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_dir2_trace.h"
51 #include "xfs_quota.h"
53 #include "xfs_filestream.h"
54 #include "xfs_vnodeops.h"
56 kmem_zone_t *xfs_ifork_zone;
57 kmem_zone_t *xfs_inode_zone;
60 * Used in xfs_itruncate(). This is the maximum number of extents
61 * freed from a file in a single transaction.
63 #define XFS_ITRUNC_MAX_EXTENTS 2
65 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
66 STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
67 STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
68 STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
72 * Make sure that the extents in the given memory buffer
82 xfs_bmbt_rec_host_t rec;
85 for (i = 0; i < nrecs; i++) {
86 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
87 rec.l0 = get_unaligned(&ep->l0);
88 rec.l1 = get_unaligned(&ep->l1);
89 xfs_bmbt_get_all(&rec, &irec);
90 if (fmt == XFS_EXTFMT_NOSTATE)
91 ASSERT(irec.br_state == XFS_EXT_NORM);
95 #define xfs_validate_extents(ifp, nrecs, fmt)
99 * Check that none of the inode's in the buffer have a next
100 * unlinked field of 0.
112 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
114 for (i = 0; i < j; i++) {
115 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
116 i * mp->m_sb.sb_inodesize);
117 if (!dip->di_next_unlinked) {
118 xfs_fs_cmn_err(CE_ALERT, mp,
119 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
121 ASSERT(dip->di_next_unlinked);
128 * Find the buffer associated with the given inode map
129 * We do basic validation checks on the buffer once it has been
130 * retrieved from disk.
136 struct xfs_imap *imap,
146 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno,
147 (int)imap->im_len, buf_flags, &bp);
149 if (error != EAGAIN) {
151 "xfs_imap_to_bp: xfs_trans_read_buf()returned "
152 "an error %d on %s. Returning error.",
153 error, mp->m_fsname);
155 ASSERT(buf_flags & XFS_BUF_TRYLOCK);
161 * Validate the magic number and version of every inode in the buffer
162 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
165 ni = BBTOB(imap->im_len) >> mp->m_sb.sb_inodelog;
166 #else /* usual case */
170 for (i = 0; i < ni; i++) {
174 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
175 (i << mp->m_sb.sb_inodelog));
176 di_ok = be16_to_cpu(dip->di_magic) == XFS_DINODE_MAGIC &&
177 XFS_DINODE_GOOD_VERSION(dip->di_version);
178 if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
179 XFS_ERRTAG_ITOBP_INOTOBP,
180 XFS_RANDOM_ITOBP_INOTOBP))) {
181 if (iget_flags & XFS_IGET_BULKSTAT) {
182 xfs_trans_brelse(tp, bp);
183 return XFS_ERROR(EINVAL);
185 XFS_CORRUPTION_ERROR("xfs_imap_to_bp",
186 XFS_ERRLEVEL_HIGH, mp, dip);
189 "Device %s - bad inode magic/vsn "
190 "daddr %lld #%d (magic=%x)",
191 XFS_BUFTARG_NAME(mp->m_ddev_targp),
192 (unsigned long long)imap->im_blkno, i,
193 be16_to_cpu(dip->di_magic));
195 xfs_trans_brelse(tp, bp);
196 return XFS_ERROR(EFSCORRUPTED);
200 xfs_inobp_check(mp, bp);
203 * Mark the buffer as an inode buffer now that it looks good
205 XFS_BUF_SET_VTYPE(bp, B_FS_INO);
212 * This routine is called to map an inode number within a file
213 * system to the buffer containing the on-disk version of the
214 * inode. It returns a pointer to the buffer containing the
215 * on-disk inode in the bpp parameter, and in the dip parameter
216 * it returns a pointer to the on-disk inode within that buffer.
218 * If a non-zero error is returned, then the contents of bpp and
219 * dipp are undefined.
221 * Use xfs_imap() to determine the size and location of the
222 * buffer to read from disk.
234 struct xfs_imap imap;
239 error = xfs_imap(mp, tp, ino, &imap, imap_flags);
243 error = xfs_imap_to_bp(mp, tp, &imap, &bp, XFS_BUF_LOCK, imap_flags);
247 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
249 *offset = imap.im_boffset;
255 * This routine is called to map an inode to the buffer containing
256 * the on-disk version of the inode. It returns a pointer to the
257 * buffer containing the on-disk inode in the bpp parameter, and in
258 * the dip parameter it returns a pointer to the on-disk inode within
261 * If a non-zero error is returned, then the contents of bpp and
262 * dipp are undefined.
264 * The inode is expected to already been mapped to its buffer and read
265 * in once, thus we can use the mapping information stored in the inode
266 * rather than calling xfs_imap(). This allows us to avoid the overhead
267 * of looking at the inode btree for small block file systems
282 ASSERT(ip->i_imap.im_blkno != 0);
284 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp, buf_flags, 0);
289 ASSERT(buf_flags & XFS_BUF_TRYLOCK);
295 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
301 * Move inode type and inode format specific information from the
302 * on-disk inode to the in-core inode. For fifos, devs, and sockets
303 * this means set if_rdev to the proper value. For files, directories,
304 * and symlinks this means to bring in the in-line data or extent
305 * pointers. For a file in B-tree format, only the root is immediately
306 * brought in-core. The rest will be in-lined in if_extents when it
307 * is first referenced (see xfs_iread_extents()).
314 xfs_attr_shortform_t *atp;
318 ip->i_df.if_ext_max =
319 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
322 if (unlikely(be32_to_cpu(dip->di_nextents) +
323 be16_to_cpu(dip->di_anextents) >
324 be64_to_cpu(dip->di_nblocks))) {
325 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
326 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
327 (unsigned long long)ip->i_ino,
328 (int)(be32_to_cpu(dip->di_nextents) +
329 be16_to_cpu(dip->di_anextents)),
331 be64_to_cpu(dip->di_nblocks));
332 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
334 return XFS_ERROR(EFSCORRUPTED);
337 if (unlikely(dip->di_forkoff > ip->i_mount->m_sb.sb_inodesize)) {
338 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
339 "corrupt dinode %Lu, forkoff = 0x%x.",
340 (unsigned long long)ip->i_ino,
342 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
344 return XFS_ERROR(EFSCORRUPTED);
347 switch (ip->i_d.di_mode & S_IFMT) {
352 if (unlikely(dip->di_format != XFS_DINODE_FMT_DEV)) {
353 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
355 return XFS_ERROR(EFSCORRUPTED);
359 ip->i_df.if_u2.if_rdev = xfs_dinode_get_rdev(dip);
365 switch (dip->di_format) {
366 case XFS_DINODE_FMT_LOCAL:
368 * no local regular files yet
370 if (unlikely((be16_to_cpu(dip->di_mode) & S_IFMT) == S_IFREG)) {
371 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
373 "(local format for regular file).",
374 (unsigned long long) ip->i_ino);
375 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
378 return XFS_ERROR(EFSCORRUPTED);
381 di_size = be64_to_cpu(dip->di_size);
382 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
383 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
385 "(bad size %Ld for local inode).",
386 (unsigned long long) ip->i_ino,
387 (long long) di_size);
388 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
391 return XFS_ERROR(EFSCORRUPTED);
395 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
397 case XFS_DINODE_FMT_EXTENTS:
398 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
400 case XFS_DINODE_FMT_BTREE:
401 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
404 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
406 return XFS_ERROR(EFSCORRUPTED);
411 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
412 return XFS_ERROR(EFSCORRUPTED);
417 if (!XFS_DFORK_Q(dip))
419 ASSERT(ip->i_afp == NULL);
420 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
421 ip->i_afp->if_ext_max =
422 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
423 switch (dip->di_aformat) {
424 case XFS_DINODE_FMT_LOCAL:
425 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
426 size = be16_to_cpu(atp->hdr.totsize);
427 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
429 case XFS_DINODE_FMT_EXTENTS:
430 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
432 case XFS_DINODE_FMT_BTREE:
433 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
436 error = XFS_ERROR(EFSCORRUPTED);
440 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
442 xfs_idestroy_fork(ip, XFS_DATA_FORK);
448 * The file is in-lined in the on-disk inode.
449 * If it fits into if_inline_data, then copy
450 * it there, otherwise allocate a buffer for it
451 * and copy the data there. Either way, set
452 * if_data to point at the data.
453 * If we allocate a buffer for the data, make
454 * sure that its size is a multiple of 4 and
455 * record the real size in i_real_bytes.
468 * If the size is unreasonable, then something
469 * is wrong and we just bail out rather than crash in
470 * kmem_alloc() or memcpy() below.
472 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
473 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
475 "(bad size %d for local fork, size = %d).",
476 (unsigned long long) ip->i_ino, size,
477 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
478 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
480 return XFS_ERROR(EFSCORRUPTED);
482 ifp = XFS_IFORK_PTR(ip, whichfork);
485 ifp->if_u1.if_data = NULL;
486 else if (size <= sizeof(ifp->if_u2.if_inline_data))
487 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
489 real_size = roundup(size, 4);
490 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
492 ifp->if_bytes = size;
493 ifp->if_real_bytes = real_size;
495 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
496 ifp->if_flags &= ~XFS_IFEXTENTS;
497 ifp->if_flags |= XFS_IFINLINE;
502 * The file consists of a set of extents all
503 * of which fit into the on-disk inode.
504 * If there are few enough extents to fit into
505 * the if_inline_ext, then copy them there.
506 * Otherwise allocate a buffer for them and copy
507 * them into it. Either way, set if_extents
508 * to point at the extents.
522 ifp = XFS_IFORK_PTR(ip, whichfork);
523 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
524 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
527 * If the number of extents is unreasonable, then something
528 * is wrong and we just bail out rather than crash in
529 * kmem_alloc() or memcpy() below.
531 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
532 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
533 "corrupt inode %Lu ((a)extents = %d).",
534 (unsigned long long) ip->i_ino, nex);
535 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
537 return XFS_ERROR(EFSCORRUPTED);
540 ifp->if_real_bytes = 0;
542 ifp->if_u1.if_extents = NULL;
543 else if (nex <= XFS_INLINE_EXTS)
544 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
546 xfs_iext_add(ifp, 0, nex);
548 ifp->if_bytes = size;
550 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
551 xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip));
552 for (i = 0; i < nex; i++, dp++) {
553 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
554 ep->l0 = get_unaligned_be64(&dp->l0);
555 ep->l1 = get_unaligned_be64(&dp->l1);
557 XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork);
558 if (whichfork != XFS_DATA_FORK ||
559 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
560 if (unlikely(xfs_check_nostate_extents(
562 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
565 return XFS_ERROR(EFSCORRUPTED);
568 ifp->if_flags |= XFS_IFEXTENTS;
573 * The file has too many extents to fit into
574 * the inode, so they are in B-tree format.
575 * Allocate a buffer for the root of the B-tree
576 * and copy the root into it. The i_extents
577 * field will remain NULL until all of the
578 * extents are read in (when they are needed).
586 xfs_bmdr_block_t *dfp;
592 ifp = XFS_IFORK_PTR(ip, whichfork);
593 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
594 size = XFS_BMAP_BROOT_SPACE(dfp);
595 nrecs = be16_to_cpu(dfp->bb_numrecs);
598 * blow out if -- fork has less extents than can fit in
599 * fork (fork shouldn't be a btree format), root btree
600 * block has more records than can fit into the fork,
601 * or the number of extents is greater than the number of
604 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
605 || XFS_BMDR_SPACE_CALC(nrecs) >
606 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
607 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
608 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
609 "corrupt inode %Lu (btree).",
610 (unsigned long long) ip->i_ino);
611 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
613 return XFS_ERROR(EFSCORRUPTED);
616 ifp->if_broot_bytes = size;
617 ifp->if_broot = kmem_alloc(size, KM_SLEEP);
618 ASSERT(ifp->if_broot != NULL);
620 * Copy and convert from the on-disk structure
621 * to the in-memory structure.
623 xfs_bmdr_to_bmbt(ip->i_mount, dfp,
624 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
625 ifp->if_broot, size);
626 ifp->if_flags &= ~XFS_IFEXTENTS;
627 ifp->if_flags |= XFS_IFBROOT;
633 xfs_dinode_from_disk(
637 to->di_magic = be16_to_cpu(from->di_magic);
638 to->di_mode = be16_to_cpu(from->di_mode);
639 to->di_version = from ->di_version;
640 to->di_format = from->di_format;
641 to->di_onlink = be16_to_cpu(from->di_onlink);
642 to->di_uid = be32_to_cpu(from->di_uid);
643 to->di_gid = be32_to_cpu(from->di_gid);
644 to->di_nlink = be32_to_cpu(from->di_nlink);
645 to->di_projid = be16_to_cpu(from->di_projid);
646 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
647 to->di_flushiter = be16_to_cpu(from->di_flushiter);
648 to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec);
649 to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec);
650 to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec);
651 to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec);
652 to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec);
653 to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec);
654 to->di_size = be64_to_cpu(from->di_size);
655 to->di_nblocks = be64_to_cpu(from->di_nblocks);
656 to->di_extsize = be32_to_cpu(from->di_extsize);
657 to->di_nextents = be32_to_cpu(from->di_nextents);
658 to->di_anextents = be16_to_cpu(from->di_anextents);
659 to->di_forkoff = from->di_forkoff;
660 to->di_aformat = from->di_aformat;
661 to->di_dmevmask = be32_to_cpu(from->di_dmevmask);
662 to->di_dmstate = be16_to_cpu(from->di_dmstate);
663 to->di_flags = be16_to_cpu(from->di_flags);
664 to->di_gen = be32_to_cpu(from->di_gen);
670 xfs_icdinode_t *from)
672 to->di_magic = cpu_to_be16(from->di_magic);
673 to->di_mode = cpu_to_be16(from->di_mode);
674 to->di_version = from ->di_version;
675 to->di_format = from->di_format;
676 to->di_onlink = cpu_to_be16(from->di_onlink);
677 to->di_uid = cpu_to_be32(from->di_uid);
678 to->di_gid = cpu_to_be32(from->di_gid);
679 to->di_nlink = cpu_to_be32(from->di_nlink);
680 to->di_projid = cpu_to_be16(from->di_projid);
681 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
682 to->di_flushiter = cpu_to_be16(from->di_flushiter);
683 to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec);
684 to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec);
685 to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec);
686 to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec);
687 to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec);
688 to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec);
689 to->di_size = cpu_to_be64(from->di_size);
690 to->di_nblocks = cpu_to_be64(from->di_nblocks);
691 to->di_extsize = cpu_to_be32(from->di_extsize);
692 to->di_nextents = cpu_to_be32(from->di_nextents);
693 to->di_anextents = cpu_to_be16(from->di_anextents);
694 to->di_forkoff = from->di_forkoff;
695 to->di_aformat = from->di_aformat;
696 to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
697 to->di_dmstate = cpu_to_be16(from->di_dmstate);
698 to->di_flags = cpu_to_be16(from->di_flags);
699 to->di_gen = cpu_to_be32(from->di_gen);
708 if (di_flags & XFS_DIFLAG_ANY) {
709 if (di_flags & XFS_DIFLAG_REALTIME)
710 flags |= XFS_XFLAG_REALTIME;
711 if (di_flags & XFS_DIFLAG_PREALLOC)
712 flags |= XFS_XFLAG_PREALLOC;
713 if (di_flags & XFS_DIFLAG_IMMUTABLE)
714 flags |= XFS_XFLAG_IMMUTABLE;
715 if (di_flags & XFS_DIFLAG_APPEND)
716 flags |= XFS_XFLAG_APPEND;
717 if (di_flags & XFS_DIFLAG_SYNC)
718 flags |= XFS_XFLAG_SYNC;
719 if (di_flags & XFS_DIFLAG_NOATIME)
720 flags |= XFS_XFLAG_NOATIME;
721 if (di_flags & XFS_DIFLAG_NODUMP)
722 flags |= XFS_XFLAG_NODUMP;
723 if (di_flags & XFS_DIFLAG_RTINHERIT)
724 flags |= XFS_XFLAG_RTINHERIT;
725 if (di_flags & XFS_DIFLAG_PROJINHERIT)
726 flags |= XFS_XFLAG_PROJINHERIT;
727 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
728 flags |= XFS_XFLAG_NOSYMLINKS;
729 if (di_flags & XFS_DIFLAG_EXTSIZE)
730 flags |= XFS_XFLAG_EXTSIZE;
731 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
732 flags |= XFS_XFLAG_EXTSZINHERIT;
733 if (di_flags & XFS_DIFLAG_NODEFRAG)
734 flags |= XFS_XFLAG_NODEFRAG;
735 if (di_flags & XFS_DIFLAG_FILESTREAM)
736 flags |= XFS_XFLAG_FILESTREAM;
746 xfs_icdinode_t *dic = &ip->i_d;
748 return _xfs_dic2xflags(dic->di_flags) |
749 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
756 return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) |
757 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
761 * Read the disk inode attributes into the in-core inode structure.
776 * Fill in the location information in the in-core inode.
778 ip->i_imap.im_blkno = bno;
779 error = xfs_imap(mp, tp, ip->i_ino, &ip->i_imap, iget_flags);
782 ASSERT(bno == 0 || bno == ip->i_imap.im_blkno);
785 * Get pointers to the on-disk inode and the buffer containing it.
787 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp,
788 XFS_BUF_LOCK, iget_flags);
791 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
794 * If we got something that isn't an inode it means someone
795 * (nfs or dmi) has a stale handle.
797 if (be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC) {
799 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
800 "dip->di_magic (0x%x) != "
801 "XFS_DINODE_MAGIC (0x%x)",
802 be16_to_cpu(dip->di_magic),
805 error = XFS_ERROR(EINVAL);
810 * If the on-disk inode is already linked to a directory
811 * entry, copy all of the inode into the in-core inode.
812 * xfs_iformat() handles copying in the inode format
813 * specific information.
814 * Otherwise, just get the truly permanent information.
817 xfs_dinode_from_disk(&ip->i_d, dip);
818 error = xfs_iformat(ip, dip);
821 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
822 "xfs_iformat() returned error %d",
828 ip->i_d.di_magic = be16_to_cpu(dip->di_magic);
829 ip->i_d.di_version = dip->di_version;
830 ip->i_d.di_gen = be32_to_cpu(dip->di_gen);
831 ip->i_d.di_flushiter = be16_to_cpu(dip->di_flushiter);
833 * Make sure to pull in the mode here as well in
834 * case the inode is released without being used.
835 * This ensures that xfs_inactive() will see that
836 * the inode is already free and not try to mess
837 * with the uninitialized part of it.
841 * Initialize the per-fork minima and maxima for a new
842 * inode here. xfs_iformat will do it for old inodes.
844 ip->i_df.if_ext_max =
845 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
849 * The inode format changed when we moved the link count and
850 * made it 32 bits long. If this is an old format inode,
851 * convert it in memory to look like a new one. If it gets
852 * flushed to disk we will convert back before flushing or
853 * logging it. We zero out the new projid field and the old link
854 * count field. We'll handle clearing the pad field (the remains
855 * of the old uuid field) when we actually convert the inode to
856 * the new format. We don't change the version number so that we
857 * can distinguish this from a real new format inode.
859 if (ip->i_d.di_version == 1) {
860 ip->i_d.di_nlink = ip->i_d.di_onlink;
861 ip->i_d.di_onlink = 0;
862 ip->i_d.di_projid = 0;
865 ip->i_delayed_blks = 0;
866 ip->i_size = ip->i_d.di_size;
869 * Mark the buffer containing the inode as something to keep
870 * around for a while. This helps to keep recently accessed
871 * meta-data in-core longer.
873 XFS_BUF_SET_REF(bp, XFS_INO_REF);
876 * Use xfs_trans_brelse() to release the buffer containing the
877 * on-disk inode, because it was acquired with xfs_trans_read_buf()
878 * in xfs_itobp() above. If tp is NULL, this is just a normal
879 * brelse(). If we're within a transaction, then xfs_trans_brelse()
880 * will only release the buffer if it is not dirty within the
881 * transaction. It will be OK to release the buffer in this case,
882 * because inodes on disk are never destroyed and we will be
883 * locking the new in-core inode before putting it in the hash
884 * table where other processes can find it. Thus we don't have
885 * to worry about the inode being changed just because we released
889 xfs_trans_brelse(tp, bp);
894 * Read in extents from a btree-format inode.
895 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
905 xfs_extnum_t nextents;
908 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
909 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
911 return XFS_ERROR(EFSCORRUPTED);
913 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
914 size = nextents * sizeof(xfs_bmbt_rec_t);
915 ifp = XFS_IFORK_PTR(ip, whichfork);
918 * We know that the size is valid (it's checked in iformat_btree)
920 ifp->if_lastex = NULLEXTNUM;
921 ifp->if_bytes = ifp->if_real_bytes = 0;
922 ifp->if_flags |= XFS_IFEXTENTS;
923 xfs_iext_add(ifp, 0, nextents);
924 error = xfs_bmap_read_extents(tp, ip, whichfork);
926 xfs_iext_destroy(ifp);
927 ifp->if_flags &= ~XFS_IFEXTENTS;
930 xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip));
935 * Allocate an inode on disk and return a copy of its in-core version.
936 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
937 * appropriately within the inode. The uid and gid for the inode are
938 * set according to the contents of the given cred structure.
940 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
941 * has a free inode available, call xfs_iget()
942 * to obtain the in-core version of the allocated inode. Finally,
943 * fill in the inode and log its initial contents. In this case,
944 * ialloc_context would be set to NULL and call_again set to false.
946 * If xfs_dialloc() does not have an available inode,
947 * it will replenish its supply by doing an allocation. Since we can
948 * only do one allocation within a transaction without deadlocks, we
949 * must commit the current transaction before returning the inode itself.
950 * In this case, therefore, we will set call_again to true and return.
951 * The caller should then commit the current transaction, start a new
952 * transaction, and call xfs_ialloc() again to actually get the inode.
954 * To ensure that some other process does not grab the inode that
955 * was allocated during the first call to xfs_ialloc(), this routine
956 * also returns the [locked] bp pointing to the head of the freelist
957 * as ialloc_context. The caller should hold this buffer across
958 * the commit and pass it back into this routine on the second call.
960 * If we are allocating quota inodes, we do not have a parent inode
961 * to attach to or associate with (i.e. pip == NULL) because they
962 * are not linked into the directory structure - they are attached
963 * directly to the superblock - and so have no parent.
975 xfs_buf_t **ialloc_context,
976 boolean_t *call_again,
987 * Call the space management code to pick
988 * the on-disk inode to be allocated.
990 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
991 ialloc_context, call_again, &ino);
994 if (*call_again || ino == NULLFSINO) {
998 ASSERT(*ialloc_context == NULL);
1001 * Get the in-core inode with the lock held exclusively.
1002 * This is because we're setting fields here we need
1003 * to prevent others from looking at until we're done.
1005 error = xfs_trans_iget(tp->t_mountp, tp, ino,
1006 XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
1011 ip->i_d.di_mode = (__uint16_t)mode;
1012 ip->i_d.di_onlink = 0;
1013 ip->i_d.di_nlink = nlink;
1014 ASSERT(ip->i_d.di_nlink == nlink);
1015 ip->i_d.di_uid = current_fsuid();
1016 ip->i_d.di_gid = current_fsgid();
1017 ip->i_d.di_projid = prid;
1018 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1021 * If the superblock version is up to where we support new format
1022 * inodes and this is currently an old format inode, then change
1023 * the inode version number now. This way we only do the conversion
1024 * here rather than here and in the flush/logging code.
1026 if (xfs_sb_version_hasnlink(&tp->t_mountp->m_sb) &&
1027 ip->i_d.di_version == 1) {
1028 ip->i_d.di_version = 2;
1030 * We've already zeroed the old link count, the projid field,
1031 * and the pad field.
1036 * Project ids won't be stored on disk if we are using a version 1 inode.
1038 if ((prid != 0) && (ip->i_d.di_version == 1))
1039 xfs_bump_ino_vers2(tp, ip);
1041 if (pip && XFS_INHERIT_GID(pip)) {
1042 ip->i_d.di_gid = pip->i_d.di_gid;
1043 if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1044 ip->i_d.di_mode |= S_ISGID;
1049 * If the group ID of the new file does not match the effective group
1050 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1051 * (and only if the irix_sgid_inherit compatibility variable is set).
1053 if ((irix_sgid_inherit) &&
1054 (ip->i_d.di_mode & S_ISGID) &&
1055 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1056 ip->i_d.di_mode &= ~S_ISGID;
1059 ip->i_d.di_size = 0;
1061 ip->i_d.di_nextents = 0;
1062 ASSERT(ip->i_d.di_nblocks == 0);
1065 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
1066 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
1067 ip->i_d.di_atime = ip->i_d.di_mtime;
1068 ip->i_d.di_ctime = ip->i_d.di_mtime;
1071 * di_gen will have been taken care of in xfs_iread.
1073 ip->i_d.di_extsize = 0;
1074 ip->i_d.di_dmevmask = 0;
1075 ip->i_d.di_dmstate = 0;
1076 ip->i_d.di_flags = 0;
1077 flags = XFS_ILOG_CORE;
1078 switch (mode & S_IFMT) {
1083 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1084 ip->i_df.if_u2.if_rdev = rdev;
1085 ip->i_df.if_flags = 0;
1086 flags |= XFS_ILOG_DEV;
1090 * we can't set up filestreams until after the VFS inode
1091 * is set up properly.
1093 if (pip && xfs_inode_is_filestream(pip))
1097 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
1100 if ((mode & S_IFMT) == S_IFDIR) {
1101 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1102 di_flags |= XFS_DIFLAG_RTINHERIT;
1103 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1104 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1105 ip->i_d.di_extsize = pip->i_d.di_extsize;
1107 } else if ((mode & S_IFMT) == S_IFREG) {
1108 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1109 di_flags |= XFS_DIFLAG_REALTIME;
1110 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1111 di_flags |= XFS_DIFLAG_EXTSIZE;
1112 ip->i_d.di_extsize = pip->i_d.di_extsize;
1115 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1116 xfs_inherit_noatime)
1117 di_flags |= XFS_DIFLAG_NOATIME;
1118 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1120 di_flags |= XFS_DIFLAG_NODUMP;
1121 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1123 di_flags |= XFS_DIFLAG_SYNC;
1124 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1125 xfs_inherit_nosymlinks)
1126 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1127 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1128 di_flags |= XFS_DIFLAG_PROJINHERIT;
1129 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1130 xfs_inherit_nodefrag)
1131 di_flags |= XFS_DIFLAG_NODEFRAG;
1132 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
1133 di_flags |= XFS_DIFLAG_FILESTREAM;
1134 ip->i_d.di_flags |= di_flags;
1138 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1139 ip->i_df.if_flags = XFS_IFEXTENTS;
1140 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1141 ip->i_df.if_u1.if_extents = NULL;
1147 * Attribute fork settings for new inode.
1149 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1150 ip->i_d.di_anextents = 0;
1153 * Log the new values stuffed into the inode.
1155 xfs_trans_log_inode(tp, ip, flags);
1157 /* now that we have an i_mode we can setup inode ops and unlock */
1158 xfs_setup_inode(ip);
1160 /* now we have set up the vfs inode we can associate the filestream */
1162 error = xfs_filestream_associate(pip, ip);
1166 xfs_iflags_set(ip, XFS_IFILESTREAM);
1174 * Check to make sure that there are no blocks allocated to the
1175 * file beyond the size of the file. We don't check this for
1176 * files with fixed size extents or real time extents, but we
1177 * at least do it for regular files.
1186 xfs_fileoff_t map_first;
1188 xfs_bmbt_irec_t imaps[2];
1190 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1193 if (XFS_IS_REALTIME_INODE(ip))
1196 if (ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE)
1200 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1202 * The filesystem could be shutting down, so bmapi may return
1205 if (xfs_bmapi(NULL, ip, map_first,
1207 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1209 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1212 ASSERT(nimaps == 1);
1213 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1218 * Calculate the last possible buffered byte in a file. This must
1219 * include data that was buffered beyond the EOF by the write code.
1220 * This also needs to deal with overflowing the xfs_fsize_t type
1221 * which can happen for sizes near the limit.
1223 * We also need to take into account any blocks beyond the EOF. It
1224 * may be the case that they were buffered by a write which failed.
1225 * In that case the pages will still be in memory, but the inode size
1226 * will never have been updated.
1233 xfs_fsize_t last_byte;
1234 xfs_fileoff_t last_block;
1235 xfs_fileoff_t size_last_block;
1238 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED));
1242 * Only check for blocks beyond the EOF if the extents have
1243 * been read in. This eliminates the need for the inode lock,
1244 * and it also saves us from looking when it really isn't
1247 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1248 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1256 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_size);
1257 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1259 last_byte = XFS_FSB_TO_B(mp, last_block);
1260 if (last_byte < 0) {
1261 return XFS_MAXIOFFSET(mp);
1263 last_byte += (1 << mp->m_writeio_log);
1264 if (last_byte < 0) {
1265 return XFS_MAXIOFFSET(mp);
1270 #if defined(XFS_RW_TRACE)
1276 xfs_fsize_t new_size,
1277 xfs_off_t toss_start,
1278 xfs_off_t toss_finish)
1280 if (ip->i_rwtrace == NULL) {
1284 ktrace_enter(ip->i_rwtrace,
1287 (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
1288 (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
1289 (void*)((long)flag),
1290 (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
1291 (void*)(unsigned long)(new_size & 0xffffffff),
1292 (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
1293 (void*)(unsigned long)(toss_start & 0xffffffff),
1294 (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
1295 (void*)(unsigned long)(toss_finish & 0xffffffff),
1296 (void*)(unsigned long)current_cpu(),
1297 (void*)(unsigned long)current_pid(),
1303 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1307 * Start the truncation of the file to new_size. The new size
1308 * must be smaller than the current size. This routine will
1309 * clear the buffer and page caches of file data in the removed
1310 * range, and xfs_itruncate_finish() will remove the underlying
1313 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1314 * must NOT have the inode lock held at all. This is because we're
1315 * calling into the buffer/page cache code and we can't hold the
1316 * inode lock when we do so.
1318 * We need to wait for any direct I/Os in flight to complete before we
1319 * proceed with the truncate. This is needed to prevent the extents
1320 * being read or written by the direct I/Os from being removed while the
1321 * I/O is in flight as there is no other method of synchronising
1322 * direct I/O with the truncate operation. Also, because we hold
1323 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1324 * started until the truncate completes and drops the lock. Essentially,
1325 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1326 * between direct I/Os and the truncate operation.
1328 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1329 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1330 * in the case that the caller is locking things out of order and
1331 * may not be able to call xfs_itruncate_finish() with the inode lock
1332 * held without dropping the I/O lock. If the caller must drop the
1333 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1334 * must be called again with all the same restrictions as the initial
1338 xfs_itruncate_start(
1341 xfs_fsize_t new_size)
1343 xfs_fsize_t last_byte;
1344 xfs_off_t toss_start;
1348 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1349 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1350 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1351 (flags == XFS_ITRUNC_MAYBE));
1355 /* wait for the completion of any pending DIOs */
1356 if (new_size == 0 || new_size < ip->i_size)
1360 * Call toss_pages or flushinval_pages to get rid of pages
1361 * overlapping the region being removed. We have to use
1362 * the less efficient flushinval_pages in the case that the
1363 * caller may not be able to finish the truncate without
1364 * dropping the inode's I/O lock. Make sure
1365 * to catch any pages brought in by buffers overlapping
1366 * the EOF by searching out beyond the isize by our
1367 * block size. We round new_size up to a block boundary
1368 * so that we don't toss things on the same block as
1369 * new_size but before it.
1371 * Before calling toss_page or flushinval_pages, make sure to
1372 * call remapf() over the same region if the file is mapped.
1373 * This frees up mapped file references to the pages in the
1374 * given range and for the flushinval_pages case it ensures
1375 * that we get the latest mapped changes flushed out.
1377 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1378 toss_start = XFS_FSB_TO_B(mp, toss_start);
1379 if (toss_start < 0) {
1381 * The place to start tossing is beyond our maximum
1382 * file size, so there is no way that the data extended
1387 last_byte = xfs_file_last_byte(ip);
1388 xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
1390 if (last_byte > toss_start) {
1391 if (flags & XFS_ITRUNC_DEFINITE) {
1392 xfs_tosspages(ip, toss_start,
1393 -1, FI_REMAPF_LOCKED);
1395 error = xfs_flushinval_pages(ip, toss_start,
1396 -1, FI_REMAPF_LOCKED);
1401 if (new_size == 0) {
1402 ASSERT(VN_CACHED(VFS_I(ip)) == 0);
1409 * Shrink the file to the given new_size. The new size must be smaller than
1410 * the current size. This will free up the underlying blocks in the removed
1411 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
1413 * The transaction passed to this routine must have made a permanent log
1414 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1415 * given transaction and start new ones, so make sure everything involved in
1416 * the transaction is tidy before calling here. Some transaction will be
1417 * returned to the caller to be committed. The incoming transaction must
1418 * already include the inode, and both inode locks must be held exclusively.
1419 * The inode must also be "held" within the transaction. On return the inode
1420 * will be "held" within the returned transaction. This routine does NOT
1421 * require any disk space to be reserved for it within the transaction.
1423 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1424 * indicates the fork which is to be truncated. For the attribute fork we only
1425 * support truncation to size 0.
1427 * We use the sync parameter to indicate whether or not the first transaction
1428 * we perform might have to be synchronous. For the attr fork, it needs to be
1429 * so if the unlink of the inode is not yet known to be permanent in the log.
1430 * This keeps us from freeing and reusing the blocks of the attribute fork
1431 * before the unlink of the inode becomes permanent.
1433 * For the data fork, we normally have to run synchronously if we're being
1434 * called out of the inactive path or we're being called out of the create path
1435 * where we're truncating an existing file. Either way, the truncate needs to
1436 * be sync so blocks don't reappear in the file with altered data in case of a
1437 * crash. wsync filesystems can run the first case async because anything that
1438 * shrinks the inode has to run sync so by the time we're called here from
1439 * inactive, the inode size is permanently set to 0.
1441 * Calls from the truncate path always need to be sync unless we're in a wsync
1442 * filesystem and the file has already been unlinked.
1444 * The caller is responsible for correctly setting the sync parameter. It gets
1445 * too hard for us to guess here which path we're being called out of just
1446 * based on inode state.
1448 * If we get an error, we must return with the inode locked and linked into the
1449 * current transaction. This keeps things simple for the higher level code,
1450 * because it always knows that the inode is locked and held in the transaction
1451 * that returns to it whether errors occur or not. We don't mark the inode
1452 * dirty on error so that transactions can be easily aborted if possible.
1455 xfs_itruncate_finish(
1458 xfs_fsize_t new_size,
1462 xfs_fsblock_t first_block;
1463 xfs_fileoff_t first_unmap_block;
1464 xfs_fileoff_t last_block;
1465 xfs_filblks_t unmap_len=0;
1470 xfs_bmap_free_t free_list;
1473 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
1474 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1475 ASSERT(*tp != NULL);
1476 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1477 ASSERT(ip->i_transp == *tp);
1478 ASSERT(ip->i_itemp != NULL);
1479 ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
1483 mp = (ntp)->t_mountp;
1484 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1487 * We only support truncating the entire attribute fork.
1489 if (fork == XFS_ATTR_FORK) {
1492 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1493 xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
1495 * The first thing we do is set the size to new_size permanently
1496 * on disk. This way we don't have to worry about anyone ever
1497 * being able to look at the data being freed even in the face
1498 * of a crash. What we're getting around here is the case where
1499 * we free a block, it is allocated to another file, it is written
1500 * to, and then we crash. If the new data gets written to the
1501 * file but the log buffers containing the free and reallocation
1502 * don't, then we'd end up with garbage in the blocks being freed.
1503 * As long as we make the new_size permanent before actually
1504 * freeing any blocks it doesn't matter if they get writtten to.
1506 * The callers must signal into us whether or not the size
1507 * setting here must be synchronous. There are a few cases
1508 * where it doesn't have to be synchronous. Those cases
1509 * occur if the file is unlinked and we know the unlink is
1510 * permanent or if the blocks being truncated are guaranteed
1511 * to be beyond the inode eof (regardless of the link count)
1512 * and the eof value is permanent. Both of these cases occur
1513 * only on wsync-mounted filesystems. In those cases, we're
1514 * guaranteed that no user will ever see the data in the blocks
1515 * that are being truncated so the truncate can run async.
1516 * In the free beyond eof case, the file may wind up with
1517 * more blocks allocated to it than it needs if we crash
1518 * and that won't get fixed until the next time the file
1519 * is re-opened and closed but that's ok as that shouldn't
1520 * be too many blocks.
1522 * However, we can't just make all wsync xactions run async
1523 * because there's one call out of the create path that needs
1524 * to run sync where it's truncating an existing file to size
1525 * 0 whose size is > 0.
1527 * It's probably possible to come up with a test in this
1528 * routine that would correctly distinguish all the above
1529 * cases from the values of the function parameters and the
1530 * inode state but for sanity's sake, I've decided to let the
1531 * layers above just tell us. It's simpler to correctly figure
1532 * out in the layer above exactly under what conditions we
1533 * can run async and I think it's easier for others read and
1534 * follow the logic in case something has to be changed.
1535 * cscope is your friend -- rcc.
1537 * The attribute fork is much simpler.
1539 * For the attribute fork we allow the caller to tell us whether
1540 * the unlink of the inode that led to this call is yet permanent
1541 * in the on disk log. If it is not and we will be freeing extents
1542 * in this inode then we make the first transaction synchronous
1543 * to make sure that the unlink is permanent by the time we free
1546 if (fork == XFS_DATA_FORK) {
1547 if (ip->i_d.di_nextents > 0) {
1549 * If we are not changing the file size then do
1550 * not update the on-disk file size - we may be
1551 * called from xfs_inactive_free_eofblocks(). If we
1552 * update the on-disk file size and then the system
1553 * crashes before the contents of the file are
1554 * flushed to disk then the files may be full of
1555 * holes (ie NULL files bug).
1557 if (ip->i_size != new_size) {
1558 ip->i_d.di_size = new_size;
1559 ip->i_size = new_size;
1560 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1564 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1565 if (ip->i_d.di_anextents > 0)
1566 xfs_trans_set_sync(ntp);
1568 ASSERT(fork == XFS_DATA_FORK ||
1569 (fork == XFS_ATTR_FORK &&
1570 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1571 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1574 * Since it is possible for space to become allocated beyond
1575 * the end of the file (in a crash where the space is allocated
1576 * but the inode size is not yet updated), simply remove any
1577 * blocks which show up between the new EOF and the maximum
1578 * possible file size. If the first block to be removed is
1579 * beyond the maximum file size (ie it is the same as last_block),
1580 * then there is nothing to do.
1582 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1583 ASSERT(first_unmap_block <= last_block);
1585 if (last_block == first_unmap_block) {
1588 unmap_len = last_block - first_unmap_block + 1;
1592 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1593 * will tell us whether it freed the entire range or
1594 * not. If this is a synchronous mount (wsync),
1595 * then we can tell bunmapi to keep all the
1596 * transactions asynchronous since the unlink
1597 * transaction that made this inode inactive has
1598 * already hit the disk. There's no danger of
1599 * the freed blocks being reused, there being a
1600 * crash, and the reused blocks suddenly reappearing
1601 * in this file with garbage in them once recovery
1604 XFS_BMAP_INIT(&free_list, &first_block);
1605 error = xfs_bunmapi(ntp, ip,
1606 first_unmap_block, unmap_len,
1607 XFS_BMAPI_AFLAG(fork) |
1608 (sync ? 0 : XFS_BMAPI_ASYNC),
1609 XFS_ITRUNC_MAX_EXTENTS,
1610 &first_block, &free_list,
1614 * If the bunmapi call encounters an error,
1615 * return to the caller where the transaction
1616 * can be properly aborted. We just need to
1617 * make sure we're not holding any resources
1618 * that we were not when we came in.
1620 xfs_bmap_cancel(&free_list);
1625 * Duplicate the transaction that has the permanent
1626 * reservation and commit the old transaction.
1628 error = xfs_bmap_finish(tp, &free_list, &committed);
1631 /* link the inode into the next xact in the chain */
1632 xfs_trans_ijoin(ntp, ip,
1633 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1634 xfs_trans_ihold(ntp, ip);
1639 * If the bmap finish call encounters an error, return
1640 * to the caller where the transaction can be properly
1641 * aborted. We just need to make sure we're not
1642 * holding any resources that we were not when we came
1645 * Aborting from this point might lose some blocks in
1646 * the file system, but oh well.
1648 xfs_bmap_cancel(&free_list);
1654 * Mark the inode dirty so it will be logged and
1655 * moved forward in the log as part of every commit.
1657 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1660 ntp = xfs_trans_dup(ntp);
1661 error = xfs_trans_commit(*tp, 0);
1664 /* link the inode into the next transaction in the chain */
1665 xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1666 xfs_trans_ihold(ntp, ip);
1671 * transaction commit worked ok so we can drop the extra ticket
1672 * reference that we gained in xfs_trans_dup()
1674 xfs_log_ticket_put(ntp->t_ticket);
1675 error = xfs_trans_reserve(ntp, 0,
1676 XFS_ITRUNCATE_LOG_RES(mp), 0,
1677 XFS_TRANS_PERM_LOG_RES,
1678 XFS_ITRUNCATE_LOG_COUNT);
1683 * Only update the size in the case of the data fork, but
1684 * always re-log the inode so that our permanent transaction
1685 * can keep on rolling it forward in the log.
1687 if (fork == XFS_DATA_FORK) {
1688 xfs_isize_check(mp, ip, new_size);
1690 * If we are not changing the file size then do
1691 * not update the on-disk file size - we may be
1692 * called from xfs_inactive_free_eofblocks(). If we
1693 * update the on-disk file size and then the system
1694 * crashes before the contents of the file are
1695 * flushed to disk then the files may be full of
1696 * holes (ie NULL files bug).
1698 if (ip->i_size != new_size) {
1699 ip->i_d.di_size = new_size;
1700 ip->i_size = new_size;
1703 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1704 ASSERT((new_size != 0) ||
1705 (fork == XFS_ATTR_FORK) ||
1706 (ip->i_delayed_blks == 0));
1707 ASSERT((new_size != 0) ||
1708 (fork == XFS_ATTR_FORK) ||
1709 (ip->i_d.di_nextents == 0));
1710 xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
1715 * This is called when the inode's link count goes to 0.
1716 * We place the on-disk inode on a list in the AGI. It
1717 * will be pulled from this list when the inode is freed.
1734 ASSERT(ip->i_d.di_nlink == 0);
1735 ASSERT(ip->i_d.di_mode != 0);
1736 ASSERT(ip->i_transp == tp);
1741 * Get the agi buffer first. It ensures lock ordering
1744 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1747 agi = XFS_BUF_TO_AGI(agibp);
1750 * Get the index into the agi hash table for the
1751 * list this inode will go on.
1753 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1755 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1756 ASSERT(agi->agi_unlinked[bucket_index]);
1757 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1759 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
1761 * There is already another inode in the bucket we need
1762 * to add ourselves to. Add us at the front of the list.
1763 * Here we put the head pointer into our next pointer,
1764 * and then we fall through to point the head at us.
1766 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XFS_BUF_LOCK);
1770 ASSERT(be32_to_cpu(dip->di_next_unlinked) == NULLAGINO);
1771 /* both on-disk, don't endian flip twice */
1772 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1773 offset = ip->i_imap.im_boffset +
1774 offsetof(xfs_dinode_t, di_next_unlinked);
1775 xfs_trans_inode_buf(tp, ibp);
1776 xfs_trans_log_buf(tp, ibp, offset,
1777 (offset + sizeof(xfs_agino_t) - 1));
1778 xfs_inobp_check(mp, ibp);
1782 * Point the bucket head pointer at the inode being inserted.
1785 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1786 offset = offsetof(xfs_agi_t, agi_unlinked) +
1787 (sizeof(xfs_agino_t) * bucket_index);
1788 xfs_trans_log_buf(tp, agibp, offset,
1789 (offset + sizeof(xfs_agino_t) - 1));
1794 * Pull the on-disk inode from the AGI unlinked list.
1807 xfs_agnumber_t agno;
1809 xfs_agino_t next_agino;
1810 xfs_buf_t *last_ibp;
1811 xfs_dinode_t *last_dip = NULL;
1813 int offset, last_offset = 0;
1817 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1820 * Get the agi buffer first. It ensures lock ordering
1823 error = xfs_read_agi(mp, tp, agno, &agibp);
1827 agi = XFS_BUF_TO_AGI(agibp);
1830 * Get the index into the agi hash table for the
1831 * list this inode will go on.
1833 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1835 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1836 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
1837 ASSERT(agi->agi_unlinked[bucket_index]);
1839 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
1841 * We're at the head of the list. Get the inode's
1842 * on-disk buffer to see if there is anyone after us
1843 * on the list. Only modify our next pointer if it
1844 * is not already NULLAGINO. This saves us the overhead
1845 * of dealing with the buffer when there is no need to
1848 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XFS_BUF_LOCK);
1851 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
1852 error, mp->m_fsname);
1855 next_agino = be32_to_cpu(dip->di_next_unlinked);
1856 ASSERT(next_agino != 0);
1857 if (next_agino != NULLAGINO) {
1858 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
1859 offset = ip->i_imap.im_boffset +
1860 offsetof(xfs_dinode_t, di_next_unlinked);
1861 xfs_trans_inode_buf(tp, ibp);
1862 xfs_trans_log_buf(tp, ibp, offset,
1863 (offset + sizeof(xfs_agino_t) - 1));
1864 xfs_inobp_check(mp, ibp);
1866 xfs_trans_brelse(tp, ibp);
1869 * Point the bucket head pointer at the next inode.
1871 ASSERT(next_agino != 0);
1872 ASSERT(next_agino != agino);
1873 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
1874 offset = offsetof(xfs_agi_t, agi_unlinked) +
1875 (sizeof(xfs_agino_t) * bucket_index);
1876 xfs_trans_log_buf(tp, agibp, offset,
1877 (offset + sizeof(xfs_agino_t) - 1));
1880 * We need to search the list for the inode being freed.
1882 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
1884 while (next_agino != agino) {
1886 * If the last inode wasn't the one pointing to
1887 * us, then release its buffer since we're not
1888 * going to do anything with it.
1890 if (last_ibp != NULL) {
1891 xfs_trans_brelse(tp, last_ibp);
1893 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
1894 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
1895 &last_ibp, &last_offset, 0);
1898 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
1899 error, mp->m_fsname);
1902 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
1903 ASSERT(next_agino != NULLAGINO);
1904 ASSERT(next_agino != 0);
1907 * Now last_ibp points to the buffer previous to us on
1908 * the unlinked list. Pull us from the list.
1910 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XFS_BUF_LOCK);
1913 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
1914 error, mp->m_fsname);
1917 next_agino = be32_to_cpu(dip->di_next_unlinked);
1918 ASSERT(next_agino != 0);
1919 ASSERT(next_agino != agino);
1920 if (next_agino != NULLAGINO) {
1921 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
1922 offset = ip->i_imap.im_boffset +
1923 offsetof(xfs_dinode_t, di_next_unlinked);
1924 xfs_trans_inode_buf(tp, ibp);
1925 xfs_trans_log_buf(tp, ibp, offset,
1926 (offset + sizeof(xfs_agino_t) - 1));
1927 xfs_inobp_check(mp, ibp);
1929 xfs_trans_brelse(tp, ibp);
1932 * Point the previous inode on the list to the next inode.
1934 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
1935 ASSERT(next_agino != 0);
1936 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
1937 xfs_trans_inode_buf(tp, last_ibp);
1938 xfs_trans_log_buf(tp, last_ibp, offset,
1939 (offset + sizeof(xfs_agino_t) - 1));
1940 xfs_inobp_check(mp, last_ibp);
1947 xfs_inode_t *free_ip,
1951 xfs_mount_t *mp = free_ip->i_mount;
1952 int blks_per_cluster;
1955 int i, j, found, pre_flushed;
1958 xfs_inode_t *ip, **ip_found;
1959 xfs_inode_log_item_t *iip;
1960 xfs_log_item_t *lip;
1961 xfs_perag_t *pag = xfs_get_perag(mp, inum);
1963 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
1964 blks_per_cluster = 1;
1965 ninodes = mp->m_sb.sb_inopblock;
1966 nbufs = XFS_IALLOC_BLOCKS(mp);
1968 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
1969 mp->m_sb.sb_blocksize;
1970 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
1971 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
1974 ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
1976 for (j = 0; j < nbufs; j++, inum += ninodes) {
1977 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
1978 XFS_INO_TO_AGBNO(mp, inum));
1982 * Look for each inode in memory and attempt to lock it,
1983 * we can be racing with flush and tail pushing here.
1984 * any inode we get the locks on, add to an array of
1985 * inode items to process later.
1987 * The get the buffer lock, we could beat a flush
1988 * or tail pushing thread to the lock here, in which
1989 * case they will go looking for the inode buffer
1990 * and fail, we need some other form of interlock
1994 for (i = 0; i < ninodes; i++) {
1995 read_lock(&pag->pag_ici_lock);
1996 ip = radix_tree_lookup(&pag->pag_ici_root,
1997 XFS_INO_TO_AGINO(mp, (inum + i)));
1999 /* Inode not in memory or we found it already,
2002 if (!ip || xfs_iflags_test(ip, XFS_ISTALE)) {
2003 read_unlock(&pag->pag_ici_lock);
2007 if (xfs_inode_clean(ip)) {
2008 read_unlock(&pag->pag_ici_lock);
2012 /* If we can get the locks then add it to the
2013 * list, otherwise by the time we get the bp lock
2014 * below it will already be attached to the
2018 /* This inode will already be locked - by us, lets
2022 if (ip == free_ip) {
2023 if (xfs_iflock_nowait(ip)) {
2024 xfs_iflags_set(ip, XFS_ISTALE);
2025 if (xfs_inode_clean(ip)) {
2028 ip_found[found++] = ip;
2031 read_unlock(&pag->pag_ici_lock);
2035 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2036 if (xfs_iflock_nowait(ip)) {
2037 xfs_iflags_set(ip, XFS_ISTALE);
2039 if (xfs_inode_clean(ip)) {
2041 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2043 ip_found[found++] = ip;
2046 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2049 read_unlock(&pag->pag_ici_lock);
2052 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2053 mp->m_bsize * blks_per_cluster,
2057 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
2059 if (lip->li_type == XFS_LI_INODE) {
2060 iip = (xfs_inode_log_item_t *)lip;
2061 ASSERT(iip->ili_logged == 1);
2062 lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
2063 xfs_trans_ail_copy_lsn(mp->m_ail,
2064 &iip->ili_flush_lsn,
2065 &iip->ili_item.li_lsn);
2066 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2069 lip = lip->li_bio_list;
2072 for (i = 0; i < found; i++) {
2077 ip->i_update_core = 0;
2079 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2083 iip->ili_last_fields = iip->ili_format.ilf_fields;
2084 iip->ili_format.ilf_fields = 0;
2085 iip->ili_logged = 1;
2086 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2087 &iip->ili_item.li_lsn);
2089 xfs_buf_attach_iodone(bp,
2090 (void(*)(xfs_buf_t*,xfs_log_item_t*))
2091 xfs_istale_done, (xfs_log_item_t *)iip);
2092 if (ip != free_ip) {
2093 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2097 if (found || pre_flushed)
2098 xfs_trans_stale_inode_buf(tp, bp);
2099 xfs_trans_binval(tp, bp);
2102 kmem_free(ip_found);
2103 xfs_put_perag(mp, pag);
2107 * This is called to return an inode to the inode free list.
2108 * The inode should already be truncated to 0 length and have
2109 * no pages associated with it. This routine also assumes that
2110 * the inode is already a part of the transaction.
2112 * The on-disk copy of the inode will have been added to the list
2113 * of unlinked inodes in the AGI. We need to remove the inode from
2114 * that list atomically with respect to freeing it here.
2120 xfs_bmap_free_t *flist)
2124 xfs_ino_t first_ino;
2128 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2129 ASSERT(ip->i_transp == tp);
2130 ASSERT(ip->i_d.di_nlink == 0);
2131 ASSERT(ip->i_d.di_nextents == 0);
2132 ASSERT(ip->i_d.di_anextents == 0);
2133 ASSERT((ip->i_d.di_size == 0 && ip->i_size == 0) ||
2134 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2135 ASSERT(ip->i_d.di_nblocks == 0);
2138 * Pull the on-disk inode from the AGI unlinked list.
2140 error = xfs_iunlink_remove(tp, ip);
2145 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2149 ip->i_d.di_mode = 0; /* mark incore inode as free */
2150 ip->i_d.di_flags = 0;
2151 ip->i_d.di_dmevmask = 0;
2152 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2153 ip->i_df.if_ext_max =
2154 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2155 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2156 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2158 * Bump the generation count so no one will be confused
2159 * by reincarnations of this inode.
2163 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2165 error = xfs_itobp(ip->i_mount, tp, ip, &dip, &ibp, XFS_BUF_LOCK);
2170 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2171 * from picking up this inode when it is reclaimed (its incore state
2172 * initialzed but not flushed to disk yet). The in-core di_mode is
2173 * already cleared and a corresponding transaction logged.
2174 * The hack here just synchronizes the in-core to on-disk
2175 * di_mode value in advance before the actual inode sync to disk.
2176 * This is OK because the inode is already unlinked and would never
2177 * change its di_mode again for this inode generation.
2178 * This is a temporary hack that would require a proper fix
2184 xfs_ifree_cluster(ip, tp, first_ino);
2191 * Reallocate the space for if_broot based on the number of records
2192 * being added or deleted as indicated in rec_diff. Move the records
2193 * and pointers in if_broot to fit the new size. When shrinking this
2194 * will eliminate holes between the records and pointers created by
2195 * the caller. When growing this will create holes to be filled in
2198 * The caller must not request to add more records than would fit in
2199 * the on-disk inode root. If the if_broot is currently NULL, then
2200 * if we adding records one will be allocated. The caller must also
2201 * not request that the number of records go below zero, although
2202 * it can go to zero.
2204 * ip -- the inode whose if_broot area is changing
2205 * ext_diff -- the change in the number of records, positive or negative,
2206 * requested for the if_broot array.
2214 struct xfs_mount *mp = ip->i_mount;
2217 struct xfs_btree_block *new_broot;
2224 * Handle the degenerate case quietly.
2226 if (rec_diff == 0) {
2230 ifp = XFS_IFORK_PTR(ip, whichfork);
2233 * If there wasn't any memory allocated before, just
2234 * allocate it now and get out.
2236 if (ifp->if_broot_bytes == 0) {
2237 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
2238 ifp->if_broot = kmem_alloc(new_size, KM_SLEEP);
2239 ifp->if_broot_bytes = (int)new_size;
2244 * If there is already an existing if_broot, then we need
2245 * to realloc() it and shift the pointers to their new
2246 * location. The records don't change location because
2247 * they are kept butted up against the btree block header.
2249 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
2250 new_max = cur_max + rec_diff;
2251 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2252 ifp->if_broot = kmem_realloc(ifp->if_broot, new_size,
2253 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2255 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2256 ifp->if_broot_bytes);
2257 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2259 ifp->if_broot_bytes = (int)new_size;
2260 ASSERT(ifp->if_broot_bytes <=
2261 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2262 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2267 * rec_diff is less than 0. In this case, we are shrinking the
2268 * if_broot buffer. It must already exist. If we go to zero
2269 * records, just get rid of the root and clear the status bit.
2271 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
2272 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
2273 new_max = cur_max + rec_diff;
2274 ASSERT(new_max >= 0);
2276 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2280 new_broot = kmem_alloc(new_size, KM_SLEEP);
2282 * First copy over the btree block header.
2284 memcpy(new_broot, ifp->if_broot, XFS_BTREE_LBLOCK_LEN);
2287 ifp->if_flags &= ~XFS_IFBROOT;
2291 * Only copy the records and pointers if there are any.
2295 * First copy the records.
2297 op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1);
2298 np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1);
2299 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2302 * Then copy the pointers.
2304 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2305 ifp->if_broot_bytes);
2306 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1,
2308 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2310 kmem_free(ifp->if_broot);
2311 ifp->if_broot = new_broot;
2312 ifp->if_broot_bytes = (int)new_size;
2313 ASSERT(ifp->if_broot_bytes <=
2314 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2320 * This is called when the amount of space needed for if_data
2321 * is increased or decreased. The change in size is indicated by
2322 * the number of bytes that need to be added or deleted in the
2323 * byte_diff parameter.
2325 * If the amount of space needed has decreased below the size of the
2326 * inline buffer, then switch to using the inline buffer. Otherwise,
2327 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2328 * to what is needed.
2330 * ip -- the inode whose if_data area is changing
2331 * byte_diff -- the change in the number of bytes, positive or negative,
2332 * requested for the if_data array.
2344 if (byte_diff == 0) {
2348 ifp = XFS_IFORK_PTR(ip, whichfork);
2349 new_size = (int)ifp->if_bytes + byte_diff;
2350 ASSERT(new_size >= 0);
2352 if (new_size == 0) {
2353 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2354 kmem_free(ifp->if_u1.if_data);
2356 ifp->if_u1.if_data = NULL;
2358 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2360 * If the valid extents/data can fit in if_inline_ext/data,
2361 * copy them from the malloc'd vector and free it.
2363 if (ifp->if_u1.if_data == NULL) {
2364 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2365 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2366 ASSERT(ifp->if_real_bytes != 0);
2367 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2369 kmem_free(ifp->if_u1.if_data);
2370 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2375 * Stuck with malloc/realloc.
2376 * For inline data, the underlying buffer must be
2377 * a multiple of 4 bytes in size so that it can be
2378 * logged and stay on word boundaries. We enforce
2381 real_size = roundup(new_size, 4);
2382 if (ifp->if_u1.if_data == NULL) {
2383 ASSERT(ifp->if_real_bytes == 0);
2384 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2385 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2387 * Only do the realloc if the underlying size
2388 * is really changing.
2390 if (ifp->if_real_bytes != real_size) {
2391 ifp->if_u1.if_data =
2392 kmem_realloc(ifp->if_u1.if_data,
2398 ASSERT(ifp->if_real_bytes == 0);
2399 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2400 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2404 ifp->if_real_bytes = real_size;
2405 ifp->if_bytes = new_size;
2406 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2416 ifp = XFS_IFORK_PTR(ip, whichfork);
2417 if (ifp->if_broot != NULL) {
2418 kmem_free(ifp->if_broot);
2419 ifp->if_broot = NULL;
2423 * If the format is local, then we can't have an extents
2424 * array so just look for an inline data array. If we're
2425 * not local then we may or may not have an extents list,
2426 * so check and free it up if we do.
2428 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2429 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2430 (ifp->if_u1.if_data != NULL)) {
2431 ASSERT(ifp->if_real_bytes != 0);
2432 kmem_free(ifp->if_u1.if_data);
2433 ifp->if_u1.if_data = NULL;
2434 ifp->if_real_bytes = 0;
2436 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2437 ((ifp->if_flags & XFS_IFEXTIREC) ||
2438 ((ifp->if_u1.if_extents != NULL) &&
2439 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
2440 ASSERT(ifp->if_real_bytes != 0);
2441 xfs_iext_destroy(ifp);
2443 ASSERT(ifp->if_u1.if_extents == NULL ||
2444 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2445 ASSERT(ifp->if_real_bytes == 0);
2446 if (whichfork == XFS_ATTR_FORK) {
2447 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2453 * Increment the pin count of the given buffer.
2454 * This value is protected by ipinlock spinlock in the mount structure.
2460 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2462 atomic_inc(&ip->i_pincount);
2466 * Decrement the pin count of the given inode, and wake up
2467 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2468 * inode must have been previously pinned with a call to xfs_ipin().
2474 ASSERT(atomic_read(&ip->i_pincount) > 0);
2476 if (atomic_dec_and_test(&ip->i_pincount))
2477 wake_up(&ip->i_ipin_wait);
2481 * This is called to unpin an inode. It can be directed to wait or to return
2482 * immediately without waiting for the inode to be unpinned. The caller must
2483 * have the inode locked in at least shared mode so that the buffer cannot be
2484 * subsequently pinned once someone is waiting for it to be unpinned.
2491 xfs_inode_log_item_t *iip = ip->i_itemp;
2493 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2494 if (atomic_read(&ip->i_pincount) == 0)
2497 /* Give the log a push to start the unpinning I/O */
2498 xfs_log_force(ip->i_mount, (iip && iip->ili_last_lsn) ?
2499 iip->ili_last_lsn : 0, XFS_LOG_FORCE);
2501 wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
2508 __xfs_iunpin_wait(ip, 1);
2515 __xfs_iunpin_wait(ip, 0);
2520 * xfs_iextents_copy()
2522 * This is called to copy the REAL extents (as opposed to the delayed
2523 * allocation extents) from the inode into the given buffer. It
2524 * returns the number of bytes copied into the buffer.
2526 * If there are no delayed allocation extents, then we can just
2527 * memcpy() the extents into the buffer. Otherwise, we need to
2528 * examine each extent in turn and skip those which are delayed.
2540 xfs_fsblock_t start_block;
2542 ifp = XFS_IFORK_PTR(ip, whichfork);
2543 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2544 ASSERT(ifp->if_bytes > 0);
2546 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2547 XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork);
2551 * There are some delayed allocation extents in the
2552 * inode, so copy the extents one at a time and skip
2553 * the delayed ones. There must be at least one
2554 * non-delayed extent.
2557 for (i = 0; i < nrecs; i++) {
2558 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
2559 start_block = xfs_bmbt_get_startblock(ep);
2560 if (ISNULLSTARTBLOCK(start_block)) {
2562 * It's a delayed allocation extent, so skip it.
2567 /* Translate to on disk format */
2568 put_unaligned(cpu_to_be64(ep->l0), &dp->l0);
2569 put_unaligned(cpu_to_be64(ep->l1), &dp->l1);
2573 ASSERT(copied != 0);
2574 xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip));
2576 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2580 * Each of the following cases stores data into the same region
2581 * of the on-disk inode, so only one of them can be valid at
2582 * any given time. While it is possible to have conflicting formats
2583 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2584 * in EXTENTS format, this can only happen when the fork has
2585 * changed formats after being modified but before being flushed.
2586 * In these cases, the format always takes precedence, because the
2587 * format indicates the current state of the fork.
2594 xfs_inode_log_item_t *iip,
2601 #ifdef XFS_TRANS_DEBUG
2604 static const short brootflag[2] =
2605 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2606 static const short dataflag[2] =
2607 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2608 static const short extflag[2] =
2609 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2613 ifp = XFS_IFORK_PTR(ip, whichfork);
2615 * This can happen if we gave up in iformat in an error path,
2616 * for the attribute fork.
2619 ASSERT(whichfork == XFS_ATTR_FORK);
2622 cp = XFS_DFORK_PTR(dip, whichfork);
2624 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2625 case XFS_DINODE_FMT_LOCAL:
2626 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2627 (ifp->if_bytes > 0)) {
2628 ASSERT(ifp->if_u1.if_data != NULL);
2629 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2630 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2634 case XFS_DINODE_FMT_EXTENTS:
2635 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2636 !(iip->ili_format.ilf_fields & extflag[whichfork]));
2637 ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
2638 (ifp->if_bytes == 0));
2639 ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
2640 (ifp->if_bytes > 0));
2641 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
2642 (ifp->if_bytes > 0)) {
2643 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2644 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2649 case XFS_DINODE_FMT_BTREE:
2650 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
2651 (ifp->if_broot_bytes > 0)) {
2652 ASSERT(ifp->if_broot != NULL);
2653 ASSERT(ifp->if_broot_bytes <=
2654 (XFS_IFORK_SIZE(ip, whichfork) +
2655 XFS_BROOT_SIZE_ADJ));
2656 xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes,
2657 (xfs_bmdr_block_t *)cp,
2658 XFS_DFORK_SIZE(dip, mp, whichfork));
2662 case XFS_DINODE_FMT_DEV:
2663 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
2664 ASSERT(whichfork == XFS_DATA_FORK);
2665 xfs_dinode_put_rdev(dip, ip->i_df.if_u2.if_rdev);
2669 case XFS_DINODE_FMT_UUID:
2670 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
2671 ASSERT(whichfork == XFS_DATA_FORK);
2672 memcpy(XFS_DFORK_DPTR(dip),
2673 &ip->i_df.if_u2.if_uuid,
2689 xfs_mount_t *mp = ip->i_mount;
2690 xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino);
2691 unsigned long first_index, mask;
2692 unsigned long inodes_per_cluster;
2694 xfs_inode_t **ilist;
2701 ASSERT(pag->pagi_inodeok);
2702 ASSERT(pag->pag_ici_init);
2704 inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog;
2705 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
2706 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
2710 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
2711 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
2712 read_lock(&pag->pag_ici_lock);
2713 /* really need a gang lookup range call here */
2714 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
2715 first_index, inodes_per_cluster);
2719 for (i = 0; i < nr_found; i++) {
2723 /* if the inode lies outside this cluster, we're done. */
2724 if ((XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index)
2727 * Do an un-protected check to see if the inode is dirty and
2728 * is a candidate for flushing. These checks will be repeated
2729 * later after the appropriate locks are acquired.
2731 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
2735 * Try to get locks. If any are unavailable or it is pinned,
2736 * then this inode cannot be flushed and is skipped.
2739 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
2741 if (!xfs_iflock_nowait(iq)) {
2742 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2745 if (xfs_ipincount(iq)) {
2747 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2752 * arriving here means that this inode can be flushed. First
2753 * re-check that it's dirty before flushing.
2755 if (!xfs_inode_clean(iq)) {
2757 error = xfs_iflush_int(iq, bp);
2759 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2760 goto cluster_corrupt_out;
2766 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2770 XFS_STATS_INC(xs_icluster_flushcnt);
2771 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
2775 read_unlock(&pag->pag_ici_lock);
2780 cluster_corrupt_out:
2782 * Corruption detected in the clustering loop. Invalidate the
2783 * inode buffer and shut down the filesystem.
2785 read_unlock(&pag->pag_ici_lock);
2787 * Clean up the buffer. If it was B_DELWRI, just release it --
2788 * brelse can handle it with no problems. If not, shut down the
2789 * filesystem before releasing the buffer.
2791 bufwasdelwri = XFS_BUF_ISDELAYWRITE(bp);
2795 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2797 if (!bufwasdelwri) {
2799 * Just like incore_relse: if we have b_iodone functions,
2800 * mark the buffer as an error and call them. Otherwise
2801 * mark it as stale and brelse.
2803 if (XFS_BUF_IODONE_FUNC(bp)) {
2804 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
2808 XFS_BUF_ERROR(bp,EIO);
2817 * Unlocks the flush lock
2819 xfs_iflush_abort(iq);
2821 return XFS_ERROR(EFSCORRUPTED);
2825 * xfs_iflush() will write a modified inode's changes out to the
2826 * inode's on disk home. The caller must have the inode lock held
2827 * in at least shared mode and the inode flush completion must be
2828 * active as well. The inode lock will still be held upon return from
2829 * the call and the caller is free to unlock it.
2830 * The inode flush will be completed when the inode reaches the disk.
2831 * The flags indicate how the inode's buffer should be written out.
2838 xfs_inode_log_item_t *iip;
2843 int noblock = (flags == XFS_IFLUSH_ASYNC_NOBLOCK);
2844 enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
2846 XFS_STATS_INC(xs_iflush_count);
2848 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2849 ASSERT(!completion_done(&ip->i_flush));
2850 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2851 ip->i_d.di_nextents > ip->i_df.if_ext_max);
2857 * If the inode isn't dirty, then just release the inode
2858 * flush lock and do nothing.
2860 if (xfs_inode_clean(ip)) {
2866 * We can't flush the inode until it is unpinned, so wait for it if we
2867 * are allowed to block. We know noone new can pin it, because we are
2868 * holding the inode lock shared and you need to hold it exclusively to
2871 * If we are not allowed to block, force the log out asynchronously so
2872 * that when we come back the inode will be unpinned. If other inodes
2873 * in the same cluster are dirty, they will probably write the inode
2874 * out for us if they occur after the log force completes.
2876 if (noblock && xfs_ipincount(ip)) {
2877 xfs_iunpin_nowait(ip);
2881 xfs_iunpin_wait(ip);
2884 * This may have been unpinned because the filesystem is shutting
2885 * down forcibly. If that's the case we must not write this inode
2886 * to disk, because the log record didn't make it to disk!
2888 if (XFS_FORCED_SHUTDOWN(mp)) {
2889 ip->i_update_core = 0;
2891 iip->ili_format.ilf_fields = 0;
2893 return XFS_ERROR(EIO);
2897 * Decide how buffer will be flushed out. This is done before
2898 * the call to xfs_iflush_int because this field is zeroed by it.
2900 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
2902 * Flush out the inode buffer according to the directions
2903 * of the caller. In the cases where the caller has given
2904 * us a choice choose the non-delwri case. This is because
2905 * the inode is in the AIL and we need to get it out soon.
2908 case XFS_IFLUSH_SYNC:
2909 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
2912 case XFS_IFLUSH_ASYNC_NOBLOCK:
2913 case XFS_IFLUSH_ASYNC:
2914 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
2917 case XFS_IFLUSH_DELWRI:
2927 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
2928 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
2929 case XFS_IFLUSH_DELWRI:
2932 case XFS_IFLUSH_ASYNC_NOBLOCK:
2933 case XFS_IFLUSH_ASYNC:
2936 case XFS_IFLUSH_SYNC:
2947 * Get the buffer containing the on-disk inode.
2949 error = xfs_itobp(mp, NULL, ip, &dip, &bp,
2950 noblock ? XFS_BUF_TRYLOCK : XFS_BUF_LOCK);
2957 * First flush out the inode that xfs_iflush was called with.
2959 error = xfs_iflush_int(ip, bp);
2964 * If the buffer is pinned then push on the log now so we won't
2965 * get stuck waiting in the write for too long.
2967 if (XFS_BUF_ISPINNED(bp))
2968 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
2972 * see if other inodes can be gathered into this write
2974 error = xfs_iflush_cluster(ip, bp);
2976 goto cluster_corrupt_out;
2978 if (flags & INT_DELWRI) {
2979 xfs_bdwrite(mp, bp);
2980 } else if (flags & INT_ASYNC) {
2981 error = xfs_bawrite(mp, bp);
2983 error = xfs_bwrite(mp, bp);
2989 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2990 cluster_corrupt_out:
2992 * Unlocks the flush lock
2994 xfs_iflush_abort(ip);
2995 return XFS_ERROR(EFSCORRUPTED);
3004 xfs_inode_log_item_t *iip;
3007 #ifdef XFS_TRANS_DEBUG
3011 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3012 ASSERT(!completion_done(&ip->i_flush));
3013 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3014 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3021 * If the inode isn't dirty, then just release the inode
3022 * flush lock and do nothing.
3024 if (xfs_inode_clean(ip)) {
3029 /* set *dip = inode's place in the buffer */
3030 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
3033 * Clear i_update_core before copying out the data.
3034 * This is for coordination with our timestamp updates
3035 * that don't hold the inode lock. They will always
3036 * update the timestamps BEFORE setting i_update_core,
3037 * so if we clear i_update_core after they set it we
3038 * are guaranteed to see their updates to the timestamps.
3039 * I believe that this depends on strongly ordered memory
3040 * semantics, but we have that. We use the SYNCHRONIZE
3041 * macro to make sure that the compiler does not reorder
3042 * the i_update_core access below the data copy below.
3044 ip->i_update_core = 0;
3048 * Make sure to get the latest atime from the Linux inode.
3050 xfs_synchronize_atime(ip);
3052 if (XFS_TEST_ERROR(be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC,
3053 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3054 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3055 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3056 ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3059 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3060 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3061 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3062 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3063 ip->i_ino, ip, ip->i_d.di_magic);
3066 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3068 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3069 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3070 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3071 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3072 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3076 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
3078 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3079 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3080 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3081 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3082 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3083 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3088 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3089 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3090 XFS_RANDOM_IFLUSH_5)) {
3091 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3092 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3094 ip->i_d.di_nextents + ip->i_d.di_anextents,
3099 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3100 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3101 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3102 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3103 ip->i_ino, ip->i_d.di_forkoff, ip);
3107 * bump the flush iteration count, used to detect flushes which
3108 * postdate a log record during recovery.
3111 ip->i_d.di_flushiter++;
3114 * Copy the dirty parts of the inode into the on-disk
3115 * inode. We always copy out the core of the inode,
3116 * because if the inode is dirty at all the core must
3119 xfs_dinode_to_disk(dip, &ip->i_d);
3121 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3122 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3123 ip->i_d.di_flushiter = 0;
3126 * If this is really an old format inode and the superblock version
3127 * has not been updated to support only new format inodes, then
3128 * convert back to the old inode format. If the superblock version
3129 * has been updated, then make the conversion permanent.
3131 ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
3132 if (ip->i_d.di_version == 1) {
3133 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
3137 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3138 dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink);
3141 * The superblock version has already been bumped,
3142 * so just make the conversion to the new inode
3145 ip->i_d.di_version = 2;
3146 dip->di_version = 2;
3147 ip->i_d.di_onlink = 0;
3149 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3150 memset(&(dip->di_pad[0]), 0,
3151 sizeof(dip->di_pad));
3152 ASSERT(ip->i_d.di_projid == 0);
3156 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
3157 if (XFS_IFORK_Q(ip))
3158 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3159 xfs_inobp_check(mp, bp);
3162 * We've recorded everything logged in the inode, so we'd
3163 * like to clear the ilf_fields bits so we don't log and
3164 * flush things unnecessarily. However, we can't stop
3165 * logging all this information until the data we've copied
3166 * into the disk buffer is written to disk. If we did we might
3167 * overwrite the copy of the inode in the log with all the
3168 * data after re-logging only part of it, and in the face of
3169 * a crash we wouldn't have all the data we need to recover.
3171 * What we do is move the bits to the ili_last_fields field.
3172 * When logging the inode, these bits are moved back to the
3173 * ilf_fields field. In the xfs_iflush_done() routine we
3174 * clear ili_last_fields, since we know that the information
3175 * those bits represent is permanently on disk. As long as
3176 * the flush completes before the inode is logged again, then
3177 * both ilf_fields and ili_last_fields will be cleared.
3179 * We can play with the ilf_fields bits here, because the inode
3180 * lock must be held exclusively in order to set bits there
3181 * and the flush lock protects the ili_last_fields bits.
3182 * Set ili_logged so the flush done
3183 * routine can tell whether or not to look in the AIL.
3184 * Also, store the current LSN of the inode so that we can tell
3185 * whether the item has moved in the AIL from xfs_iflush_done().
3186 * In order to read the lsn we need the AIL lock, because
3187 * it is a 64 bit value that cannot be read atomically.
3189 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3190 iip->ili_last_fields = iip->ili_format.ilf_fields;
3191 iip->ili_format.ilf_fields = 0;
3192 iip->ili_logged = 1;
3194 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3195 &iip->ili_item.li_lsn);
3198 * Attach the function xfs_iflush_done to the inode's
3199 * buffer. This will remove the inode from the AIL
3200 * and unlock the inode's flush lock when the inode is
3201 * completely written to disk.
3203 xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
3204 xfs_iflush_done, (xfs_log_item_t *)iip);
3206 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3207 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3210 * We're flushing an inode which is not in the AIL and has
3211 * not been logged but has i_update_core set. For this
3212 * case we can use a B_DELWRI flush and immediately drop
3213 * the inode flush lock because we can avoid the whole
3214 * AIL state thing. It's OK to drop the flush lock now,
3215 * because we've already locked the buffer and to do anything
3216 * you really need both.
3219 ASSERT(iip->ili_logged == 0);
3220 ASSERT(iip->ili_last_fields == 0);
3221 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3229 return XFS_ERROR(EFSCORRUPTED);
3234 #ifdef XFS_ILOCK_TRACE
3235 ktrace_t *xfs_ilock_trace_buf;
3238 xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
3240 ktrace_enter(ip->i_lock_trace,
3242 (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
3243 (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
3244 (void *)ra, /* caller of ilock */
3245 (void *)(unsigned long)current_cpu(),
3246 (void *)(unsigned long)current_pid(),
3247 NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL);
3252 * Return a pointer to the extent record at file index idx.
3254 xfs_bmbt_rec_host_t *
3256 xfs_ifork_t *ifp, /* inode fork pointer */
3257 xfs_extnum_t idx) /* index of target extent */
3260 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
3261 return ifp->if_u1.if_ext_irec->er_extbuf;
3262 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3263 xfs_ext_irec_t *erp; /* irec pointer */
3264 int erp_idx = 0; /* irec index */
3265 xfs_extnum_t page_idx = idx; /* ext index in target list */
3267 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3268 return &erp->er_extbuf[page_idx];
3269 } else if (ifp->if_bytes) {
3270 return &ifp->if_u1.if_extents[idx];
3277 * Insert new item(s) into the extent records for incore inode
3278 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3282 xfs_ifork_t *ifp, /* inode fork pointer */
3283 xfs_extnum_t idx, /* starting index of new items */
3284 xfs_extnum_t count, /* number of inserted items */
3285 xfs_bmbt_irec_t *new) /* items to insert */
3287 xfs_extnum_t i; /* extent record index */
3289 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3290 xfs_iext_add(ifp, idx, count);
3291 for (i = idx; i < idx + count; i++, new++)
3292 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new);
3296 * This is called when the amount of space required for incore file
3297 * extents needs to be increased. The ext_diff parameter stores the
3298 * number of new extents being added and the idx parameter contains
3299 * the extent index where the new extents will be added. If the new
3300 * extents are being appended, then we just need to (re)allocate and
3301 * initialize the space. Otherwise, if the new extents are being
3302 * inserted into the middle of the existing entries, a bit more work
3303 * is required to make room for the new extents to be inserted. The
3304 * caller is responsible for filling in the new extent entries upon
3309 xfs_ifork_t *ifp, /* inode fork pointer */
3310 xfs_extnum_t idx, /* index to begin adding exts */
3311 int ext_diff) /* number of extents to add */
3313 int byte_diff; /* new bytes being added */
3314 int new_size; /* size of extents after adding */
3315 xfs_extnum_t nextents; /* number of extents in file */
3317 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3318 ASSERT((idx >= 0) && (idx <= nextents));
3319 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
3320 new_size = ifp->if_bytes + byte_diff;
3322 * If the new number of extents (nextents + ext_diff)
3323 * fits inside the inode, then continue to use the inline
3326 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
3327 if (idx < nextents) {
3328 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
3329 &ifp->if_u2.if_inline_ext[idx],
3330 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3331 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
3333 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3334 ifp->if_real_bytes = 0;
3335 ifp->if_lastex = nextents + ext_diff;
3338 * Otherwise use a linear (direct) extent list.
3339 * If the extents are currently inside the inode,
3340 * xfs_iext_realloc_direct will switch us from
3341 * inline to direct extent allocation mode.
3343 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
3344 xfs_iext_realloc_direct(ifp, new_size);
3345 if (idx < nextents) {
3346 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
3347 &ifp->if_u1.if_extents[idx],
3348 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3349 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
3352 /* Indirection array */
3354 xfs_ext_irec_t *erp;
3358 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
3359 if (ifp->if_flags & XFS_IFEXTIREC) {
3360 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3362 xfs_iext_irec_init(ifp);
3363 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3364 erp = ifp->if_u1.if_ext_irec;
3366 /* Extents fit in target extent page */
3367 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3368 if (page_idx < erp->er_extcount) {
3369 memmove(&erp->er_extbuf[page_idx + ext_diff],
3370 &erp->er_extbuf[page_idx],
3371 (erp->er_extcount - page_idx) *
3372 sizeof(xfs_bmbt_rec_t));
3373 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3375 erp->er_extcount += ext_diff;
3376 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3378 /* Insert a new extent page */
3380 xfs_iext_add_indirect_multi(ifp,
3381 erp_idx, page_idx, ext_diff);
3384 * If extent(s) are being appended to the last page in
3385 * the indirection array and the new extent(s) don't fit
3386 * in the page, then erp is NULL and erp_idx is set to
3387 * the next index needed in the indirection array.
3390 int count = ext_diff;
3393 erp = xfs_iext_irec_new(ifp, erp_idx);
3394 erp->er_extcount = count;
3395 count -= MIN(count, (int)XFS_LINEAR_EXTS);
3402 ifp->if_bytes = new_size;
3406 * This is called when incore extents are being added to the indirection
3407 * array and the new extents do not fit in the target extent list. The
3408 * erp_idx parameter contains the irec index for the target extent list
3409 * in the indirection array, and the idx parameter contains the extent
3410 * index within the list. The number of extents being added is stored
3411 * in the count parameter.
3413 * |-------| |-------|
3414 * | | | | idx - number of extents before idx
3416 * | | | | count - number of extents being inserted at idx
3417 * |-------| |-------|
3418 * | count | | nex2 | nex2 - number of extents after idx + count
3419 * |-------| |-------|
3422 xfs_iext_add_indirect_multi(
3423 xfs_ifork_t *ifp, /* inode fork pointer */
3424 int erp_idx, /* target extent irec index */
3425 xfs_extnum_t idx, /* index within target list */
3426 int count) /* new extents being added */
3428 int byte_diff; /* new bytes being added */
3429 xfs_ext_irec_t *erp; /* pointer to irec entry */
3430 xfs_extnum_t ext_diff; /* number of extents to add */
3431 xfs_extnum_t ext_cnt; /* new extents still needed */
3432 xfs_extnum_t nex2; /* extents after idx + count */
3433 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3434 int nlists; /* number of irec's (lists) */
3436 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3437 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3438 nex2 = erp->er_extcount - idx;
3439 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3442 * Save second part of target extent list
3443 * (all extents past */
3445 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3446 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS);
3447 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3448 erp->er_extcount -= nex2;
3449 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3450 memset(&erp->er_extbuf[idx], 0, byte_diff);
3454 * Add the new extents to the end of the target
3455 * list, then allocate new irec record(s) and
3456 * extent buffer(s) as needed to store the rest
3457 * of the new extents.
3460 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3462 erp->er_extcount += ext_diff;
3463 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3464 ext_cnt -= ext_diff;
3468 erp = xfs_iext_irec_new(ifp, erp_idx);
3469 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3470 erp->er_extcount = ext_diff;
3471 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3472 ext_cnt -= ext_diff;
3475 /* Add nex2 extents back to indirection array */
3477 xfs_extnum_t ext_avail;
3480 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3481 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3484 * If nex2 extents fit in the current page, append
3485 * nex2_ep after the new extents.
3487 if (nex2 <= ext_avail) {
3488 i = erp->er_extcount;
3491 * Otherwise, check if space is available in the
3494 else if ((erp_idx < nlists - 1) &&
3495 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3496 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3499 /* Create a hole for nex2 extents */
3500 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3501 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3504 * Final choice, create a new extent page for
3509 erp = xfs_iext_irec_new(ifp, erp_idx);
3511 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
3513 erp->er_extcount += nex2;
3514 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3519 * This is called when the amount of space required for incore file
3520 * extents needs to be decreased. The ext_diff parameter stores the
3521 * number of extents to be removed and the idx parameter contains
3522 * the extent index where the extents will be removed from.
3524 * If the amount of space needed has decreased below the linear
3525 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3526 * extent array. Otherwise, use kmem_realloc() to adjust the
3527 * size to what is needed.
3531 xfs_ifork_t *ifp, /* inode fork pointer */
3532 xfs_extnum_t idx, /* index to begin removing exts */
3533 int ext_diff) /* number of extents to remove */
3535 xfs_extnum_t nextents; /* number of extents in file */
3536 int new_size; /* size of extents after removal */
3538 ASSERT(ext_diff > 0);
3539 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3540 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3542 if (new_size == 0) {
3543 xfs_iext_destroy(ifp);
3544 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3545 xfs_iext_remove_indirect(ifp, idx, ext_diff);
3546 } else if (ifp->if_real_bytes) {
3547 xfs_iext_remove_direct(ifp, idx, ext_diff);
3549 xfs_iext_remove_inline(ifp, idx, ext_diff);
3551 ifp->if_bytes = new_size;
3555 * This removes ext_diff extents from the inline buffer, beginning
3556 * at extent index idx.
3559 xfs_iext_remove_inline(
3560 xfs_ifork_t *ifp, /* inode fork pointer */
3561 xfs_extnum_t idx, /* index to begin removing exts */
3562 int ext_diff) /* number of extents to remove */
3564 int nextents; /* number of extents in file */
3566 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3567 ASSERT(idx < XFS_INLINE_EXTS);
3568 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3569 ASSERT(((nextents - ext_diff) > 0) &&
3570 (nextents - ext_diff) < XFS_INLINE_EXTS);
3572 if (idx + ext_diff < nextents) {
3573 memmove(&ifp->if_u2.if_inline_ext[idx],
3574 &ifp->if_u2.if_inline_ext[idx + ext_diff],
3575 (nextents - (idx + ext_diff)) *
3576 sizeof(xfs_bmbt_rec_t));
3577 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
3578 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3580 memset(&ifp->if_u2.if_inline_ext[idx], 0,
3581 ext_diff * sizeof(xfs_bmbt_rec_t));
3586 * This removes ext_diff extents from a linear (direct) extent list,
3587 * beginning at extent index idx. If the extents are being removed
3588 * from the end of the list (ie. truncate) then we just need to re-
3589 * allocate the list to remove the extra space. Otherwise, if the
3590 * extents are being removed from the middle of the existing extent
3591 * entries, then we first need to move the extent records beginning
3592 * at idx + ext_diff up in the list to overwrite the records being
3593 * removed, then remove the extra space via kmem_realloc.
3596 xfs_iext_remove_direct(
3597 xfs_ifork_t *ifp, /* inode fork pointer */
3598 xfs_extnum_t idx, /* index to begin removing exts */
3599 int ext_diff) /* number of extents to remove */
3601 xfs_extnum_t nextents; /* number of extents in file */
3602 int new_size; /* size of extents after removal */
3604 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3605 new_size = ifp->if_bytes -
3606 (ext_diff * sizeof(xfs_bmbt_rec_t));
3607 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3609 if (new_size == 0) {
3610 xfs_iext_destroy(ifp);
3613 /* Move extents up in the list (if needed) */
3614 if (idx + ext_diff < nextents) {
3615 memmove(&ifp->if_u1.if_extents[idx],
3616 &ifp->if_u1.if_extents[idx + ext_diff],
3617 (nextents - (idx + ext_diff)) *
3618 sizeof(xfs_bmbt_rec_t));
3620 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
3621 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3623 * Reallocate the direct extent list. If the extents
3624 * will fit inside the inode then xfs_iext_realloc_direct
3625 * will switch from direct to inline extent allocation
3628 xfs_iext_realloc_direct(ifp, new_size);
3629 ifp->if_bytes = new_size;
3633 * This is called when incore extents are being removed from the
3634 * indirection array and the extents being removed span multiple extent
3635 * buffers. The idx parameter contains the file extent index where we
3636 * want to begin removing extents, and the count parameter contains
3637 * how many extents need to be removed.
3639 * |-------| |-------|
3640 * | nex1 | | | nex1 - number of extents before idx
3641 * |-------| | count |
3642 * | | | | count - number of extents being removed at idx
3643 * | count | |-------|
3644 * | | | nex2 | nex2 - number of extents after idx + count
3645 * |-------| |-------|
3648 xfs_iext_remove_indirect(
3649 xfs_ifork_t *ifp, /* inode fork pointer */
3650 xfs_extnum_t idx, /* index to begin removing extents */
3651 int count) /* number of extents to remove */
3653 xfs_ext_irec_t *erp; /* indirection array pointer */
3654 int erp_idx = 0; /* indirection array index */
3655 xfs_extnum_t ext_cnt; /* extents left to remove */
3656 xfs_extnum_t ext_diff; /* extents to remove in current list */
3657 xfs_extnum_t nex1; /* number of extents before idx */
3658 xfs_extnum_t nex2; /* extents after idx + count */
3659 int nlists; /* entries in indirection array */
3660 int page_idx = idx; /* index in target extent list */
3662 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3663 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3664 ASSERT(erp != NULL);
3665 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3669 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
3670 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
3672 * Check for deletion of entire list;
3673 * xfs_iext_irec_remove() updates extent offsets.
3675 if (ext_diff == erp->er_extcount) {
3676 xfs_iext_irec_remove(ifp, erp_idx);
3677 ext_cnt -= ext_diff;
3680 ASSERT(erp_idx < ifp->if_real_bytes /
3682 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3689 /* Move extents up (if needed) */
3691 memmove(&erp->er_extbuf[nex1],
3692 &erp->er_extbuf[nex1 + ext_diff],
3693 nex2 * sizeof(xfs_bmbt_rec_t));
3695 /* Zero out rest of page */
3696 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
3697 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
3698 /* Update remaining counters */
3699 erp->er_extcount -= ext_diff;
3700 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
3701 ext_cnt -= ext_diff;
3706 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
3707 xfs_iext_irec_compact(ifp);
3711 * Create, destroy, or resize a linear (direct) block of extents.
3714 xfs_iext_realloc_direct(
3715 xfs_ifork_t *ifp, /* inode fork pointer */
3716 int new_size) /* new size of extents */
3718 int rnew_size; /* real new size of extents */
3720 rnew_size = new_size;
3722 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
3723 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
3724 (new_size != ifp->if_real_bytes)));
3726 /* Free extent records */
3727 if (new_size == 0) {
3728 xfs_iext_destroy(ifp);
3730 /* Resize direct extent list and zero any new bytes */
3731 else if (ifp->if_real_bytes) {
3732 /* Check if extents will fit inside the inode */
3733 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
3734 xfs_iext_direct_to_inline(ifp, new_size /
3735 (uint)sizeof(xfs_bmbt_rec_t));
3736 ifp->if_bytes = new_size;
3739 if (!is_power_of_2(new_size)){
3740 rnew_size = roundup_pow_of_two(new_size);
3742 if (rnew_size != ifp->if_real_bytes) {
3743 ifp->if_u1.if_extents =
3744 kmem_realloc(ifp->if_u1.if_extents,
3746 ifp->if_real_bytes, KM_NOFS);
3748 if (rnew_size > ifp->if_real_bytes) {
3749 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
3750 (uint)sizeof(xfs_bmbt_rec_t)], 0,
3751 rnew_size - ifp->if_real_bytes);
3755 * Switch from the inline extent buffer to a direct
3756 * extent list. Be sure to include the inline extent
3757 * bytes in new_size.
3760 new_size += ifp->if_bytes;
3761 if (!is_power_of_2(new_size)) {
3762 rnew_size = roundup_pow_of_two(new_size);
3764 xfs_iext_inline_to_direct(ifp, rnew_size);
3766 ifp->if_real_bytes = rnew_size;
3767 ifp->if_bytes = new_size;
3771 * Switch from linear (direct) extent records to inline buffer.
3774 xfs_iext_direct_to_inline(
3775 xfs_ifork_t *ifp, /* inode fork pointer */
3776 xfs_extnum_t nextents) /* number of extents in file */
3778 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3779 ASSERT(nextents <= XFS_INLINE_EXTS);
3781 * The inline buffer was zeroed when we switched
3782 * from inline to direct extent allocation mode,
3783 * so we don't need to clear it here.
3785 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
3786 nextents * sizeof(xfs_bmbt_rec_t));
3787 kmem_free(ifp->if_u1.if_extents);
3788 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3789 ifp->if_real_bytes = 0;
3793 * Switch from inline buffer to linear (direct) extent records.
3794 * new_size should already be rounded up to the next power of 2
3795 * by the caller (when appropriate), so use new_size as it is.
3796 * However, since new_size may be rounded up, we can't update
3797 * if_bytes here. It is the caller's responsibility to update
3798 * if_bytes upon return.
3801 xfs_iext_inline_to_direct(
3802 xfs_ifork_t *ifp, /* inode fork pointer */
3803 int new_size) /* number of extents in file */
3805 ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS);
3806 memset(ifp->if_u1.if_extents, 0, new_size);
3807 if (ifp->if_bytes) {
3808 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
3810 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3811 sizeof(xfs_bmbt_rec_t));
3813 ifp->if_real_bytes = new_size;
3817 * Resize an extent indirection array to new_size bytes.
3820 xfs_iext_realloc_indirect(
3821 xfs_ifork_t *ifp, /* inode fork pointer */
3822 int new_size) /* new indirection array size */
3824 int nlists; /* number of irec's (ex lists) */
3825 int size; /* current indirection array size */
3827 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3828 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3829 size = nlists * sizeof(xfs_ext_irec_t);
3830 ASSERT(ifp->if_real_bytes);
3831 ASSERT((new_size >= 0) && (new_size != size));
3832 if (new_size == 0) {
3833 xfs_iext_destroy(ifp);
3835 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
3836 kmem_realloc(ifp->if_u1.if_ext_irec,
3837 new_size, size, KM_NOFS);
3842 * Switch from indirection array to linear (direct) extent allocations.
3845 xfs_iext_indirect_to_direct(
3846 xfs_ifork_t *ifp) /* inode fork pointer */
3848 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
3849 xfs_extnum_t nextents; /* number of extents in file */
3850 int size; /* size of file extents */
3852 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3853 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3854 ASSERT(nextents <= XFS_LINEAR_EXTS);
3855 size = nextents * sizeof(xfs_bmbt_rec_t);
3857 xfs_iext_irec_compact_pages(ifp);
3858 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
3860 ep = ifp->if_u1.if_ext_irec->er_extbuf;
3861 kmem_free(ifp->if_u1.if_ext_irec);
3862 ifp->if_flags &= ~XFS_IFEXTIREC;
3863 ifp->if_u1.if_extents = ep;
3864 ifp->if_bytes = size;
3865 if (nextents < XFS_LINEAR_EXTS) {
3866 xfs_iext_realloc_direct(ifp, size);
3871 * Free incore file extents.
3875 xfs_ifork_t *ifp) /* inode fork pointer */
3877 if (ifp->if_flags & XFS_IFEXTIREC) {
3881 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3882 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
3883 xfs_iext_irec_remove(ifp, erp_idx);
3885 ifp->if_flags &= ~XFS_IFEXTIREC;
3886 } else if (ifp->if_real_bytes) {
3887 kmem_free(ifp->if_u1.if_extents);
3888 } else if (ifp->if_bytes) {
3889 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3890 sizeof(xfs_bmbt_rec_t));
3892 ifp->if_u1.if_extents = NULL;
3893 ifp->if_real_bytes = 0;
3898 * Return a pointer to the extent record for file system block bno.
3900 xfs_bmbt_rec_host_t * /* pointer to found extent record */
3901 xfs_iext_bno_to_ext(
3902 xfs_ifork_t *ifp, /* inode fork pointer */
3903 xfs_fileoff_t bno, /* block number to search for */
3904 xfs_extnum_t *idxp) /* index of target extent */
3906 xfs_bmbt_rec_host_t *base; /* pointer to first extent */
3907 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
3908 xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */
3909 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
3910 int high; /* upper boundary in search */
3911 xfs_extnum_t idx = 0; /* index of target extent */
3912 int low; /* lower boundary in search */
3913 xfs_extnum_t nextents; /* number of file extents */
3914 xfs_fileoff_t startoff = 0; /* start offset of extent */
3916 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3917 if (nextents == 0) {
3922 if (ifp->if_flags & XFS_IFEXTIREC) {
3923 /* Find target extent list */
3925 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
3926 base = erp->er_extbuf;
3927 high = erp->er_extcount - 1;
3929 base = ifp->if_u1.if_extents;
3930 high = nextents - 1;
3932 /* Binary search extent records */
3933 while (low <= high) {
3934 idx = (low + high) >> 1;
3936 startoff = xfs_bmbt_get_startoff(ep);
3937 blockcount = xfs_bmbt_get_blockcount(ep);
3938 if (bno < startoff) {
3940 } else if (bno >= startoff + blockcount) {
3943 /* Convert back to file-based extent index */
3944 if (ifp->if_flags & XFS_IFEXTIREC) {
3945 idx += erp->er_extoff;
3951 /* Convert back to file-based extent index */
3952 if (ifp->if_flags & XFS_IFEXTIREC) {
3953 idx += erp->er_extoff;
3955 if (bno >= startoff + blockcount) {
3956 if (++idx == nextents) {
3959 ep = xfs_iext_get_ext(ifp, idx);
3967 * Return a pointer to the indirection array entry containing the
3968 * extent record for filesystem block bno. Store the index of the
3969 * target irec in *erp_idxp.
3971 xfs_ext_irec_t * /* pointer to found extent record */
3972 xfs_iext_bno_to_irec(
3973 xfs_ifork_t *ifp, /* inode fork pointer */
3974 xfs_fileoff_t bno, /* block number to search for */
3975 int *erp_idxp) /* irec index of target ext list */
3977 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
3978 xfs_ext_irec_t *erp_next; /* next indirection array entry */
3979 int erp_idx; /* indirection array index */
3980 int nlists; /* number of extent irec's (lists) */
3981 int high; /* binary search upper limit */
3982 int low; /* binary search lower limit */
3984 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3985 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3989 while (low <= high) {
3990 erp_idx = (low + high) >> 1;
3991 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3992 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
3993 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
3995 } else if (erp_next && bno >=
3996 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
4002 *erp_idxp = erp_idx;
4007 * Return a pointer to the indirection array entry containing the
4008 * extent record at file extent index *idxp. Store the index of the
4009 * target irec in *erp_idxp and store the page index of the target
4010 * extent record in *idxp.
4013 xfs_iext_idx_to_irec(
4014 xfs_ifork_t *ifp, /* inode fork pointer */
4015 xfs_extnum_t *idxp, /* extent index (file -> page) */
4016 int *erp_idxp, /* pointer to target irec */
4017 int realloc) /* new bytes were just added */
4019 xfs_ext_irec_t *prev; /* pointer to previous irec */
4020 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
4021 int erp_idx; /* indirection array index */
4022 int nlists; /* number of irec's (ex lists) */
4023 int high; /* binary search upper limit */
4024 int low; /* binary search lower limit */
4025 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
4027 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4028 ASSERT(page_idx >= 0 && page_idx <=
4029 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
4030 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4035 /* Binary search extent irec's */
4036 while (low <= high) {
4037 erp_idx = (low + high) >> 1;
4038 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4039 prev = erp_idx > 0 ? erp - 1 : NULL;
4040 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
4041 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
4043 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
4044 (page_idx == erp->er_extoff + erp->er_extcount &&
4047 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
4048 erp->er_extcount == XFS_LINEAR_EXTS) {
4052 erp = erp_idx < nlists ? erp + 1 : NULL;
4055 page_idx -= erp->er_extoff;
4060 *erp_idxp = erp_idx;
4065 * Allocate and initialize an indirection array once the space needed
4066 * for incore extents increases above XFS_IEXT_BUFSZ.
4070 xfs_ifork_t *ifp) /* inode fork pointer */
4072 xfs_ext_irec_t *erp; /* indirection array pointer */
4073 xfs_extnum_t nextents; /* number of extents in file */
4075 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4076 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4077 ASSERT(nextents <= XFS_LINEAR_EXTS);
4079 erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS);
4081 if (nextents == 0) {
4082 ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
4083 } else if (!ifp->if_real_bytes) {
4084 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
4085 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
4086 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
4088 erp->er_extbuf = ifp->if_u1.if_extents;
4089 erp->er_extcount = nextents;
4092 ifp->if_flags |= XFS_IFEXTIREC;
4093 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
4094 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
4095 ifp->if_u1.if_ext_irec = erp;
4101 * Allocate and initialize a new entry in the indirection array.
4105 xfs_ifork_t *ifp, /* inode fork pointer */
4106 int erp_idx) /* index for new irec */
4108 xfs_ext_irec_t *erp; /* indirection array pointer */
4109 int i; /* loop counter */
4110 int nlists; /* number of irec's (ex lists) */
4112 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4113 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4115 /* Resize indirection array */
4116 xfs_iext_realloc_indirect(ifp, ++nlists *
4117 sizeof(xfs_ext_irec_t));
4119 * Move records down in the array so the
4120 * new page can use erp_idx.
4122 erp = ifp->if_u1.if_ext_irec;
4123 for (i = nlists - 1; i > erp_idx; i--) {
4124 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
4126 ASSERT(i == erp_idx);
4128 /* Initialize new extent record */
4129 erp = ifp->if_u1.if_ext_irec;
4130 erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
4131 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4132 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
4133 erp[erp_idx].er_extcount = 0;
4134 erp[erp_idx].er_extoff = erp_idx > 0 ?
4135 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
4136 return (&erp[erp_idx]);
4140 * Remove a record from the indirection array.
4143 xfs_iext_irec_remove(
4144 xfs_ifork_t *ifp, /* inode fork pointer */
4145 int erp_idx) /* irec index to remove */
4147 xfs_ext_irec_t *erp; /* indirection array pointer */
4148 int i; /* loop counter */
4149 int nlists; /* number of irec's (ex lists) */
4151 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4152 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4153 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4154 if (erp->er_extbuf) {
4155 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
4157 kmem_free(erp->er_extbuf);
4159 /* Compact extent records */
4160 erp = ifp->if_u1.if_ext_irec;
4161 for (i = erp_idx; i < nlists - 1; i++) {
4162 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
4165 * Manually free the last extent record from the indirection
4166 * array. A call to xfs_iext_realloc_indirect() with a size
4167 * of zero would result in a call to xfs_iext_destroy() which
4168 * would in turn call this function again, creating a nasty
4172 xfs_iext_realloc_indirect(ifp,
4173 nlists * sizeof(xfs_ext_irec_t));
4175 kmem_free(ifp->if_u1.if_ext_irec);
4177 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4181 * This is called to clean up large amounts of unused memory allocated
4182 * by the indirection array. Before compacting anything though, verify
4183 * that the indirection array is still needed and switch back to the
4184 * linear extent list (or even the inline buffer) if possible. The
4185 * compaction policy is as follows:
4187 * Full Compaction: Extents fit into a single page (or inline buffer)
4188 * Partial Compaction: Extents occupy less than 50% of allocated space
4189 * No Compaction: Extents occupy at least 50% of allocated space
4192 xfs_iext_irec_compact(
4193 xfs_ifork_t *ifp) /* inode fork pointer */
4195 xfs_extnum_t nextents; /* number of extents in file */
4196 int nlists; /* number of irec's (ex lists) */
4198 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4199 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4200 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4202 if (nextents == 0) {
4203 xfs_iext_destroy(ifp);
4204 } else if (nextents <= XFS_INLINE_EXTS) {
4205 xfs_iext_indirect_to_direct(ifp);
4206 xfs_iext_direct_to_inline(ifp, nextents);
4207 } else if (nextents <= XFS_LINEAR_EXTS) {
4208 xfs_iext_indirect_to_direct(ifp);
4209 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
4210 xfs_iext_irec_compact_pages(ifp);
4215 * Combine extents from neighboring extent pages.
4218 xfs_iext_irec_compact_pages(
4219 xfs_ifork_t *ifp) /* inode fork pointer */
4221 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
4222 int erp_idx = 0; /* indirection array index */
4223 int nlists; /* number of irec's (ex lists) */
4225 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4226 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4227 while (erp_idx < nlists - 1) {
4228 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4230 if (erp_next->er_extcount <=
4231 (XFS_LINEAR_EXTS - erp->er_extcount)) {
4232 memcpy(&erp->er_extbuf[erp->er_extcount],
4233 erp_next->er_extbuf, erp_next->er_extcount *
4234 sizeof(xfs_bmbt_rec_t));
4235 erp->er_extcount += erp_next->er_extcount;
4237 * Free page before removing extent record
4238 * so er_extoffs don't get modified in
4239 * xfs_iext_irec_remove.
4241 kmem_free(erp_next->er_extbuf);
4242 erp_next->er_extbuf = NULL;
4243 xfs_iext_irec_remove(ifp, erp_idx + 1);
4244 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4252 * This is called to update the er_extoff field in the indirection
4253 * array when extents have been added or removed from one of the
4254 * extent lists. erp_idx contains the irec index to begin updating
4255 * at and ext_diff contains the number of extents that were added
4259 xfs_iext_irec_update_extoffs(
4260 xfs_ifork_t *ifp, /* inode fork pointer */
4261 int erp_idx, /* irec index to update */
4262 int ext_diff) /* number of new extents */
4264 int i; /* loop counter */
4265 int nlists; /* number of irec's (ex lists */
4267 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4268 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4269 for (i = erp_idx; i < nlists; i++) {
4270 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;