| blob | cdc4c28926d0cf62d972e807323f9ce3d92de4c0 |
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 */
53 #include <linux/log2.h>
59 /*
60 * Used in xfs_itruncate(). This is the maximum number of extents
61 * freed from a file in a single transaction.
62 */
63 #define XFS_ITRUNC_MAX_EXTENTS 2
71 #ifdef DEBUG
72 /*
73 * Make sure that the extents in the given memory buffer
74 * are valid.
75 */
76 STATIC void
81 xfs_exntfmt_t fmt)
82 {
84 xfs_bmbt_irec_t irec;
85 xfs_bmbt_rec_t rec;
94 else
98 }
99 }
101 #define xfs_validate_extents(ifp, nrecs, disk, fmt)
104 /*
105 * Check that none of the inode's in the buffer have a next
106 * unlinked field of 0.
107 */
108 #if defined(DEBUG)
109 void
113 {
125 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
126 bp);
128 }
129 }
130 }
131 #endif
133 /*
134 * This routine is called to map an inode number within a file
135 * system to the buffer containing the on-disk version of the
136 * inode. It returns a pointer to the buffer containing the
137 * on-disk inode in the bpp parameter, and in the dip parameter
138 * it returns a pointer to the on-disk inode within that buffer.
139 *
140 * If a non-zero error is returned, then the contents of bpp and
141 * dipp are undefined.
142 *
143 * Use xfs_imap() to determine the size and location of the
144 * buffer to read from disk.
145 */
146 STATIC int
150 xfs_ino_t ino,
154 {
156 xfs_imap_t imap;
161 /*
162 * Call the space management code to find the location of the
163 * inode on disk.
164 */
169 "xfs_inotobp: xfs_imap() returned an "
172 }
174 /*
175 * If the inode number maps to a block outside the bounds of the
176 * file system then return NULL rather than calling read_buf
177 * and panicing when we get an error from the driver.
178 */
182 "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
187 }
189 /*
190 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
191 * default to just a read_buf() call.
192 */
198 "xfs_inotobp: xfs_trans_read_buf() returned an "
201 }
203 di_ok =
207 XFS_RANDOM_ITOBP_INOTOBP))) {
211 "xfs_inotobp: XFS_TEST_ERROR() returned an "
214 }
218 /*
219 * Set *dipp to point to the on-disk inode in the buffer.
220 */
225 }
228 /*
229 * This routine is called to map an inode to the buffer containing
230 * the on-disk version of the inode. It returns a pointer to the
231 * buffer containing the on-disk inode in the bpp parameter, and in
232 * the dip parameter it returns a pointer to the on-disk inode within
233 * that buffer.
234 *
235 * If a non-zero error is returned, then the contents of bpp and
236 * dipp are undefined.
237 *
238 * If the inode is new and has not yet been initialized, use xfs_imap()
239 * to determine the size and location of the buffer to read from disk.
240 * If the inode has already been mapped to its buffer and read in once,
241 * then use the mapping information stored in the inode rather than
242 * calling xfs_imap(). This allows us to avoid the overhead of looking
243 * at the inode btree for small block file systems (see xfs_dilocate()).
244 * We can tell whether the inode has been mapped in before by comparing
245 * its disk block address to 0. Only uninitialized inodes will have
246 * 0 for the disk block address.
247 */
248 int
255 xfs_daddr_t bno,
256 uint imap_flags)
257 {
258 xfs_imap_t imap;
265 /*
266 * Call the space management code to find the location of the
267 * inode on disk.
268 */
274 /*
275 * If the inode number maps to a block outside the bounds
276 * of the file system then return NULL rather than calling
277 * read_buf and panicing when we get an error from the
278 * driver.
279 */
282 #ifdef DEBUG
284 "(imap.im_blkno (0x%llx) "
285 "+ imap.im_len (0x%llx)) > "
286 " XFS_FSB_TO_BB(mp, "
293 }
295 /*
296 * Fill in the fields in the inode that will be used to
297 * map the inode to its buffer from now on.
298 */
303 /*
304 * We've already mapped the inode once, so just use the
305 * mapping that we saved the first time.
306 */
310 }
313 /*
314 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
315 * default to just a read_buf() call.
316 */
320 #ifdef DEBUG
322 "xfs_trans_read_buf() returned error %d, "
328 }
330 /*
331 * Validate the magic number and version of every inode in the buffer
332 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
333 * No validation is done here in userspace (xfs_repair).
334 */
335 #if !defined(__KERNEL__)
337 #elif defined(DEBUG)
341 #endif
352 XFS_ERRTAG_ITOBP_INOTOBP,
353 XFS_RANDOM_ITOBP_INOTOBP))) {
357 }
358 #ifdef DEBUG
360 "Device %s - bad inode magic/vsn "
365 #endif
370 }
371 }
375 /*
376 * Mark the buffer as an inode buffer now that it looks good
377 */
380 /*
381 * Set *dipp to point to the on-disk inode in the buffer.
382 */
386 }
388 /*
389 * Move inode type and inode format specific information from the
390 * on-disk inode to the in-core inode. For fifos, devs, and sockets
391 * this means set if_rdev to the proper value. For files, directories,
392 * and symlinks this means to bring in the in-line data or extent
393 * pointers. For a file in B-tree format, only the root is immediately
394 * brought in-core. The rest will be in-lined in if_extents when it
395 * is first referenced (see xfs_iread_extents()).
396 */
397 STATIC int
401 {
405 xfs_fsize_t di_size;
424 }
426 if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) {
434 }
445 }
456 /*
457 * no local regular files yet
458 */
461 "corrupt inode %Lu "
465 XFS_ERRLEVEL_LOW,
468 }
473 "corrupt inode %Lu "
478 XFS_ERRLEVEL_LOW,
481 }
496 }
502 }
505 }
527 }
532 }
534 }
536 /*
537 * The file is in-lined in the on-disk inode.
538 * If it fits into if_inline_data, then copy
539 * it there, otherwise allocate a buffer for it
540 * and copy the data there. Either way, set
541 * if_data to point at the data.
542 * If we allocate a buffer for the data, make
543 * sure that its size is a multiple of 4 and
544 * record the real size in i_real_bytes.
545 */
546 STATIC int
552 {
556 /*
557 * If the size is unreasonable, then something
558 * is wrong and we just bail out rather than crash in
559 * kmem_alloc() or memcpy() below.
560 */
563 "corrupt inode %Lu "
570 }
580 }
588 }
590 /*
591 * The file consists of a set of extents all
592 * of which fit into the on-disk inode.
593 * If there are few enough extents to fit into
594 * the if_inline_ext, then copy them there.
595 * Otherwise allocate a buffer for them and copy
596 * them into it. Either way, set if_extents
597 * to point at the extents.
598 */
599 STATIC int
604 {
615 /*
616 * If the number of extents is unreasonable, then something
617 * is wrong and we just bail out rather than crash in
618 * kmem_alloc() or memcpy() below.
619 */
627 }
634 else
644 ARCH_CONVERT);
646 ARCH_CONVERT);
647 }
654 XFS_ERRLEVEL_LOW,
657 }
658 }
661 }
663 /*
664 * The file has too many extents to fit into
665 * the inode, so they are in B-tree format.
666 * Allocate a buffer for the root of the B-tree
667 * and copy the root into it. The i_extents
668 * field will remain NULL until all of the
669 * extents are read in (when they are needed).
670 */
671 STATIC int
676 {
679 /* REFERENCED */
688 /*
689 * blow out if -- fork has less extents than can fit in
690 * fork (fork shouldn't be a btree format), root btree
691 * block has more records than can fit into the fork,
692 * or the number of extents is greater than the number of
693 * blocks.
694 */
705 }
710 /*
711 * Copy and convert from the on-disk structure
712 * to the in-memory structure.
713 */
720 }
722 /*
723 * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
724 * and native format
725 *
726 * buf = on-disk representation
727 * dip = native representation
728 * dir = direction - +ve -> disk to native
729 * -ve -> native to disk
730 */
731 void
733 xfs_caddr_t buf,
736 {
759 }
786 }
788 STATIC uint
790 __uint16_t di_flags)
791 {
823 }
826 }
828 uint
831 {
836 }
838 uint
841 {
844 }
846 /*
847 * Given a mount structure and an inode number, return a pointer
848 * to a newly allocated in-core inode corresponding to the given
849 * inode number.
850 *
851 * Initialize the inode's attributes and extent pointers if it
852 * already has them (it will not if the inode has no links).
853 */
854 int
858 xfs_ino_t ino,
860 xfs_daddr_t bno,
861 uint imap_flags)
862 {
875 /*
876 * Get pointer's to the on-disk inode and the buffer containing it.
877 * If the inode number refers to a block outside the file system
878 * then xfs_itobp() will return NULL. In this case we should
879 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
880 * know that this is a new incore inode.
881 */
886 }
888 /*
889 * Initialize inode's trace buffers.
890 * Do this before xfs_iformat in case it adds entries.
891 */
892 #ifdef XFS_BMAP_TRACE
894 #endif
895 #ifdef XFS_BMBT_TRACE
897 #endif
898 #ifdef XFS_RW_TRACE
900 #endif
901 #ifdef XFS_ILOCK_TRACE
903 #endif
904 #ifdef XFS_DIR2_TRACE
906 #endif
908 /*
909 * If we got something that isn't an inode it means someone
910 * (nfs or dmi) has a stale handle.
911 */
915 #ifdef DEBUG
917 "dip->di_core.di_magic (0x%x) != "
920 XFS_DINODE_MAGIC);
923 }
925 /*
926 * If the on-disk inode is already linked to a directory
927 * entry, copy all of the inode into the in-core inode.
928 * xfs_iformat() handles copying in the inode format
929 * specific information.
930 * Otherwise, just get the truly permanent information.
931 */
939 #ifdef DEBUG
942 error);
945 }
951 /*
952 * Make sure to pull in the mode here as well in
953 * case the inode is released without being used.
954 * This ensures that xfs_inactive() will see that
955 * the inode is already free and not try to mess
956 * with the uninitialized part of it.
957 */
959 /*
960 * Initialize the per-fork minima and maxima for a new
961 * inode here. xfs_iformat will do it for old inodes.
962 */
965 }
969 /*
970 * The inode format changed when we moved the link count and
971 * made it 32 bits long. If this is an old format inode,
972 * convert it in memory to look like a new one. If it gets
973 * flushed to disk we will convert back before flushing or
974 * logging it. We zero out the new projid field and the old link
975 * count field. We'll handle clearing the pad field (the remains
976 * of the old uuid field) when we actually convert the inode to
977 * the new format. We don't change the version number so that we
978 * can distinguish this from a real new format inode.
979 */
984 }
989 /*
990 * Mark the buffer containing the inode as something to keep
991 * around for a while. This helps to keep recently accessed
992 * meta-data in-core longer.
993 */
996 /*
997 * Use xfs_trans_brelse() to release the buffer containing the
998 * on-disk inode, because it was acquired with xfs_trans_read_buf()
999 * in xfs_itobp() above. If tp is NULL, this is just a normal
1000 * brelse(). If we're within a transaction, then xfs_trans_brelse()
1001 * will only release the buffer if it is not dirty within the
1002 * transaction. It will be OK to release the buffer in this case,
1003 * because inodes on disk are never destroyed and we will be
1004 * locking the new in-core inode before putting it in the hash
1005 * table where other processes can find it. Thus we don't have
1006 * to worry about the inode being changed just because we released
1007 * the buffer.
1008 */
1012 }
1014 /*
1015 * Read in extents from a btree-format inode.
1016 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1017 */
1018 int
1023 {
1026 xfs_extnum_t nextents;
1033 }
1038 /*
1039 * We know that the size is valid (it's checked in iformat_btree)
1040 */
1050 }
1053 }
1055 /*
1056 * Allocate an inode on disk and return a copy of its in-core version.
1057 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1058 * appropriately within the inode. The uid and gid for the inode are
1059 * set according to the contents of the given cred structure.
1060 *
1061 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1062 * has a free inode available, call xfs_iget()
1063 * to obtain the in-core version of the allocated inode. Finally,
1064 * fill in the inode and log its initial contents. In this case,
1065 * ialloc_context would be set to NULL and call_again set to false.
1066 *
1067 * If xfs_dialloc() does not have an available inode,
1068 * it will replenish its supply by doing an allocation. Since we can
1069 * only do one allocation within a transaction without deadlocks, we
1070 * must commit the current transaction before returning the inode itself.
1071 * In this case, therefore, we will set call_again to true and return.
1072 * The caller should then commit the current transaction, start a new
1073 * transaction, and call xfs_ialloc() again to actually get the inode.
1074 *
1075 * To ensure that some other process does not grab the inode that
1076 * was allocated during the first call to xfs_ialloc(), this routine
1077 * also returns the [locked] bp pointing to the head of the freelist
1078 * as ialloc_context. The caller should hold this buffer across
1079 * the commit and pass it back into this routine on the second call.
1080 *
1081 * If we are allocating quota inodes, we do not have a parent inode
1082 * to attach to or associate with (i.e. pip == NULL) because they
1083 * are not linked into the directory structure - they are attached
1084 * directly to the superblock - and so have no parent.
1085 */
1086 int
1090 mode_t mode,
1091 xfs_nlink_t nlink,
1092 xfs_dev_t rdev,
1094 xfs_prid_t prid,
1099 {
1100 xfs_ino_t ino;
1103 uint flags;
1106 /*
1107 * Call the space management code to pick
1108 * the on-disk inode to be allocated.
1109 */
1114 }
1118 }
1121 /*
1122 * Get the in-core inode with the lock held exclusively.
1123 * This is because we're setting fields here we need
1124 * to prevent others from looking at until we're done.
1125 */
1130 }
1143 /*
1144 * If the superblock version is up to where we support new format
1145 * inodes and this is currently an old format inode, then change
1146 * the inode version number now. This way we only do the conversion
1147 * here rather than here and in the flush/logging code.
1148 */
1152 /*
1153 * We've already zeroed the old link count, the projid field,
1154 * and the pad field.
1155 */
1156 }
1158 /*
1159 * Project ids won't be stored on disk if we are using a version 1 inode.
1160 */
1168 }
1169 }
1171 /*
1172 * If the group ID of the new file does not match the effective group
1173 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1174 * (and only if the irix_sgid_inherit compatibility variable is set).
1175 */
1180 }
1187 /*
1188 * di_gen will have been taken care of in xfs_iread.
1189 */
1212 }
1213 /* fall through */
1224 }
1229 }
1233 }
1234 }
1236 xfs_inherit_noatime)
1239 xfs_inherit_nodump)
1242 xfs_inherit_sync)
1245 xfs_inherit_nosymlinks)
1250 xfs_inherit_nodefrag)
1255 }
1256 /* FALLTHROUGH */
1265 }
1266 /*
1267 * Attribute fork settings for new inode.
1268 */
1272 /*
1273 * Log the new values stuffed into the inode.
1274 */
1277 /* now that we have an i_mode we can setup inode ops and unlock */
1282 }
1284 /*
1285 * Check to make sure that there are no blocks allocated to the
1286 * file beyond the size of the file. We don't check this for
1287 * files with fixed size extents or real time extents, but we
1288 * at least do it for regular files.
1289 */
1290 #ifdef DEBUG
1291 void
1295 xfs_fsize_t isize)
1296 {
1297 xfs_fileoff_t map_first;
1309 /*
1310 * The filesystem could be shutting down, so bmapi may return
1311 * an error.
1312 */
1316 map_first),
1322 }
1325 /*
1326 * Calculate the last possible buffered byte in a file. This must
1327 * include data that was buffered beyond the EOF by the write code.
1328 * This also needs to deal with overflowing the xfs_fsize_t type
1329 * which can happen for sizes near the limit.
1330 *
1331 * We also need to take into account any blocks beyond the EOF. It
1332 * may be the case that they were buffered by a write which failed.
1333 * In that case the pages will still be in memory, but the inode size
1334 * will never have been updated.
1335 */
1336 xfs_fsize_t
1339 {
1341 xfs_fsize_t last_byte;
1342 xfs_fileoff_t last_block;
1343 xfs_fileoff_t size_last_block;
1349 /*
1350 * Only check for blocks beyond the EOF if the extents have
1351 * been read in. This eliminates the need for the inode lock,
1352 * and it also saves us from looking when it really isn't
1353 * necessary.
1354 */
1357 XFS_DATA_FORK);
1360 }
1363 }
1370 }
1374 }
1376 }
1378 #if defined(XFS_RW_TRACE)
1379 STATIC void
1384 xfs_fsize_t new_size,
1385 xfs_off_t toss_start,
1386 xfs_off_t toss_finish)
1387 {
1390 }
1409 }
1410 #else
1411 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1412 #endif
1414 /*
1415 * Start the truncation of the file to new_size. The new size
1416 * must be smaller than the current size. This routine will
1417 * clear the buffer and page caches of file data in the removed
1418 * range, and xfs_itruncate_finish() will remove the underlying
1419 * disk blocks.
1420 *
1421 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1422 * must NOT have the inode lock held at all. This is because we're
1423 * calling into the buffer/page cache code and we can't hold the
1424 * inode lock when we do so.
1425 *
1426 * We need to wait for any direct I/Os in flight to complete before we
1427 * proceed with the truncate. This is needed to prevent the extents
1428 * being read or written by the direct I/Os from being removed while the
1429 * I/O is in flight as there is no other method of synchronising
1430 * direct I/O with the truncate operation. Also, because we hold
1431 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1432 * started until the truncate completes and drops the lock. Essentially,
1433 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1434 * between direct I/Os and the truncate operation.
1435 *
1436 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1437 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1438 * in the case that the caller is locking things out of order and
1439 * may not be able to call xfs_itruncate_finish() with the inode lock
1440 * held without dropping the I/O lock. If the caller must drop the
1441 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1442 * must be called again with all the same restrictions as the initial
1443 * call.
1444 */
1445 int
1448 uint flags,
1449 xfs_fsize_t new_size)
1450 {
1451 xfs_fsize_t last_byte;
1452 xfs_off_t toss_start;
1467 /*
1468 * Call toss_pages or flushinval_pages to get rid of pages
1469 * overlapping the region being removed. We have to use
1470 * the less efficient flushinval_pages in the case that the
1471 * caller may not be able to finish the truncate without
1472 * dropping the inode's I/O lock. Make sure
1473 * to catch any pages brought in by buffers overlapping
1474 * the EOF by searching out beyond the isize by our
1475 * block size. We round new_size up to a block boundary
1476 * so that we don't toss things on the same block as
1477 * new_size but before it.
1478 *
1479 * Before calling toss_page or flushinval_pages, make sure to
1480 * call remapf() over the same region if the file is mapped.
1481 * This frees up mapped file references to the pages in the
1482 * given range and for the flushinval_pages case it ensures
1483 * that we get the latest mapped changes flushed out.
1484 */
1488 /*
1489 * The place to start tossing is beyond our maximum
1490 * file size, so there is no way that the data extended
1491 * out there.
1492 */
1494 }
1497 last_byte);
1503 }
1504 }
1506 #ifdef DEBUG
1509 }
1510 #endif
1512 }
1514 /*
1515 * Shrink the file to the given new_size. The new
1516 * size must be smaller than the current size.
1517 * This will free up the underlying blocks
1518 * in the removed range after a call to xfs_itruncate_start()
1519 * or xfs_atruncate_start().
1520 *
1521 * The transaction passed to this routine must have made
1522 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1523 * This routine may commit the given transaction and
1524 * start new ones, so make sure everything involved in
1525 * the transaction is tidy before calling here.
1526 * Some transaction will be returned to the caller to be
1527 * committed. The incoming transaction must already include
1528 * the inode, and both inode locks must be held exclusively.
1529 * The inode must also be "held" within the transaction. On
1530 * return the inode will be "held" within the returned transaction.
1531 * This routine does NOT require any disk space to be reserved
1532 * for it within the transaction.
1533 *
1534 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1535 * and it indicates the fork which is to be truncated. For the
1536 * attribute fork we only support truncation to size 0.
1537 *
1538 * We use the sync parameter to indicate whether or not the first
1539 * transaction we perform might have to be synchronous. For the attr fork,
1540 * it needs to be so if the unlink of the inode is not yet known to be
1541 * permanent in the log. This keeps us from freeing and reusing the
1542 * blocks of the attribute fork before the unlink of the inode becomes
1543 * permanent.
1544 *
1545 * For the data fork, we normally have to run synchronously if we're
1546 * being called out of the inactive path or we're being called
1547 * out of the create path where we're truncating an existing file.
1548 * Either way, the truncate needs to be sync so blocks don't reappear
1549 * in the file with altered data in case of a crash. wsync filesystems
1550 * can run the first case async because anything that shrinks the inode
1551 * has to run sync so by the time we're called here from inactive, the
1552 * inode size is permanently set to 0.
1553 *
1554 * Calls from the truncate path always need to be sync unless we're
1555 * in a wsync filesystem and the file has already been unlinked.
1556 *
1557 * The caller is responsible for correctly setting the sync parameter.
1558 * It gets too hard for us to guess here which path we're being called
1559 * out of just based on inode state.
1560 */
1561 int
1565 xfs_fsize_t new_size,
1568 {
1569 xfs_fsblock_t first_block;
1570 xfs_fileoff_t first_unmap_block;
1571 xfs_fileoff_t last_block;
1577 xfs_bmap_free_t free_list;
1594 /*
1595 * We only support truncating the entire attribute fork.
1596 */
1599 }
1602 /*
1603 * The first thing we do is set the size to new_size permanently
1604 * on disk. This way we don't have to worry about anyone ever
1605 * being able to look at the data being freed even in the face
1606 * of a crash. What we're getting around here is the case where
1607 * we free a block, it is allocated to another file, it is written
1608 * to, and then we crash. If the new data gets written to the
1609 * file but the log buffers containing the free and reallocation
1610 * don't, then we'd end up with garbage in the blocks being freed.
1611 * As long as we make the new_size permanent before actually
1612 * freeing any blocks it doesn't matter if they get writtten to.
1613 *
1614 * The callers must signal into us whether or not the size
1615 * setting here must be synchronous. There are a few cases
1616 * where it doesn't have to be synchronous. Those cases
1617 * occur if the file is unlinked and we know the unlink is
1618 * permanent or if the blocks being truncated are guaranteed
1619 * to be beyond the inode eof (regardless of the link count)
1620 * and the eof value is permanent. Both of these cases occur
1621 * only on wsync-mounted filesystems. In those cases, we're
1622 * guaranteed that no user will ever see the data in the blocks
1623 * that are being truncated so the truncate can run async.
1624 * In the free beyond eof case, the file may wind up with
1625 * more blocks allocated to it than it needs if we crash
1626 * and that won't get fixed until the next time the file
1627 * is re-opened and closed but that's ok as that shouldn't
1628 * be too many blocks.
1629 *
1630 * However, we can't just make all wsync xactions run async
1631 * because there's one call out of the create path that needs
1632 * to run sync where it's truncating an existing file to size
1633 * 0 whose size is > 0.
1634 *
1635 * It's probably possible to come up with a test in this
1636 * routine that would correctly distinguish all the above
1637 * cases from the values of the function parameters and the
1638 * inode state but for sanity's sake, I've decided to let the
1639 * layers above just tell us. It's simpler to correctly figure
1640 * out in the layer above exactly under what conditions we
1641 * can run async and I think it's easier for others read and
1642 * follow the logic in case something has to be changed.
1643 * cscope is your friend -- rcc.
1644 *
1645 * The attribute fork is much simpler.
1646 *
1647 * For the attribute fork we allow the caller to tell us whether
1648 * the unlink of the inode that led to this call is yet permanent
1649 * in the on disk log. If it is not and we will be freeing extents
1650 * in this inode then we make the first transaction synchronous
1651 * to make sure that the unlink is permanent by the time we free
1652 * the blocks.
1653 */
1656 /*
1657 * If we are not changing the file size then do
1658 * not update the on-disk file size - we may be
1659 * called from xfs_inactive_free_eofblocks(). If we
1660 * update the on-disk file size and then the system
1661 * crashes before the contents of the file are
1662 * flushed to disk then the files may be full of
1663 * holes (ie NULL files bug).
1664 */
1669 }
1670 }
1675 }
1681 /*
1682 * Since it is possible for space to become allocated beyond
1683 * the end of the file (in a crash where the space is allocated
1684 * but the inode size is not yet updated), simply remove any
1685 * blocks which show up between the new EOF and the maximum
1686 * possible file size. If the first block to be removed is
1687 * beyond the maximum file size (ie it is the same as last_block),
1688 * then there is nothing to do.
1689 */
1697 }
1699 /*
1700 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1701 * will tell us whether it freed the entire range or
1702 * not. If this is a synchronous mount (wsync),
1703 * then we can tell bunmapi to keep all the
1704 * transactions asynchronous since the unlink
1705 * transaction that made this inode inactive has
1706 * already hit the disk. There's no danger of
1707 * the freed blocks being reused, there being a
1708 * crash, and the reused blocks suddenly reappearing
1709 * in this file with garbage in them once recovery
1710 * runs.
1711 */
1717 XFS_ITRUNC_MAX_EXTENTS,
1721 /*
1722 * If the bunmapi call encounters an error,
1723 * return to the caller where the transaction
1724 * can be properly aborted. We just need to
1725 * make sure we're not holding any resources
1726 * that we were not when we came in.
1727 */
1730 }
1732 /*
1733 * Duplicate the transaction that has the permanent
1734 * reservation and commit the old transaction.
1735 */
1739 /*
1740 * If the bmap finish call encounters an error,
1741 * return to the caller where the transaction
1742 * can be properly aborted. We just need to
1743 * make sure we're not holding any resources
1744 * that we were not when we came in.
1745 *
1746 * Aborting from this point might lose some
1747 * blocks in the file system, but oh well.
1748 */
1751 /*
1752 * If the passed in transaction committed
1753 * in xfs_bmap_finish(), then we want to
1754 * add the inode to this one before returning.
1755 * This keeps things simple for the higher
1756 * level code, because it always knows that
1757 * the inode is locked and held in the
1758 * transaction that returns to it whether
1759 * errors occur or not. We don't mark the
1760 * inode dirty so that this transaction can
1761 * be easily aborted if possible.
1762 */
1766 }
1768 }
1771 /*
1772 * The first xact was committed,
1773 * so add the inode to the new one.
1774 * Mark it dirty so it will be logged
1775 * and moved forward in the log as
1776 * part of every commit.
1777 */
1782 }
1787 XFS_TRANS_PERM_LOG_RES,
1788 XFS_ITRUNCATE_LOG_COUNT);
1789 /*
1790 * Add the inode being truncated to the next chained
1791 * transaction.
1792 */
1797 }
1798 /*
1799 * Only update the size in the case of the data fork, but
1800 * always re-log the inode so that our permanent transaction
1801 * can keep on rolling it forward in the log.
1802 */
1805 /*
1806 * If we are not changing the file size then do
1807 * not update the on-disk file size - we may be
1808 * called from xfs_inactive_free_eofblocks(). If we
1809 * update the on-disk file size and then the system
1810 * crashes before the contents of the file are
1811 * flushed to disk then the files may be full of
1812 * holes (ie NULL files bug).
1813 */
1817 }
1818 }
1828 }
1831 /*
1832 * xfs_igrow_start
1833 *
1834 * Do the first part of growing a file: zero any data in the last
1835 * block that is beyond the old EOF. We need to do this before
1836 * the inode is joined to the transaction to modify the i_size.
1837 * That way we can drop the inode lock and call into the buffer
1838 * cache to get the buffer mapping the EOF.
1839 */
1840 int
1843 xfs_fsize_t new_size,
1845 {
1852 /*
1853 * Zero any pages that may have been created by
1854 * xfs_write_file() beyond the end of the file
1855 * and any blocks between the old and new file sizes.
1856 */
1860 }
1862 /*
1863 * xfs_igrow_finish
1864 *
1865 * This routine is called to extend the size of a file.
1866 * The inode must have both the iolock and the ilock locked
1867 * for update and it must be a part of the current transaction.
1868 * The xfs_igrow_start() function must have been called previously.
1869 * If the change_flag is not zero, the inode change timestamp will
1870 * be updated.
1871 */
1872 void
1876 xfs_fsize_t new_size,
1878 {
1884 /*
1885 * Update the file size. Update the inode change timestamp
1886 * if change_flag set.
1887 */
1894 }
1897 /*
1898 * This is called when the inode's link count goes to 0.
1899 * We place the on-disk inode on a list in the AGI. It
1900 * will be pulled from this list when the inode is freed.
1901 */
1902 int
1906 {
1912 xfs_agnumber_t agno;
1913 xfs_daddr_t agdaddr;
1914 xfs_agino_t agino;
1929 /*
1930 * Get the agi buffer first. It ensures lock ordering
1931 * on the list.
1932 */
1937 }
1938 /*
1939 * Validate the magic number of the agi block.
1940 */
1942 agi_ok =
1946 XFS_RANDOM_IUNLINK))) {
1950 }
1951 /*
1952 * Get the index into the agi hash table for the
1953 * list this inode will go on.
1954 */
1962 /*
1963 * There is already another inode in the bucket we need
1964 * to add ourselves to. Add us at the front of the list.
1965 * Here we put the head pointer into our next pointer,
1966 * and then we fall through to point the head at us.
1967 */
1971 }
1974 /* both on-disk, don't endian flip twice */
1982 }
1984 /*
1985 * Point the bucket head pointer at the inode being inserted.
1986 */
1994 }
1996 /*
1997 * Pull the on-disk inode from the AGI unlinked list.
1998 */
1999 STATIC int
2003 {
2004 xfs_ino_t next_ino;
2010 xfs_agnumber_t agno;
2011 xfs_daddr_t agdaddr;
2012 xfs_agino_t agino;
2013 xfs_agino_t next_agino;
2021 /*
2022 * First pull the on-disk inode from the AGI unlinked list.
2023 */
2029 /*
2030 * Get the agi buffer first. It ensures lock ordering
2031 * on the list.
2032 */
2040 }
2041 /*
2042 * Validate the magic number of the agi block.
2043 */
2045 agi_ok =
2049 XFS_RANDOM_IUNLINK_REMOVE))) {
2057 }
2058 /*
2059 * Get the index into the agi hash table for the
2060 * list this inode will go on.
2061 */
2069 /*
2070 * We're at the head of the list. Get the inode's
2071 * on-disk buffer to see if there is anyone after us
2072 * on the list. Only modify our next pointer if it
2073 * is not already NULLAGINO. This saves us the overhead
2074 * of dealing with the buffer when there is no need to
2075 * change it.
2076 */
2083 }
2096 }
2097 /*
2098 * Point the bucket head pointer at the next inode.
2099 */
2108 /*
2109 * We need to search the list for the inode being freed.
2110 */
2114 /*
2115 * If the last inode wasn't the one pointing to
2116 * us, then release its buffer since we're not
2117 * going to do anything with it.
2118 */
2121 }
2130 }
2134 }
2135 /*
2136 * Now last_ibp points to the buffer previous to us on
2137 * the unlinked list. Pull us from the list.
2138 */
2145 }
2159 }
2160 /*
2161 * Point the previous inode on the list to the next inode.
2162 */
2170 }
2172 }
2175 {
2179 }
2181 STATIC void
2185 xfs_ino_t inum)
2186 {
2192 xfs_daddr_t blkno;
2209 }
2218 /*
2219 * Look for each inode in memory and attempt to lock it,
2220 * we can be racing with flush and tail pushing here.
2221 * any inode we get the locks on, add to an array of
2222 * inode items to process later.
2223 *
2224 * The get the buffer lock, we could beat a flush
2225 * or tail pushing thread to the lock here, in which
2226 * case they will go looking for the inode buffer
2227 * and fail, we need some other form of interlock
2228 * here.
2229 */
2237 }
2239 /* Inode not in memory or we found it already,
2240 * nothing to do
2241 */
2245 }
2250 }
2252 /* If we can get the locks then add it to the
2253 * list, otherwise by the time we get the bp lock
2254 * below it will already be attached to the
2255 * inode buffer.
2256 */
2258 /* This inode will already be locked - by us, lets
2259 * keep it that way.
2260 */
2269 }
2270 }
2273 }
2284 }
2287 }
2288 }
2291 }
2295 XFS_BUF_LOCK);
2308 pre_flushed++;
2309 }
2311 }
2322 }
2336 }
2337 }
2342 }
2345 }
2347 /*
2348 * This is called to return an inode to the inode free list.
2349 * The inode should already be truncated to 0 length and have
2350 * no pages associated with it. This routine also assumes that
2351 * the inode is already a part of the transaction.
2352 *
2353 * The on-disk copy of the inode will have been added to the list
2354 * of unlinked inodes in the AGI. We need to remove the inode from
2355 * that list atomically with respect to freeing it here.
2356 */
2357 int
2362 {
2365 xfs_ino_t first_ino;
2376 /*
2377 * Pull the on-disk inode from the AGI unlinked list.
2378 */
2382 }
2387 }
2396 /*
2397 * Bump the generation count so no one will be confused
2398 * by reincarnations of this inode.
2399 */
2405 }
2408 }
2410 /*
2411 * Reallocate the space for if_broot based on the number of records
2412 * being added or deleted as indicated in rec_diff. Move the records
2413 * and pointers in if_broot to fit the new size. When shrinking this
2414 * will eliminate holes between the records and pointers created by
2415 * the caller. When growing this will create holes to be filled in
2416 * by the caller.
2417 *
2418 * The caller must not request to add more records than would fit in
2419 * the on-disk inode root. If the if_broot is currently NULL, then
2420 * if we adding records one will be allocated. The caller must also
2421 * not request that the number of records go below zero, although
2422 * it can go to zero.
2423 *
2424 * ip -- the inode whose if_broot area is changing
2425 * ext_diff -- the change in the number of records, positive or negative,
2426 * requested for the if_broot array.
2427 */
2428 void
2433 {
2442 /*
2443 * Handle the degenerate case quietly.
2444 */
2447 }
2451 /*
2452 * If there wasn't any memory allocated before, just
2453 * allocate it now and get out.
2454 */
2458 KM_SLEEP);
2461 }
2463 /*
2464 * If there is already an existing if_broot, then we need
2465 * to realloc() it and shift the pointers to their new
2466 * location. The records don't change location because
2467 * they are kept butted up against the btree block header.
2468 */
2474 new_size,
2476 KM_SLEEP);
2486 }
2488 /*
2489 * rec_diff is less than 0. In this case, we are shrinking the
2490 * if_broot buffer. It must already exist. If we go to zero
2491 * records, just get rid of the root and clear the status bit.
2492 */
2499 else
2503 /*
2504 * First copy over the btree block header.
2505 */
2510 }
2512 /*
2513 * Only copy the records and pointers if there are any.
2514 */
2516 /*
2517 * First copy the records.
2518 */
2525 /*
2526 * Then copy the pointers.
2527 */
2533 }
2540 }
2543 /*
2544 * This is called when the amount of space needed for if_data
2545 * is increased or decreased. The change in size is indicated by
2546 * the number of bytes that need to be added or deleted in the
2547 * byte_diff parameter.
2548 *
2549 * If the amount of space needed has decreased below the size of the
2550 * inline buffer, then switch to using the inline buffer. Otherwise,
2551 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2552 * to what is needed.
2553 *
2554 * ip -- the inode whose if_data area is changing
2555 * byte_diff -- the change in the number of bytes, positive or negative,
2556 * requested for the if_data array.
2557 */
2558 void
2563 {
2570 }
2579 }
2583 /*
2584 * If the valid extents/data can fit in if_inline_ext/data,
2585 * copy them from the malloc'd vector and free it.
2586 */
2592 new_size);
2595 }
2598 /*
2599 * Stuck with malloc/realloc.
2600 * For inline data, the underlying buffer must be
2601 * a multiple of 4 bytes in size so that it can be
2602 * logged and stay on word boundaries. We enforce
2603 * that here.
2604 */
2610 /*
2611 * Only do the realloc if the underlying size
2612 * is really changing.
2613 */
2617 real_size,
2619 KM_SLEEP);
2620 }
2626 }
2627 }
2631 }
2636 /*
2637 * Map inode to disk block and offset.
2638 *
2639 * mp -- the mount point structure for the current file system
2640 * tp -- the current transaction
2641 * ino -- the inode number of the inode to be located
2642 * imap -- this structure is filled in with the information necessary
2643 * to retrieve the given inode from disk
2644 * flags -- flags to pass to xfs_dilocate indicating whether or not
2645 * lookups in the inode btree were OK or not
2646 */
2647 int
2651 xfs_ino_t ino,
2653 uint flags)
2654 {
2655 xfs_fsblock_t fsbno;
2665 }
2672 }
2674 void
2678 {
2685 }
2687 /*
2688 * If the format is local, then we can't have an extents
2689 * array so just look for an inline data array. If we're
2690 * not local then we may or may not have an extents list,
2691 * so check and free it up if we do.
2692 */
2700 }
2707 }
2714 }
2715 }
2717 /*
2718 * This is called free all the memory associated with an inode.
2719 * It must free the inode itself and any buffers allocated for
2720 * if_extents/if_data and if_broot. It must also free the lock
2721 * associated with the inode.
2722 */
2723 void
2726 {
2734 }
2740 #ifdef XFS_BMAP_TRACE
2742 #endif
2743 #ifdef XFS_BMBT_TRACE
2745 #endif
2746 #ifdef XFS_RW_TRACE
2748 #endif
2749 #ifdef XFS_ILOCK_TRACE
2751 #endif
2752 #ifdef XFS_DIR2_TRACE
2754 #endif
2756 /*
2757 * Only if we are shutting down the fs will we see an
2758 * inode still in the AIL. If it is there, we should remove
2759 * it to prevent a use-after-free from occurring.
2760 */
2771 else
2773 }
2775 }
2777 }
2780 /*
2781 * Increment the pin count of the given buffer.
2782 * This value is protected by ipinlock spinlock in the mount structure.
2783 */
2784 void
2787 {
2791 }
2793 /*
2794 * Decrement the pin count of the given inode, and wake up
2795 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2796 * inode must have been previously pinned with a call to xfs_ipin().
2797 */
2798 void
2801 {
2806 /*
2807 * If the inode is currently being reclaimed, the link between
2808 * the bhv_vnode and the xfs_inode will be broken after the
2809 * XFS_IRECLAIM* flag is set. Hence, if these flags are not
2810 * set, then we can move forward and mark the linux inode dirty
2811 * knowing that it is still valid as it won't freed until after
2812 * the bhv_vnode<->xfs_inode link is broken in xfs_reclaim. The
2813 * i_flags_lock is used to synchronise the setting of the
2814 * XFS_IRECLAIM* flags and the breaking of the link, and so we
2815 * can execute atomically w.r.t to reclaim by holding this lock
2816 * here.
2817 *
2818 * However, we still need to issue the unpin wakeup call as the
2819 * inode reclaim may be blocked waiting for the inode to become
2820 * unpinned.
2821 */
2831 /* make sync come back and flush this inode */
2834 }
2837 }
2838 }
2840 /*
2841 * This is called to wait for the given inode to be unpinned.
2842 * It will sleep until this happens. The caller must have the
2843 * inode locked in at least shared mode so that the buffer cannot
2844 * be subsequently pinned once someone is waiting for it to be
2845 * unpinned.
2846 */
2847 STATIC void
2850 {
2852 xfs_lsn_t lsn;
2858 }
2865 }
2867 /*
2868 * Give the log a push so we don't wait here too long.
2869 */
2873 }
2876 /*
2877 * xfs_iextents_copy()
2878 *
2879 * This is called to copy the REAL extents (as opposed to the delayed
2880 * allocation extents) from the inode into the given buffer. It
2881 * returns the number of bytes copied into the buffer.
2882 *
2883 * If there are no delayed allocation extents, then we can just
2884 * memcpy() the extents into the buffer. Otherwise, we need to
2885 * examine each extent in turn and skip those which are delayed.
2886 */
2887 int
2892 {
2899 xfs_fsblock_t start_block;
2909 /*
2910 * There are some delayed allocation extents in the
2911 * inode, so copy the extents one at a time and skip
2912 * the delayed ones. There must be at least one
2913 * non-delayed extent.
2914 */
2921 /*
2922 * It's a delayed allocation extent, so skip it.
2923 */
2925 }
2927 /* Translate to on disk format */
2932 dest_ep++;
2933 copied++;
2934 }
2939 }
2941 /*
2942 * Each of the following cases stores data into the same region
2943 * of the on-disk inode, so only one of them can be valid at
2944 * any given time. While it is possible to have conflicting formats
2945 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2946 * in EXTENTS format, this can only happen when the fork has
2947 * changed formats after being modified but before being flushed.
2948 * In these cases, the format always takes precedence, because the
2949 * format indicates the current state of the fork.
2950 */
2951 /*ARGSUSED*/
2952 STATIC int
2959 {
2963 #ifdef XFS_TRANS_DEBUG
2965 #endif
2976 /*
2977 * This can happen if we gave up in iformat in an error path,
2978 * for the attribute fork.
2979 */
2983 }
2993 }
3007 whichfork);
3008 }
3017 XFS_BROOT_SIZE_ADJ));
3021 }
3028 }
3036 }
3042 }
3045 }
3047 /*
3048 * xfs_iflush() will write a modified inode's changes out to the
3049 * inode's on disk home. The caller must have the inode lock held
3050 * in at least shared mode and the inode flush semaphore must be
3051 * held as well. The inode lock will still be held upon return from
3052 * the call and the caller is free to unlock it.
3053 * The inode flush lock will be unlocked when the inode reaches the disk.
3054 * The flags indicate how the inode's buffer should be written out.
3055 */
3056 int
3059 uint flags)
3060 {
3066 /* REFERENCED */
3084 /*
3085 * If the inode isn't dirty, then just release the inode
3086 * flush lock and do nothing.
3087 */
3094 }
3096 /*
3097 * We can't flush the inode until it is unpinned, so
3098 * wait for it. We know noone new can pin it, because
3099 * we are holding the inode lock shared and you need
3100 * to hold it exclusively to pin the inode.
3101 */
3104 /*
3105 * This may have been unpinned because the filesystem is shutting
3106 * down forcibly. If that's the case we must not write this inode
3107 * to disk, because the log record didn't make it to disk!
3108 */
3115 }
3117 /*
3118 * Get the buffer containing the on-disk inode.
3119 */
3124 }
3126 /*
3127 * Decide how buffer will be flushed out. This is done before
3128 * the call to xfs_iflush_int because this field is zeroed by it.
3129 */
3131 /*
3132 * Flush out the inode buffer according to the directions
3133 * of the caller. In the cases where the caller has given
3134 * us a choice choose the non-delwri case. This is because
3135 * the inode is in the AIL and we need to get it out soon.
3136 */
3153 }
3171 }
3172 }
3174 /*
3175 * First flush out the inode that xfs_iflush was called with.
3176 */
3180 }
3182 /*
3183 * inode clustering:
3184 * see if other inodes can be gathered into this write
3185 */
3194 /*
3195 * Do an un-protected check to see if the inode is dirty and
3196 * is a candidate for flushing. These checks will be repeated
3197 * later after the appropriate locks are acquired.
3198 */
3205 }
3207 /*
3208 * Try to get locks. If any are unavailable,
3209 * then this inode cannot be flushed and is skipped.
3210 */
3212 /* get inode locks (just i_lock) */
3214 /* get inode flush lock */
3216 /* check if pinned */
3218 /* arriving here means that
3219 * this inode can be flushed.
3220 * first re-check that it's
3221 * dirty
3222 */
3227 clcount++;
3231 XFS_ILOCK_SHARED);
3233 }
3236 }
3239 }
3240 }
3242 }
3243 }
3249 }
3251 /*
3252 * If the buffer is pinned then push on the log so we won't
3253 * get stuck waiting in the write for too long.
3254 */
3257 }
3265 }
3268 corrupt_out:
3272 /*
3273 * Unlocks the flush lock
3274 */
3277 cluster_corrupt_out:
3278 /* Corruption detected in the clustering loop. Invalidate the
3279 * inode buffer and shut down the filesystem.
3280 */
3283 /*
3284 * Clean up the buffer. If it was B_DELWRI, just release it --
3285 * brelse can handle it with no problems. If not, shut down the
3286 * filesystem before releasing the buffer.
3287 */
3290 }
3295 /*
3296 * Just like incore_relse: if we have b_iodone functions,
3297 * mark the buffer as an error and call them. Otherwise
3298 * mark it as stale and brelse.
3299 */
3310 }
3311 }
3314 /*
3315 * Unlocks the flush lock
3316 */
3318 }
3321 STATIC int
3325 {
3329 #ifdef XFS_TRANS_DEBUG
3331 #endif
3343 /*
3344 * If the inode isn't dirty, then just release the inode
3345 * flush lock and do nothing.
3346 */
3351 }
3353 /* set *dip = inode's place in the buffer */
3356 /*
3357 * Clear i_update_core before copying out the data.
3358 * This is for coordination with our timestamp updates
3359 * that don't hold the inode lock. They will always
3360 * update the timestamps BEFORE setting i_update_core,
3361 * so if we clear i_update_core after they set it we
3362 * are guaranteed to see their updates to the timestamps.
3363 * I believe that this depends on strongly ordered memory
3364 * semantics, but we have that. We use the SYNCHRONIZE
3365 * macro to make sure that the compiler does not reorder
3366 * the i_update_core access below the data copy below.
3367 */
3371 /*
3372 * Make sure to get the latest atime from the Linux inode.
3373 */
3382 }
3389 }
3399 }
3410 }
3411 }
3414 XFS_RANDOM_IFLUSH_5)) {
3420 ip);
3422 }
3429 }
3430 /*
3431 * bump the flush iteration count, used to detect flushes which
3432 * postdate a log record during recovery.
3433 */
3437 /*
3438 * Copy the dirty parts of the inode into the on-disk
3439 * inode. We always copy out the core of the inode,
3440 * because if the inode is dirty at all the core must
3441 * be.
3442 */
3445 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3449 /*
3450 * If this is really an old format inode and the superblock version
3451 * has not been updated to support only new format inodes, then
3452 * convert back to the old inode format. If the superblock version
3453 * has been updated, then make the conversion permanent.
3454 */
3459 /*
3460 * Convert it back.
3461 */
3465 /*
3466 * The superblock version has already been bumped,
3467 * so just make the conversion to the new inode
3468 * format permanent.
3469 */
3478 }
3479 }
3483 }
3486 /*
3487 * The only error from xfs_iflush_fork is on the data fork.
3488 */
3490 }
3493 /*
3494 * We've recorded everything logged in the inode, so we'd
3495 * like to clear the ilf_fields bits so we don't log and
3496 * flush things unnecessarily. However, we can't stop
3497 * logging all this information until the data we've copied
3498 * into the disk buffer is written to disk. If we did we might
3499 * overwrite the copy of the inode in the log with all the
3500 * data after re-logging only part of it, and in the face of
3501 * a crash we wouldn't have all the data we need to recover.
3502 *
3503 * What we do is move the bits to the ili_last_fields field.
3504 * When logging the inode, these bits are moved back to the
3505 * ilf_fields field. In the xfs_iflush_done() routine we
3506 * clear ili_last_fields, since we know that the information
3507 * those bits represent is permanently on disk. As long as
3508 * the flush completes before the inode is logged again, then
3509 * both ilf_fields and ili_last_fields will be cleared.
3510 *
3511 * We can play with the ilf_fields bits here, because the inode
3512 * lock must be held exclusively in order to set bits there
3513 * and the flush lock protects the ili_last_fields bits.
3514 * Set ili_logged so the flush done
3515 * routine can tell whether or not to look in the AIL.
3516 * Also, store the current LSN of the inode so that we can tell
3517 * whether the item has moved in the AIL from xfs_iflush_done().
3518 * In order to read the lsn we need the AIL lock, because
3519 * it is a 64 bit value that cannot be read atomically.
3520 */
3531 /*
3532 * Attach the function xfs_iflush_done to the inode's
3533 * buffer. This will remove the inode from the AIL
3534 * and unlock the inode's flush lock when the inode is
3535 * completely written to disk.
3536 */
3543 /*
3544 * We're flushing an inode which is not in the AIL and has
3545 * not been logged but has i_update_core set. For this
3546 * case we can use a B_DELWRI flush and immediately drop
3547 * the inode flush lock because we can avoid the whole
3548 * AIL state thing. It's OK to drop the flush lock now,
3549 * because we've already locked the buffer and to do anything
3550 * you really need both.
3551 */
3556 }
3558 }
3562 corrupt_out:
3564 }
3567 /*
3568 * Flush all inactive inodes in mp.
3569 */
3570 void
3573 {
3577 again:
3584 /* Make sure we skip markers inserted by sync */
3588 }
3595 }
3601 out:
3603 }
3605 /*
3606 * xfs_iaccess: check accessibility of inode for mode.
3607 */
3608 int
3611 mode_t mode,
3613 {
3627 }
3629 /*
3630 * If there's an Access Control List it's used instead of
3631 * the mode bits.
3632 */
3640 }
3642 /*
3643 * If the DACs are ok we don't need any capability check.
3644 */
3647 /*
3648 * Read/write DACs are always overridable.
3649 * Executable DACs are overridable if at least one exec bit is set.
3650 */
3660 #ifdef NOISE
3664 }
3666 }
3668 /*
3669 * xfs_iroundup: round up argument to next power of two
3670 */
3671 uint
3673 uint v)
3674 {
3676 uint m;
3689 }
3692 }
3694 #ifdef XFS_ILOCK_TRACE
3697 void
3699 {
3708 }
3709 #endif
3711 /*
3712 * Return a pointer to the extent record at file index idx.
3713 */
3714 xfs_bmbt_rec_t *
3718 {
3733 }
3734 }
3736 /*
3737 * Insert new item(s) into the extent records for incore inode
3738 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3739 */
3740 void
3746 {
3755 }
3756 }
3758 /*
3759 * This is called when the amount of space required for incore file
3760 * extents needs to be increased. The ext_diff parameter stores the
3761 * number of new extents being added and the idx parameter contains
3762 * the extent index where the new extents will be added. If the new
3763 * extents are being appended, then we just need to (re)allocate and
3764 * initialize the space. Otherwise, if the new extents are being
3765 * inserted into the middle of the existing entries, a bit more work
3766 * is required to make room for the new extents to be inserted. The
3767 * caller is responsible for filling in the new extent entries upon
3768 * return.
3769 */
3770 void
3775 {
3784 /*
3785 * If the new number of extents (nextents + ext_diff)
3786 * fits inside the inode, then continue to use the inline
3787 * extent buffer.
3788 */
3795 }
3799 }
3800 /*
3801 * Otherwise use a linear (direct) extent list.
3802 * If the extents are currently inside the inode,
3803 * xfs_iext_realloc_direct will switch us from
3804 * inline to direct extent allocation mode.
3805 */
3813 }
3814 }
3815 /* Indirection array */
3828 }
3829 /* Extents fit in target extent page */
3837 }
3840 }
3841 /* Insert a new extent page */
3845 }
3846 /*
3847 * If extent(s) are being appended to the last page in
3848 * the indirection array and the new extent(s) don't fit
3849 * in the page, then erp is NULL and erp_idx is set to
3850 * the next index needed in the indirection array.
3851 */
3860 erp_idx++;
3861 }
3862 }
3863 }
3864 }
3866 }
3868 /*
3869 * This is called when incore extents are being added to the indirection
3870 * array and the new extents do not fit in the target extent list. The
3871 * erp_idx parameter contains the irec index for the target extent list
3872 * in the indirection array, and the idx parameter contains the extent
3873 * index within the list. The number of extents being added is stored
3874 * in the count parameter.
3875 *
3876 * |-------| |-------|
3877 * | | | | idx - number of extents before idx
3878 * | idx | | count |
3879 * | | | | count - number of extents being inserted at idx
3880 * |-------| |-------|
3881 * | count | | nex2 | nex2 - number of extents after idx + count
3882 * |-------| |-------|
3883 */
3884 void
3890 {
3904 /*
3905 * Save second part of target extent list
3906 * (all extents past */
3914 }
3916 /*
3917 * Add the new extents to the end of the target
3918 * list, then allocate new irec record(s) and
3919 * extent buffer(s) as needed to store the rest
3920 * of the new extents.
3921 */
3928 }
3930 erp_idx++;
3936 }
3938 /* Add nex2 extents back to indirection array */
3940 xfs_extnum_t ext_avail;
3946 /*
3947 * If nex2 extents fit in the current page, append
3948 * nex2_ep after the new extents.
3949 */
3952 }
3953 /*
3954 * Otherwise, check if space is available in the
3955 * next page.
3956 */
3960 erp_idx++;
3961 erp++;
3962 /* Create a hole for nex2 extents */
3965 }
3966 /*
3967 * Final choice, create a new extent page for
3968 * nex2 extents.
3969 */
3971 erp_idx++;
3973 }
3978 }
3979 }
3981 /*
3982 * This is called when the amount of space required for incore file
3983 * extents needs to be decreased. The ext_diff parameter stores the
3984 * number of extents to be removed and the idx parameter contains
3985 * the extent index where the extents will be removed from.
3986 *
3987 * If the amount of space needed has decreased below the linear
3988 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3989 * extent array. Otherwise, use kmem_realloc() to adjust the
3990 * size to what is needed.
3991 */
3992 void
3997 {
4013 }
4015 }
4017 /*
4018 * This removes ext_diff extents from the inline buffer, beginning
4019 * at extent index idx.
4020 */
4021 void
4026 {
4045 }
4046 }
4048 /*
4049 * This removes ext_diff extents from a linear (direct) extent list,
4050 * beginning at extent index idx. If the extents are being removed
4051 * from the end of the list (ie. truncate) then we just need to re-
4052 * allocate the list to remove the extra space. Otherwise, if the
4053 * extents are being removed from the middle of the existing extent
4054 * entries, then we first need to move the extent records beginning
4055 * at idx + ext_diff up in the list to overwrite the records being
4056 * removed, then remove the extra space via kmem_realloc.
4057 */
4058 void
4063 {
4075 }
4076 /* Move extents up in the list (if needed) */
4082 }
4085 /*
4086 * Reallocate the direct extent list. If the extents
4087 * will fit inside the inode then xfs_iext_realloc_direct
4088 * will switch from direct to inline extent allocation
4089 * mode for us.
4090 */
4093 }
4095 /*
4096 * This is called when incore extents are being removed from the
4097 * indirection array and the extents being removed span multiple extent
4098 * buffers. The idx parameter contains the file extent index where we
4099 * want to begin removing extents, and the count parameter contains
4100 * how many extents need to be removed.
4101 *
4102 * |-------| |-------|
4103 * | nex1 | | | nex1 - number of extents before idx
4104 * |-------| | count |
4105 * | | | | count - number of extents being removed at idx
4106 * | count | |-------|
4107 * | | | nex2 | nex2 - number of extents after idx + count
4108 * |-------| |-------|
4109 */
4110 void
4115 {
4134 /*
4135 * Check for deletion of entire list;
4136 * xfs_iext_irec_remove() updates extent offsets.
4137 */
4144 XFS_IEXT_BUFSZ);
4150 }
4151 }
4152 /* Move extents up (if needed) */
4157 }
4158 /* Zero out rest of page */
4161 /* Update remaining counters */
4166 erp_idx++;
4167 erp++;
4168 }
4171 }
4173 /*
4174 * Create, destroy, or resize a linear (direct) block of extents.
4175 */
4176 void
4180 {
4189 /* Free extent records */
4192 }
4193 /* Resize direct extent list and zero any new bytes */
4195 /* Check if extents will fit inside the inode */
4201 }
4204 }
4208 rnew_size,
4210 KM_SLEEP);
4211 }
4216 }
4217 }
4218 /*
4219 * Switch from the inline extent buffer to a direct
4220 * extent list. Be sure to include the inline extent
4221 * bytes in new_size.
4222 */
4227 }
4229 }
4232 }
4234 /*
4235 * Switch from linear (direct) extent records to inline buffer.
4236 */
4237 void
4241 {
4244 /*
4245 * The inline buffer was zeroed when we switched
4246 * from inline to direct extent allocation mode,
4247 * so we don't need to clear it here.
4248 */
4254 }
4256 /*
4257 * Switch from inline buffer to linear (direct) extent records.
4258 * new_size should already be rounded up to the next power of 2
4259 * by the caller (when appropriate), so use new_size as it is.
4260 * However, since new_size may be rounded up, we can't update
4261 * if_bytes here. It is the caller's responsibility to update
4262 * if_bytes upon return.
4263 */
4264 void
4268 {
4277 }
4279 }
4281 /*
4282 * Resize an extent indirection array to new_size bytes.
4283 */
4284 void
4288 {
4303 }
4304 }
4306 /*
4307 * Switch from indirection array to linear (direct) extent allocations.
4308 */
4309 void
4312 {
4332 }
4333 }
4335 /*
4336 * Free incore file extents.
4337 */
4338 void
4341 {
4349 }
4356 }
4360 }
4362 /*
4363 * Return a pointer to the extent record for file system block bno.
4364 */
4370 {
4385 }
4388 /* Find target extent list */
4396 }
4397 /* Binary search extent records */
4408 /* Convert back to file-based extent index */
4411 }
4414 }
4415 }
4416 /* Convert back to file-based extent index */
4419 }
4425 }
4426 }
4429 }
4431 /*
4432 * Return a pointer to the indirection array entry containing the
4433 * extent record for filesystem block bno. Store the index of the
4434 * target irec in *erp_idxp.
4435 */
4441 {
4465 }
4466 }
4469 }
4471 /*
4472 * Return a pointer to the indirection array entry containing the
4473 * extent record at file extent index *idxp. Store the index of the
4474 * target irec in *erp_idxp and store the page index of the target
4475 * extent record in *idxp.
4476 */
4477 xfs_ext_irec_t *
4483 {
4500 /* Binary search extent irec's */
4516 erp_idx++;
4522 }
4523 }
4527 }
4529 /*
4530 * Allocate and initialize an indirection array once the space needed
4531 * for incore extents increases above XFS_IEXT_BUFSZ.
4532 */
4533 void
4536 {
4554 }
4565 }
4567 /*
4568 * Allocate and initialize a new entry in the indirection array.
4569 */
4570 xfs_ext_irec_t *
4574 {
4582 /* Resize indirection array */
4585 /*
4586 * Move records down in the array so the
4587 * new page can use erp_idx.
4588 */
4592 }
4595 /* Initialize new extent record */
4605 }
4607 /*
4608 * Remove a record from the indirection array.
4609 */
4610 void
4614 {
4626 }
4627 /* Compact extent records */
4631 }
4632 /*
4633 * Manually free the last extent record from the indirection
4634 * array. A call to xfs_iext_realloc_indirect() with a size
4635 * of zero would result in a call to xfs_iext_destroy() which
4636 * would in turn call this function again, creating a nasty
4637 * infinite loop.
4638 */
4645 }
4647 }
4649 /*
4650 * This is called to clean up large amounts of unused memory allocated
4651 * by the indirection array. Before compacting anything though, verify
4652 * that the indirection array is still needed and switch back to the
4653 * linear extent list (or even the inline buffer) if possible. The
4654 * compaction policy is as follows:
4655 *
4656 * Full Compaction: Extents fit into a single page (or inline buffer)
4657 * Full Compaction: Extents occupy less than 10% of allocated space
4658 * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
4659 * No Compaction: Extents occupy at least 50% of allocated space
4660 */
4661 void
4664 {
4683 }
4684 }
4686 /*
4687 * Combine extents from neighboring extent pages.
4688 */
4689 void
4692 {
4708 /*
4709 * Free page before removing extent record
4710 * so er_extoffs don't get modified in
4711 * xfs_iext_irec_remove.
4712 */
4718 erp_idx++;
4719 }
4720 }
4721 }
4723 /*
4724 * Fully compact the extent records managed by the indirection array.
4725 */
4726 void
4729 {
4749 /* Remove next page */
4751 /*
4752 * Free page before removing extent record
4753 * so er_extoffs don't get modified in
4754 * xfs_iext_irec_remove.
4755 */
4762 /* Update next page */
4764 /* Move rest of page up to become next new page */
4771 }
4773 erp_idx++;
4776 else
4778 }
4782 }
4783 }
4785 /*
4786 * This is called to update the er_extoff field in the indirection
4787 * array when extents have been added or removed from one of the
4788 * extent lists. erp_idx contains the irec index to begin updating
4789 * at and ext_diff contains the number of extents that were added
4790 * or removed.
4791 */
4792 void
4797 {
4805 }
4806 }