2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
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.
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.
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
18 #include <linux/log2.h>
22 #include "xfs_types.h"
26 #include "xfs_trans.h"
27 #include "xfs_trans_priv.h"
31 #include "xfs_dmapi.h"
32 #include "xfs_mount.h"
33 #include "xfs_bmap_btree.h"
34 #include "xfs_alloc_btree.h"
35 #include "xfs_ialloc_btree.h"
36 #include "xfs_dir2_sf.h"
37 #include "xfs_attr_sf.h"
38 #include "xfs_dinode.h"
39 #include "xfs_inode.h"
40 #include "xfs_buf_item.h"
41 #include "xfs_inode_item.h"
42 #include "xfs_btree.h"
43 #include "xfs_btree_trace.h"
44 #include "xfs_alloc.h"
45 #include "xfs_ialloc.h"
48 #include "xfs_error.h"
49 #include "xfs_utils.h"
50 #include "xfs_quota.h"
51 #include "xfs_filestream.h"
52 #include "xfs_vnodeops.h"
53 #include "xfs_trace.h"
55 kmem_zone_t *xfs_ifork_zone;
56 kmem_zone_t *xfs_inode_zone;
59 * Used in xfs_itruncate(). This is the maximum number of extents
60 * freed from a file in a single transaction.
62 #define XFS_ITRUNC_MAX_EXTENTS 2
64 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
65 STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
66 STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
67 STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
71 * Make sure that the extents in the given memory buffer
81 xfs_bmbt_rec_host_t rec;
84 for (i = 0; i < nrecs; i++) {
85 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
86 rec.l0 = get_unaligned(&ep->l0);
87 rec.l1 = get_unaligned(&ep->l1);
88 xfs_bmbt_get_all(&rec, &irec);
89 if (fmt == XFS_EXTFMT_NOSTATE)
90 ASSERT(irec.br_state == XFS_EXT_NORM);
94 #define xfs_validate_extents(ifp, nrecs, fmt)
98 * Check that none of the inode's in the buffer have a next
99 * unlinked field of 0.
111 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
113 for (i = 0; i < j; i++) {
114 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
115 i * mp->m_sb.sb_inodesize);
116 if (!dip->di_next_unlinked) {
117 xfs_fs_cmn_err(CE_ALERT, mp,
118 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
120 ASSERT(dip->di_next_unlinked);
127 * Find the buffer associated with the given inode map
128 * We do basic validation checks on the buffer once it has been
129 * retrieved from disk.
135 struct xfs_imap *imap,
145 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno,
146 (int)imap->im_len, buf_flags, &bp);
148 if (error != EAGAIN) {
150 "xfs_imap_to_bp: xfs_trans_read_buf()returned "
151 "an error %d on %s. Returning error.",
152 error, mp->m_fsname);
154 ASSERT(buf_flags & XFS_BUF_TRYLOCK);
160 * Validate the magic number and version of every inode in the buffer
161 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
164 ni = BBTOB(imap->im_len) >> mp->m_sb.sb_inodelog;
165 #else /* usual case */
169 for (i = 0; i < ni; i++) {
173 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
174 (i << mp->m_sb.sb_inodelog));
175 di_ok = be16_to_cpu(dip->di_magic) == XFS_DINODE_MAGIC &&
176 XFS_DINODE_GOOD_VERSION(dip->di_version);
177 if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
178 XFS_ERRTAG_ITOBP_INOTOBP,
179 XFS_RANDOM_ITOBP_INOTOBP))) {
180 if (iget_flags & XFS_IGET_BULKSTAT) {
181 xfs_trans_brelse(tp, bp);
182 return XFS_ERROR(EINVAL);
184 XFS_CORRUPTION_ERROR("xfs_imap_to_bp",
185 XFS_ERRLEVEL_HIGH, mp, dip);
188 "Device %s - bad inode magic/vsn "
189 "daddr %lld #%d (magic=%x)",
190 XFS_BUFTARG_NAME(mp->m_ddev_targp),
191 (unsigned long long)imap->im_blkno, i,
192 be16_to_cpu(dip->di_magic));
194 xfs_trans_brelse(tp, bp);
195 return XFS_ERROR(EFSCORRUPTED);
199 xfs_inobp_check(mp, bp);
202 * Mark the buffer as an inode buffer now that it looks good
204 XFS_BUF_SET_VTYPE(bp, B_FS_INO);
211 * This routine is called to map an inode number within a file
212 * system to the buffer containing the on-disk version of the
213 * inode. It returns a pointer to the buffer containing the
214 * on-disk inode in the bpp parameter, and in the dip parameter
215 * it returns a pointer to the on-disk inode within that buffer.
217 * If a non-zero error is returned, then the contents of bpp and
218 * dipp are undefined.
220 * Use xfs_imap() to determine the size and location of the
221 * buffer to read from disk.
233 struct xfs_imap imap;
238 error = xfs_imap(mp, tp, ino, &imap, imap_flags);
242 error = xfs_imap_to_bp(mp, tp, &imap, &bp, XFS_BUF_LOCK, imap_flags);
246 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
248 *offset = imap.im_boffset;
254 * This routine is called to map an inode to the buffer containing
255 * the on-disk version of the inode. It returns a pointer to the
256 * buffer containing the on-disk inode in the bpp parameter, and in
257 * the dip parameter it returns a pointer to the on-disk inode within
260 * If a non-zero error is returned, then the contents of bpp and
261 * dipp are undefined.
263 * The inode is expected to already been mapped to its buffer and read
264 * in once, thus we can use the mapping information stored in the inode
265 * rather than calling xfs_imap(). This allows us to avoid the overhead
266 * of looking at the inode btree for small block file systems
281 ASSERT(ip->i_imap.im_blkno != 0);
283 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp, buf_flags, 0);
288 ASSERT(buf_flags & XFS_BUF_TRYLOCK);
294 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
300 * Move inode type and inode format specific information from the
301 * on-disk inode to the in-core inode. For fifos, devs, and sockets
302 * this means set if_rdev to the proper value. For files, directories,
303 * and symlinks this means to bring in the in-line data or extent
304 * pointers. For a file in B-tree format, only the root is immediately
305 * brought in-core. The rest will be in-lined in if_extents when it
306 * is first referenced (see xfs_iread_extents()).
313 xfs_attr_shortform_t *atp;
317 ip->i_df.if_ext_max =
318 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
321 if (unlikely(be32_to_cpu(dip->di_nextents) +
322 be16_to_cpu(dip->di_anextents) >
323 be64_to_cpu(dip->di_nblocks))) {
324 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
325 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
326 (unsigned long long)ip->i_ino,
327 (int)(be32_to_cpu(dip->di_nextents) +
328 be16_to_cpu(dip->di_anextents)),
330 be64_to_cpu(dip->di_nblocks));
331 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
333 return XFS_ERROR(EFSCORRUPTED);
336 if (unlikely(dip->di_forkoff > ip->i_mount->m_sb.sb_inodesize)) {
337 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
338 "corrupt dinode %Lu, forkoff = 0x%x.",
339 (unsigned long long)ip->i_ino,
341 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
343 return XFS_ERROR(EFSCORRUPTED);
346 if (unlikely((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) &&
347 !ip->i_mount->m_rtdev_targp)) {
348 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
349 "corrupt dinode %Lu, has realtime flag set.",
351 XFS_CORRUPTION_ERROR("xfs_iformat(realtime)",
352 XFS_ERRLEVEL_LOW, ip->i_mount, dip);
353 return XFS_ERROR(EFSCORRUPTED);
356 switch (ip->i_d.di_mode & S_IFMT) {
361 if (unlikely(dip->di_format != XFS_DINODE_FMT_DEV)) {
362 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
364 return XFS_ERROR(EFSCORRUPTED);
368 ip->i_df.if_u2.if_rdev = xfs_dinode_get_rdev(dip);
374 switch (dip->di_format) {
375 case XFS_DINODE_FMT_LOCAL:
377 * no local regular files yet
379 if (unlikely((be16_to_cpu(dip->di_mode) & S_IFMT) == S_IFREG)) {
380 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
382 "(local format for regular file).",
383 (unsigned long long) ip->i_ino);
384 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
387 return XFS_ERROR(EFSCORRUPTED);
390 di_size = be64_to_cpu(dip->di_size);
391 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
392 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
394 "(bad size %Ld for local inode).",
395 (unsigned long long) ip->i_ino,
396 (long long) di_size);
397 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
400 return XFS_ERROR(EFSCORRUPTED);
404 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
406 case XFS_DINODE_FMT_EXTENTS:
407 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
409 case XFS_DINODE_FMT_BTREE:
410 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
413 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
415 return XFS_ERROR(EFSCORRUPTED);
420 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
421 return XFS_ERROR(EFSCORRUPTED);
426 if (!XFS_DFORK_Q(dip))
428 ASSERT(ip->i_afp == NULL);
429 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
430 ip->i_afp->if_ext_max =
431 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
432 switch (dip->di_aformat) {
433 case XFS_DINODE_FMT_LOCAL:
434 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
435 size = be16_to_cpu(atp->hdr.totsize);
437 if (unlikely(size < sizeof(struct xfs_attr_sf_hdr))) {
438 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
440 "(bad attr fork size %Ld).",
441 (unsigned long long) ip->i_ino,
443 XFS_CORRUPTION_ERROR("xfs_iformat(8)",
446 return XFS_ERROR(EFSCORRUPTED);
449 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
451 case XFS_DINODE_FMT_EXTENTS:
452 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
454 case XFS_DINODE_FMT_BTREE:
455 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
458 error = XFS_ERROR(EFSCORRUPTED);
462 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
464 xfs_idestroy_fork(ip, XFS_DATA_FORK);
470 * The file is in-lined in the on-disk inode.
471 * If it fits into if_inline_data, then copy
472 * it there, otherwise allocate a buffer for it
473 * and copy the data there. Either way, set
474 * if_data to point at the data.
475 * If we allocate a buffer for the data, make
476 * sure that its size is a multiple of 4 and
477 * record the real size in i_real_bytes.
490 * If the size is unreasonable, then something
491 * is wrong and we just bail out rather than crash in
492 * kmem_alloc() or memcpy() below.
494 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
495 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
497 "(bad size %d for local fork, size = %d).",
498 (unsigned long long) ip->i_ino, size,
499 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
500 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
502 return XFS_ERROR(EFSCORRUPTED);
504 ifp = XFS_IFORK_PTR(ip, whichfork);
507 ifp->if_u1.if_data = NULL;
508 else if (size <= sizeof(ifp->if_u2.if_inline_data))
509 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
511 real_size = roundup(size, 4);
512 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
514 ifp->if_bytes = size;
515 ifp->if_real_bytes = real_size;
517 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
518 ifp->if_flags &= ~XFS_IFEXTENTS;
519 ifp->if_flags |= XFS_IFINLINE;
524 * The file consists of a set of extents all
525 * of which fit into the on-disk inode.
526 * If there are few enough extents to fit into
527 * the if_inline_ext, then copy them there.
528 * Otherwise allocate a buffer for them and copy
529 * them into it. Either way, set if_extents
530 * to point at the extents.
544 ifp = XFS_IFORK_PTR(ip, whichfork);
545 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
546 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
549 * If the number of extents is unreasonable, then something
550 * is wrong and we just bail out rather than crash in
551 * kmem_alloc() or memcpy() below.
553 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
554 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
555 "corrupt inode %Lu ((a)extents = %d).",
556 (unsigned long long) ip->i_ino, nex);
557 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
559 return XFS_ERROR(EFSCORRUPTED);
562 ifp->if_real_bytes = 0;
564 ifp->if_u1.if_extents = NULL;
565 else if (nex <= XFS_INLINE_EXTS)
566 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
568 xfs_iext_add(ifp, 0, nex);
570 ifp->if_bytes = size;
572 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
573 xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip));
574 for (i = 0; i < nex; i++, dp++) {
575 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
576 ep->l0 = get_unaligned_be64(&dp->l0);
577 ep->l1 = get_unaligned_be64(&dp->l1);
579 XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork);
580 if (whichfork != XFS_DATA_FORK ||
581 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
582 if (unlikely(xfs_check_nostate_extents(
584 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
587 return XFS_ERROR(EFSCORRUPTED);
590 ifp->if_flags |= XFS_IFEXTENTS;
595 * The file has too many extents to fit into
596 * the inode, so they are in B-tree format.
597 * Allocate a buffer for the root of the B-tree
598 * and copy the root into it. The i_extents
599 * field will remain NULL until all of the
600 * extents are read in (when they are needed).
608 xfs_bmdr_block_t *dfp;
614 ifp = XFS_IFORK_PTR(ip, whichfork);
615 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
616 size = XFS_BMAP_BROOT_SPACE(dfp);
617 nrecs = be16_to_cpu(dfp->bb_numrecs);
620 * blow out if -- fork has less extents than can fit in
621 * fork (fork shouldn't be a btree format), root btree
622 * block has more records than can fit into the fork,
623 * or the number of extents is greater than the number of
626 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
627 || XFS_BMDR_SPACE_CALC(nrecs) >
628 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
629 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
630 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
631 "corrupt inode %Lu (btree).",
632 (unsigned long long) ip->i_ino);
633 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
635 return XFS_ERROR(EFSCORRUPTED);
638 ifp->if_broot_bytes = size;
639 ifp->if_broot = kmem_alloc(size, KM_SLEEP);
640 ASSERT(ifp->if_broot != NULL);
642 * Copy and convert from the on-disk structure
643 * to the in-memory structure.
645 xfs_bmdr_to_bmbt(ip->i_mount, dfp,
646 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
647 ifp->if_broot, size);
648 ifp->if_flags &= ~XFS_IFEXTENTS;
649 ifp->if_flags |= XFS_IFBROOT;
655 xfs_dinode_from_disk(
659 to->di_magic = be16_to_cpu(from->di_magic);
660 to->di_mode = be16_to_cpu(from->di_mode);
661 to->di_version = from ->di_version;
662 to->di_format = from->di_format;
663 to->di_onlink = be16_to_cpu(from->di_onlink);
664 to->di_uid = be32_to_cpu(from->di_uid);
665 to->di_gid = be32_to_cpu(from->di_gid);
666 to->di_nlink = be32_to_cpu(from->di_nlink);
667 to->di_projid = be16_to_cpu(from->di_projid);
668 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
669 to->di_flushiter = be16_to_cpu(from->di_flushiter);
670 to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec);
671 to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec);
672 to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec);
673 to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec);
674 to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec);
675 to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec);
676 to->di_size = be64_to_cpu(from->di_size);
677 to->di_nblocks = be64_to_cpu(from->di_nblocks);
678 to->di_extsize = be32_to_cpu(from->di_extsize);
679 to->di_nextents = be32_to_cpu(from->di_nextents);
680 to->di_anextents = be16_to_cpu(from->di_anextents);
681 to->di_forkoff = from->di_forkoff;
682 to->di_aformat = from->di_aformat;
683 to->di_dmevmask = be32_to_cpu(from->di_dmevmask);
684 to->di_dmstate = be16_to_cpu(from->di_dmstate);
685 to->di_flags = be16_to_cpu(from->di_flags);
686 to->di_gen = be32_to_cpu(from->di_gen);
692 xfs_icdinode_t *from)
694 to->di_magic = cpu_to_be16(from->di_magic);
695 to->di_mode = cpu_to_be16(from->di_mode);
696 to->di_version = from ->di_version;
697 to->di_format = from->di_format;
698 to->di_onlink = cpu_to_be16(from->di_onlink);
699 to->di_uid = cpu_to_be32(from->di_uid);
700 to->di_gid = cpu_to_be32(from->di_gid);
701 to->di_nlink = cpu_to_be32(from->di_nlink);
702 to->di_projid = cpu_to_be16(from->di_projid);
703 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
704 to->di_flushiter = cpu_to_be16(from->di_flushiter);
705 to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec);
706 to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec);
707 to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec);
708 to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec);
709 to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec);
710 to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec);
711 to->di_size = cpu_to_be64(from->di_size);
712 to->di_nblocks = cpu_to_be64(from->di_nblocks);
713 to->di_extsize = cpu_to_be32(from->di_extsize);
714 to->di_nextents = cpu_to_be32(from->di_nextents);
715 to->di_anextents = cpu_to_be16(from->di_anextents);
716 to->di_forkoff = from->di_forkoff;
717 to->di_aformat = from->di_aformat;
718 to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
719 to->di_dmstate = cpu_to_be16(from->di_dmstate);
720 to->di_flags = cpu_to_be16(from->di_flags);
721 to->di_gen = cpu_to_be32(from->di_gen);
730 if (di_flags & XFS_DIFLAG_ANY) {
731 if (di_flags & XFS_DIFLAG_REALTIME)
732 flags |= XFS_XFLAG_REALTIME;
733 if (di_flags & XFS_DIFLAG_PREALLOC)
734 flags |= XFS_XFLAG_PREALLOC;
735 if (di_flags & XFS_DIFLAG_IMMUTABLE)
736 flags |= XFS_XFLAG_IMMUTABLE;
737 if (di_flags & XFS_DIFLAG_APPEND)
738 flags |= XFS_XFLAG_APPEND;
739 if (di_flags & XFS_DIFLAG_SYNC)
740 flags |= XFS_XFLAG_SYNC;
741 if (di_flags & XFS_DIFLAG_NOATIME)
742 flags |= XFS_XFLAG_NOATIME;
743 if (di_flags & XFS_DIFLAG_NODUMP)
744 flags |= XFS_XFLAG_NODUMP;
745 if (di_flags & XFS_DIFLAG_RTINHERIT)
746 flags |= XFS_XFLAG_RTINHERIT;
747 if (di_flags & XFS_DIFLAG_PROJINHERIT)
748 flags |= XFS_XFLAG_PROJINHERIT;
749 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
750 flags |= XFS_XFLAG_NOSYMLINKS;
751 if (di_flags & XFS_DIFLAG_EXTSIZE)
752 flags |= XFS_XFLAG_EXTSIZE;
753 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
754 flags |= XFS_XFLAG_EXTSZINHERIT;
755 if (di_flags & XFS_DIFLAG_NODEFRAG)
756 flags |= XFS_XFLAG_NODEFRAG;
757 if (di_flags & XFS_DIFLAG_FILESTREAM)
758 flags |= XFS_XFLAG_FILESTREAM;
768 xfs_icdinode_t *dic = &ip->i_d;
770 return _xfs_dic2xflags(dic->di_flags) |
771 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
778 return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) |
779 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
783 * Read the disk inode attributes into the in-core inode structure.
798 * Fill in the location information in the in-core inode.
800 ip->i_imap.im_blkno = bno;
801 error = xfs_imap(mp, tp, ip->i_ino, &ip->i_imap, iget_flags);
804 ASSERT(bno == 0 || bno == ip->i_imap.im_blkno);
807 * Get pointers to the on-disk inode and the buffer containing it.
809 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp,
810 XFS_BUF_LOCK, iget_flags);
813 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
816 * If we got something that isn't an inode it means someone
817 * (nfs or dmi) has a stale handle.
819 if (be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC) {
821 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
822 "dip->di_magic (0x%x) != "
823 "XFS_DINODE_MAGIC (0x%x)",
824 be16_to_cpu(dip->di_magic),
827 error = XFS_ERROR(EINVAL);
832 * If the on-disk inode is already linked to a directory
833 * entry, copy all of the inode into the in-core inode.
834 * xfs_iformat() handles copying in the inode format
835 * specific information.
836 * Otherwise, just get the truly permanent information.
839 xfs_dinode_from_disk(&ip->i_d, dip);
840 error = xfs_iformat(ip, dip);
843 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
844 "xfs_iformat() returned error %d",
850 ip->i_d.di_magic = be16_to_cpu(dip->di_magic);
851 ip->i_d.di_version = dip->di_version;
852 ip->i_d.di_gen = be32_to_cpu(dip->di_gen);
853 ip->i_d.di_flushiter = be16_to_cpu(dip->di_flushiter);
855 * Make sure to pull in the mode here as well in
856 * case the inode is released without being used.
857 * This ensures that xfs_inactive() will see that
858 * the inode is already free and not try to mess
859 * with the uninitialized part of it.
863 * Initialize the per-fork minima and maxima for a new
864 * inode here. xfs_iformat will do it for old inodes.
866 ip->i_df.if_ext_max =
867 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
871 * The inode format changed when we moved the link count and
872 * made it 32 bits long. If this is an old format inode,
873 * convert it in memory to look like a new one. If it gets
874 * flushed to disk we will convert back before flushing or
875 * logging it. We zero out the new projid field and the old link
876 * count field. We'll handle clearing the pad field (the remains
877 * of the old uuid field) when we actually convert the inode to
878 * the new format. We don't change the version number so that we
879 * can distinguish this from a real new format inode.
881 if (ip->i_d.di_version == 1) {
882 ip->i_d.di_nlink = ip->i_d.di_onlink;
883 ip->i_d.di_onlink = 0;
884 ip->i_d.di_projid = 0;
887 ip->i_delayed_blks = 0;
888 ip->i_size = ip->i_d.di_size;
891 * Mark the buffer containing the inode as something to keep
892 * around for a while. This helps to keep recently accessed
893 * meta-data in-core longer.
895 XFS_BUF_SET_REF(bp, XFS_INO_REF);
898 * Use xfs_trans_brelse() to release the buffer containing the
899 * on-disk inode, because it was acquired with xfs_trans_read_buf()
900 * in xfs_itobp() above. If tp is NULL, this is just a normal
901 * brelse(). If we're within a transaction, then xfs_trans_brelse()
902 * will only release the buffer if it is not dirty within the
903 * transaction. It will be OK to release the buffer in this case,
904 * because inodes on disk are never destroyed and we will be
905 * locking the new in-core inode before putting it in the hash
906 * table where other processes can find it. Thus we don't have
907 * to worry about the inode being changed just because we released
911 xfs_trans_brelse(tp, bp);
916 * Read in extents from a btree-format inode.
917 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
927 xfs_extnum_t nextents;
930 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
931 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
933 return XFS_ERROR(EFSCORRUPTED);
935 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
936 size = nextents * sizeof(xfs_bmbt_rec_t);
937 ifp = XFS_IFORK_PTR(ip, whichfork);
940 * We know that the size is valid (it's checked in iformat_btree)
942 ifp->if_lastex = NULLEXTNUM;
943 ifp->if_bytes = ifp->if_real_bytes = 0;
944 ifp->if_flags |= XFS_IFEXTENTS;
945 xfs_iext_add(ifp, 0, nextents);
946 error = xfs_bmap_read_extents(tp, ip, whichfork);
948 xfs_iext_destroy(ifp);
949 ifp->if_flags &= ~XFS_IFEXTENTS;
952 xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip));
957 * Allocate an inode on disk and return a copy of its in-core version.
958 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
959 * appropriately within the inode. The uid and gid for the inode are
960 * set according to the contents of the given cred structure.
962 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
963 * has a free inode available, call xfs_iget()
964 * to obtain the in-core version of the allocated inode. Finally,
965 * fill in the inode and log its initial contents. In this case,
966 * ialloc_context would be set to NULL and call_again set to false.
968 * If xfs_dialloc() does not have an available inode,
969 * it will replenish its supply by doing an allocation. Since we can
970 * only do one allocation within a transaction without deadlocks, we
971 * must commit the current transaction before returning the inode itself.
972 * In this case, therefore, we will set call_again to true and return.
973 * The caller should then commit the current transaction, start a new
974 * transaction, and call xfs_ialloc() again to actually get the inode.
976 * To ensure that some other process does not grab the inode that
977 * was allocated during the first call to xfs_ialloc(), this routine
978 * also returns the [locked] bp pointing to the head of the freelist
979 * as ialloc_context. The caller should hold this buffer across
980 * the commit and pass it back into this routine on the second call.
982 * If we are allocating quota inodes, we do not have a parent inode
983 * to attach to or associate with (i.e. pip == NULL) because they
984 * are not linked into the directory structure - they are attached
985 * directly to the superblock - and so have no parent.
997 xfs_buf_t **ialloc_context,
998 boolean_t *call_again,
1006 int filestreams = 0;
1009 * Call the space management code to pick
1010 * the on-disk inode to be allocated.
1012 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
1013 ialloc_context, call_again, &ino);
1016 if (*call_again || ino == NULLFSINO) {
1020 ASSERT(*ialloc_context == NULL);
1023 * Get the in-core inode with the lock held exclusively.
1024 * This is because we're setting fields here we need
1025 * to prevent others from looking at until we're done.
1027 error = xfs_trans_iget(tp->t_mountp, tp, ino,
1028 XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
1033 ip->i_d.di_mode = (__uint16_t)mode;
1034 ip->i_d.di_onlink = 0;
1035 ip->i_d.di_nlink = nlink;
1036 ASSERT(ip->i_d.di_nlink == nlink);
1037 ip->i_d.di_uid = current_fsuid();
1038 ip->i_d.di_gid = current_fsgid();
1039 ip->i_d.di_projid = prid;
1040 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1043 * If the superblock version is up to where we support new format
1044 * inodes and this is currently an old format inode, then change
1045 * the inode version number now. This way we only do the conversion
1046 * here rather than here and in the flush/logging code.
1048 if (xfs_sb_version_hasnlink(&tp->t_mountp->m_sb) &&
1049 ip->i_d.di_version == 1) {
1050 ip->i_d.di_version = 2;
1052 * We've already zeroed the old link count, the projid field,
1053 * and the pad field.
1058 * Project ids won't be stored on disk if we are using a version 1 inode.
1060 if ((prid != 0) && (ip->i_d.di_version == 1))
1061 xfs_bump_ino_vers2(tp, ip);
1063 if (pip && XFS_INHERIT_GID(pip)) {
1064 ip->i_d.di_gid = pip->i_d.di_gid;
1065 if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1066 ip->i_d.di_mode |= S_ISGID;
1071 * If the group ID of the new file does not match the effective group
1072 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1073 * (and only if the irix_sgid_inherit compatibility variable is set).
1075 if ((irix_sgid_inherit) &&
1076 (ip->i_d.di_mode & S_ISGID) &&
1077 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1078 ip->i_d.di_mode &= ~S_ISGID;
1081 ip->i_d.di_size = 0;
1083 ip->i_d.di_nextents = 0;
1084 ASSERT(ip->i_d.di_nblocks == 0);
1087 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
1088 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
1089 ip->i_d.di_atime = ip->i_d.di_mtime;
1090 ip->i_d.di_ctime = ip->i_d.di_mtime;
1093 * di_gen will have been taken care of in xfs_iread.
1095 ip->i_d.di_extsize = 0;
1096 ip->i_d.di_dmevmask = 0;
1097 ip->i_d.di_dmstate = 0;
1098 ip->i_d.di_flags = 0;
1099 flags = XFS_ILOG_CORE;
1100 switch (mode & S_IFMT) {
1105 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1106 ip->i_df.if_u2.if_rdev = rdev;
1107 ip->i_df.if_flags = 0;
1108 flags |= XFS_ILOG_DEV;
1112 * we can't set up filestreams until after the VFS inode
1113 * is set up properly.
1115 if (pip && xfs_inode_is_filestream(pip))
1119 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
1122 if ((mode & S_IFMT) == S_IFDIR) {
1123 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1124 di_flags |= XFS_DIFLAG_RTINHERIT;
1125 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1126 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1127 ip->i_d.di_extsize = pip->i_d.di_extsize;
1129 } else if ((mode & S_IFMT) == S_IFREG) {
1130 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1131 di_flags |= XFS_DIFLAG_REALTIME;
1132 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1133 di_flags |= XFS_DIFLAG_EXTSIZE;
1134 ip->i_d.di_extsize = pip->i_d.di_extsize;
1137 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1138 xfs_inherit_noatime)
1139 di_flags |= XFS_DIFLAG_NOATIME;
1140 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1142 di_flags |= XFS_DIFLAG_NODUMP;
1143 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1145 di_flags |= XFS_DIFLAG_SYNC;
1146 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1147 xfs_inherit_nosymlinks)
1148 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1149 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1150 di_flags |= XFS_DIFLAG_PROJINHERIT;
1151 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1152 xfs_inherit_nodefrag)
1153 di_flags |= XFS_DIFLAG_NODEFRAG;
1154 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
1155 di_flags |= XFS_DIFLAG_FILESTREAM;
1156 ip->i_d.di_flags |= di_flags;
1160 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1161 ip->i_df.if_flags = XFS_IFEXTENTS;
1162 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1163 ip->i_df.if_u1.if_extents = NULL;
1169 * Attribute fork settings for new inode.
1171 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1172 ip->i_d.di_anextents = 0;
1175 * Log the new values stuffed into the inode.
1177 xfs_trans_log_inode(tp, ip, flags);
1179 /* now that we have an i_mode we can setup inode ops and unlock */
1180 xfs_setup_inode(ip);
1182 /* now we have set up the vfs inode we can associate the filestream */
1184 error = xfs_filestream_associate(pip, ip);
1188 xfs_iflags_set(ip, XFS_IFILESTREAM);
1196 * Check to make sure that there are no blocks allocated to the
1197 * file beyond the size of the file. We don't check this for
1198 * files with fixed size extents or real time extents, but we
1199 * at least do it for regular files.
1208 xfs_fileoff_t map_first;
1210 xfs_bmbt_irec_t imaps[2];
1212 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1215 if (XFS_IS_REALTIME_INODE(ip))
1218 if (ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE)
1222 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1224 * The filesystem could be shutting down, so bmapi may return
1227 if (xfs_bmapi(NULL, ip, map_first,
1229 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1231 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1234 ASSERT(nimaps == 1);
1235 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1240 * Calculate the last possible buffered byte in a file. This must
1241 * include data that was buffered beyond the EOF by the write code.
1242 * This also needs to deal with overflowing the xfs_fsize_t type
1243 * which can happen for sizes near the limit.
1245 * We also need to take into account any blocks beyond the EOF. It
1246 * may be the case that they were buffered by a write which failed.
1247 * In that case the pages will still be in memory, but the inode size
1248 * will never have been updated.
1255 xfs_fsize_t last_byte;
1256 xfs_fileoff_t last_block;
1257 xfs_fileoff_t size_last_block;
1260 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED));
1264 * Only check for blocks beyond the EOF if the extents have
1265 * been read in. This eliminates the need for the inode lock,
1266 * and it also saves us from looking when it really isn't
1269 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1270 xfs_ilock(ip, XFS_ILOCK_SHARED);
1271 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1273 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1280 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_size);
1281 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1283 last_byte = XFS_FSB_TO_B(mp, last_block);
1284 if (last_byte < 0) {
1285 return XFS_MAXIOFFSET(mp);
1287 last_byte += (1 << mp->m_writeio_log);
1288 if (last_byte < 0) {
1289 return XFS_MAXIOFFSET(mp);
1295 * Start the truncation of the file to new_size. The new size
1296 * must be smaller than the current size. This routine will
1297 * clear the buffer and page caches of file data in the removed
1298 * range, and xfs_itruncate_finish() will remove the underlying
1301 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1302 * must NOT have the inode lock held at all. This is because we're
1303 * calling into the buffer/page cache code and we can't hold the
1304 * inode lock when we do so.
1306 * We need to wait for any direct I/Os in flight to complete before we
1307 * proceed with the truncate. This is needed to prevent the extents
1308 * being read or written by the direct I/Os from being removed while the
1309 * I/O is in flight as there is no other method of synchronising
1310 * direct I/O with the truncate operation. Also, because we hold
1311 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1312 * started until the truncate completes and drops the lock. Essentially,
1313 * the xfs_ioend_wait() call forms an I/O barrier that provides strict
1314 * ordering between direct I/Os and the truncate operation.
1316 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1317 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1318 * in the case that the caller is locking things out of order and
1319 * may not be able to call xfs_itruncate_finish() with the inode lock
1320 * held without dropping the I/O lock. If the caller must drop the
1321 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1322 * must be called again with all the same restrictions as the initial
1326 xfs_itruncate_start(
1329 xfs_fsize_t new_size)
1331 xfs_fsize_t last_byte;
1332 xfs_off_t toss_start;
1336 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1337 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1338 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1339 (flags == XFS_ITRUNC_MAYBE));
1343 /* wait for the completion of any pending DIOs */
1344 if (new_size == 0 || new_size < ip->i_size)
1348 * Call toss_pages or flushinval_pages to get rid of pages
1349 * overlapping the region being removed. We have to use
1350 * the less efficient flushinval_pages in the case that the
1351 * caller may not be able to finish the truncate without
1352 * dropping the inode's I/O lock. Make sure
1353 * to catch any pages brought in by buffers overlapping
1354 * the EOF by searching out beyond the isize by our
1355 * block size. We round new_size up to a block boundary
1356 * so that we don't toss things on the same block as
1357 * new_size but before it.
1359 * Before calling toss_page or flushinval_pages, make sure to
1360 * call remapf() over the same region if the file is mapped.
1361 * This frees up mapped file references to the pages in the
1362 * given range and for the flushinval_pages case it ensures
1363 * that we get the latest mapped changes flushed out.
1365 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1366 toss_start = XFS_FSB_TO_B(mp, toss_start);
1367 if (toss_start < 0) {
1369 * The place to start tossing is beyond our maximum
1370 * file size, so there is no way that the data extended
1375 last_byte = xfs_file_last_byte(ip);
1376 trace_xfs_itruncate_start(ip, flags, new_size, toss_start, last_byte);
1377 if (last_byte > toss_start) {
1378 if (flags & XFS_ITRUNC_DEFINITE) {
1379 xfs_tosspages(ip, toss_start,
1380 -1, FI_REMAPF_LOCKED);
1382 error = xfs_flushinval_pages(ip, toss_start,
1383 -1, FI_REMAPF_LOCKED);
1388 if (new_size == 0) {
1389 ASSERT(VN_CACHED(VFS_I(ip)) == 0);
1396 * Shrink the file to the given new_size. The new size must be smaller than
1397 * the current size. This will free up the underlying blocks in the removed
1398 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
1400 * The transaction passed to this routine must have made a permanent log
1401 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1402 * given transaction and start new ones, so make sure everything involved in
1403 * the transaction is tidy before calling here. Some transaction will be
1404 * returned to the caller to be committed. The incoming transaction must
1405 * already include the inode, and both inode locks must be held exclusively.
1406 * The inode must also be "held" within the transaction. On return the inode
1407 * will be "held" within the returned transaction. This routine does NOT
1408 * require any disk space to be reserved for it within the transaction.
1410 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1411 * indicates the fork which is to be truncated. For the attribute fork we only
1412 * support truncation to size 0.
1414 * We use the sync parameter to indicate whether or not the first transaction
1415 * we perform might have to be synchronous. For the attr fork, it needs to be
1416 * so if the unlink of the inode is not yet known to be permanent in the log.
1417 * This keeps us from freeing and reusing the blocks of the attribute fork
1418 * before the unlink of the inode becomes permanent.
1420 * For the data fork, we normally have to run synchronously if we're being
1421 * called out of the inactive path or we're being called out of the create path
1422 * where we're truncating an existing file. Either way, the truncate needs to
1423 * be sync so blocks don't reappear in the file with altered data in case of a
1424 * crash. wsync filesystems can run the first case async because anything that
1425 * shrinks the inode has to run sync so by the time we're called here from
1426 * inactive, the inode size is permanently set to 0.
1428 * Calls from the truncate path always need to be sync unless we're in a wsync
1429 * filesystem and the file has already been unlinked.
1431 * The caller is responsible for correctly setting the sync parameter. It gets
1432 * too hard for us to guess here which path we're being called out of just
1433 * based on inode state.
1435 * If we get an error, we must return with the inode locked and linked into the
1436 * current transaction. This keeps things simple for the higher level code,
1437 * because it always knows that the inode is locked and held in the transaction
1438 * that returns to it whether errors occur or not. We don't mark the inode
1439 * dirty on error so that transactions can be easily aborted if possible.
1442 xfs_itruncate_finish(
1445 xfs_fsize_t new_size,
1449 xfs_fsblock_t first_block;
1450 xfs_fileoff_t first_unmap_block;
1451 xfs_fileoff_t last_block;
1452 xfs_filblks_t unmap_len=0;
1457 xfs_bmap_free_t free_list;
1460 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
1461 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1462 ASSERT(*tp != NULL);
1463 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1464 ASSERT(ip->i_transp == *tp);
1465 ASSERT(ip->i_itemp != NULL);
1466 ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
1470 mp = (ntp)->t_mountp;
1471 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1474 * We only support truncating the entire attribute fork.
1476 if (fork == XFS_ATTR_FORK) {
1479 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1480 trace_xfs_itruncate_finish_start(ip, new_size);
1483 * The first thing we do is set the size to new_size permanently
1484 * on disk. This way we don't have to worry about anyone ever
1485 * being able to look at the data being freed even in the face
1486 * of a crash. What we're getting around here is the case where
1487 * we free a block, it is allocated to another file, it is written
1488 * to, and then we crash. If the new data gets written to the
1489 * file but the log buffers containing the free and reallocation
1490 * don't, then we'd end up with garbage in the blocks being freed.
1491 * As long as we make the new_size permanent before actually
1492 * freeing any blocks it doesn't matter if they get writtten to.
1494 * The callers must signal into us whether or not the size
1495 * setting here must be synchronous. There are a few cases
1496 * where it doesn't have to be synchronous. Those cases
1497 * occur if the file is unlinked and we know the unlink is
1498 * permanent or if the blocks being truncated are guaranteed
1499 * to be beyond the inode eof (regardless of the link count)
1500 * and the eof value is permanent. Both of these cases occur
1501 * only on wsync-mounted filesystems. In those cases, we're
1502 * guaranteed that no user will ever see the data in the blocks
1503 * that are being truncated so the truncate can run async.
1504 * In the free beyond eof case, the file may wind up with
1505 * more blocks allocated to it than it needs if we crash
1506 * and that won't get fixed until the next time the file
1507 * is re-opened and closed but that's ok as that shouldn't
1508 * be too many blocks.
1510 * However, we can't just make all wsync xactions run async
1511 * because there's one call out of the create path that needs
1512 * to run sync where it's truncating an existing file to size
1513 * 0 whose size is > 0.
1515 * It's probably possible to come up with a test in this
1516 * routine that would correctly distinguish all the above
1517 * cases from the values of the function parameters and the
1518 * inode state but for sanity's sake, I've decided to let the
1519 * layers above just tell us. It's simpler to correctly figure
1520 * out in the layer above exactly under what conditions we
1521 * can run async and I think it's easier for others read and
1522 * follow the logic in case something has to be changed.
1523 * cscope is your friend -- rcc.
1525 * The attribute fork is much simpler.
1527 * For the attribute fork we allow the caller to tell us whether
1528 * the unlink of the inode that led to this call is yet permanent
1529 * in the on disk log. If it is not and we will be freeing extents
1530 * in this inode then we make the first transaction synchronous
1531 * to make sure that the unlink is permanent by the time we free
1534 if (fork == XFS_DATA_FORK) {
1535 if (ip->i_d.di_nextents > 0) {
1537 * If we are not changing the file size then do
1538 * not update the on-disk file size - we may be
1539 * called from xfs_inactive_free_eofblocks(). If we
1540 * update the on-disk file size and then the system
1541 * crashes before the contents of the file are
1542 * flushed to disk then the files may be full of
1543 * holes (ie NULL files bug).
1545 if (ip->i_size != new_size) {
1546 ip->i_d.di_size = new_size;
1547 ip->i_size = new_size;
1548 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1552 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1553 if (ip->i_d.di_anextents > 0)
1554 xfs_trans_set_sync(ntp);
1556 ASSERT(fork == XFS_DATA_FORK ||
1557 (fork == XFS_ATTR_FORK &&
1558 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1559 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1562 * Since it is possible for space to become allocated beyond
1563 * the end of the file (in a crash where the space is allocated
1564 * but the inode size is not yet updated), simply remove any
1565 * blocks which show up between the new EOF and the maximum
1566 * possible file size. If the first block to be removed is
1567 * beyond the maximum file size (ie it is the same as last_block),
1568 * then there is nothing to do.
1570 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1571 ASSERT(first_unmap_block <= last_block);
1573 if (last_block == first_unmap_block) {
1576 unmap_len = last_block - first_unmap_block + 1;
1580 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1581 * will tell us whether it freed the entire range or
1582 * not. If this is a synchronous mount (wsync),
1583 * then we can tell bunmapi to keep all the
1584 * transactions asynchronous since the unlink
1585 * transaction that made this inode inactive has
1586 * already hit the disk. There's no danger of
1587 * the freed blocks being reused, there being a
1588 * crash, and the reused blocks suddenly reappearing
1589 * in this file with garbage in them once recovery
1592 xfs_bmap_init(&free_list, &first_block);
1593 error = xfs_bunmapi(ntp, ip,
1594 first_unmap_block, unmap_len,
1595 xfs_bmapi_aflag(fork) |
1596 (sync ? 0 : XFS_BMAPI_ASYNC),
1597 XFS_ITRUNC_MAX_EXTENTS,
1598 &first_block, &free_list,
1602 * If the bunmapi call encounters an error,
1603 * return to the caller where the transaction
1604 * can be properly aborted. We just need to
1605 * make sure we're not holding any resources
1606 * that we were not when we came in.
1608 xfs_bmap_cancel(&free_list);
1613 * Duplicate the transaction that has the permanent
1614 * reservation and commit the old transaction.
1616 error = xfs_bmap_finish(tp, &free_list, &committed);
1619 /* link the inode into the next xact in the chain */
1620 xfs_trans_ijoin(ntp, ip,
1621 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1622 xfs_trans_ihold(ntp, ip);
1627 * If the bmap finish call encounters an error, return
1628 * to the caller where the transaction can be properly
1629 * aborted. We just need to make sure we're not
1630 * holding any resources that we were not when we came
1633 * Aborting from this point might lose some blocks in
1634 * the file system, but oh well.
1636 xfs_bmap_cancel(&free_list);
1642 * Mark the inode dirty so it will be logged and
1643 * moved forward in the log as part of every commit.
1645 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1648 ntp = xfs_trans_dup(ntp);
1649 error = xfs_trans_commit(*tp, 0);
1652 /* link the inode into the next transaction in the chain */
1653 xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1654 xfs_trans_ihold(ntp, ip);
1659 * transaction commit worked ok so we can drop the extra ticket
1660 * reference that we gained in xfs_trans_dup()
1662 xfs_log_ticket_put(ntp->t_ticket);
1663 error = xfs_trans_reserve(ntp, 0,
1664 XFS_ITRUNCATE_LOG_RES(mp), 0,
1665 XFS_TRANS_PERM_LOG_RES,
1666 XFS_ITRUNCATE_LOG_COUNT);
1671 * Only update the size in the case of the data fork, but
1672 * always re-log the inode so that our permanent transaction
1673 * can keep on rolling it forward in the log.
1675 if (fork == XFS_DATA_FORK) {
1676 xfs_isize_check(mp, ip, new_size);
1678 * If we are not changing the file size then do
1679 * not update the on-disk file size - we may be
1680 * called from xfs_inactive_free_eofblocks(). If we
1681 * update the on-disk file size and then the system
1682 * crashes before the contents of the file are
1683 * flushed to disk then the files may be full of
1684 * holes (ie NULL files bug).
1686 if (ip->i_size != new_size) {
1687 ip->i_d.di_size = new_size;
1688 ip->i_size = new_size;
1691 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1692 ASSERT((new_size != 0) ||
1693 (fork == XFS_ATTR_FORK) ||
1694 (ip->i_delayed_blks == 0));
1695 ASSERT((new_size != 0) ||
1696 (fork == XFS_ATTR_FORK) ||
1697 (ip->i_d.di_nextents == 0));
1698 trace_xfs_itruncate_finish_end(ip, new_size);
1703 * This is called when the inode's link count goes to 0.
1704 * We place the on-disk inode on a list in the AGI. It
1705 * will be pulled from this list when the inode is freed.
1722 ASSERT(ip->i_d.di_nlink == 0);
1723 ASSERT(ip->i_d.di_mode != 0);
1724 ASSERT(ip->i_transp == tp);
1729 * Get the agi buffer first. It ensures lock ordering
1732 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1735 agi = XFS_BUF_TO_AGI(agibp);
1738 * Get the index into the agi hash table for the
1739 * list this inode will go on.
1741 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1743 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1744 ASSERT(agi->agi_unlinked[bucket_index]);
1745 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1747 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
1749 * There is already another inode in the bucket we need
1750 * to add ourselves to. Add us at the front of the list.
1751 * Here we put the head pointer into our next pointer,
1752 * and then we fall through to point the head at us.
1754 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XFS_BUF_LOCK);
1758 ASSERT(be32_to_cpu(dip->di_next_unlinked) == NULLAGINO);
1759 /* both on-disk, don't endian flip twice */
1760 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1761 offset = ip->i_imap.im_boffset +
1762 offsetof(xfs_dinode_t, di_next_unlinked);
1763 xfs_trans_inode_buf(tp, ibp);
1764 xfs_trans_log_buf(tp, ibp, offset,
1765 (offset + sizeof(xfs_agino_t) - 1));
1766 xfs_inobp_check(mp, ibp);
1770 * Point the bucket head pointer at the inode being inserted.
1773 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1774 offset = offsetof(xfs_agi_t, agi_unlinked) +
1775 (sizeof(xfs_agino_t) * bucket_index);
1776 xfs_trans_log_buf(tp, agibp, offset,
1777 (offset + sizeof(xfs_agino_t) - 1));
1782 * Pull the on-disk inode from the AGI unlinked list.
1795 xfs_agnumber_t agno;
1797 xfs_agino_t next_agino;
1798 xfs_buf_t *last_ibp;
1799 xfs_dinode_t *last_dip = NULL;
1801 int offset, last_offset = 0;
1805 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1808 * Get the agi buffer first. It ensures lock ordering
1811 error = xfs_read_agi(mp, tp, agno, &agibp);
1815 agi = XFS_BUF_TO_AGI(agibp);
1818 * Get the index into the agi hash table for the
1819 * list this inode will go on.
1821 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1823 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1824 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
1825 ASSERT(agi->agi_unlinked[bucket_index]);
1827 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
1829 * We're at the head of the list. Get the inode's
1830 * on-disk buffer to see if there is anyone after us
1831 * on the list. Only modify our next pointer if it
1832 * is not already NULLAGINO. This saves us the overhead
1833 * of dealing with the buffer when there is no need to
1836 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XFS_BUF_LOCK);
1839 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
1840 error, mp->m_fsname);
1843 next_agino = be32_to_cpu(dip->di_next_unlinked);
1844 ASSERT(next_agino != 0);
1845 if (next_agino != NULLAGINO) {
1846 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
1847 offset = ip->i_imap.im_boffset +
1848 offsetof(xfs_dinode_t, di_next_unlinked);
1849 xfs_trans_inode_buf(tp, ibp);
1850 xfs_trans_log_buf(tp, ibp, offset,
1851 (offset + sizeof(xfs_agino_t) - 1));
1852 xfs_inobp_check(mp, ibp);
1854 xfs_trans_brelse(tp, ibp);
1857 * Point the bucket head pointer at the next inode.
1859 ASSERT(next_agino != 0);
1860 ASSERT(next_agino != agino);
1861 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
1862 offset = offsetof(xfs_agi_t, agi_unlinked) +
1863 (sizeof(xfs_agino_t) * bucket_index);
1864 xfs_trans_log_buf(tp, agibp, offset,
1865 (offset + sizeof(xfs_agino_t) - 1));
1868 * We need to search the list for the inode being freed.
1870 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
1872 while (next_agino != agino) {
1874 * If the last inode wasn't the one pointing to
1875 * us, then release its buffer since we're not
1876 * going to do anything with it.
1878 if (last_ibp != NULL) {
1879 xfs_trans_brelse(tp, last_ibp);
1881 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
1882 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
1883 &last_ibp, &last_offset, 0);
1886 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
1887 error, mp->m_fsname);
1890 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
1891 ASSERT(next_agino != NULLAGINO);
1892 ASSERT(next_agino != 0);
1895 * Now last_ibp points to the buffer previous to us on
1896 * the unlinked list. Pull us from the list.
1898 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XFS_BUF_LOCK);
1901 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
1902 error, mp->m_fsname);
1905 next_agino = be32_to_cpu(dip->di_next_unlinked);
1906 ASSERT(next_agino != 0);
1907 ASSERT(next_agino != agino);
1908 if (next_agino != NULLAGINO) {
1909 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
1910 offset = ip->i_imap.im_boffset +
1911 offsetof(xfs_dinode_t, di_next_unlinked);
1912 xfs_trans_inode_buf(tp, ibp);
1913 xfs_trans_log_buf(tp, ibp, offset,
1914 (offset + sizeof(xfs_agino_t) - 1));
1915 xfs_inobp_check(mp, ibp);
1917 xfs_trans_brelse(tp, ibp);
1920 * Point the previous inode on the list to the next inode.
1922 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
1923 ASSERT(next_agino != 0);
1924 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
1925 xfs_trans_inode_buf(tp, last_ibp);
1926 xfs_trans_log_buf(tp, last_ibp, offset,
1927 (offset + sizeof(xfs_agino_t) - 1));
1928 xfs_inobp_check(mp, last_ibp);
1935 xfs_inode_t *free_ip,
1939 xfs_mount_t *mp = free_ip->i_mount;
1940 int blks_per_cluster;
1943 int i, j, found, pre_flushed;
1946 xfs_inode_t *ip, **ip_found;
1947 xfs_inode_log_item_t *iip;
1948 xfs_log_item_t *lip;
1949 struct xfs_perag *pag;
1951 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
1952 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
1953 blks_per_cluster = 1;
1954 ninodes = mp->m_sb.sb_inopblock;
1955 nbufs = XFS_IALLOC_BLOCKS(mp);
1957 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
1958 mp->m_sb.sb_blocksize;
1959 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
1960 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
1963 ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
1965 for (j = 0; j < nbufs; j++, inum += ninodes) {
1966 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
1967 XFS_INO_TO_AGBNO(mp, inum));
1971 * Look for each inode in memory and attempt to lock it,
1972 * we can be racing with flush and tail pushing here.
1973 * any inode we get the locks on, add to an array of
1974 * inode items to process later.
1976 * The get the buffer lock, we could beat a flush
1977 * or tail pushing thread to the lock here, in which
1978 * case they will go looking for the inode buffer
1979 * and fail, we need some other form of interlock
1983 for (i = 0; i < ninodes; i++) {
1984 read_lock(&pag->pag_ici_lock);
1985 ip = radix_tree_lookup(&pag->pag_ici_root,
1986 XFS_INO_TO_AGINO(mp, (inum + i)));
1988 /* Inode not in memory or we found it already,
1991 if (!ip || xfs_iflags_test(ip, XFS_ISTALE)) {
1992 read_unlock(&pag->pag_ici_lock);
1996 if (xfs_inode_clean(ip)) {
1997 read_unlock(&pag->pag_ici_lock);
2001 /* If we can get the locks then add it to the
2002 * list, otherwise by the time we get the bp lock
2003 * below it will already be attached to the
2007 /* This inode will already be locked - by us, lets
2011 if (ip == free_ip) {
2012 if (xfs_iflock_nowait(ip)) {
2013 xfs_iflags_set(ip, XFS_ISTALE);
2014 if (xfs_inode_clean(ip)) {
2017 ip_found[found++] = ip;
2020 read_unlock(&pag->pag_ici_lock);
2024 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2025 if (xfs_iflock_nowait(ip)) {
2026 xfs_iflags_set(ip, XFS_ISTALE);
2028 if (xfs_inode_clean(ip)) {
2030 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2032 ip_found[found++] = ip;
2035 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2038 read_unlock(&pag->pag_ici_lock);
2041 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2042 mp->m_bsize * blks_per_cluster,
2046 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
2048 if (lip->li_type == XFS_LI_INODE) {
2049 iip = (xfs_inode_log_item_t *)lip;
2050 ASSERT(iip->ili_logged == 1);
2051 lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
2052 xfs_trans_ail_copy_lsn(mp->m_ail,
2053 &iip->ili_flush_lsn,
2054 &iip->ili_item.li_lsn);
2055 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2058 lip = lip->li_bio_list;
2061 for (i = 0; i < found; i++) {
2066 ip->i_update_core = 0;
2068 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2072 iip->ili_last_fields = iip->ili_format.ilf_fields;
2073 iip->ili_format.ilf_fields = 0;
2074 iip->ili_logged = 1;
2075 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2076 &iip->ili_item.li_lsn);
2078 xfs_buf_attach_iodone(bp,
2079 (void(*)(xfs_buf_t*,xfs_log_item_t*))
2080 xfs_istale_done, (xfs_log_item_t *)iip);
2081 if (ip != free_ip) {
2082 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2086 if (found || pre_flushed)
2087 xfs_trans_stale_inode_buf(tp, bp);
2088 xfs_trans_binval(tp, bp);
2091 kmem_free(ip_found);
2096 * This is called to return an inode to the inode free list.
2097 * The inode should already be truncated to 0 length and have
2098 * no pages associated with it. This routine also assumes that
2099 * the inode is already a part of the transaction.
2101 * The on-disk copy of the inode will have been added to the list
2102 * of unlinked inodes in the AGI. We need to remove the inode from
2103 * that list atomically with respect to freeing it here.
2109 xfs_bmap_free_t *flist)
2113 xfs_ino_t first_ino;
2117 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2118 ASSERT(ip->i_transp == tp);
2119 ASSERT(ip->i_d.di_nlink == 0);
2120 ASSERT(ip->i_d.di_nextents == 0);
2121 ASSERT(ip->i_d.di_anextents == 0);
2122 ASSERT((ip->i_d.di_size == 0 && ip->i_size == 0) ||
2123 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2124 ASSERT(ip->i_d.di_nblocks == 0);
2127 * Pull the on-disk inode from the AGI unlinked list.
2129 error = xfs_iunlink_remove(tp, ip);
2134 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2138 ip->i_d.di_mode = 0; /* mark incore inode as free */
2139 ip->i_d.di_flags = 0;
2140 ip->i_d.di_dmevmask = 0;
2141 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2142 ip->i_df.if_ext_max =
2143 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2144 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2145 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2147 * Bump the generation count so no one will be confused
2148 * by reincarnations of this inode.
2152 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2154 error = xfs_itobp(ip->i_mount, tp, ip, &dip, &ibp, XFS_BUF_LOCK);
2159 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2160 * from picking up this inode when it is reclaimed (its incore state
2161 * initialzed but not flushed to disk yet). The in-core di_mode is
2162 * already cleared and a corresponding transaction logged.
2163 * The hack here just synchronizes the in-core to on-disk
2164 * di_mode value in advance before the actual inode sync to disk.
2165 * This is OK because the inode is already unlinked and would never
2166 * change its di_mode again for this inode generation.
2167 * This is a temporary hack that would require a proper fix
2173 xfs_ifree_cluster(ip, tp, first_ino);
2180 * Reallocate the space for if_broot based on the number of records
2181 * being added or deleted as indicated in rec_diff. Move the records
2182 * and pointers in if_broot to fit the new size. When shrinking this
2183 * will eliminate holes between the records and pointers created by
2184 * the caller. When growing this will create holes to be filled in
2187 * The caller must not request to add more records than would fit in
2188 * the on-disk inode root. If the if_broot is currently NULL, then
2189 * if we adding records one will be allocated. The caller must also
2190 * not request that the number of records go below zero, although
2191 * it can go to zero.
2193 * ip -- the inode whose if_broot area is changing
2194 * ext_diff -- the change in the number of records, positive or negative,
2195 * requested for the if_broot array.
2203 struct xfs_mount *mp = ip->i_mount;
2206 struct xfs_btree_block *new_broot;
2213 * Handle the degenerate case quietly.
2215 if (rec_diff == 0) {
2219 ifp = XFS_IFORK_PTR(ip, whichfork);
2222 * If there wasn't any memory allocated before, just
2223 * allocate it now and get out.
2225 if (ifp->if_broot_bytes == 0) {
2226 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
2227 ifp->if_broot = kmem_alloc(new_size, KM_SLEEP);
2228 ifp->if_broot_bytes = (int)new_size;
2233 * If there is already an existing if_broot, then we need
2234 * to realloc() it and shift the pointers to their new
2235 * location. The records don't change location because
2236 * they are kept butted up against the btree block header.
2238 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
2239 new_max = cur_max + rec_diff;
2240 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2241 ifp->if_broot = kmem_realloc(ifp->if_broot, new_size,
2242 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2244 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2245 ifp->if_broot_bytes);
2246 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2248 ifp->if_broot_bytes = (int)new_size;
2249 ASSERT(ifp->if_broot_bytes <=
2250 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2251 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2256 * rec_diff is less than 0. In this case, we are shrinking the
2257 * if_broot buffer. It must already exist. If we go to zero
2258 * records, just get rid of the root and clear the status bit.
2260 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
2261 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
2262 new_max = cur_max + rec_diff;
2263 ASSERT(new_max >= 0);
2265 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2269 new_broot = kmem_alloc(new_size, KM_SLEEP);
2271 * First copy over the btree block header.
2273 memcpy(new_broot, ifp->if_broot, XFS_BTREE_LBLOCK_LEN);
2276 ifp->if_flags &= ~XFS_IFBROOT;
2280 * Only copy the records and pointers if there are any.
2284 * First copy the records.
2286 op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1);
2287 np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1);
2288 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2291 * Then copy the pointers.
2293 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2294 ifp->if_broot_bytes);
2295 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1,
2297 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2299 kmem_free(ifp->if_broot);
2300 ifp->if_broot = new_broot;
2301 ifp->if_broot_bytes = (int)new_size;
2302 ASSERT(ifp->if_broot_bytes <=
2303 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2309 * This is called when the amount of space needed for if_data
2310 * is increased or decreased. The change in size is indicated by
2311 * the number of bytes that need to be added or deleted in the
2312 * byte_diff parameter.
2314 * If the amount of space needed has decreased below the size of the
2315 * inline buffer, then switch to using the inline buffer. Otherwise,
2316 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2317 * to what is needed.
2319 * ip -- the inode whose if_data area is changing
2320 * byte_diff -- the change in the number of bytes, positive or negative,
2321 * requested for the if_data array.
2333 if (byte_diff == 0) {
2337 ifp = XFS_IFORK_PTR(ip, whichfork);
2338 new_size = (int)ifp->if_bytes + byte_diff;
2339 ASSERT(new_size >= 0);
2341 if (new_size == 0) {
2342 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2343 kmem_free(ifp->if_u1.if_data);
2345 ifp->if_u1.if_data = NULL;
2347 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2349 * If the valid extents/data can fit in if_inline_ext/data,
2350 * copy them from the malloc'd vector and free it.
2352 if (ifp->if_u1.if_data == NULL) {
2353 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2354 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2355 ASSERT(ifp->if_real_bytes != 0);
2356 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2358 kmem_free(ifp->if_u1.if_data);
2359 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2364 * Stuck with malloc/realloc.
2365 * For inline data, the underlying buffer must be
2366 * a multiple of 4 bytes in size so that it can be
2367 * logged and stay on word boundaries. We enforce
2370 real_size = roundup(new_size, 4);
2371 if (ifp->if_u1.if_data == NULL) {
2372 ASSERT(ifp->if_real_bytes == 0);
2373 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2374 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2376 * Only do the realloc if the underlying size
2377 * is really changing.
2379 if (ifp->if_real_bytes != real_size) {
2380 ifp->if_u1.if_data =
2381 kmem_realloc(ifp->if_u1.if_data,
2387 ASSERT(ifp->if_real_bytes == 0);
2388 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2389 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2393 ifp->if_real_bytes = real_size;
2394 ifp->if_bytes = new_size;
2395 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2405 ifp = XFS_IFORK_PTR(ip, whichfork);
2406 if (ifp->if_broot != NULL) {
2407 kmem_free(ifp->if_broot);
2408 ifp->if_broot = NULL;
2412 * If the format is local, then we can't have an extents
2413 * array so just look for an inline data array. If we're
2414 * not local then we may or may not have an extents list,
2415 * so check and free it up if we do.
2417 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2418 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2419 (ifp->if_u1.if_data != NULL)) {
2420 ASSERT(ifp->if_real_bytes != 0);
2421 kmem_free(ifp->if_u1.if_data);
2422 ifp->if_u1.if_data = NULL;
2423 ifp->if_real_bytes = 0;
2425 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2426 ((ifp->if_flags & XFS_IFEXTIREC) ||
2427 ((ifp->if_u1.if_extents != NULL) &&
2428 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
2429 ASSERT(ifp->if_real_bytes != 0);
2430 xfs_iext_destroy(ifp);
2432 ASSERT(ifp->if_u1.if_extents == NULL ||
2433 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2434 ASSERT(ifp->if_real_bytes == 0);
2435 if (whichfork == XFS_ATTR_FORK) {
2436 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2442 * Increment the pin count of the given buffer.
2443 * This value is protected by ipinlock spinlock in the mount structure.
2449 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2451 atomic_inc(&ip->i_pincount);
2455 * Decrement the pin count of the given inode, and wake up
2456 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2457 * inode must have been previously pinned with a call to xfs_ipin().
2463 ASSERT(atomic_read(&ip->i_pincount) > 0);
2465 if (atomic_dec_and_test(&ip->i_pincount))
2466 wake_up(&ip->i_ipin_wait);
2470 * This is called to unpin an inode. It can be directed to wait or to return
2471 * immediately without waiting for the inode to be unpinned. The caller must
2472 * have the inode locked in at least shared mode so that the buffer cannot be
2473 * subsequently pinned once someone is waiting for it to be unpinned.
2480 xfs_inode_log_item_t *iip = ip->i_itemp;
2482 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2483 if (atomic_read(&ip->i_pincount) == 0)
2486 /* Give the log a push to start the unpinning I/O */
2487 xfs_log_force(ip->i_mount, (iip && iip->ili_last_lsn) ?
2488 iip->ili_last_lsn : 0, XFS_LOG_FORCE);
2490 wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
2497 __xfs_iunpin_wait(ip, 1);
2504 __xfs_iunpin_wait(ip, 0);
2509 * xfs_iextents_copy()
2511 * This is called to copy the REAL extents (as opposed to the delayed
2512 * allocation extents) from the inode into the given buffer. It
2513 * returns the number of bytes copied into the buffer.
2515 * If there are no delayed allocation extents, then we can just
2516 * memcpy() the extents into the buffer. Otherwise, we need to
2517 * examine each extent in turn and skip those which are delayed.
2529 xfs_fsblock_t start_block;
2531 ifp = XFS_IFORK_PTR(ip, whichfork);
2532 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2533 ASSERT(ifp->if_bytes > 0);
2535 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2536 XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork);
2540 * There are some delayed allocation extents in the
2541 * inode, so copy the extents one at a time and skip
2542 * the delayed ones. There must be at least one
2543 * non-delayed extent.
2546 for (i = 0; i < nrecs; i++) {
2547 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
2548 start_block = xfs_bmbt_get_startblock(ep);
2549 if (isnullstartblock(start_block)) {
2551 * It's a delayed allocation extent, so skip it.
2556 /* Translate to on disk format */
2557 put_unaligned(cpu_to_be64(ep->l0), &dp->l0);
2558 put_unaligned(cpu_to_be64(ep->l1), &dp->l1);
2562 ASSERT(copied != 0);
2563 xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip));
2565 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2569 * Each of the following cases stores data into the same region
2570 * of the on-disk inode, so only one of them can be valid at
2571 * any given time. While it is possible to have conflicting formats
2572 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2573 * in EXTENTS format, this can only happen when the fork has
2574 * changed formats after being modified but before being flushed.
2575 * In these cases, the format always takes precedence, because the
2576 * format indicates the current state of the fork.
2583 xfs_inode_log_item_t *iip,
2590 #ifdef XFS_TRANS_DEBUG
2593 static const short brootflag[2] =
2594 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2595 static const short dataflag[2] =
2596 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2597 static const short extflag[2] =
2598 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2602 ifp = XFS_IFORK_PTR(ip, whichfork);
2604 * This can happen if we gave up in iformat in an error path,
2605 * for the attribute fork.
2608 ASSERT(whichfork == XFS_ATTR_FORK);
2611 cp = XFS_DFORK_PTR(dip, whichfork);
2613 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2614 case XFS_DINODE_FMT_LOCAL:
2615 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2616 (ifp->if_bytes > 0)) {
2617 ASSERT(ifp->if_u1.if_data != NULL);
2618 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2619 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2623 case XFS_DINODE_FMT_EXTENTS:
2624 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2625 !(iip->ili_format.ilf_fields & extflag[whichfork]));
2626 ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
2627 (ifp->if_bytes == 0));
2628 ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
2629 (ifp->if_bytes > 0));
2630 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
2631 (ifp->if_bytes > 0)) {
2632 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2633 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2638 case XFS_DINODE_FMT_BTREE:
2639 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
2640 (ifp->if_broot_bytes > 0)) {
2641 ASSERT(ifp->if_broot != NULL);
2642 ASSERT(ifp->if_broot_bytes <=
2643 (XFS_IFORK_SIZE(ip, whichfork) +
2644 XFS_BROOT_SIZE_ADJ));
2645 xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes,
2646 (xfs_bmdr_block_t *)cp,
2647 XFS_DFORK_SIZE(dip, mp, whichfork));
2651 case XFS_DINODE_FMT_DEV:
2652 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
2653 ASSERT(whichfork == XFS_DATA_FORK);
2654 xfs_dinode_put_rdev(dip, ip->i_df.if_u2.if_rdev);
2658 case XFS_DINODE_FMT_UUID:
2659 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
2660 ASSERT(whichfork == XFS_DATA_FORK);
2661 memcpy(XFS_DFORK_DPTR(dip),
2662 &ip->i_df.if_u2.if_uuid,
2678 xfs_mount_t *mp = ip->i_mount;
2679 struct xfs_perag *pag;
2680 unsigned long first_index, mask;
2681 unsigned long inodes_per_cluster;
2683 xfs_inode_t **ilist;
2690 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2691 ASSERT(pag->pagi_inodeok);
2692 ASSERT(pag->pag_ici_init);
2694 inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog;
2695 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
2696 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
2700 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
2701 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
2702 read_lock(&pag->pag_ici_lock);
2703 /* really need a gang lookup range call here */
2704 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
2705 first_index, inodes_per_cluster);
2709 for (i = 0; i < nr_found; i++) {
2713 /* if the inode lies outside this cluster, we're done. */
2714 if ((XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index)
2717 * Do an un-protected check to see if the inode is dirty and
2718 * is a candidate for flushing. These checks will be repeated
2719 * later after the appropriate locks are acquired.
2721 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
2725 * Try to get locks. If any are unavailable or it is pinned,
2726 * then this inode cannot be flushed and is skipped.
2729 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
2731 if (!xfs_iflock_nowait(iq)) {
2732 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2735 if (xfs_ipincount(iq)) {
2737 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2742 * arriving here means that this inode can be flushed. First
2743 * re-check that it's dirty before flushing.
2745 if (!xfs_inode_clean(iq)) {
2747 error = xfs_iflush_int(iq, bp);
2749 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2750 goto cluster_corrupt_out;
2756 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2760 XFS_STATS_INC(xs_icluster_flushcnt);
2761 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
2765 read_unlock(&pag->pag_ici_lock);
2772 cluster_corrupt_out:
2774 * Corruption detected in the clustering loop. Invalidate the
2775 * inode buffer and shut down the filesystem.
2777 read_unlock(&pag->pag_ici_lock);
2779 * Clean up the buffer. If it was B_DELWRI, just release it --
2780 * brelse can handle it with no problems. If not, shut down the
2781 * filesystem before releasing the buffer.
2783 bufwasdelwri = XFS_BUF_ISDELAYWRITE(bp);
2787 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2789 if (!bufwasdelwri) {
2791 * Just like incore_relse: if we have b_iodone functions,
2792 * mark the buffer as an error and call them. Otherwise
2793 * mark it as stale and brelse.
2795 if (XFS_BUF_IODONE_FUNC(bp)) {
2796 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
2799 XFS_BUF_ERROR(bp,EIO);
2808 * Unlocks the flush lock
2810 xfs_iflush_abort(iq);
2813 return XFS_ERROR(EFSCORRUPTED);
2817 * xfs_iflush() will write a modified inode's changes out to the
2818 * inode's on disk home. The caller must have the inode lock held
2819 * in at least shared mode and the inode flush completion must be
2820 * active as well. The inode lock will still be held upon return from
2821 * the call and the caller is free to unlock it.
2822 * The inode flush will be completed when the inode reaches the disk.
2823 * The flags indicate how the inode's buffer should be written out.
2830 xfs_inode_log_item_t *iip;
2835 int noblock = (flags == XFS_IFLUSH_ASYNC_NOBLOCK);
2836 enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
2838 XFS_STATS_INC(xs_iflush_count);
2840 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2841 ASSERT(!completion_done(&ip->i_flush));
2842 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2843 ip->i_d.di_nextents > ip->i_df.if_ext_max);
2849 * If the inode isn't dirty, then just release the inode flush lock and
2852 if (xfs_inode_clean(ip)) {
2858 * We can't flush the inode until it is unpinned, so wait for it if we
2859 * are allowed to block. We know noone new can pin it, because we are
2860 * holding the inode lock shared and you need to hold it exclusively to
2863 * If we are not allowed to block, force the log out asynchronously so
2864 * that when we come back the inode will be unpinned. If other inodes
2865 * in the same cluster are dirty, they will probably write the inode
2866 * out for us if they occur after the log force completes.
2868 if (noblock && xfs_ipincount(ip)) {
2869 xfs_iunpin_nowait(ip);
2873 xfs_iunpin_wait(ip);
2876 * For stale inodes we cannot rely on the backing buffer remaining
2877 * stale in cache for the remaining life of the stale inode and so
2878 * xfs_itobp() below may give us a buffer that no longer contains
2879 * inodes below. We have to check this after ensuring the inode is
2880 * unpinned so that it is safe to reclaim the stale inode after the
2883 if (xfs_iflags_test(ip, XFS_ISTALE)) {
2889 * This may have been unpinned because the filesystem is shutting
2890 * down forcibly. If that's the case we must not write this inode
2891 * to disk, because the log record didn't make it to disk!
2893 if (XFS_FORCED_SHUTDOWN(mp)) {
2894 ip->i_update_core = 0;
2896 iip->ili_format.ilf_fields = 0;
2898 return XFS_ERROR(EIO);
2902 * Decide how buffer will be flushed out. This is done before
2903 * the call to xfs_iflush_int because this field is zeroed by it.
2905 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
2907 * Flush out the inode buffer according to the directions
2908 * of the caller. In the cases where the caller has given
2909 * us a choice choose the non-delwri case. This is because
2910 * the inode is in the AIL and we need to get it out soon.
2913 case XFS_IFLUSH_SYNC:
2914 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
2917 case XFS_IFLUSH_ASYNC_NOBLOCK:
2918 case XFS_IFLUSH_ASYNC:
2919 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
2922 case XFS_IFLUSH_DELWRI:
2932 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
2933 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
2934 case XFS_IFLUSH_DELWRI:
2937 case XFS_IFLUSH_ASYNC_NOBLOCK:
2938 case XFS_IFLUSH_ASYNC:
2941 case XFS_IFLUSH_SYNC:
2952 * Get the buffer containing the on-disk inode.
2954 error = xfs_itobp(mp, NULL, ip, &dip, &bp,
2955 noblock ? XFS_BUF_TRYLOCK : XFS_BUF_LOCK);
2962 * First flush out the inode that xfs_iflush was called with.
2964 error = xfs_iflush_int(ip, bp);
2969 * If the buffer is pinned then push on the log now so we won't
2970 * get stuck waiting in the write for too long.
2972 if (XFS_BUF_ISPINNED(bp))
2973 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
2977 * see if other inodes can be gathered into this write
2979 error = xfs_iflush_cluster(ip, bp);
2981 goto cluster_corrupt_out;
2983 if (flags & INT_DELWRI) {
2984 xfs_bdwrite(mp, bp);
2985 } else if (flags & INT_ASYNC) {
2986 error = xfs_bawrite(mp, bp);
2988 error = xfs_bwrite(mp, bp);
2994 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2995 cluster_corrupt_out:
2997 * Unlocks the flush lock
2999 xfs_iflush_abort(ip);
3000 return XFS_ERROR(EFSCORRUPTED);
3009 xfs_inode_log_item_t *iip;
3012 #ifdef XFS_TRANS_DEBUG
3016 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3017 ASSERT(!completion_done(&ip->i_flush));
3018 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3019 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3026 * If the inode isn't dirty, then just release the inode
3027 * flush lock and do nothing.
3029 if (xfs_inode_clean(ip)) {
3034 /* set *dip = inode's place in the buffer */
3035 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
3038 * Clear i_update_core before copying out the data.
3039 * This is for coordination with our timestamp updates
3040 * that don't hold the inode lock. They will always
3041 * update the timestamps BEFORE setting i_update_core,
3042 * so if we clear i_update_core after they set it we
3043 * are guaranteed to see their updates to the timestamps.
3044 * I believe that this depends on strongly ordered memory
3045 * semantics, but we have that. We use the SYNCHRONIZE
3046 * macro to make sure that the compiler does not reorder
3047 * the i_update_core access below the data copy below.
3049 ip->i_update_core = 0;
3053 * Make sure to get the latest timestamps from the Linux inode.
3055 xfs_synchronize_times(ip);
3057 if (XFS_TEST_ERROR(be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC,
3058 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3059 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3060 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3061 ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3064 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3065 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3066 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3067 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3068 ip->i_ino, ip, ip->i_d.di_magic);
3071 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3073 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3074 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3075 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3076 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3077 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3081 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
3083 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3084 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3085 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3086 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3087 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3088 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3093 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3094 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3095 XFS_RANDOM_IFLUSH_5)) {
3096 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3097 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3099 ip->i_d.di_nextents + ip->i_d.di_anextents,
3104 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3105 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3106 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3107 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3108 ip->i_ino, ip->i_d.di_forkoff, ip);
3112 * bump the flush iteration count, used to detect flushes which
3113 * postdate a log record during recovery.
3116 ip->i_d.di_flushiter++;
3119 * Copy the dirty parts of the inode into the on-disk
3120 * inode. We always copy out the core of the inode,
3121 * because if the inode is dirty at all the core must
3124 xfs_dinode_to_disk(dip, &ip->i_d);
3126 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3127 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3128 ip->i_d.di_flushiter = 0;
3131 * If this is really an old format inode and the superblock version
3132 * has not been updated to support only new format inodes, then
3133 * convert back to the old inode format. If the superblock version
3134 * has been updated, then make the conversion permanent.
3136 ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
3137 if (ip->i_d.di_version == 1) {
3138 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
3142 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3143 dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink);
3146 * The superblock version has already been bumped,
3147 * so just make the conversion to the new inode
3150 ip->i_d.di_version = 2;
3151 dip->di_version = 2;
3152 ip->i_d.di_onlink = 0;
3154 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3155 memset(&(dip->di_pad[0]), 0,
3156 sizeof(dip->di_pad));
3157 ASSERT(ip->i_d.di_projid == 0);
3161 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
3162 if (XFS_IFORK_Q(ip))
3163 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3164 xfs_inobp_check(mp, bp);
3167 * We've recorded everything logged in the inode, so we'd
3168 * like to clear the ilf_fields bits so we don't log and
3169 * flush things unnecessarily. However, we can't stop
3170 * logging all this information until the data we've copied
3171 * into the disk buffer is written to disk. If we did we might
3172 * overwrite the copy of the inode in the log with all the
3173 * data after re-logging only part of it, and in the face of
3174 * a crash we wouldn't have all the data we need to recover.
3176 * What we do is move the bits to the ili_last_fields field.
3177 * When logging the inode, these bits are moved back to the
3178 * ilf_fields field. In the xfs_iflush_done() routine we
3179 * clear ili_last_fields, since we know that the information
3180 * those bits represent is permanently on disk. As long as
3181 * the flush completes before the inode is logged again, then
3182 * both ilf_fields and ili_last_fields will be cleared.
3184 * We can play with the ilf_fields bits here, because the inode
3185 * lock must be held exclusively in order to set bits there
3186 * and the flush lock protects the ili_last_fields bits.
3187 * Set ili_logged so the flush done
3188 * routine can tell whether or not to look in the AIL.
3189 * Also, store the current LSN of the inode so that we can tell
3190 * whether the item has moved in the AIL from xfs_iflush_done().
3191 * In order to read the lsn we need the AIL lock, because
3192 * it is a 64 bit value that cannot be read atomically.
3194 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3195 iip->ili_last_fields = iip->ili_format.ilf_fields;
3196 iip->ili_format.ilf_fields = 0;
3197 iip->ili_logged = 1;
3199 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3200 &iip->ili_item.li_lsn);
3203 * Attach the function xfs_iflush_done to the inode's
3204 * buffer. This will remove the inode from the AIL
3205 * and unlock the inode's flush lock when the inode is
3206 * completely written to disk.
3208 xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
3209 xfs_iflush_done, (xfs_log_item_t *)iip);
3211 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3212 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3215 * We're flushing an inode which is not in the AIL and has
3216 * not been logged but has i_update_core set. For this
3217 * case we can use a B_DELWRI flush and immediately drop
3218 * the inode flush lock because we can avoid the whole
3219 * AIL state thing. It's OK to drop the flush lock now,
3220 * because we've already locked the buffer and to do anything
3221 * you really need both.
3224 ASSERT(iip->ili_logged == 0);
3225 ASSERT(iip->ili_last_fields == 0);
3226 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3234 return XFS_ERROR(EFSCORRUPTED);
3238 * Return a pointer to the extent record at file index idx.
3240 xfs_bmbt_rec_host_t *
3242 xfs_ifork_t *ifp, /* inode fork pointer */
3243 xfs_extnum_t idx) /* index of target extent */
3246 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
3247 return ifp->if_u1.if_ext_irec->er_extbuf;
3248 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3249 xfs_ext_irec_t *erp; /* irec pointer */
3250 int erp_idx = 0; /* irec index */
3251 xfs_extnum_t page_idx = idx; /* ext index in target list */
3253 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3254 return &erp->er_extbuf[page_idx];
3255 } else if (ifp->if_bytes) {
3256 return &ifp->if_u1.if_extents[idx];
3263 * Insert new item(s) into the extent records for incore inode
3264 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3268 xfs_inode_t *ip, /* incore inode pointer */
3269 xfs_extnum_t idx, /* starting index of new items */
3270 xfs_extnum_t count, /* number of inserted items */
3271 xfs_bmbt_irec_t *new, /* items to insert */
3272 int state) /* type of extent conversion */
3274 xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
3275 xfs_extnum_t i; /* extent record index */
3277 trace_xfs_iext_insert(ip, idx, new, state, _RET_IP_);
3279 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3280 xfs_iext_add(ifp, idx, count);
3281 for (i = idx; i < idx + count; i++, new++)
3282 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new);
3286 * This is called when the amount of space required for incore file
3287 * extents needs to be increased. The ext_diff parameter stores the
3288 * number of new extents being added and the idx parameter contains
3289 * the extent index where the new extents will be added. If the new
3290 * extents are being appended, then we just need to (re)allocate and
3291 * initialize the space. Otherwise, if the new extents are being
3292 * inserted into the middle of the existing entries, a bit more work
3293 * is required to make room for the new extents to be inserted. The
3294 * caller is responsible for filling in the new extent entries upon
3299 xfs_ifork_t *ifp, /* inode fork pointer */
3300 xfs_extnum_t idx, /* index to begin adding exts */
3301 int ext_diff) /* number of extents to add */
3303 int byte_diff; /* new bytes being added */
3304 int new_size; /* size of extents after adding */
3305 xfs_extnum_t nextents; /* number of extents in file */
3307 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3308 ASSERT((idx >= 0) && (idx <= nextents));
3309 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
3310 new_size = ifp->if_bytes + byte_diff;
3312 * If the new number of extents (nextents + ext_diff)
3313 * fits inside the inode, then continue to use the inline
3316 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
3317 if (idx < nextents) {
3318 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
3319 &ifp->if_u2.if_inline_ext[idx],
3320 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3321 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
3323 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3324 ifp->if_real_bytes = 0;
3325 ifp->if_lastex = nextents + ext_diff;
3328 * Otherwise use a linear (direct) extent list.
3329 * If the extents are currently inside the inode,
3330 * xfs_iext_realloc_direct will switch us from
3331 * inline to direct extent allocation mode.
3333 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
3334 xfs_iext_realloc_direct(ifp, new_size);
3335 if (idx < nextents) {
3336 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
3337 &ifp->if_u1.if_extents[idx],
3338 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3339 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
3342 /* Indirection array */
3344 xfs_ext_irec_t *erp;
3348 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
3349 if (ifp->if_flags & XFS_IFEXTIREC) {
3350 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3352 xfs_iext_irec_init(ifp);
3353 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3354 erp = ifp->if_u1.if_ext_irec;
3356 /* Extents fit in target extent page */
3357 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3358 if (page_idx < erp->er_extcount) {
3359 memmove(&erp->er_extbuf[page_idx + ext_diff],
3360 &erp->er_extbuf[page_idx],
3361 (erp->er_extcount - page_idx) *
3362 sizeof(xfs_bmbt_rec_t));
3363 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3365 erp->er_extcount += ext_diff;
3366 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3368 /* Insert a new extent page */
3370 xfs_iext_add_indirect_multi(ifp,
3371 erp_idx, page_idx, ext_diff);
3374 * If extent(s) are being appended to the last page in
3375 * the indirection array and the new extent(s) don't fit
3376 * in the page, then erp is NULL and erp_idx is set to
3377 * the next index needed in the indirection array.
3380 int count = ext_diff;
3383 erp = xfs_iext_irec_new(ifp, erp_idx);
3384 erp->er_extcount = count;
3385 count -= MIN(count, (int)XFS_LINEAR_EXTS);
3392 ifp->if_bytes = new_size;
3396 * This is called when incore extents are being added to the indirection
3397 * array and the new extents do not fit in the target extent list. The
3398 * erp_idx parameter contains the irec index for the target extent list
3399 * in the indirection array, and the idx parameter contains the extent
3400 * index within the list. The number of extents being added is stored
3401 * in the count parameter.
3403 * |-------| |-------|
3404 * | | | | idx - number of extents before idx
3406 * | | | | count - number of extents being inserted at idx
3407 * |-------| |-------|
3408 * | count | | nex2 | nex2 - number of extents after idx + count
3409 * |-------| |-------|
3412 xfs_iext_add_indirect_multi(
3413 xfs_ifork_t *ifp, /* inode fork pointer */
3414 int erp_idx, /* target extent irec index */
3415 xfs_extnum_t idx, /* index within target list */
3416 int count) /* new extents being added */
3418 int byte_diff; /* new bytes being added */
3419 xfs_ext_irec_t *erp; /* pointer to irec entry */
3420 xfs_extnum_t ext_diff; /* number of extents to add */
3421 xfs_extnum_t ext_cnt; /* new extents still needed */
3422 xfs_extnum_t nex2; /* extents after idx + count */
3423 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3424 int nlists; /* number of irec's (lists) */
3426 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3427 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3428 nex2 = erp->er_extcount - idx;
3429 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3432 * Save second part of target extent list
3433 * (all extents past */
3435 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3436 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS);
3437 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3438 erp->er_extcount -= nex2;
3439 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3440 memset(&erp->er_extbuf[idx], 0, byte_diff);
3444 * Add the new extents to the end of the target
3445 * list, then allocate new irec record(s) and
3446 * extent buffer(s) as needed to store the rest
3447 * of the new extents.
3450 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3452 erp->er_extcount += ext_diff;
3453 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3454 ext_cnt -= ext_diff;
3458 erp = xfs_iext_irec_new(ifp, erp_idx);
3459 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3460 erp->er_extcount = ext_diff;
3461 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3462 ext_cnt -= ext_diff;
3465 /* Add nex2 extents back to indirection array */
3467 xfs_extnum_t ext_avail;
3470 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3471 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3474 * If nex2 extents fit in the current page, append
3475 * nex2_ep after the new extents.
3477 if (nex2 <= ext_avail) {
3478 i = erp->er_extcount;
3481 * Otherwise, check if space is available in the
3484 else if ((erp_idx < nlists - 1) &&
3485 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3486 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3489 /* Create a hole for nex2 extents */
3490 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3491 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3494 * Final choice, create a new extent page for
3499 erp = xfs_iext_irec_new(ifp, erp_idx);
3501 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
3503 erp->er_extcount += nex2;
3504 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3509 * This is called when the amount of space required for incore file
3510 * extents needs to be decreased. The ext_diff parameter stores the
3511 * number of extents to be removed and the idx parameter contains
3512 * the extent index where the extents will be removed from.
3514 * If the amount of space needed has decreased below the linear
3515 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3516 * extent array. Otherwise, use kmem_realloc() to adjust the
3517 * size to what is needed.
3521 xfs_inode_t *ip, /* incore inode pointer */
3522 xfs_extnum_t idx, /* index to begin removing exts */
3523 int ext_diff, /* number of extents to remove */
3524 int state) /* type of extent conversion */
3526 xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
3527 xfs_extnum_t nextents; /* number of extents in file */
3528 int new_size; /* size of extents after removal */
3530 trace_xfs_iext_remove(ip, idx, state, _RET_IP_);
3532 ASSERT(ext_diff > 0);
3533 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3534 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3536 if (new_size == 0) {
3537 xfs_iext_destroy(ifp);
3538 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3539 xfs_iext_remove_indirect(ifp, idx, ext_diff);
3540 } else if (ifp->if_real_bytes) {
3541 xfs_iext_remove_direct(ifp, idx, ext_diff);
3543 xfs_iext_remove_inline(ifp, idx, ext_diff);
3545 ifp->if_bytes = new_size;
3549 * This removes ext_diff extents from the inline buffer, beginning
3550 * at extent index idx.
3553 xfs_iext_remove_inline(
3554 xfs_ifork_t *ifp, /* inode fork pointer */
3555 xfs_extnum_t idx, /* index to begin removing exts */
3556 int ext_diff) /* number of extents to remove */
3558 int nextents; /* number of extents in file */
3560 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3561 ASSERT(idx < XFS_INLINE_EXTS);
3562 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3563 ASSERT(((nextents - ext_diff) > 0) &&
3564 (nextents - ext_diff) < XFS_INLINE_EXTS);
3566 if (idx + ext_diff < nextents) {
3567 memmove(&ifp->if_u2.if_inline_ext[idx],
3568 &ifp->if_u2.if_inline_ext[idx + ext_diff],
3569 (nextents - (idx + ext_diff)) *
3570 sizeof(xfs_bmbt_rec_t));
3571 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
3572 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3574 memset(&ifp->if_u2.if_inline_ext[idx], 0,
3575 ext_diff * sizeof(xfs_bmbt_rec_t));
3580 * This removes ext_diff extents from a linear (direct) extent list,
3581 * beginning at extent index idx. If the extents are being removed
3582 * from the end of the list (ie. truncate) then we just need to re-
3583 * allocate the list to remove the extra space. Otherwise, if the
3584 * extents are being removed from the middle of the existing extent
3585 * entries, then we first need to move the extent records beginning
3586 * at idx + ext_diff up in the list to overwrite the records being
3587 * removed, then remove the extra space via kmem_realloc.
3590 xfs_iext_remove_direct(
3591 xfs_ifork_t *ifp, /* inode fork pointer */
3592 xfs_extnum_t idx, /* index to begin removing exts */
3593 int ext_diff) /* number of extents to remove */
3595 xfs_extnum_t nextents; /* number of extents in file */
3596 int new_size; /* size of extents after removal */
3598 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3599 new_size = ifp->if_bytes -
3600 (ext_diff * sizeof(xfs_bmbt_rec_t));
3601 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3603 if (new_size == 0) {
3604 xfs_iext_destroy(ifp);
3607 /* Move extents up in the list (if needed) */
3608 if (idx + ext_diff < nextents) {
3609 memmove(&ifp->if_u1.if_extents[idx],
3610 &ifp->if_u1.if_extents[idx + ext_diff],
3611 (nextents - (idx + ext_diff)) *
3612 sizeof(xfs_bmbt_rec_t));
3614 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
3615 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3617 * Reallocate the direct extent list. If the extents
3618 * will fit inside the inode then xfs_iext_realloc_direct
3619 * will switch from direct to inline extent allocation
3622 xfs_iext_realloc_direct(ifp, new_size);
3623 ifp->if_bytes = new_size;
3627 * This is called when incore extents are being removed from the
3628 * indirection array and the extents being removed span multiple extent
3629 * buffers. The idx parameter contains the file extent index where we
3630 * want to begin removing extents, and the count parameter contains
3631 * how many extents need to be removed.
3633 * |-------| |-------|
3634 * | nex1 | | | nex1 - number of extents before idx
3635 * |-------| | count |
3636 * | | | | count - number of extents being removed at idx
3637 * | count | |-------|
3638 * | | | nex2 | nex2 - number of extents after idx + count
3639 * |-------| |-------|
3642 xfs_iext_remove_indirect(
3643 xfs_ifork_t *ifp, /* inode fork pointer */
3644 xfs_extnum_t idx, /* index to begin removing extents */
3645 int count) /* number of extents to remove */
3647 xfs_ext_irec_t *erp; /* indirection array pointer */
3648 int erp_idx = 0; /* indirection array index */
3649 xfs_extnum_t ext_cnt; /* extents left to remove */
3650 xfs_extnum_t ext_diff; /* extents to remove in current list */
3651 xfs_extnum_t nex1; /* number of extents before idx */
3652 xfs_extnum_t nex2; /* extents after idx + count */
3653 int nlists; /* entries in indirection array */
3654 int page_idx = idx; /* index in target extent list */
3656 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3657 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3658 ASSERT(erp != NULL);
3659 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3663 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
3664 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
3666 * Check for deletion of entire list;
3667 * xfs_iext_irec_remove() updates extent offsets.
3669 if (ext_diff == erp->er_extcount) {
3670 xfs_iext_irec_remove(ifp, erp_idx);
3671 ext_cnt -= ext_diff;
3674 ASSERT(erp_idx < ifp->if_real_bytes /
3676 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3683 /* Move extents up (if needed) */
3685 memmove(&erp->er_extbuf[nex1],
3686 &erp->er_extbuf[nex1 + ext_diff],
3687 nex2 * sizeof(xfs_bmbt_rec_t));
3689 /* Zero out rest of page */
3690 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
3691 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
3692 /* Update remaining counters */
3693 erp->er_extcount -= ext_diff;
3694 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
3695 ext_cnt -= ext_diff;
3700 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
3701 xfs_iext_irec_compact(ifp);
3705 * Create, destroy, or resize a linear (direct) block of extents.
3708 xfs_iext_realloc_direct(
3709 xfs_ifork_t *ifp, /* inode fork pointer */
3710 int new_size) /* new size of extents */
3712 int rnew_size; /* real new size of extents */
3714 rnew_size = new_size;
3716 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
3717 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
3718 (new_size != ifp->if_real_bytes)));
3720 /* Free extent records */
3721 if (new_size == 0) {
3722 xfs_iext_destroy(ifp);
3724 /* Resize direct extent list and zero any new bytes */
3725 else if (ifp->if_real_bytes) {
3726 /* Check if extents will fit inside the inode */
3727 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
3728 xfs_iext_direct_to_inline(ifp, new_size /
3729 (uint)sizeof(xfs_bmbt_rec_t));
3730 ifp->if_bytes = new_size;
3733 if (!is_power_of_2(new_size)){
3734 rnew_size = roundup_pow_of_two(new_size);
3736 if (rnew_size != ifp->if_real_bytes) {
3737 ifp->if_u1.if_extents =
3738 kmem_realloc(ifp->if_u1.if_extents,
3740 ifp->if_real_bytes, KM_NOFS);
3742 if (rnew_size > ifp->if_real_bytes) {
3743 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
3744 (uint)sizeof(xfs_bmbt_rec_t)], 0,
3745 rnew_size - ifp->if_real_bytes);
3749 * Switch from the inline extent buffer to a direct
3750 * extent list. Be sure to include the inline extent
3751 * bytes in new_size.
3754 new_size += ifp->if_bytes;
3755 if (!is_power_of_2(new_size)) {
3756 rnew_size = roundup_pow_of_two(new_size);
3758 xfs_iext_inline_to_direct(ifp, rnew_size);
3760 ifp->if_real_bytes = rnew_size;
3761 ifp->if_bytes = new_size;
3765 * Switch from linear (direct) extent records to inline buffer.
3768 xfs_iext_direct_to_inline(
3769 xfs_ifork_t *ifp, /* inode fork pointer */
3770 xfs_extnum_t nextents) /* number of extents in file */
3772 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3773 ASSERT(nextents <= XFS_INLINE_EXTS);
3775 * The inline buffer was zeroed when we switched
3776 * from inline to direct extent allocation mode,
3777 * so we don't need to clear it here.
3779 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
3780 nextents * sizeof(xfs_bmbt_rec_t));
3781 kmem_free(ifp->if_u1.if_extents);
3782 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3783 ifp->if_real_bytes = 0;
3787 * Switch from inline buffer to linear (direct) extent records.
3788 * new_size should already be rounded up to the next power of 2
3789 * by the caller (when appropriate), so use new_size as it is.
3790 * However, since new_size may be rounded up, we can't update
3791 * if_bytes here. It is the caller's responsibility to update
3792 * if_bytes upon return.
3795 xfs_iext_inline_to_direct(
3796 xfs_ifork_t *ifp, /* inode fork pointer */
3797 int new_size) /* number of extents in file */
3799 ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS);
3800 memset(ifp->if_u1.if_extents, 0, new_size);
3801 if (ifp->if_bytes) {
3802 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
3804 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3805 sizeof(xfs_bmbt_rec_t));
3807 ifp->if_real_bytes = new_size;
3811 * Resize an extent indirection array to new_size bytes.
3814 xfs_iext_realloc_indirect(
3815 xfs_ifork_t *ifp, /* inode fork pointer */
3816 int new_size) /* new indirection array size */
3818 int nlists; /* number of irec's (ex lists) */
3819 int size; /* current indirection array size */
3821 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3822 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3823 size = nlists * sizeof(xfs_ext_irec_t);
3824 ASSERT(ifp->if_real_bytes);
3825 ASSERT((new_size >= 0) && (new_size != size));
3826 if (new_size == 0) {
3827 xfs_iext_destroy(ifp);
3829 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
3830 kmem_realloc(ifp->if_u1.if_ext_irec,
3831 new_size, size, KM_NOFS);
3836 * Switch from indirection array to linear (direct) extent allocations.
3839 xfs_iext_indirect_to_direct(
3840 xfs_ifork_t *ifp) /* inode fork pointer */
3842 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
3843 xfs_extnum_t nextents; /* number of extents in file */
3844 int size; /* size of file extents */
3846 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3847 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3848 ASSERT(nextents <= XFS_LINEAR_EXTS);
3849 size = nextents * sizeof(xfs_bmbt_rec_t);
3851 xfs_iext_irec_compact_pages(ifp);
3852 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
3854 ep = ifp->if_u1.if_ext_irec->er_extbuf;
3855 kmem_free(ifp->if_u1.if_ext_irec);
3856 ifp->if_flags &= ~XFS_IFEXTIREC;
3857 ifp->if_u1.if_extents = ep;
3858 ifp->if_bytes = size;
3859 if (nextents < XFS_LINEAR_EXTS) {
3860 xfs_iext_realloc_direct(ifp, size);
3865 * Free incore file extents.
3869 xfs_ifork_t *ifp) /* inode fork pointer */
3871 if (ifp->if_flags & XFS_IFEXTIREC) {
3875 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3876 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
3877 xfs_iext_irec_remove(ifp, erp_idx);
3879 ifp->if_flags &= ~XFS_IFEXTIREC;
3880 } else if (ifp->if_real_bytes) {
3881 kmem_free(ifp->if_u1.if_extents);
3882 } else if (ifp->if_bytes) {
3883 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3884 sizeof(xfs_bmbt_rec_t));
3886 ifp->if_u1.if_extents = NULL;
3887 ifp->if_real_bytes = 0;
3892 * Return a pointer to the extent record for file system block bno.
3894 xfs_bmbt_rec_host_t * /* pointer to found extent record */
3895 xfs_iext_bno_to_ext(
3896 xfs_ifork_t *ifp, /* inode fork pointer */
3897 xfs_fileoff_t bno, /* block number to search for */
3898 xfs_extnum_t *idxp) /* index of target extent */
3900 xfs_bmbt_rec_host_t *base; /* pointer to first extent */
3901 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
3902 xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */
3903 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
3904 int high; /* upper boundary in search */
3905 xfs_extnum_t idx = 0; /* index of target extent */
3906 int low; /* lower boundary in search */
3907 xfs_extnum_t nextents; /* number of file extents */
3908 xfs_fileoff_t startoff = 0; /* start offset of extent */
3910 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3911 if (nextents == 0) {
3916 if (ifp->if_flags & XFS_IFEXTIREC) {
3917 /* Find target extent list */
3919 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
3920 base = erp->er_extbuf;
3921 high = erp->er_extcount - 1;
3923 base = ifp->if_u1.if_extents;
3924 high = nextents - 1;
3926 /* Binary search extent records */
3927 while (low <= high) {
3928 idx = (low + high) >> 1;
3930 startoff = xfs_bmbt_get_startoff(ep);
3931 blockcount = xfs_bmbt_get_blockcount(ep);
3932 if (bno < startoff) {
3934 } else if (bno >= startoff + blockcount) {
3937 /* Convert back to file-based extent index */
3938 if (ifp->if_flags & XFS_IFEXTIREC) {
3939 idx += erp->er_extoff;
3945 /* Convert back to file-based extent index */
3946 if (ifp->if_flags & XFS_IFEXTIREC) {
3947 idx += erp->er_extoff;
3949 if (bno >= startoff + blockcount) {
3950 if (++idx == nextents) {
3953 ep = xfs_iext_get_ext(ifp, idx);
3961 * Return a pointer to the indirection array entry containing the
3962 * extent record for filesystem block bno. Store the index of the
3963 * target irec in *erp_idxp.
3965 xfs_ext_irec_t * /* pointer to found extent record */
3966 xfs_iext_bno_to_irec(
3967 xfs_ifork_t *ifp, /* inode fork pointer */
3968 xfs_fileoff_t bno, /* block number to search for */
3969 int *erp_idxp) /* irec index of target ext list */
3971 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
3972 xfs_ext_irec_t *erp_next; /* next indirection array entry */
3973 int erp_idx; /* indirection array index */
3974 int nlists; /* number of extent irec's (lists) */
3975 int high; /* binary search upper limit */
3976 int low; /* binary search lower limit */
3978 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3979 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3983 while (low <= high) {
3984 erp_idx = (low + high) >> 1;
3985 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3986 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
3987 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
3989 } else if (erp_next && bno >=
3990 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
3996 *erp_idxp = erp_idx;
4001 * Return a pointer to the indirection array entry containing the
4002 * extent record at file extent index *idxp. Store the index of the
4003 * target irec in *erp_idxp and store the page index of the target
4004 * extent record in *idxp.
4007 xfs_iext_idx_to_irec(
4008 xfs_ifork_t *ifp, /* inode fork pointer */
4009 xfs_extnum_t *idxp, /* extent index (file -> page) */
4010 int *erp_idxp, /* pointer to target irec */
4011 int realloc) /* new bytes were just added */
4013 xfs_ext_irec_t *prev; /* pointer to previous irec */
4014 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
4015 int erp_idx; /* indirection array index */
4016 int nlists; /* number of irec's (ex lists) */
4017 int high; /* binary search upper limit */
4018 int low; /* binary search lower limit */
4019 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
4021 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4022 ASSERT(page_idx >= 0 && page_idx <=
4023 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
4024 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4029 /* Binary search extent irec's */
4030 while (low <= high) {
4031 erp_idx = (low + high) >> 1;
4032 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4033 prev = erp_idx > 0 ? erp - 1 : NULL;
4034 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
4035 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
4037 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
4038 (page_idx == erp->er_extoff + erp->er_extcount &&
4041 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
4042 erp->er_extcount == XFS_LINEAR_EXTS) {
4046 erp = erp_idx < nlists ? erp + 1 : NULL;
4049 page_idx -= erp->er_extoff;
4054 *erp_idxp = erp_idx;
4059 * Allocate and initialize an indirection array once the space needed
4060 * for incore extents increases above XFS_IEXT_BUFSZ.
4064 xfs_ifork_t *ifp) /* inode fork pointer */
4066 xfs_ext_irec_t *erp; /* indirection array pointer */
4067 xfs_extnum_t nextents; /* number of extents in file */
4069 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4070 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4071 ASSERT(nextents <= XFS_LINEAR_EXTS);
4073 erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS);
4075 if (nextents == 0) {
4076 ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
4077 } else if (!ifp->if_real_bytes) {
4078 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
4079 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
4080 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
4082 erp->er_extbuf = ifp->if_u1.if_extents;
4083 erp->er_extcount = nextents;
4086 ifp->if_flags |= XFS_IFEXTIREC;
4087 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
4088 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
4089 ifp->if_u1.if_ext_irec = erp;
4095 * Allocate and initialize a new entry in the indirection array.
4099 xfs_ifork_t *ifp, /* inode fork pointer */
4100 int erp_idx) /* index for new irec */
4102 xfs_ext_irec_t *erp; /* indirection array pointer */
4103 int i; /* loop counter */
4104 int nlists; /* number of irec's (ex lists) */
4106 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4107 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4109 /* Resize indirection array */
4110 xfs_iext_realloc_indirect(ifp, ++nlists *
4111 sizeof(xfs_ext_irec_t));
4113 * Move records down in the array so the
4114 * new page can use erp_idx.
4116 erp = ifp->if_u1.if_ext_irec;
4117 for (i = nlists - 1; i > erp_idx; i--) {
4118 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
4120 ASSERT(i == erp_idx);
4122 /* Initialize new extent record */
4123 erp = ifp->if_u1.if_ext_irec;
4124 erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
4125 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4126 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
4127 erp[erp_idx].er_extcount = 0;
4128 erp[erp_idx].er_extoff = erp_idx > 0 ?
4129 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
4130 return (&erp[erp_idx]);
4134 * Remove a record from the indirection array.
4137 xfs_iext_irec_remove(
4138 xfs_ifork_t *ifp, /* inode fork pointer */
4139 int erp_idx) /* irec index to remove */
4141 xfs_ext_irec_t *erp; /* indirection array pointer */
4142 int i; /* loop counter */
4143 int nlists; /* number of irec's (ex lists) */
4145 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4146 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4147 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4148 if (erp->er_extbuf) {
4149 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
4151 kmem_free(erp->er_extbuf);
4153 /* Compact extent records */
4154 erp = ifp->if_u1.if_ext_irec;
4155 for (i = erp_idx; i < nlists - 1; i++) {
4156 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
4159 * Manually free the last extent record from the indirection
4160 * array. A call to xfs_iext_realloc_indirect() with a size
4161 * of zero would result in a call to xfs_iext_destroy() which
4162 * would in turn call this function again, creating a nasty
4166 xfs_iext_realloc_indirect(ifp,
4167 nlists * sizeof(xfs_ext_irec_t));
4169 kmem_free(ifp->if_u1.if_ext_irec);
4171 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4175 * This is called to clean up large amounts of unused memory allocated
4176 * by the indirection array. Before compacting anything though, verify
4177 * that the indirection array is still needed and switch back to the
4178 * linear extent list (or even the inline buffer) if possible. The
4179 * compaction policy is as follows:
4181 * Full Compaction: Extents fit into a single page (or inline buffer)
4182 * Partial Compaction: Extents occupy less than 50% of allocated space
4183 * No Compaction: Extents occupy at least 50% of allocated space
4186 xfs_iext_irec_compact(
4187 xfs_ifork_t *ifp) /* inode fork pointer */
4189 xfs_extnum_t nextents; /* number of extents in file */
4190 int nlists; /* number of irec's (ex lists) */
4192 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4193 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4194 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4196 if (nextents == 0) {
4197 xfs_iext_destroy(ifp);
4198 } else if (nextents <= XFS_INLINE_EXTS) {
4199 xfs_iext_indirect_to_direct(ifp);
4200 xfs_iext_direct_to_inline(ifp, nextents);
4201 } else if (nextents <= XFS_LINEAR_EXTS) {
4202 xfs_iext_indirect_to_direct(ifp);
4203 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
4204 xfs_iext_irec_compact_pages(ifp);
4209 * Combine extents from neighboring extent pages.
4212 xfs_iext_irec_compact_pages(
4213 xfs_ifork_t *ifp) /* inode fork pointer */
4215 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
4216 int erp_idx = 0; /* indirection array index */
4217 int nlists; /* number of irec's (ex lists) */
4219 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4220 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4221 while (erp_idx < nlists - 1) {
4222 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4224 if (erp_next->er_extcount <=
4225 (XFS_LINEAR_EXTS - erp->er_extcount)) {
4226 memcpy(&erp->er_extbuf[erp->er_extcount],
4227 erp_next->er_extbuf, erp_next->er_extcount *
4228 sizeof(xfs_bmbt_rec_t));
4229 erp->er_extcount += erp_next->er_extcount;
4231 * Free page before removing extent record
4232 * so er_extoffs don't get modified in
4233 * xfs_iext_irec_remove.
4235 kmem_free(erp_next->er_extbuf);
4236 erp_next->er_extbuf = NULL;
4237 xfs_iext_irec_remove(ifp, erp_idx + 1);
4238 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4246 * This is called to update the er_extoff field in the indirection
4247 * array when extents have been added or removed from one of the
4248 * extent lists. erp_idx contains the irec index to begin updating
4249 * at and ext_diff contains the number of extents that were added
4253 xfs_iext_irec_update_extoffs(
4254 xfs_ifork_t *ifp, /* inode fork pointer */
4255 int erp_idx, /* irec index to update */
4256 int ext_diff) /* number of new extents */
4258 int i; /* loop counter */
4259 int nlists; /* number of irec's (ex lists */
4261 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4262 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4263 for (i = erp_idx; i < nlists; i++) {
4264 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;