACs are ok we don't need any capability check.
3629 */
3632 /*
3633 * Read/write DACs are always overridable.
3634 * Executable DACs are overridable if at least one exec bit is set.
3635 */
3645 #ifdef NOISE
3649 }
3651 }
3653 /*
3654 * xfs_iroundup: round up argument to next power of two
3655 */
3656 uint
3658 uint v)
3659 {
3661 uint m;
3674 }
3677 }
3679 #ifdef XFS_ILOCK_TRACE
3682 void
3684 {
3693 }
3694 #endif
3696 /*
3697 * Return a pointer to the extent record at file index idx.
3698 */
3699 xfs_bmbt_rec_host_t *
3703 {
3718 }
3719 }
3721 /*
3722 * Insert new item(s) into the extent records for incore inode
3723 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3724 */
3725 void
3731 {
3738 }
3740 /*
3741 * This is called when the amount of space required for incore file
3742 * extents needs to be increased. The ext_diff parameter stores the
3743 * number of new extents being added and the idx parameter contains
3744 * the extent index where the new extents will be added. If the new
3745 * extents are being appended, then we just need to (re)allocate and
3746 * initialize the space. Otherwise, if the new extents are being
3747 * inserted into the middle of the existing entries, a bit more work
3748 * is required to make room for the new extents to be inserted. The
3749 * caller is responsible for filling in the new extent entries upon
3750 * return.
3751 */
3752 void
3757 {
3766 /*
3767 * If the new number of extents (nextents + ext_diff)
3768 * fits inside the inode, then continue to use the inline
3769 * extent buffer.
3770 */
3777 }
3781 }
3782 /*
3783 * Otherwise use a linear (direct) extent list.
3784 * If the extents are currently inside the inode,
3785 * xfs_iext_realloc_direct will switch us from
3786 * inline to direct extent allocation mode.
3787 */
3795 }
3796 }
3797 /* Indirection array */
3810 }
3811 /* Extents fit in target extent page */
3819 }
3822 }
3823 /* Insert a new extent page */
3827 }
3828 /*
3829 * If extent(s) are being appended to the last page in
3830 * the indirection array and the new extent(s) don't fit
3831 * in the page, then erp is NULL and erp_idx is set to
3832 * the next index needed in the indirection array.
3833 */
3842 erp_idx++;
3843 }
3844 }
3845 }
3846 }
3848 }
3850 /*
3851 * This is called when incore extents are being added to the indirection
3852 * array and the new extents do not fit in the target extent list. The
3853 * erp_idx parameter contains the irec index for the target extent list
3854 * in the indirection array, and the idx parameter contains the extent
3855 * index within the list. The number of extents being added is stored
3856 * in the count parameter.
3857 *
3858 * |-------| |-------|
3859 * | | | | idx - number of extents before idx
3860 * | idx | | count |
3861 * | | | | count - number of extents being inserted at idx
3862 * |-------| |-------|
3863 * | count | | nex2 | nex2 - number of extents after idx + count
3864 * |-------| |-------|
3865 */
3866 void
3872 {
3886 /*
3887 * Save second part of target extent list
3888 * (all extents past */
3896 }
3898 /*
3899 * Add the new extents to the end of the target
3900 * list, then allocate new irec record(s) and
3901 * extent buffer(s) as needed to store the rest
3902 * of the new extents.
3903 */
3910 }
3912 erp_idx++;
3918 }
3920 /* Add nex2 extents back to indirection array */
3922 xfs_extnum_t ext_avail;
3928 /*
3929 * If nex2 extents fit in the current page, append
3930 * nex2_ep after the new extents.
3931 */
3934 }
3935 /*
3936 * Otherwise, check if space is available in the
3937 * next page.
3938 */
3942 erp_idx++;
3943 erp++;
3944 /* Create a hole for nex2 extents */
3947 }
3948 /*
3949 * Final choice, create a new extent page for
3950 * nex2 extents.
3951 */
3953 erp_idx++;
3955 }
3960 }
3961 }
3963 /*
3964 * This is called when the amount of space required for incore file
3965 * extents needs to be decreased. The ext_diff parameter stores the
3966 * number of extents to be removed and the idx parameter contains
3967 * the extent index where the extents will be removed from.
3968 *
3969 * If the amount of space needed has decreased below the linear
3970 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3971 * extent array. Otherwise, use kmem_realloc() to adjust the
3972 * size to what is needed.
3973 */
3974 void
3979 {
3995 }
3997 }
3999 /*
4000 * This removes ext_diff extents from the inline buffer, beginning
4001 * at extent index idx.
4002 */
4003 void
4008 {
4027 }
4028 }
4030 /*
4031 * This removes ext_diff extents from a linear (direct) extent list,
4032 * beginning at extent index idx. If the extents are being removed
4033 * from the end of the list (ie. truncate) then we just need to re-
4034 * allocate the list to remove the extra space. Otherwise, if the
4035 * extents are being removed from the middle of the existing extent
4036 * entries, then we first need to move the extent records beginning
4037 * at idx + ext_diff up in the list to overwrite the records being
4038 * removed, then remove the extra space via kmem_realloc.
4039 */
4040 void
4045 {
4057 }
4058 /* Move extents up in the list (if needed) */
4064 }
4067 /*
4068 * Reallocate the direct extent list. If the extents
4069 * will fit inside the inode then xfs_iext_realloc_direct
4070 * will switch from direct to inline extent allocation
4071 * mode for us.
4072 */
4075 }
4077 /*
4078 * This is called when incore extents are being removed from the
4079 * indirection array and the extents being removed span multiple extent
4080 * buffers. The idx parameter contains the file extent index where we
4081 * want to begin removing extents, and the count parameter contains
4082 * how many extents need to be removed.
4083 *
4084 * |-------| |-------|
4085 * | nex1 | | | nex1 - number of extents before idx
4086 * |-------| | count |
4087 * | | | | count - number of extents being removed at idx
4088 * | count | |-------|
4089 * | | | nex2 | nex2 - number of extents after idx + count
4090 * |-------| |-------|
4091 */
4092 void
4097 {
4116 /*
4117 * Check for deletion of entire list;
4118 * xfs_iext_irec_remove() updates extent offsets.
4119 */
4126 XFS_IEXT_BUFSZ);
4132 }
4133 }
4134 /* Move extents up (if needed) */
4139 }
4140 /* Zero out rest of page */
4143 /* Update remaining counters */
4148 erp_idx++;
4149 erp++;
4150 }
4153 }
4155 /*
4156 * Create, destroy, or resize a linear (direct) block of extents.
4157 */
4158 void
4162 {
4171 /* Free extent records */
4174 }
4175 /* Resize direct extent list and zero any new bytes */
4177 /* Check if extents will fit inside the inode */
4183 }
4186 }
4190 rnew_size,
4192 KM_SLEEP);
4193 }
4198 }
4199 }
4200 /*
4201 * Switch from the inline extent buffer to a direct
4202 * extent list. Be sure to include the inline extent
4203 * bytes in new_size.
4204 */
4209 }
4211 }
4214 }
4216 /*
4217 * Switch from linear (direct) extent records to inline buffer.
4218 */
4219 void
4223 {
4226 /*
4227 * The inline buffer was zeroed when we switched
4228 * from inline to direct extent allocation mode,
4229 * so we don't need to clear it here.
4230 */
4236 }
4238 /*
4239 * Switch from inline buffer to linear (direct) extent records.
4240 * new_size should already be rounded up to the next power of 2
4241 * by the caller (when appropriate), so use new_size as it is.
4242 * However, since new_size may be rounded up, we can't update
4243 * if_bytes here. It is the caller's responsibility to update
4244 * if_bytes upon return.
4245 */
4246 void
4250 {
4258 }
4260 }
4262 /*
4263 * Resize an extent indirection array to new_size bytes.
4264 */
4265 void
4269 {
4284 }
4285 }
4287 /*
4288 * Switch from indirection array to linear (direct) extent allocations.
4289 */
4290 void
4293 {
4313 }
4314 }
4316 /*
4317 * Free incore file extents.
4318 */
4319 void
4322 {
4330 }
4337 }
4341 }
4343 /*
4344 * Return a pointer to the extent record for file system block bno.
4345 */
4351 {
4366 }
4369 /* Find target extent list */
4377 }
4378 /* Binary search extent records */
4389 /* Convert back to file-based extent index */
4392 }
4395 }
4396 }
4397 /* Convert back to file-based extent index */
4400 }
4406 }
4407 }
4410 }
4412 /*
4413 * Return a pointer to the indirection array entry containing the
4414 * extent record for filesystem block bno. Store the index of the
4415 * target irec in *erp_idxp.
4416 */
4422 {
4446 }
4447 }
4450 }
4452 /*
4453 * Return a pointer to the indirection array entry containing the
4454 * extent record at file extent index *idxp. Store the index of the
4455 * target irec in *erp_idxp and store the page index of the target
4456 * extent record in *idxp.
4457 */
4458 xfs_ext_irec_t *
4464 {
4481 /* Binary search extent irec's */
4497 erp_idx++;
4503 }
4504 }
4508 }
4510 /*
4511 * Allocate and initialize an indirection array once the space needed
4512 * for incore extents increases above XFS_IEXT_BUFSZ.
4513 */
4514 void
4517 {
4534 }
4545 }
4547 /*
4548 * Allocate and initialize a new entry in the indirection array.
4549 */
4550 xfs_ext_irec_t *
4554 {
4562 /* Resize indirection array */
4565 /*
4566 * Move records down in the array so the
4567 * new page can use erp_idx.
4568 */
4572 }
4575 /* Initialize new extent record */
4584 }
4586 /*
4587 * Remove a record from the indirection array.
4588 */
4589 void
4593 {
4605 }
4606 /* Compact extent records */
4610 }
4611 /*
4612 * Manually free the last extent record from the indirection
4613 * array. A call to xfs_iext_realloc_indirect() with a size
4614 * of zero would result in a call to xfs_iext_destroy() which
4615 * would in turn call this function again, creating a nasty
4616 * infinite loop.
4617 */
4624 }
4626 }
4628 /*
4629 * This is called to clean up large amounts of unused memory allocated
4630 * by the indirection array. Before compacting anything though, verify
4631 * that the indirection array is still needed and switch back to the
4632 * linear extent list (or even the inline buffer) if possible. The
4633 * compaction policy is as follows:
4634 *
4635 * Full Compaction: Extents fit into a single page (or inline buffer)
4636 * Full Compaction: Extents occupy less than 10% of allocated space
4637 * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
4638 * No Compaction: Extents occupy at least 50% of allocated space
4639 */
4640 void
4643 {
4662 }
4663 }
4665 /*
4666 * Combine extents from neighboring extent pages.
4667 */
4668 void
4671 {
4687 /*
4688 * Free page before removing extent record
4689 * so er_extoffs don't get modified in
4690 * xfs_iext_irec_remove.
4691 */
4697 erp_idx++;
4698 }
4699 }
4700 }
4702 /*
4703 * Fully compact the extent records managed by the indirection array.
4704 */
4705 void
4708 {
4728 /* Remove next page */
4730 /*
4731 * Free page before removing extent record
4732 * so er_extoffs don't get modified in
4733 * xfs_iext_irec_remove.
4734 */
4741 /* Update next page */
4743 /* Move rest of page up to become next new page */
4750 }
4752 erp_idx++;
4755 else
4757 }
4761 }
4762 }
4764 /*
4765 * This is called to update the er_extoff field in the indirection
4766 * array when extents have been added or removed from one of the
4767 * extent lists. erp_idx contains the irec index to begin updating
4768 * at and ext_diff contains the number of extents that were added
4769 * or removed.
4770 */
4771 void
4776 {
4784 }
4785 }