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"
30 #include "xfs_mount.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_attr_sf.h"
35 #include "xfs_dinode.h"
36 #include "xfs_inode.h"
37 #include "xfs_buf_item.h"
38 #include "xfs_inode_item.h"
39 #include "xfs_btree.h"
40 #include "xfs_btree_trace.h"
41 #include "xfs_alloc.h"
42 #include "xfs_ialloc.h"
44 #include "xfs_error.h"
45 #include "xfs_utils.h"
46 #include "xfs_quota.h"
47 #include "xfs_filestream.h"
48 #include "xfs_vnodeops.h"
49 #include "xfs_trace.h"
51 kmem_zone_t *xfs_ifork_zone;
52 kmem_zone_t *xfs_inode_zone;
55 * Used in xfs_itruncate_extents(). This is the maximum number of extents
56 * freed from a file in a single transaction.
58 #define XFS_ITRUNC_MAX_EXTENTS 2
60 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
61 STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
62 STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
63 STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
67 * Make sure that the extents in the given memory buffer
77 xfs_bmbt_rec_host_t rec;
80 for (i = 0; i < nrecs; i++) {
81 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
82 rec.l0 = get_unaligned(&ep->l0);
83 rec.l1 = get_unaligned(&ep->l1);
84 xfs_bmbt_get_all(&rec, &irec);
85 if (fmt == XFS_EXTFMT_NOSTATE)
86 ASSERT(irec.br_state == XFS_EXT_NORM);
90 #define xfs_validate_extents(ifp, nrecs, fmt)
94 * Check that none of the inode's in the buffer have a next
95 * unlinked field of 0.
107 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
109 for (i = 0; i < j; i++) {
110 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
111 i * mp->m_sb.sb_inodesize);
112 if (!dip->di_next_unlinked) {
114 "Detected bogus zero next_unlinked field in incore inode buffer 0x%p.",
116 ASSERT(dip->di_next_unlinked);
123 * Find the buffer associated with the given inode map
124 * We do basic validation checks on the buffer once it has been
125 * retrieved from disk.
131 struct xfs_imap *imap,
141 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno,
142 (int)imap->im_len, buf_flags, &bp);
144 if (error != EAGAIN) {
146 "%s: xfs_trans_read_buf() returned error %d.",
149 ASSERT(buf_flags & XBF_TRYLOCK);
155 * Validate the magic number and version of every inode in the buffer
156 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
159 ni = BBTOB(imap->im_len) >> mp->m_sb.sb_inodelog;
160 #else /* usual case */
164 for (i = 0; i < ni; i++) {
168 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
169 (i << mp->m_sb.sb_inodelog));
170 di_ok = dip->di_magic == cpu_to_be16(XFS_DINODE_MAGIC) &&
171 XFS_DINODE_GOOD_VERSION(dip->di_version);
172 if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
173 XFS_ERRTAG_ITOBP_INOTOBP,
174 XFS_RANDOM_ITOBP_INOTOBP))) {
175 if (iget_flags & XFS_IGET_UNTRUSTED) {
176 xfs_trans_brelse(tp, bp);
177 return XFS_ERROR(EINVAL);
179 XFS_CORRUPTION_ERROR("xfs_imap_to_bp",
180 XFS_ERRLEVEL_HIGH, mp, dip);
183 "bad inode magic/vsn daddr %lld #%d (magic=%x)",
184 (unsigned long long)imap->im_blkno, i,
185 be16_to_cpu(dip->di_magic));
188 xfs_trans_brelse(tp, bp);
189 return XFS_ERROR(EFSCORRUPTED);
193 xfs_inobp_check(mp, bp);
196 * Mark the buffer as an inode buffer now that it looks good
198 XFS_BUF_SET_VTYPE(bp, B_FS_INO);
205 * This routine is called to map an inode number within a file
206 * system to the buffer containing the on-disk version of the
207 * inode. It returns a pointer to the buffer containing the
208 * on-disk inode in the bpp parameter, and in the dip parameter
209 * it returns a pointer to the on-disk inode within that buffer.
211 * If a non-zero error is returned, then the contents of bpp and
212 * dipp are undefined.
214 * Use xfs_imap() to determine the size and location of the
215 * buffer to read from disk.
227 struct xfs_imap imap;
232 error = xfs_imap(mp, tp, ino, &imap, imap_flags);
236 error = xfs_imap_to_bp(mp, tp, &imap, &bp, XBF_LOCK, imap_flags);
240 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
242 *offset = imap.im_boffset;
248 * This routine is called to map an inode to the buffer containing
249 * the on-disk version of the inode. It returns a pointer to the
250 * buffer containing the on-disk inode in the bpp parameter, and in
251 * the dip parameter it returns a pointer to the on-disk inode within
254 * If a non-zero error is returned, then the contents of bpp and
255 * dipp are undefined.
257 * The inode is expected to already been mapped to its buffer and read
258 * in once, thus we can use the mapping information stored in the inode
259 * rather than calling xfs_imap(). This allows us to avoid the overhead
260 * of looking at the inode btree for small block file systems
275 ASSERT(ip->i_imap.im_blkno != 0);
277 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp, buf_flags, 0);
282 ASSERT(buf_flags & XBF_TRYLOCK);
288 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
294 * Move inode type and inode format specific information from the
295 * on-disk inode to the in-core inode. For fifos, devs, and sockets
296 * this means set if_rdev to the proper value. For files, directories,
297 * and symlinks this means to bring in the in-line data or extent
298 * pointers. For a file in B-tree format, only the root is immediately
299 * brought in-core. The rest will be in-lined in if_extents when it
300 * is first referenced (see xfs_iread_extents()).
307 xfs_attr_shortform_t *atp;
311 ip->i_df.if_ext_max =
312 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
315 if (unlikely(be32_to_cpu(dip->di_nextents) +
316 be16_to_cpu(dip->di_anextents) >
317 be64_to_cpu(dip->di_nblocks))) {
318 xfs_warn(ip->i_mount,
319 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
320 (unsigned long long)ip->i_ino,
321 (int)(be32_to_cpu(dip->di_nextents) +
322 be16_to_cpu(dip->di_anextents)),
324 be64_to_cpu(dip->di_nblocks));
325 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
327 return XFS_ERROR(EFSCORRUPTED);
330 if (unlikely(dip->di_forkoff > ip->i_mount->m_sb.sb_inodesize)) {
331 xfs_warn(ip->i_mount, "corrupt dinode %Lu, forkoff = 0x%x.",
332 (unsigned long long)ip->i_ino,
334 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
336 return XFS_ERROR(EFSCORRUPTED);
339 if (unlikely((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) &&
340 !ip->i_mount->m_rtdev_targp)) {
341 xfs_warn(ip->i_mount,
342 "corrupt dinode %Lu, has realtime flag set.",
344 XFS_CORRUPTION_ERROR("xfs_iformat(realtime)",
345 XFS_ERRLEVEL_LOW, ip->i_mount, dip);
346 return XFS_ERROR(EFSCORRUPTED);
349 switch (ip->i_d.di_mode & S_IFMT) {
354 if (unlikely(dip->di_format != XFS_DINODE_FMT_DEV)) {
355 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
357 return XFS_ERROR(EFSCORRUPTED);
361 ip->i_df.if_u2.if_rdev = xfs_dinode_get_rdev(dip);
367 switch (dip->di_format) {
368 case XFS_DINODE_FMT_LOCAL:
370 * no local regular files yet
372 if (unlikely((be16_to_cpu(dip->di_mode) & S_IFMT) == S_IFREG)) {
373 xfs_warn(ip->i_mount,
374 "corrupt inode %Lu (local format for regular file).",
375 (unsigned long long) ip->i_ino);
376 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
379 return XFS_ERROR(EFSCORRUPTED);
382 di_size = be64_to_cpu(dip->di_size);
383 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
384 xfs_warn(ip->i_mount,
385 "corrupt inode %Lu (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 | KM_NOFS);
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);
428 if (unlikely(size < sizeof(struct xfs_attr_sf_hdr))) {
429 xfs_warn(ip->i_mount,
430 "corrupt inode %Lu (bad attr fork size %Ld).",
431 (unsigned long long) ip->i_ino,
433 XFS_CORRUPTION_ERROR("xfs_iformat(8)",
436 return XFS_ERROR(EFSCORRUPTED);
439 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
441 case XFS_DINODE_FMT_EXTENTS:
442 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
444 case XFS_DINODE_FMT_BTREE:
445 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
448 error = XFS_ERROR(EFSCORRUPTED);
452 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
454 xfs_idestroy_fork(ip, XFS_DATA_FORK);
460 * The file is in-lined in the on-disk inode.
461 * If it fits into if_inline_data, then copy
462 * it there, otherwise allocate a buffer for it
463 * and copy the data there. Either way, set
464 * if_data to point at the data.
465 * If we allocate a buffer for the data, make
466 * sure that its size is a multiple of 4 and
467 * record the real size in i_real_bytes.
480 * If the size is unreasonable, then something
481 * is wrong and we just bail out rather than crash in
482 * kmem_alloc() or memcpy() below.
484 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
485 xfs_warn(ip->i_mount,
486 "corrupt inode %Lu (bad size %d for local fork, size = %d).",
487 (unsigned long long) ip->i_ino, size,
488 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
489 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
491 return XFS_ERROR(EFSCORRUPTED);
493 ifp = XFS_IFORK_PTR(ip, whichfork);
496 ifp->if_u1.if_data = NULL;
497 else if (size <= sizeof(ifp->if_u2.if_inline_data))
498 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
500 real_size = roundup(size, 4);
501 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP | KM_NOFS);
503 ifp->if_bytes = size;
504 ifp->if_real_bytes = real_size;
506 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
507 ifp->if_flags &= ~XFS_IFEXTENTS;
508 ifp->if_flags |= XFS_IFINLINE;
513 * The file consists of a set of extents all
514 * of which fit into the on-disk inode.
515 * If there are few enough extents to fit into
516 * the if_inline_ext, then copy them there.
517 * Otherwise allocate a buffer for them and copy
518 * them into it. Either way, set if_extents
519 * to point at the extents.
533 ifp = XFS_IFORK_PTR(ip, whichfork);
534 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
535 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
538 * If the number of extents is unreasonable, then something
539 * is wrong and we just bail out rather than crash in
540 * kmem_alloc() or memcpy() below.
542 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
543 xfs_warn(ip->i_mount, "corrupt inode %Lu ((a)extents = %d).",
544 (unsigned long long) ip->i_ino, nex);
545 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
547 return XFS_ERROR(EFSCORRUPTED);
550 ifp->if_real_bytes = 0;
552 ifp->if_u1.if_extents = NULL;
553 else if (nex <= XFS_INLINE_EXTS)
554 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
556 xfs_iext_add(ifp, 0, nex);
558 ifp->if_bytes = size;
560 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
561 xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip));
562 for (i = 0; i < nex; i++, dp++) {
563 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
564 ep->l0 = get_unaligned_be64(&dp->l0);
565 ep->l1 = get_unaligned_be64(&dp->l1);
567 XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork);
568 if (whichfork != XFS_DATA_FORK ||
569 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
570 if (unlikely(xfs_check_nostate_extents(
572 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
575 return XFS_ERROR(EFSCORRUPTED);
578 ifp->if_flags |= XFS_IFEXTENTS;
583 * The file has too many extents to fit into
584 * the inode, so they are in B-tree format.
585 * Allocate a buffer for the root of the B-tree
586 * and copy the root into it. The i_extents
587 * field will remain NULL until all of the
588 * extents are read in (when they are needed).
596 xfs_bmdr_block_t *dfp;
602 ifp = XFS_IFORK_PTR(ip, whichfork);
603 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
604 size = XFS_BMAP_BROOT_SPACE(dfp);
605 nrecs = be16_to_cpu(dfp->bb_numrecs);
608 * blow out if -- fork has less extents than can fit in
609 * fork (fork shouldn't be a btree format), root btree
610 * block has more records than can fit into the fork,
611 * or the number of extents is greater than the number of
614 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
615 || XFS_BMDR_SPACE_CALC(nrecs) >
616 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
617 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
618 xfs_warn(ip->i_mount, "corrupt inode %Lu (btree).",
619 (unsigned long long) ip->i_ino);
620 XFS_CORRUPTION_ERROR("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
622 return XFS_ERROR(EFSCORRUPTED);
625 ifp->if_broot_bytes = size;
626 ifp->if_broot = kmem_alloc(size, KM_SLEEP | KM_NOFS);
627 ASSERT(ifp->if_broot != NULL);
629 * Copy and convert from the on-disk structure
630 * to the in-memory structure.
632 xfs_bmdr_to_bmbt(ip->i_mount, dfp,
633 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
634 ifp->if_broot, size);
635 ifp->if_flags &= ~XFS_IFEXTENTS;
636 ifp->if_flags |= XFS_IFBROOT;
642 xfs_dinode_from_disk(
646 to->di_magic = be16_to_cpu(from->di_magic);
647 to->di_mode = be16_to_cpu(from->di_mode);
648 to->di_version = from ->di_version;
649 to->di_format = from->di_format;
650 to->di_onlink = be16_to_cpu(from->di_onlink);
651 to->di_uid = be32_to_cpu(from->di_uid);
652 to->di_gid = be32_to_cpu(from->di_gid);
653 to->di_nlink = be32_to_cpu(from->di_nlink);
654 to->di_projid_lo = be16_to_cpu(from->di_projid_lo);
655 to->di_projid_hi = be16_to_cpu(from->di_projid_hi);
656 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
657 to->di_flushiter = be16_to_cpu(from->di_flushiter);
658 to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec);
659 to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec);
660 to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec);
661 to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec);
662 to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec);
663 to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec);
664 to->di_size = be64_to_cpu(from->di_size);
665 to->di_nblocks = be64_to_cpu(from->di_nblocks);
666 to->di_extsize = be32_to_cpu(from->di_extsize);
667 to->di_nextents = be32_to_cpu(from->di_nextents);
668 to->di_anextents = be16_to_cpu(from->di_anextents);
669 to->di_forkoff = from->di_forkoff;
670 to->di_aformat = from->di_aformat;
671 to->di_dmevmask = be32_to_cpu(from->di_dmevmask);
672 to->di_dmstate = be16_to_cpu(from->di_dmstate);
673 to->di_flags = be16_to_cpu(from->di_flags);
674 to->di_gen = be32_to_cpu(from->di_gen);
680 xfs_icdinode_t *from)
682 to->di_magic = cpu_to_be16(from->di_magic);
683 to->di_mode = cpu_to_be16(from->di_mode);
684 to->di_version = from ->di_version;
685 to->di_format = from->di_format;
686 to->di_onlink = cpu_to_be16(from->di_onlink);
687 to->di_uid = cpu_to_be32(from->di_uid);
688 to->di_gid = cpu_to_be32(from->di_gid);
689 to->di_nlink = cpu_to_be32(from->di_nlink);
690 to->di_projid_lo = cpu_to_be16(from->di_projid_lo);
691 to->di_projid_hi = cpu_to_be16(from->di_projid_hi);
692 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
693 to->di_flushiter = cpu_to_be16(from->di_flushiter);
694 to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec);
695 to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec);
696 to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec);
697 to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec);
698 to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec);
699 to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec);
700 to->di_size = cpu_to_be64(from->di_size);
701 to->di_nblocks = cpu_to_be64(from->di_nblocks);
702 to->di_extsize = cpu_to_be32(from->di_extsize);
703 to->di_nextents = cpu_to_be32(from->di_nextents);
704 to->di_anextents = cpu_to_be16(from->di_anextents);
705 to->di_forkoff = from->di_forkoff;
706 to->di_aformat = from->di_aformat;
707 to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
708 to->di_dmstate = cpu_to_be16(from->di_dmstate);
709 to->di_flags = cpu_to_be16(from->di_flags);
710 to->di_gen = cpu_to_be32(from->di_gen);
719 if (di_flags & XFS_DIFLAG_ANY) {
720 if (di_flags & XFS_DIFLAG_REALTIME)
721 flags |= XFS_XFLAG_REALTIME;
722 if (di_flags & XFS_DIFLAG_PREALLOC)
723 flags |= XFS_XFLAG_PREALLOC;
724 if (di_flags & XFS_DIFLAG_IMMUTABLE)
725 flags |= XFS_XFLAG_IMMUTABLE;
726 if (di_flags & XFS_DIFLAG_APPEND)
727 flags |= XFS_XFLAG_APPEND;
728 if (di_flags & XFS_DIFLAG_SYNC)
729 flags |= XFS_XFLAG_SYNC;
730 if (di_flags & XFS_DIFLAG_NOATIME)
731 flags |= XFS_XFLAG_NOATIME;
732 if (di_flags & XFS_DIFLAG_NODUMP)
733 flags |= XFS_XFLAG_NODUMP;
734 if (di_flags & XFS_DIFLAG_RTINHERIT)
735 flags |= XFS_XFLAG_RTINHERIT;
736 if (di_flags & XFS_DIFLAG_PROJINHERIT)
737 flags |= XFS_XFLAG_PROJINHERIT;
738 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
739 flags |= XFS_XFLAG_NOSYMLINKS;
740 if (di_flags & XFS_DIFLAG_EXTSIZE)
741 flags |= XFS_XFLAG_EXTSIZE;
742 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
743 flags |= XFS_XFLAG_EXTSZINHERIT;
744 if (di_flags & XFS_DIFLAG_NODEFRAG)
745 flags |= XFS_XFLAG_NODEFRAG;
746 if (di_flags & XFS_DIFLAG_FILESTREAM)
747 flags |= XFS_XFLAG_FILESTREAM;
757 xfs_icdinode_t *dic = &ip->i_d;
759 return _xfs_dic2xflags(dic->di_flags) |
760 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
767 return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) |
768 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
772 * Read the disk inode attributes into the in-core inode structure.
786 * Fill in the location information in the in-core inode.
788 error = xfs_imap(mp, tp, ip->i_ino, &ip->i_imap, iget_flags);
793 * Get pointers to the on-disk inode and the buffer containing it.
795 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp,
796 XBF_LOCK, iget_flags);
799 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
802 * If we got something that isn't an inode it means someone
803 * (nfs or dmi) has a stale handle.
805 if (dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC)) {
808 "%s: dip->di_magic (0x%x) != XFS_DINODE_MAGIC (0x%x)",
809 __func__, be16_to_cpu(dip->di_magic), XFS_DINODE_MAGIC);
811 error = XFS_ERROR(EINVAL);
816 * If the on-disk inode is already linked to a directory
817 * entry, copy all of the inode into the in-core inode.
818 * xfs_iformat() handles copying in the inode format
819 * specific information.
820 * Otherwise, just get the truly permanent information.
823 xfs_dinode_from_disk(&ip->i_d, dip);
824 error = xfs_iformat(ip, dip);
827 xfs_alert(mp, "%s: xfs_iformat() returned error %d",
833 ip->i_d.di_magic = be16_to_cpu(dip->di_magic);
834 ip->i_d.di_version = dip->di_version;
835 ip->i_d.di_gen = be32_to_cpu(dip->di_gen);
836 ip->i_d.di_flushiter = be16_to_cpu(dip->di_flushiter);
838 * Make sure to pull in the mode here as well in
839 * case the inode is released without being used.
840 * This ensures that xfs_inactive() will see that
841 * the inode is already free and not try to mess
842 * with the uninitialized part of it.
846 * Initialize the per-fork minima and maxima for a new
847 * inode here. xfs_iformat will do it for old inodes.
849 ip->i_df.if_ext_max =
850 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
854 * The inode format changed when we moved the link count and
855 * made it 32 bits long. If this is an old format inode,
856 * convert it in memory to look like a new one. If it gets
857 * flushed to disk we will convert back before flushing or
858 * logging it. We zero out the new projid field and the old link
859 * count field. We'll handle clearing the pad field (the remains
860 * of the old uuid field) when we actually convert the inode to
861 * the new format. We don't change the version number so that we
862 * can distinguish this from a real new format inode.
864 if (ip->i_d.di_version == 1) {
865 ip->i_d.di_nlink = ip->i_d.di_onlink;
866 ip->i_d.di_onlink = 0;
867 xfs_set_projid(ip, 0);
870 ip->i_delayed_blks = 0;
871 ip->i_size = ip->i_d.di_size;
874 * Mark the buffer containing the inode as something to keep
875 * around for a while. This helps to keep recently accessed
876 * meta-data in-core longer.
878 xfs_buf_set_ref(bp, XFS_INO_REF);
881 * Use xfs_trans_brelse() to release the buffer containing the
882 * on-disk inode, because it was acquired with xfs_trans_read_buf()
883 * in xfs_itobp() above. If tp is NULL, this is just a normal
884 * brelse(). If we're within a transaction, then xfs_trans_brelse()
885 * will only release the buffer if it is not dirty within the
886 * transaction. It will be OK to release the buffer in this case,
887 * because inodes on disk are never destroyed and we will be
888 * locking the new in-core inode before putting it in the hash
889 * table where other processes can find it. Thus we don't have
890 * to worry about the inode being changed just because we released
894 xfs_trans_brelse(tp, bp);
899 * Read in extents from a btree-format inode.
900 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
910 xfs_extnum_t nextents;
912 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
913 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
915 return XFS_ERROR(EFSCORRUPTED);
917 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
918 ifp = XFS_IFORK_PTR(ip, whichfork);
921 * We know that the size is valid (it's checked in iformat_btree)
923 ifp->if_bytes = ifp->if_real_bytes = 0;
924 ifp->if_flags |= XFS_IFEXTENTS;
925 xfs_iext_add(ifp, 0, nextents);
926 error = xfs_bmap_read_extents(tp, ip, whichfork);
928 xfs_iext_destroy(ifp);
929 ifp->if_flags &= ~XFS_IFEXTENTS;
932 xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip));
937 * Allocate an inode on disk and return a copy of its in-core version.
938 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
939 * appropriately within the inode. The uid and gid for the inode are
940 * set according to the contents of the given cred structure.
942 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
943 * has a free inode available, call xfs_iget()
944 * to obtain the in-core version of the allocated inode. Finally,
945 * fill in the inode and log its initial contents. In this case,
946 * ialloc_context would be set to NULL and call_again set to false.
948 * If xfs_dialloc() does not have an available inode,
949 * it will replenish its supply by doing an allocation. Since we can
950 * only do one allocation within a transaction without deadlocks, we
951 * must commit the current transaction before returning the inode itself.
952 * In this case, therefore, we will set call_again to true and return.
953 * The caller should then commit the current transaction, start a new
954 * transaction, and call xfs_ialloc() again to actually get the inode.
956 * To ensure that some other process does not grab the inode that
957 * was allocated during the first call to xfs_ialloc(), this routine
958 * also returns the [locked] bp pointing to the head of the freelist
959 * as ialloc_context. The caller should hold this buffer across
960 * the commit and pass it back into this routine on the second call.
962 * If we are allocating quota inodes, we do not have a parent inode
963 * to attach to or associate with (i.e. pip == NULL) because they
964 * are not linked into the directory structure - they are attached
965 * directly to the superblock - and so have no parent.
976 xfs_buf_t **ialloc_context,
977 boolean_t *call_again,
988 * Call the space management code to pick
989 * the on-disk inode to be allocated.
991 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
992 ialloc_context, call_again, &ino);
995 if (*call_again || ino == NULLFSINO) {
999 ASSERT(*ialloc_context == NULL);
1002 * Get the in-core inode with the lock held exclusively.
1003 * This is because we're setting fields here we need
1004 * to prevent others from looking at until we're done.
1006 error = xfs_iget(tp->t_mountp, tp, ino, XFS_IGET_CREATE,
1007 XFS_ILOCK_EXCL, &ip);
1012 ip->i_d.di_mode = (__uint16_t)mode;
1013 ip->i_d.di_onlink = 0;
1014 ip->i_d.di_nlink = nlink;
1015 ASSERT(ip->i_d.di_nlink == nlink);
1016 ip->i_d.di_uid = current_fsuid();
1017 ip->i_d.di_gid = current_fsgid();
1018 xfs_set_projid(ip, prid);
1019 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1022 * If the superblock version is up to where we support new format
1023 * inodes and this is currently an old format inode, then change
1024 * the inode version number now. This way we only do the conversion
1025 * here rather than here and in the flush/logging code.
1027 if (xfs_sb_version_hasnlink(&tp->t_mountp->m_sb) &&
1028 ip->i_d.di_version == 1) {
1029 ip->i_d.di_version = 2;
1031 * We've already zeroed the old link count, the projid field,
1032 * and the pad field.
1037 * Project ids won't be stored on disk if we are using a version 1 inode.
1039 if ((prid != 0) && (ip->i_d.di_version == 1))
1040 xfs_bump_ino_vers2(tp, ip);
1042 if (pip && XFS_INHERIT_GID(pip)) {
1043 ip->i_d.di_gid = pip->i_d.di_gid;
1044 if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1045 ip->i_d.di_mode |= S_ISGID;
1050 * If the group ID of the new file does not match the effective group
1051 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1052 * (and only if the irix_sgid_inherit compatibility variable is set).
1054 if ((irix_sgid_inherit) &&
1055 (ip->i_d.di_mode & S_ISGID) &&
1056 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1057 ip->i_d.di_mode &= ~S_ISGID;
1060 ip->i_d.di_size = 0;
1062 ip->i_d.di_nextents = 0;
1063 ASSERT(ip->i_d.di_nblocks == 0);
1066 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
1067 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
1068 ip->i_d.di_atime = ip->i_d.di_mtime;
1069 ip->i_d.di_ctime = ip->i_d.di_mtime;
1072 * di_gen will have been taken care of in xfs_iread.
1074 ip->i_d.di_extsize = 0;
1075 ip->i_d.di_dmevmask = 0;
1076 ip->i_d.di_dmstate = 0;
1077 ip->i_d.di_flags = 0;
1078 flags = XFS_ILOG_CORE;
1079 switch (mode & S_IFMT) {
1084 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1085 ip->i_df.if_u2.if_rdev = rdev;
1086 ip->i_df.if_flags = 0;
1087 flags |= XFS_ILOG_DEV;
1091 * we can't set up filestreams until after the VFS inode
1092 * is set up properly.
1094 if (pip && xfs_inode_is_filestream(pip))
1098 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
1101 if ((mode & S_IFMT) == S_IFDIR) {
1102 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1103 di_flags |= XFS_DIFLAG_RTINHERIT;
1104 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1105 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1106 ip->i_d.di_extsize = pip->i_d.di_extsize;
1108 } else if ((mode & S_IFMT) == S_IFREG) {
1109 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1110 di_flags |= XFS_DIFLAG_REALTIME;
1111 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1112 di_flags |= XFS_DIFLAG_EXTSIZE;
1113 ip->i_d.di_extsize = pip->i_d.di_extsize;
1116 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1117 xfs_inherit_noatime)
1118 di_flags |= XFS_DIFLAG_NOATIME;
1119 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1121 di_flags |= XFS_DIFLAG_NODUMP;
1122 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1124 di_flags |= XFS_DIFLAG_SYNC;
1125 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1126 xfs_inherit_nosymlinks)
1127 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1128 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1129 di_flags |= XFS_DIFLAG_PROJINHERIT;
1130 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1131 xfs_inherit_nodefrag)
1132 di_flags |= XFS_DIFLAG_NODEFRAG;
1133 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
1134 di_flags |= XFS_DIFLAG_FILESTREAM;
1135 ip->i_d.di_flags |= di_flags;
1139 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1140 ip->i_df.if_flags = XFS_IFEXTENTS;
1141 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1142 ip->i_df.if_u1.if_extents = NULL;
1148 * Attribute fork settings for new inode.
1150 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1151 ip->i_d.di_anextents = 0;
1154 * Log the new values stuffed into the inode.
1156 xfs_trans_ijoin_ref(tp, ip, XFS_ILOCK_EXCL);
1157 xfs_trans_log_inode(tp, ip, flags);
1159 /* now that we have an i_mode we can setup inode ops and unlock */
1160 xfs_setup_inode(ip);
1162 /* now we have set up the vfs inode we can associate the filestream */
1164 error = xfs_filestream_associate(pip, ip);
1168 xfs_iflags_set(ip, XFS_IFILESTREAM);
1176 * Check to make sure that there are no blocks allocated to the
1177 * file beyond the size of the file. We don't check this for
1178 * files with fixed size extents or real time extents, but we
1179 * at least do it for regular files.
1184 struct xfs_inode *ip,
1187 struct xfs_mount *mp = ip->i_mount;
1188 xfs_fileoff_t map_first;
1190 xfs_bmbt_irec_t imaps[2];
1192 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1195 if (XFS_IS_REALTIME_INODE(ip))
1198 if (ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE)
1202 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1204 * The filesystem could be shutting down, so bmapi may return
1207 if (xfs_bmapi(NULL, ip, map_first,
1209 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1211 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1214 ASSERT(nimaps == 1);
1215 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1218 #define xfs_isize_check(ip, isize)
1222 * Free up the underlying blocks past new_size. The new size must be smaller
1223 * than the current size. This routine can be used both for the attribute and
1224 * data fork, and does not modify the inode size, which is left to the caller.
1226 * The transaction passed to this routine must have made a permanent log
1227 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1228 * given transaction and start new ones, so make sure everything involved in
1229 * the transaction is tidy before calling here. Some transaction will be
1230 * returned to the caller to be committed. The incoming transaction must
1231 * already include the inode, and both inode locks must be held exclusively.
1232 * The inode must also be "held" within the transaction. On return the inode
1233 * will be "held" within the returned transaction. This routine does NOT
1234 * require any disk space to be reserved for it within the transaction.
1236 * If we get an error, we must return with the inode locked and linked into the
1237 * current transaction. This keeps things simple for the higher level code,
1238 * because it always knows that the inode is locked and held in the transaction
1239 * that returns to it whether errors occur or not. We don't mark the inode
1240 * dirty on error so that transactions can be easily aborted if possible.
1243 xfs_itruncate_extents(
1244 struct xfs_trans **tpp,
1245 struct xfs_inode *ip,
1247 xfs_fsize_t new_size)
1249 struct xfs_mount *mp = ip->i_mount;
1250 struct xfs_trans *tp = *tpp;
1251 struct xfs_trans *ntp;
1252 xfs_bmap_free_t free_list;
1253 xfs_fsblock_t first_block;
1254 xfs_fileoff_t first_unmap_block;
1255 xfs_fileoff_t last_block;
1256 xfs_filblks_t unmap_len;
1261 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
1262 ASSERT(new_size <= ip->i_size);
1263 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1264 ASSERT(ip->i_itemp != NULL);
1265 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1266 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1269 * Since it is possible for space to become allocated beyond
1270 * the end of the file (in a crash where the space is allocated
1271 * but the inode size is not yet updated), simply remove any
1272 * blocks which show up between the new EOF and the maximum
1273 * possible file size. If the first block to be removed is
1274 * beyond the maximum file size (ie it is the same as last_block),
1275 * then there is nothing to do.
1277 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1278 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1279 if (first_unmap_block == last_block)
1282 ASSERT(first_unmap_block < last_block);
1283 unmap_len = last_block - first_unmap_block + 1;
1285 xfs_bmap_init(&free_list, &first_block);
1286 error = xfs_bunmapi(tp, ip,
1287 first_unmap_block, unmap_len,
1288 xfs_bmapi_aflag(whichfork),
1289 XFS_ITRUNC_MAX_EXTENTS,
1290 &first_block, &free_list,
1293 goto out_bmap_cancel;
1296 * Duplicate the transaction that has the permanent
1297 * reservation and commit the old transaction.
1299 error = xfs_bmap_finish(&tp, &free_list, &committed);
1301 xfs_trans_ijoin(tp, ip);
1303 goto out_bmap_cancel;
1307 * Mark the inode dirty so it will be logged and
1308 * moved forward in the log as part of every commit.
1310 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1313 ntp = xfs_trans_dup(tp);
1314 error = xfs_trans_commit(tp, 0);
1317 xfs_trans_ijoin(tp, ip);
1323 * Transaction commit worked ok so we can drop the extra ticket
1324 * reference that we gained in xfs_trans_dup()
1326 xfs_log_ticket_put(tp->t_ticket);
1327 error = xfs_trans_reserve(tp, 0,
1328 XFS_ITRUNCATE_LOG_RES(mp), 0,
1329 XFS_TRANS_PERM_LOG_RES,
1330 XFS_ITRUNCATE_LOG_COUNT);
1340 * If the bunmapi call encounters an error, return to the caller where
1341 * the transaction can be properly aborted. We just need to make sure
1342 * we're not holding any resources that we were not when we came in.
1344 xfs_bmap_cancel(&free_list);
1350 struct xfs_trans **tpp,
1351 struct xfs_inode *ip,
1352 xfs_fsize_t new_size)
1356 trace_xfs_itruncate_data_start(ip, new_size);
1359 * The first thing we do is set the size to new_size permanently on
1360 * disk. This way we don't have to worry about anyone ever being able
1361 * to look at the data being freed even in the face of a crash.
1362 * What we're getting around here is the case where we free a block, it
1363 * is allocated to another file, it is written to, and then we crash.
1364 * If the new data gets written to the file but the log buffers
1365 * containing the free and reallocation don't, then we'd end up with
1366 * garbage in the blocks being freed. As long as we make the new_size
1367 * permanent before actually freeing any blocks it doesn't matter if
1368 * they get written to.
1370 if (ip->i_d.di_nextents > 0) {
1372 * If we are not changing the file size then do not update
1373 * the on-disk file size - we may be called from
1374 * xfs_inactive_free_eofblocks(). If we update the on-disk
1375 * file size and then the system crashes before the contents
1376 * of the file are flushed to disk then the files may be
1377 * full of holes (ie NULL files bug).
1379 if (ip->i_size != new_size) {
1380 ip->i_d.di_size = new_size;
1381 ip->i_size = new_size;
1382 xfs_trans_log_inode(*tpp, ip, XFS_ILOG_CORE);
1386 error = xfs_itruncate_extents(tpp, ip, XFS_DATA_FORK, new_size);
1391 * If we are not changing the file size then do not update the on-disk
1392 * file size - we may be called from xfs_inactive_free_eofblocks().
1393 * If we update the on-disk file size and then the system crashes
1394 * before the contents of the file are flushed to disk then the files
1395 * may be full of holes (ie NULL files bug).
1397 xfs_isize_check(ip, new_size);
1398 if (ip->i_size != new_size) {
1399 ip->i_d.di_size = new_size;
1400 ip->i_size = new_size;
1403 ASSERT(new_size != 0 || ip->i_delayed_blks == 0);
1404 ASSERT(new_size != 0 || ip->i_d.di_nextents == 0);
1407 * Always re-log the inode so that our permanent transaction can keep
1408 * on rolling it forward in the log.
1410 xfs_trans_log_inode(*tpp, ip, XFS_ILOG_CORE);
1412 trace_xfs_itruncate_data_end(ip, new_size);
1417 * This is called when the inode's link count goes to 0.
1418 * We place the on-disk inode on a list in the AGI. It
1419 * will be pulled from this list when the inode is freed.
1436 ASSERT(ip->i_d.di_nlink == 0);
1437 ASSERT(ip->i_d.di_mode != 0);
1442 * Get the agi buffer first. It ensures lock ordering
1445 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1448 agi = XFS_BUF_TO_AGI(agibp);
1451 * Get the index into the agi hash table for the
1452 * list this inode will go on.
1454 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1456 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1457 ASSERT(agi->agi_unlinked[bucket_index]);
1458 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1460 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
1462 * There is already another inode in the bucket we need
1463 * to add ourselves to. Add us at the front of the list.
1464 * Here we put the head pointer into our next pointer,
1465 * and then we fall through to point the head at us.
1467 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XBF_LOCK);
1471 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
1472 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1473 offset = ip->i_imap.im_boffset +
1474 offsetof(xfs_dinode_t, di_next_unlinked);
1475 xfs_trans_inode_buf(tp, ibp);
1476 xfs_trans_log_buf(tp, ibp, offset,
1477 (offset + sizeof(xfs_agino_t) - 1));
1478 xfs_inobp_check(mp, ibp);
1482 * Point the bucket head pointer at the inode being inserted.
1485 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1486 offset = offsetof(xfs_agi_t, agi_unlinked) +
1487 (sizeof(xfs_agino_t) * bucket_index);
1488 xfs_trans_log_buf(tp, agibp, offset,
1489 (offset + sizeof(xfs_agino_t) - 1));
1494 * Pull the on-disk inode from the AGI unlinked list.
1507 xfs_agnumber_t agno;
1509 xfs_agino_t next_agino;
1510 xfs_buf_t *last_ibp;
1511 xfs_dinode_t *last_dip = NULL;
1513 int offset, last_offset = 0;
1517 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1520 * Get the agi buffer first. It ensures lock ordering
1523 error = xfs_read_agi(mp, tp, agno, &agibp);
1527 agi = XFS_BUF_TO_AGI(agibp);
1530 * Get the index into the agi hash table for the
1531 * list this inode will go on.
1533 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1535 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1536 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
1537 ASSERT(agi->agi_unlinked[bucket_index]);
1539 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
1541 * We're at the head of the list. Get the inode's
1542 * on-disk buffer to see if there is anyone after us
1543 * on the list. Only modify our next pointer if it
1544 * is not already NULLAGINO. This saves us the overhead
1545 * of dealing with the buffer when there is no need to
1548 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XBF_LOCK);
1550 xfs_warn(mp, "%s: xfs_itobp() returned error %d.",
1554 next_agino = be32_to_cpu(dip->di_next_unlinked);
1555 ASSERT(next_agino != 0);
1556 if (next_agino != NULLAGINO) {
1557 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
1558 offset = ip->i_imap.im_boffset +
1559 offsetof(xfs_dinode_t, di_next_unlinked);
1560 xfs_trans_inode_buf(tp, ibp);
1561 xfs_trans_log_buf(tp, ibp, offset,
1562 (offset + sizeof(xfs_agino_t) - 1));
1563 xfs_inobp_check(mp, ibp);
1565 xfs_trans_brelse(tp, ibp);
1568 * Point the bucket head pointer at the next inode.
1570 ASSERT(next_agino != 0);
1571 ASSERT(next_agino != agino);
1572 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
1573 offset = offsetof(xfs_agi_t, agi_unlinked) +
1574 (sizeof(xfs_agino_t) * bucket_index);
1575 xfs_trans_log_buf(tp, agibp, offset,
1576 (offset + sizeof(xfs_agino_t) - 1));
1579 * We need to search the list for the inode being freed.
1581 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
1583 while (next_agino != agino) {
1585 * If the last inode wasn't the one pointing to
1586 * us, then release its buffer since we're not
1587 * going to do anything with it.
1589 if (last_ibp != NULL) {
1590 xfs_trans_brelse(tp, last_ibp);
1592 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
1593 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
1594 &last_ibp, &last_offset, 0);
1597 "%s: xfs_inotobp() returned error %d.",
1601 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
1602 ASSERT(next_agino != NULLAGINO);
1603 ASSERT(next_agino != 0);
1606 * Now last_ibp points to the buffer previous to us on
1607 * the unlinked list. Pull us from the list.
1609 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XBF_LOCK);
1611 xfs_warn(mp, "%s: xfs_itobp(2) returned error %d.",
1615 next_agino = be32_to_cpu(dip->di_next_unlinked);
1616 ASSERT(next_agino != 0);
1617 ASSERT(next_agino != agino);
1618 if (next_agino != NULLAGINO) {
1619 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
1620 offset = ip->i_imap.im_boffset +
1621 offsetof(xfs_dinode_t, di_next_unlinked);
1622 xfs_trans_inode_buf(tp, ibp);
1623 xfs_trans_log_buf(tp, ibp, offset,
1624 (offset + sizeof(xfs_agino_t) - 1));
1625 xfs_inobp_check(mp, ibp);
1627 xfs_trans_brelse(tp, ibp);
1630 * Point the previous inode on the list to the next inode.
1632 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
1633 ASSERT(next_agino != 0);
1634 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
1635 xfs_trans_inode_buf(tp, last_ibp);
1636 xfs_trans_log_buf(tp, last_ibp, offset,
1637 (offset + sizeof(xfs_agino_t) - 1));
1638 xfs_inobp_check(mp, last_ibp);
1644 * A big issue when freeing the inode cluster is is that we _cannot_ skip any
1645 * inodes that are in memory - they all must be marked stale and attached to
1646 * the cluster buffer.
1650 xfs_inode_t *free_ip,
1654 xfs_mount_t *mp = free_ip->i_mount;
1655 int blks_per_cluster;
1662 xfs_inode_log_item_t *iip;
1663 xfs_log_item_t *lip;
1664 struct xfs_perag *pag;
1666 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
1667 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
1668 blks_per_cluster = 1;
1669 ninodes = mp->m_sb.sb_inopblock;
1670 nbufs = XFS_IALLOC_BLOCKS(mp);
1672 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
1673 mp->m_sb.sb_blocksize;
1674 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
1675 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
1678 for (j = 0; j < nbufs; j++, inum += ninodes) {
1679 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
1680 XFS_INO_TO_AGBNO(mp, inum));
1683 * We obtain and lock the backing buffer first in the process
1684 * here, as we have to ensure that any dirty inode that we
1685 * can't get the flush lock on is attached to the buffer.
1686 * If we scan the in-memory inodes first, then buffer IO can
1687 * complete before we get a lock on it, and hence we may fail
1688 * to mark all the active inodes on the buffer stale.
1690 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
1691 mp->m_bsize * blks_per_cluster,
1695 * Walk the inodes already attached to the buffer and mark them
1696 * stale. These will all have the flush locks held, so an
1697 * in-memory inode walk can't lock them. By marking them all
1698 * stale first, we will not attempt to lock them in the loop
1699 * below as the XFS_ISTALE flag will be set.
1701 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
1703 if (lip->li_type == XFS_LI_INODE) {
1704 iip = (xfs_inode_log_item_t *)lip;
1705 ASSERT(iip->ili_logged == 1);
1706 lip->li_cb = xfs_istale_done;
1707 xfs_trans_ail_copy_lsn(mp->m_ail,
1708 &iip->ili_flush_lsn,
1709 &iip->ili_item.li_lsn);
1710 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
1712 lip = lip->li_bio_list;
1717 * For each inode in memory attempt to add it to the inode
1718 * buffer and set it up for being staled on buffer IO
1719 * completion. This is safe as we've locked out tail pushing
1720 * and flushing by locking the buffer.
1722 * We have already marked every inode that was part of a
1723 * transaction stale above, which means there is no point in
1724 * even trying to lock them.
1726 for (i = 0; i < ninodes; i++) {
1729 ip = radix_tree_lookup(&pag->pag_ici_root,
1730 XFS_INO_TO_AGINO(mp, (inum + i)));
1732 /* Inode not in memory, nothing to do */
1739 * because this is an RCU protected lookup, we could
1740 * find a recently freed or even reallocated inode
1741 * during the lookup. We need to check under the
1742 * i_flags_lock for a valid inode here. Skip it if it
1743 * is not valid, the wrong inode or stale.
1745 spin_lock(&ip->i_flags_lock);
1746 if (ip->i_ino != inum + i ||
1747 __xfs_iflags_test(ip, XFS_ISTALE)) {
1748 spin_unlock(&ip->i_flags_lock);
1752 spin_unlock(&ip->i_flags_lock);
1755 * Don't try to lock/unlock the current inode, but we
1756 * _cannot_ skip the other inodes that we did not find
1757 * in the list attached to the buffer and are not
1758 * already marked stale. If we can't lock it, back off
1761 if (ip != free_ip &&
1762 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
1770 xfs_iflags_set(ip, XFS_ISTALE);
1773 * we don't need to attach clean inodes or those only
1774 * with unlogged changes (which we throw away, anyway).
1777 if (!iip || xfs_inode_clean(ip)) {
1778 ASSERT(ip != free_ip);
1779 ip->i_update_core = 0;
1781 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1785 iip->ili_last_fields = iip->ili_format.ilf_fields;
1786 iip->ili_format.ilf_fields = 0;
1787 iip->ili_logged = 1;
1788 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
1789 &iip->ili_item.li_lsn);
1791 xfs_buf_attach_iodone(bp, xfs_istale_done,
1795 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1798 xfs_trans_stale_inode_buf(tp, bp);
1799 xfs_trans_binval(tp, bp);
1806 * This is called to return an inode to the inode free list.
1807 * The inode should already be truncated to 0 length and have
1808 * no pages associated with it. This routine also assumes that
1809 * the inode is already a part of the transaction.
1811 * The on-disk copy of the inode will have been added to the list
1812 * of unlinked inodes in the AGI. We need to remove the inode from
1813 * that list atomically with respect to freeing it here.
1819 xfs_bmap_free_t *flist)
1823 xfs_ino_t first_ino;
1827 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1828 ASSERT(ip->i_d.di_nlink == 0);
1829 ASSERT(ip->i_d.di_nextents == 0);
1830 ASSERT(ip->i_d.di_anextents == 0);
1831 ASSERT((ip->i_d.di_size == 0 && ip->i_size == 0) ||
1832 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
1833 ASSERT(ip->i_d.di_nblocks == 0);
1836 * Pull the on-disk inode from the AGI unlinked list.
1838 error = xfs_iunlink_remove(tp, ip);
1843 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
1847 ip->i_d.di_mode = 0; /* mark incore inode as free */
1848 ip->i_d.di_flags = 0;
1849 ip->i_d.di_dmevmask = 0;
1850 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
1851 ip->i_df.if_ext_max =
1852 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
1853 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1854 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1856 * Bump the generation count so no one will be confused
1857 * by reincarnations of this inode.
1861 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1863 error = xfs_itobp(ip->i_mount, tp, ip, &dip, &ibp, XBF_LOCK);
1868 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
1869 * from picking up this inode when it is reclaimed (its incore state
1870 * initialzed but not flushed to disk yet). The in-core di_mode is
1871 * already cleared and a corresponding transaction logged.
1872 * The hack here just synchronizes the in-core to on-disk
1873 * di_mode value in advance before the actual inode sync to disk.
1874 * This is OK because the inode is already unlinked and would never
1875 * change its di_mode again for this inode generation.
1876 * This is a temporary hack that would require a proper fix
1882 xfs_ifree_cluster(ip, tp, first_ino);
1889 * Reallocate the space for if_broot based on the number of records
1890 * being added or deleted as indicated in rec_diff. Move the records
1891 * and pointers in if_broot to fit the new size. When shrinking this
1892 * will eliminate holes between the records and pointers created by
1893 * the caller. When growing this will create holes to be filled in
1896 * The caller must not request to add more records than would fit in
1897 * the on-disk inode root. If the if_broot is currently NULL, then
1898 * if we adding records one will be allocated. The caller must also
1899 * not request that the number of records go below zero, although
1900 * it can go to zero.
1902 * ip -- the inode whose if_broot area is changing
1903 * ext_diff -- the change in the number of records, positive or negative,
1904 * requested for the if_broot array.
1912 struct xfs_mount *mp = ip->i_mount;
1915 struct xfs_btree_block *new_broot;
1922 * Handle the degenerate case quietly.
1924 if (rec_diff == 0) {
1928 ifp = XFS_IFORK_PTR(ip, whichfork);
1931 * If there wasn't any memory allocated before, just
1932 * allocate it now and get out.
1934 if (ifp->if_broot_bytes == 0) {
1935 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
1936 ifp->if_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
1937 ifp->if_broot_bytes = (int)new_size;
1942 * If there is already an existing if_broot, then we need
1943 * to realloc() it and shift the pointers to their new
1944 * location. The records don't change location because
1945 * they are kept butted up against the btree block header.
1947 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
1948 new_max = cur_max + rec_diff;
1949 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
1950 ifp->if_broot = kmem_realloc(ifp->if_broot, new_size,
1951 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
1952 KM_SLEEP | KM_NOFS);
1953 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
1954 ifp->if_broot_bytes);
1955 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
1957 ifp->if_broot_bytes = (int)new_size;
1958 ASSERT(ifp->if_broot_bytes <=
1959 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
1960 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
1965 * rec_diff is less than 0. In this case, we are shrinking the
1966 * if_broot buffer. It must already exist. If we go to zero
1967 * records, just get rid of the root and clear the status bit.
1969 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
1970 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
1971 new_max = cur_max + rec_diff;
1972 ASSERT(new_max >= 0);
1974 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
1978 new_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
1980 * First copy over the btree block header.
1982 memcpy(new_broot, ifp->if_broot, XFS_BTREE_LBLOCK_LEN);
1985 ifp->if_flags &= ~XFS_IFBROOT;
1989 * Only copy the records and pointers if there are any.
1993 * First copy the records.
1995 op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1);
1996 np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1);
1997 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2000 * Then copy the pointers.
2002 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2003 ifp->if_broot_bytes);
2004 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1,
2006 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2008 kmem_free(ifp->if_broot);
2009 ifp->if_broot = new_broot;
2010 ifp->if_broot_bytes = (int)new_size;
2011 ASSERT(ifp->if_broot_bytes <=
2012 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2018 * This is called when the amount of space needed for if_data
2019 * is increased or decreased. The change in size is indicated by
2020 * the number of bytes that need to be added or deleted in the
2021 * byte_diff parameter.
2023 * If the amount of space needed has decreased below the size of the
2024 * inline buffer, then switch to using the inline buffer. Otherwise,
2025 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2026 * to what is needed.
2028 * ip -- the inode whose if_data area is changing
2029 * byte_diff -- the change in the number of bytes, positive or negative,
2030 * requested for the if_data array.
2042 if (byte_diff == 0) {
2046 ifp = XFS_IFORK_PTR(ip, whichfork);
2047 new_size = (int)ifp->if_bytes + byte_diff;
2048 ASSERT(new_size >= 0);
2050 if (new_size == 0) {
2051 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2052 kmem_free(ifp->if_u1.if_data);
2054 ifp->if_u1.if_data = NULL;
2056 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2058 * If the valid extents/data can fit in if_inline_ext/data,
2059 * copy them from the malloc'd vector and free it.
2061 if (ifp->if_u1.if_data == NULL) {
2062 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2063 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2064 ASSERT(ifp->if_real_bytes != 0);
2065 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2067 kmem_free(ifp->if_u1.if_data);
2068 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2073 * Stuck with malloc/realloc.
2074 * For inline data, the underlying buffer must be
2075 * a multiple of 4 bytes in size so that it can be
2076 * logged and stay on word boundaries. We enforce
2079 real_size = roundup(new_size, 4);
2080 if (ifp->if_u1.if_data == NULL) {
2081 ASSERT(ifp->if_real_bytes == 0);
2082 ifp->if_u1.if_data = kmem_alloc(real_size,
2083 KM_SLEEP | KM_NOFS);
2084 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2086 * Only do the realloc if the underlying size
2087 * is really changing.
2089 if (ifp->if_real_bytes != real_size) {
2090 ifp->if_u1.if_data =
2091 kmem_realloc(ifp->if_u1.if_data,
2094 KM_SLEEP | KM_NOFS);
2097 ASSERT(ifp->if_real_bytes == 0);
2098 ifp->if_u1.if_data = kmem_alloc(real_size,
2099 KM_SLEEP | KM_NOFS);
2100 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2104 ifp->if_real_bytes = real_size;
2105 ifp->if_bytes = new_size;
2106 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2116 ifp = XFS_IFORK_PTR(ip, whichfork);
2117 if (ifp->if_broot != NULL) {
2118 kmem_free(ifp->if_broot);
2119 ifp->if_broot = NULL;
2123 * If the format is local, then we can't have an extents
2124 * array so just look for an inline data array. If we're
2125 * not local then we may or may not have an extents list,
2126 * so check and free it up if we do.
2128 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2129 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2130 (ifp->if_u1.if_data != NULL)) {
2131 ASSERT(ifp->if_real_bytes != 0);
2132 kmem_free(ifp->if_u1.if_data);
2133 ifp->if_u1.if_data = NULL;
2134 ifp->if_real_bytes = 0;
2136 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2137 ((ifp->if_flags & XFS_IFEXTIREC) ||
2138 ((ifp->if_u1.if_extents != NULL) &&
2139 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
2140 ASSERT(ifp->if_real_bytes != 0);
2141 xfs_iext_destroy(ifp);
2143 ASSERT(ifp->if_u1.if_extents == NULL ||
2144 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2145 ASSERT(ifp->if_real_bytes == 0);
2146 if (whichfork == XFS_ATTR_FORK) {
2147 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2153 * This is called to unpin an inode. The caller must have the inode locked
2154 * in at least shared mode so that the buffer cannot be subsequently pinned
2155 * once someone is waiting for it to be unpinned.
2159 struct xfs_inode *ip)
2161 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2163 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2165 /* Give the log a push to start the unpinning I/O */
2166 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2172 struct xfs_inode *ip)
2174 if (xfs_ipincount(ip)) {
2175 xfs_iunpin_nowait(ip);
2176 wait_event(ip->i_ipin_wait, (xfs_ipincount(ip) == 0));
2181 * xfs_iextents_copy()
2183 * This is called to copy the REAL extents (as opposed to the delayed
2184 * allocation extents) from the inode into the given buffer. It
2185 * returns the number of bytes copied into the buffer.
2187 * If there are no delayed allocation extents, then we can just
2188 * memcpy() the extents into the buffer. Otherwise, we need to
2189 * examine each extent in turn and skip those which are delayed.
2201 xfs_fsblock_t start_block;
2203 ifp = XFS_IFORK_PTR(ip, whichfork);
2204 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2205 ASSERT(ifp->if_bytes > 0);
2207 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2208 XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork);
2212 * There are some delayed allocation extents in the
2213 * inode, so copy the extents one at a time and skip
2214 * the delayed ones. There must be at least one
2215 * non-delayed extent.
2218 for (i = 0; i < nrecs; i++) {
2219 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
2220 start_block = xfs_bmbt_get_startblock(ep);
2221 if (isnullstartblock(start_block)) {
2223 * It's a delayed allocation extent, so skip it.
2228 /* Translate to on disk format */
2229 put_unaligned(cpu_to_be64(ep->l0), &dp->l0);
2230 put_unaligned(cpu_to_be64(ep->l1), &dp->l1);
2234 ASSERT(copied != 0);
2235 xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip));
2237 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2241 * Each of the following cases stores data into the same region
2242 * of the on-disk inode, so only one of them can be valid at
2243 * any given time. While it is possible to have conflicting formats
2244 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2245 * in EXTENTS format, this can only happen when the fork has
2246 * changed formats after being modified but before being flushed.
2247 * In these cases, the format always takes precedence, because the
2248 * format indicates the current state of the fork.
2255 xfs_inode_log_item_t *iip,
2262 #ifdef XFS_TRANS_DEBUG
2265 static const short brootflag[2] =
2266 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2267 static const short dataflag[2] =
2268 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2269 static const short extflag[2] =
2270 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2274 ifp = XFS_IFORK_PTR(ip, whichfork);
2276 * This can happen if we gave up in iformat in an error path,
2277 * for the attribute fork.
2280 ASSERT(whichfork == XFS_ATTR_FORK);
2283 cp = XFS_DFORK_PTR(dip, whichfork);
2285 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2286 case XFS_DINODE_FMT_LOCAL:
2287 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2288 (ifp->if_bytes > 0)) {
2289 ASSERT(ifp->if_u1.if_data != NULL);
2290 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2291 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2295 case XFS_DINODE_FMT_EXTENTS:
2296 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2297 !(iip->ili_format.ilf_fields & extflag[whichfork]));
2298 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
2299 (ifp->if_bytes > 0)) {
2300 ASSERT(xfs_iext_get_ext(ifp, 0));
2301 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2302 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2307 case XFS_DINODE_FMT_BTREE:
2308 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
2309 (ifp->if_broot_bytes > 0)) {
2310 ASSERT(ifp->if_broot != NULL);
2311 ASSERT(ifp->if_broot_bytes <=
2312 (XFS_IFORK_SIZE(ip, whichfork) +
2313 XFS_BROOT_SIZE_ADJ));
2314 xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes,
2315 (xfs_bmdr_block_t *)cp,
2316 XFS_DFORK_SIZE(dip, mp, whichfork));
2320 case XFS_DINODE_FMT_DEV:
2321 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
2322 ASSERT(whichfork == XFS_DATA_FORK);
2323 xfs_dinode_put_rdev(dip, ip->i_df.if_u2.if_rdev);
2327 case XFS_DINODE_FMT_UUID:
2328 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
2329 ASSERT(whichfork == XFS_DATA_FORK);
2330 memcpy(XFS_DFORK_DPTR(dip),
2331 &ip->i_df.if_u2.if_uuid,
2347 xfs_mount_t *mp = ip->i_mount;
2348 struct xfs_perag *pag;
2349 unsigned long first_index, mask;
2350 unsigned long inodes_per_cluster;
2352 xfs_inode_t **ilist;
2359 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2361 inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog;
2362 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
2363 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
2367 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
2368 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
2370 /* really need a gang lookup range call here */
2371 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
2372 first_index, inodes_per_cluster);
2376 for (i = 0; i < nr_found; i++) {
2382 * because this is an RCU protected lookup, we could find a
2383 * recently freed or even reallocated inode during the lookup.
2384 * We need to check under the i_flags_lock for a valid inode
2385 * here. Skip it if it is not valid or the wrong inode.
2387 spin_lock(&ip->i_flags_lock);
2389 (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) {
2390 spin_unlock(&ip->i_flags_lock);
2393 spin_unlock(&ip->i_flags_lock);
2396 * Do an un-protected check to see if the inode is dirty and
2397 * is a candidate for flushing. These checks will be repeated
2398 * later after the appropriate locks are acquired.
2400 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
2404 * Try to get locks. If any are unavailable or it is pinned,
2405 * then this inode cannot be flushed and is skipped.
2408 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
2410 if (!xfs_iflock_nowait(iq)) {
2411 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2414 if (xfs_ipincount(iq)) {
2416 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2421 * arriving here means that this inode can be flushed. First
2422 * re-check that it's dirty before flushing.
2424 if (!xfs_inode_clean(iq)) {
2426 error = xfs_iflush_int(iq, bp);
2428 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2429 goto cluster_corrupt_out;
2435 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2439 XFS_STATS_INC(xs_icluster_flushcnt);
2440 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
2451 cluster_corrupt_out:
2453 * Corruption detected in the clustering loop. Invalidate the
2454 * inode buffer and shut down the filesystem.
2458 * Clean up the buffer. If it was B_DELWRI, just release it --
2459 * brelse can handle it with no problems. If not, shut down the
2460 * filesystem before releasing the buffer.
2462 bufwasdelwri = XFS_BUF_ISDELAYWRITE(bp);
2466 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2468 if (!bufwasdelwri) {
2470 * Just like incore_relse: if we have b_iodone functions,
2471 * mark the buffer as an error and call them. Otherwise
2472 * mark it as stale and brelse.
2474 if (XFS_BUF_IODONE_FUNC(bp)) {
2477 XFS_BUF_ERROR(bp,EIO);
2478 xfs_buf_ioend(bp, 0);
2486 * Unlocks the flush lock
2488 xfs_iflush_abort(iq);
2491 return XFS_ERROR(EFSCORRUPTED);
2495 * xfs_iflush() will write a modified inode's changes out to the
2496 * inode's on disk home. The caller must have the inode lock held
2497 * in at least shared mode and the inode flush completion must be
2498 * active as well. The inode lock will still be held upon return from
2499 * the call and the caller is free to unlock it.
2500 * The inode flush will be completed when the inode reaches the disk.
2501 * The flags indicate how the inode's buffer should be written out.
2508 xfs_inode_log_item_t *iip;
2514 XFS_STATS_INC(xs_iflush_count);
2516 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2517 ASSERT(!completion_done(&ip->i_flush));
2518 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2519 ip->i_d.di_nextents > ip->i_df.if_ext_max);
2525 * We can't flush the inode until it is unpinned, so wait for it if we
2526 * are allowed to block. We know no one new can pin it, because we are
2527 * holding the inode lock shared and you need to hold it exclusively to
2530 * If we are not allowed to block, force the log out asynchronously so
2531 * that when we come back the inode will be unpinned. If other inodes
2532 * in the same cluster are dirty, they will probably write the inode
2533 * out for us if they occur after the log force completes.
2535 if (!(flags & SYNC_WAIT) && xfs_ipincount(ip)) {
2536 xfs_iunpin_nowait(ip);
2540 xfs_iunpin_wait(ip);
2543 * For stale inodes we cannot rely on the backing buffer remaining
2544 * stale in cache for the remaining life of the stale inode and so
2545 * xfs_itobp() below may give us a buffer that no longer contains
2546 * inodes below. We have to check this after ensuring the inode is
2547 * unpinned so that it is safe to reclaim the stale inode after the
2550 if (xfs_iflags_test(ip, XFS_ISTALE)) {
2556 * This may have been unpinned because the filesystem is shutting
2557 * down forcibly. If that's the case we must not write this inode
2558 * to disk, because the log record didn't make it to disk!
2560 if (XFS_FORCED_SHUTDOWN(mp)) {
2561 ip->i_update_core = 0;
2563 iip->ili_format.ilf_fields = 0;
2565 return XFS_ERROR(EIO);
2569 * Get the buffer containing the on-disk inode.
2571 error = xfs_itobp(mp, NULL, ip, &dip, &bp,
2572 (flags & SYNC_TRYLOCK) ? XBF_TRYLOCK : XBF_LOCK);
2579 * First flush out the inode that xfs_iflush was called with.
2581 error = xfs_iflush_int(ip, bp);
2586 * If the buffer is pinned then push on the log now so we won't
2587 * get stuck waiting in the write for too long.
2589 if (XFS_BUF_ISPINNED(bp))
2590 xfs_log_force(mp, 0);
2594 * see if other inodes can be gathered into this write
2596 error = xfs_iflush_cluster(ip, bp);
2598 goto cluster_corrupt_out;
2600 if (flags & SYNC_WAIT)
2601 error = xfs_bwrite(mp, bp);
2603 xfs_bdwrite(mp, bp);
2608 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2609 cluster_corrupt_out:
2611 * Unlocks the flush lock
2613 xfs_iflush_abort(ip);
2614 return XFS_ERROR(EFSCORRUPTED);
2623 xfs_inode_log_item_t *iip;
2626 #ifdef XFS_TRANS_DEBUG
2630 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2631 ASSERT(!completion_done(&ip->i_flush));
2632 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2633 ip->i_d.di_nextents > ip->i_df.if_ext_max);
2638 /* set *dip = inode's place in the buffer */
2639 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
2642 * Clear i_update_core before copying out the data.
2643 * This is for coordination with our timestamp updates
2644 * that don't hold the inode lock. They will always
2645 * update the timestamps BEFORE setting i_update_core,
2646 * so if we clear i_update_core after they set it we
2647 * are guaranteed to see their updates to the timestamps.
2648 * I believe that this depends on strongly ordered memory
2649 * semantics, but we have that. We use the SYNCHRONIZE
2650 * macro to make sure that the compiler does not reorder
2651 * the i_update_core access below the data copy below.
2653 ip->i_update_core = 0;
2657 * Make sure to get the latest timestamps from the Linux inode.
2659 xfs_synchronize_times(ip);
2661 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
2662 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
2663 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2664 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
2665 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
2668 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
2669 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
2670 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2671 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
2672 __func__, ip->i_ino, ip, ip->i_d.di_magic);
2675 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
2677 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
2678 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
2679 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
2680 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2681 "%s: Bad regular inode %Lu, ptr 0x%p",
2682 __func__, ip->i_ino, ip);
2685 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
2687 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
2688 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
2689 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
2690 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
2691 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2692 "%s: Bad directory inode %Lu, ptr 0x%p",
2693 __func__, ip->i_ino, ip);
2697 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
2698 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
2699 XFS_RANDOM_IFLUSH_5)) {
2700 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2701 "%s: detected corrupt incore inode %Lu, "
2702 "total extents = %d, nblocks = %Ld, ptr 0x%p",
2703 __func__, ip->i_ino,
2704 ip->i_d.di_nextents + ip->i_d.di_anextents,
2705 ip->i_d.di_nblocks, ip);
2708 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
2709 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
2710 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2711 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
2712 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
2716 * bump the flush iteration count, used to detect flushes which
2717 * postdate a log record during recovery.
2720 ip->i_d.di_flushiter++;
2723 * Copy the dirty parts of the inode into the on-disk
2724 * inode. We always copy out the core of the inode,
2725 * because if the inode is dirty at all the core must
2728 xfs_dinode_to_disk(dip, &ip->i_d);
2730 /* Wrap, we never let the log put out DI_MAX_FLUSH */
2731 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
2732 ip->i_d.di_flushiter = 0;
2735 * If this is really an old format inode and the superblock version
2736 * has not been updated to support only new format inodes, then
2737 * convert back to the old inode format. If the superblock version
2738 * has been updated, then make the conversion permanent.
2740 ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
2741 if (ip->i_d.di_version == 1) {
2742 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
2746 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
2747 dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink);
2750 * The superblock version has already been bumped,
2751 * so just make the conversion to the new inode
2754 ip->i_d.di_version = 2;
2755 dip->di_version = 2;
2756 ip->i_d.di_onlink = 0;
2758 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
2759 memset(&(dip->di_pad[0]), 0,
2760 sizeof(dip->di_pad));
2761 ASSERT(xfs_get_projid(ip) == 0);
2765 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
2766 if (XFS_IFORK_Q(ip))
2767 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
2768 xfs_inobp_check(mp, bp);
2771 * We've recorded everything logged in the inode, so we'd
2772 * like to clear the ilf_fields bits so we don't log and
2773 * flush things unnecessarily. However, we can't stop
2774 * logging all this information until the data we've copied
2775 * into the disk buffer is written to disk. If we did we might
2776 * overwrite the copy of the inode in the log with all the
2777 * data after re-logging only part of it, and in the face of
2778 * a crash we wouldn't have all the data we need to recover.
2780 * What we do is move the bits to the ili_last_fields field.
2781 * When logging the inode, these bits are moved back to the
2782 * ilf_fields field. In the xfs_iflush_done() routine we
2783 * clear ili_last_fields, since we know that the information
2784 * those bits represent is permanently on disk. As long as
2785 * the flush completes before the inode is logged again, then
2786 * both ilf_fields and ili_last_fields will be cleared.
2788 * We can play with the ilf_fields bits here, because the inode
2789 * lock must be held exclusively in order to set bits there
2790 * and the flush lock protects the ili_last_fields bits.
2791 * Set ili_logged so the flush done
2792 * routine can tell whether or not to look in the AIL.
2793 * Also, store the current LSN of the inode so that we can tell
2794 * whether the item has moved in the AIL from xfs_iflush_done().
2795 * In order to read the lsn we need the AIL lock, because
2796 * it is a 64 bit value that cannot be read atomically.
2798 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
2799 iip->ili_last_fields = iip->ili_format.ilf_fields;
2800 iip->ili_format.ilf_fields = 0;
2801 iip->ili_logged = 1;
2803 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2804 &iip->ili_item.li_lsn);
2807 * Attach the function xfs_iflush_done to the inode's
2808 * buffer. This will remove the inode from the AIL
2809 * and unlock the inode's flush lock when the inode is
2810 * completely written to disk.
2812 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
2814 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
2815 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
2818 * We're flushing an inode which is not in the AIL and has
2819 * not been logged but has i_update_core set. For this
2820 * case we can use a B_DELWRI flush and immediately drop
2821 * the inode flush lock because we can avoid the whole
2822 * AIL state thing. It's OK to drop the flush lock now,
2823 * because we've already locked the buffer and to do anything
2824 * you really need both.
2827 ASSERT(iip->ili_logged == 0);
2828 ASSERT(iip->ili_last_fields == 0);
2829 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
2837 return XFS_ERROR(EFSCORRUPTED);
2841 * Return a pointer to the extent record at file index idx.
2843 xfs_bmbt_rec_host_t *
2845 xfs_ifork_t *ifp, /* inode fork pointer */
2846 xfs_extnum_t idx) /* index of target extent */
2849 ASSERT(idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t));
2851 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
2852 return ifp->if_u1.if_ext_irec->er_extbuf;
2853 } else if (ifp->if_flags & XFS_IFEXTIREC) {
2854 xfs_ext_irec_t *erp; /* irec pointer */
2855 int erp_idx = 0; /* irec index */
2856 xfs_extnum_t page_idx = idx; /* ext index in target list */
2858 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
2859 return &erp->er_extbuf[page_idx];
2860 } else if (ifp->if_bytes) {
2861 return &ifp->if_u1.if_extents[idx];
2868 * Insert new item(s) into the extent records for incore inode
2869 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
2873 xfs_inode_t *ip, /* incore inode pointer */
2874 xfs_extnum_t idx, /* starting index of new items */
2875 xfs_extnum_t count, /* number of inserted items */
2876 xfs_bmbt_irec_t *new, /* items to insert */
2877 int state) /* type of extent conversion */
2879 xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
2880 xfs_extnum_t i; /* extent record index */
2882 trace_xfs_iext_insert(ip, idx, new, state, _RET_IP_);
2884 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2885 xfs_iext_add(ifp, idx, count);
2886 for (i = idx; i < idx + count; i++, new++)
2887 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new);
2891 * This is called when the amount of space required for incore file
2892 * extents needs to be increased. The ext_diff parameter stores the
2893 * number of new extents being added and the idx parameter contains
2894 * the extent index where the new extents will be added. If the new
2895 * extents are being appended, then we just need to (re)allocate and
2896 * initialize the space. Otherwise, if the new extents are being
2897 * inserted into the middle of the existing entries, a bit more work
2898 * is required to make room for the new extents to be inserted. The
2899 * caller is responsible for filling in the new extent entries upon
2904 xfs_ifork_t *ifp, /* inode fork pointer */
2905 xfs_extnum_t idx, /* index to begin adding exts */
2906 int ext_diff) /* number of extents to add */
2908 int byte_diff; /* new bytes being added */
2909 int new_size; /* size of extents after adding */
2910 xfs_extnum_t nextents; /* number of extents in file */
2912 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2913 ASSERT((idx >= 0) && (idx <= nextents));
2914 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
2915 new_size = ifp->if_bytes + byte_diff;
2917 * If the new number of extents (nextents + ext_diff)
2918 * fits inside the inode, then continue to use the inline
2921 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
2922 if (idx < nextents) {
2923 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
2924 &ifp->if_u2.if_inline_ext[idx],
2925 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
2926 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
2928 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
2929 ifp->if_real_bytes = 0;
2932 * Otherwise use a linear (direct) extent list.
2933 * If the extents are currently inside the inode,
2934 * xfs_iext_realloc_direct will switch us from
2935 * inline to direct extent allocation mode.
2937 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
2938 xfs_iext_realloc_direct(ifp, new_size);
2939 if (idx < nextents) {
2940 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
2941 &ifp->if_u1.if_extents[idx],
2942 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
2943 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
2946 /* Indirection array */
2948 xfs_ext_irec_t *erp;
2952 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
2953 if (ifp->if_flags & XFS_IFEXTIREC) {
2954 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
2956 xfs_iext_irec_init(ifp);
2957 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
2958 erp = ifp->if_u1.if_ext_irec;
2960 /* Extents fit in target extent page */
2961 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
2962 if (page_idx < erp->er_extcount) {
2963 memmove(&erp->er_extbuf[page_idx + ext_diff],
2964 &erp->er_extbuf[page_idx],
2965 (erp->er_extcount - page_idx) *
2966 sizeof(xfs_bmbt_rec_t));
2967 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
2969 erp->er_extcount += ext_diff;
2970 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
2972 /* Insert a new extent page */
2974 xfs_iext_add_indirect_multi(ifp,
2975 erp_idx, page_idx, ext_diff);
2978 * If extent(s) are being appended to the last page in
2979 * the indirection array and the new extent(s) don't fit
2980 * in the page, then erp is NULL and erp_idx is set to
2981 * the next index needed in the indirection array.
2984 int count = ext_diff;
2987 erp = xfs_iext_irec_new(ifp, erp_idx);
2988 erp->er_extcount = count;
2989 count -= MIN(count, (int)XFS_LINEAR_EXTS);
2996 ifp->if_bytes = new_size;
3000 * This is called when incore extents are being added to the indirection
3001 * array and the new extents do not fit in the target extent list. The
3002 * erp_idx parameter contains the irec index for the target extent list
3003 * in the indirection array, and the idx parameter contains the extent
3004 * index within the list. The number of extents being added is stored
3005 * in the count parameter.
3007 * |-------| |-------|
3008 * | | | | idx - number of extents before idx
3010 * | | | | count - number of extents being inserted at idx
3011 * |-------| |-------|
3012 * | count | | nex2 | nex2 - number of extents after idx + count
3013 * |-------| |-------|
3016 xfs_iext_add_indirect_multi(
3017 xfs_ifork_t *ifp, /* inode fork pointer */
3018 int erp_idx, /* target extent irec index */
3019 xfs_extnum_t idx, /* index within target list */
3020 int count) /* new extents being added */
3022 int byte_diff; /* new bytes being added */
3023 xfs_ext_irec_t *erp; /* pointer to irec entry */
3024 xfs_extnum_t ext_diff; /* number of extents to add */
3025 xfs_extnum_t ext_cnt; /* new extents still needed */
3026 xfs_extnum_t nex2; /* extents after idx + count */
3027 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3028 int nlists; /* number of irec's (lists) */
3030 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3031 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3032 nex2 = erp->er_extcount - idx;
3033 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3036 * Save second part of target extent list
3037 * (all extents past */
3039 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3040 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS);
3041 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3042 erp->er_extcount -= nex2;
3043 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3044 memset(&erp->er_extbuf[idx], 0, byte_diff);
3048 * Add the new extents to the end of the target
3049 * list, then allocate new irec record(s) and
3050 * extent buffer(s) as needed to store the rest
3051 * of the new extents.
3054 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3056 erp->er_extcount += ext_diff;
3057 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3058 ext_cnt -= ext_diff;
3062 erp = xfs_iext_irec_new(ifp, erp_idx);
3063 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3064 erp->er_extcount = ext_diff;
3065 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3066 ext_cnt -= ext_diff;
3069 /* Add nex2 extents back to indirection array */
3071 xfs_extnum_t ext_avail;
3074 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3075 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3078 * If nex2 extents fit in the current page, append
3079 * nex2_ep after the new extents.
3081 if (nex2 <= ext_avail) {
3082 i = erp->er_extcount;
3085 * Otherwise, check if space is available in the
3088 else if ((erp_idx < nlists - 1) &&
3089 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3090 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3093 /* Create a hole for nex2 extents */
3094 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3095 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3098 * Final choice, create a new extent page for
3103 erp = xfs_iext_irec_new(ifp, erp_idx);
3105 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
3107 erp->er_extcount += nex2;
3108 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3113 * This is called when the amount of space required for incore file
3114 * extents needs to be decreased. The ext_diff parameter stores the
3115 * number of extents to be removed and the idx parameter contains
3116 * the extent index where the extents will be removed from.
3118 * If the amount of space needed has decreased below the linear
3119 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3120 * extent array. Otherwise, use kmem_realloc() to adjust the
3121 * size to what is needed.
3125 xfs_inode_t *ip, /* incore inode pointer */
3126 xfs_extnum_t idx, /* index to begin removing exts */
3127 int ext_diff, /* number of extents to remove */
3128 int state) /* type of extent conversion */
3130 xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
3131 xfs_extnum_t nextents; /* number of extents in file */
3132 int new_size; /* size of extents after removal */
3134 trace_xfs_iext_remove(ip, idx, state, _RET_IP_);
3136 ASSERT(ext_diff > 0);
3137 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3138 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3140 if (new_size == 0) {
3141 xfs_iext_destroy(ifp);
3142 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3143 xfs_iext_remove_indirect(ifp, idx, ext_diff);
3144 } else if (ifp->if_real_bytes) {
3145 xfs_iext_remove_direct(ifp, idx, ext_diff);
3147 xfs_iext_remove_inline(ifp, idx, ext_diff);
3149 ifp->if_bytes = new_size;
3153 * This removes ext_diff extents from the inline buffer, beginning
3154 * at extent index idx.
3157 xfs_iext_remove_inline(
3158 xfs_ifork_t *ifp, /* inode fork pointer */
3159 xfs_extnum_t idx, /* index to begin removing exts */
3160 int ext_diff) /* number of extents to remove */
3162 int nextents; /* number of extents in file */
3164 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3165 ASSERT(idx < XFS_INLINE_EXTS);
3166 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3167 ASSERT(((nextents - ext_diff) > 0) &&
3168 (nextents - ext_diff) < XFS_INLINE_EXTS);
3170 if (idx + ext_diff < nextents) {
3171 memmove(&ifp->if_u2.if_inline_ext[idx],
3172 &ifp->if_u2.if_inline_ext[idx + ext_diff],
3173 (nextents - (idx + ext_diff)) *
3174 sizeof(xfs_bmbt_rec_t));
3175 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
3176 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3178 memset(&ifp->if_u2.if_inline_ext[idx], 0,
3179 ext_diff * sizeof(xfs_bmbt_rec_t));
3184 * This removes ext_diff extents from a linear (direct) extent list,
3185 * beginning at extent index idx. If the extents are being removed
3186 * from the end of the list (ie. truncate) then we just need to re-
3187 * allocate the list to remove the extra space. Otherwise, if the
3188 * extents are being removed from the middle of the existing extent
3189 * entries, then we first need to move the extent records beginning
3190 * at idx + ext_diff up in the list to overwrite the records being
3191 * removed, then remove the extra space via kmem_realloc.
3194 xfs_iext_remove_direct(
3195 xfs_ifork_t *ifp, /* inode fork pointer */
3196 xfs_extnum_t idx, /* index to begin removing exts */
3197 int ext_diff) /* number of extents to remove */
3199 xfs_extnum_t nextents; /* number of extents in file */
3200 int new_size; /* size of extents after removal */
3202 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3203 new_size = ifp->if_bytes -
3204 (ext_diff * sizeof(xfs_bmbt_rec_t));
3205 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3207 if (new_size == 0) {
3208 xfs_iext_destroy(ifp);
3211 /* Move extents up in the list (if needed) */
3212 if (idx + ext_diff < nextents) {
3213 memmove(&ifp->if_u1.if_extents[idx],
3214 &ifp->if_u1.if_extents[idx + ext_diff],
3215 (nextents - (idx + ext_diff)) *
3216 sizeof(xfs_bmbt_rec_t));
3218 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
3219 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3221 * Reallocate the direct extent list. If the extents
3222 * will fit inside the inode then xfs_iext_realloc_direct
3223 * will switch from direct to inline extent allocation
3226 xfs_iext_realloc_direct(ifp, new_size);
3227 ifp->if_bytes = new_size;
3231 * This is called when incore extents are being removed from the
3232 * indirection array and the extents being removed span multiple extent
3233 * buffers. The idx parameter contains the file extent index where we
3234 * want to begin removing extents, and the count parameter contains
3235 * how many extents need to be removed.
3237 * |-------| |-------|
3238 * | nex1 | | | nex1 - number of extents before idx
3239 * |-------| | count |
3240 * | | | | count - number of extents being removed at idx
3241 * | count | |-------|
3242 * | | | nex2 | nex2 - number of extents after idx + count
3243 * |-------| |-------|
3246 xfs_iext_remove_indirect(
3247 xfs_ifork_t *ifp, /* inode fork pointer */
3248 xfs_extnum_t idx, /* index to begin removing extents */
3249 int count) /* number of extents to remove */
3251 xfs_ext_irec_t *erp; /* indirection array pointer */
3252 int erp_idx = 0; /* indirection array index */
3253 xfs_extnum_t ext_cnt; /* extents left to remove */
3254 xfs_extnum_t ext_diff; /* extents to remove in current list */
3255 xfs_extnum_t nex1; /* number of extents before idx */
3256 xfs_extnum_t nex2; /* extents after idx + count */
3257 int page_idx = idx; /* index in target extent list */
3259 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3260 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3261 ASSERT(erp != NULL);
3265 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
3266 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
3268 * Check for deletion of entire list;
3269 * xfs_iext_irec_remove() updates extent offsets.
3271 if (ext_diff == erp->er_extcount) {
3272 xfs_iext_irec_remove(ifp, erp_idx);
3273 ext_cnt -= ext_diff;
3276 ASSERT(erp_idx < ifp->if_real_bytes /
3278 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3285 /* Move extents up (if needed) */
3287 memmove(&erp->er_extbuf[nex1],
3288 &erp->er_extbuf[nex1 + ext_diff],
3289 nex2 * sizeof(xfs_bmbt_rec_t));
3291 /* Zero out rest of page */
3292 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
3293 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
3294 /* Update remaining counters */
3295 erp->er_extcount -= ext_diff;
3296 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
3297 ext_cnt -= ext_diff;
3302 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
3303 xfs_iext_irec_compact(ifp);
3307 * Create, destroy, or resize a linear (direct) block of extents.
3310 xfs_iext_realloc_direct(
3311 xfs_ifork_t *ifp, /* inode fork pointer */
3312 int new_size) /* new size of extents */
3314 int rnew_size; /* real new size of extents */
3316 rnew_size = new_size;
3318 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
3319 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
3320 (new_size != ifp->if_real_bytes)));
3322 /* Free extent records */
3323 if (new_size == 0) {
3324 xfs_iext_destroy(ifp);
3326 /* Resize direct extent list and zero any new bytes */
3327 else if (ifp->if_real_bytes) {
3328 /* Check if extents will fit inside the inode */
3329 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
3330 xfs_iext_direct_to_inline(ifp, new_size /
3331 (uint)sizeof(xfs_bmbt_rec_t));
3332 ifp->if_bytes = new_size;
3335 if (!is_power_of_2(new_size)){
3336 rnew_size = roundup_pow_of_two(new_size);
3338 if (rnew_size != ifp->if_real_bytes) {
3339 ifp->if_u1.if_extents =
3340 kmem_realloc(ifp->if_u1.if_extents,
3342 ifp->if_real_bytes, KM_NOFS);
3344 if (rnew_size > ifp->if_real_bytes) {
3345 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
3346 (uint)sizeof(xfs_bmbt_rec_t)], 0,
3347 rnew_size - ifp->if_real_bytes);
3351 * Switch from the inline extent buffer to a direct
3352 * extent list. Be sure to include the inline extent
3353 * bytes in new_size.
3356 new_size += ifp->if_bytes;
3357 if (!is_power_of_2(new_size)) {
3358 rnew_size = roundup_pow_of_two(new_size);
3360 xfs_iext_inline_to_direct(ifp, rnew_size);
3362 ifp->if_real_bytes = rnew_size;
3363 ifp->if_bytes = new_size;
3367 * Switch from linear (direct) extent records to inline buffer.
3370 xfs_iext_direct_to_inline(
3371 xfs_ifork_t *ifp, /* inode fork pointer */
3372 xfs_extnum_t nextents) /* number of extents in file */
3374 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3375 ASSERT(nextents <= XFS_INLINE_EXTS);
3377 * The inline buffer was zeroed when we switched
3378 * from inline to direct extent allocation mode,
3379 * so we don't need to clear it here.
3381 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
3382 nextents * sizeof(xfs_bmbt_rec_t));
3383 kmem_free(ifp->if_u1.if_extents);
3384 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3385 ifp->if_real_bytes = 0;
3389 * Switch from inline buffer to linear (direct) extent records.
3390 * new_size should already be rounded up to the next power of 2
3391 * by the caller (when appropriate), so use new_size as it is.
3392 * However, since new_size may be rounded up, we can't update
3393 * if_bytes here. It is the caller's responsibility to update
3394 * if_bytes upon return.
3397 xfs_iext_inline_to_direct(
3398 xfs_ifork_t *ifp, /* inode fork pointer */
3399 int new_size) /* number of extents in file */
3401 ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS);
3402 memset(ifp->if_u1.if_extents, 0, new_size);
3403 if (ifp->if_bytes) {
3404 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
3406 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3407 sizeof(xfs_bmbt_rec_t));
3409 ifp->if_real_bytes = new_size;
3413 * Resize an extent indirection array to new_size bytes.
3416 xfs_iext_realloc_indirect(
3417 xfs_ifork_t *ifp, /* inode fork pointer */
3418 int new_size) /* new indirection array size */
3420 int nlists; /* number of irec's (ex lists) */
3421 int size; /* current indirection array size */
3423 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3424 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3425 size = nlists * sizeof(xfs_ext_irec_t);
3426 ASSERT(ifp->if_real_bytes);
3427 ASSERT((new_size >= 0) && (new_size != size));
3428 if (new_size == 0) {
3429 xfs_iext_destroy(ifp);
3431 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
3432 kmem_realloc(ifp->if_u1.if_ext_irec,
3433 new_size, size, KM_NOFS);
3438 * Switch from indirection array to linear (direct) extent allocations.
3441 xfs_iext_indirect_to_direct(
3442 xfs_ifork_t *ifp) /* inode fork pointer */
3444 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
3445 xfs_extnum_t nextents; /* number of extents in file */
3446 int size; /* size of file extents */
3448 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3449 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3450 ASSERT(nextents <= XFS_LINEAR_EXTS);
3451 size = nextents * sizeof(xfs_bmbt_rec_t);
3453 xfs_iext_irec_compact_pages(ifp);
3454 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
3456 ep = ifp->if_u1.if_ext_irec->er_extbuf;
3457 kmem_free(ifp->if_u1.if_ext_irec);
3458 ifp->if_flags &= ~XFS_IFEXTIREC;
3459 ifp->if_u1.if_extents = ep;
3460 ifp->if_bytes = size;
3461 if (nextents < XFS_LINEAR_EXTS) {
3462 xfs_iext_realloc_direct(ifp, size);
3467 * Free incore file extents.
3471 xfs_ifork_t *ifp) /* inode fork pointer */
3473 if (ifp->if_flags & XFS_IFEXTIREC) {
3477 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3478 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
3479 xfs_iext_irec_remove(ifp, erp_idx);
3481 ifp->if_flags &= ~XFS_IFEXTIREC;
3482 } else if (ifp->if_real_bytes) {
3483 kmem_free(ifp->if_u1.if_extents);
3484 } else if (ifp->if_bytes) {
3485 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3486 sizeof(xfs_bmbt_rec_t));
3488 ifp->if_u1.if_extents = NULL;
3489 ifp->if_real_bytes = 0;
3494 * Return a pointer to the extent record for file system block bno.
3496 xfs_bmbt_rec_host_t * /* pointer to found extent record */
3497 xfs_iext_bno_to_ext(
3498 xfs_ifork_t *ifp, /* inode fork pointer */
3499 xfs_fileoff_t bno, /* block number to search for */
3500 xfs_extnum_t *idxp) /* index of target extent */
3502 xfs_bmbt_rec_host_t *base; /* pointer to first extent */
3503 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
3504 xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */
3505 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
3506 int high; /* upper boundary in search */
3507 xfs_extnum_t idx = 0; /* index of target extent */
3508 int low; /* lower boundary in search */
3509 xfs_extnum_t nextents; /* number of file extents */
3510 xfs_fileoff_t startoff = 0; /* start offset of extent */
3512 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3513 if (nextents == 0) {
3518 if (ifp->if_flags & XFS_IFEXTIREC) {
3519 /* Find target extent list */
3521 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
3522 base = erp->er_extbuf;
3523 high = erp->er_extcount - 1;
3525 base = ifp->if_u1.if_extents;
3526 high = nextents - 1;
3528 /* Binary search extent records */
3529 while (low <= high) {
3530 idx = (low + high) >> 1;
3532 startoff = xfs_bmbt_get_startoff(ep);
3533 blockcount = xfs_bmbt_get_blockcount(ep);
3534 if (bno < startoff) {
3536 } else if (bno >= startoff + blockcount) {
3539 /* Convert back to file-based extent index */
3540 if (ifp->if_flags & XFS_IFEXTIREC) {
3541 idx += erp->er_extoff;
3547 /* Convert back to file-based extent index */
3548 if (ifp->if_flags & XFS_IFEXTIREC) {
3549 idx += erp->er_extoff;
3551 if (bno >= startoff + blockcount) {
3552 if (++idx == nextents) {
3555 ep = xfs_iext_get_ext(ifp, idx);
3563 * Return a pointer to the indirection array entry containing the
3564 * extent record for filesystem block bno. Store the index of the
3565 * target irec in *erp_idxp.
3567 xfs_ext_irec_t * /* pointer to found extent record */
3568 xfs_iext_bno_to_irec(
3569 xfs_ifork_t *ifp, /* inode fork pointer */
3570 xfs_fileoff_t bno, /* block number to search for */
3571 int *erp_idxp) /* irec index of target ext list */
3573 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
3574 xfs_ext_irec_t *erp_next; /* next indirection array entry */
3575 int erp_idx; /* indirection array index */
3576 int nlists; /* number of extent irec's (lists) */
3577 int high; /* binary search upper limit */
3578 int low; /* binary search lower limit */
3580 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3581 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3585 while (low <= high) {
3586 erp_idx = (low + high) >> 1;
3587 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3588 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
3589 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
3591 } else if (erp_next && bno >=
3592 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
3598 *erp_idxp = erp_idx;
3603 * Return a pointer to the indirection array entry containing the
3604 * extent record at file extent index *idxp. Store the index of the
3605 * target irec in *erp_idxp and store the page index of the target
3606 * extent record in *idxp.
3609 xfs_iext_idx_to_irec(
3610 xfs_ifork_t *ifp, /* inode fork pointer */
3611 xfs_extnum_t *idxp, /* extent index (file -> page) */
3612 int *erp_idxp, /* pointer to target irec */
3613 int realloc) /* new bytes were just added */
3615 xfs_ext_irec_t *prev; /* pointer to previous irec */
3616 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
3617 int erp_idx; /* indirection array index */
3618 int nlists; /* number of irec's (ex lists) */
3619 int high; /* binary search upper limit */
3620 int low; /* binary search lower limit */
3621 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
3623 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3624 ASSERT(page_idx >= 0);
3625 ASSERT(page_idx <= ifp->if_bytes / sizeof(xfs_bmbt_rec_t));
3626 ASSERT(page_idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t) || realloc);
3628 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3633 /* Binary search extent irec's */
3634 while (low <= high) {
3635 erp_idx = (low + high) >> 1;
3636 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3637 prev = erp_idx > 0 ? erp - 1 : NULL;
3638 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
3639 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
3641 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
3642 (page_idx == erp->er_extoff + erp->er_extcount &&
3645 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
3646 erp->er_extcount == XFS_LINEAR_EXTS) {
3650 erp = erp_idx < nlists ? erp + 1 : NULL;
3653 page_idx -= erp->er_extoff;
3658 *erp_idxp = erp_idx;
3663 * Allocate and initialize an indirection array once the space needed
3664 * for incore extents increases above XFS_IEXT_BUFSZ.
3668 xfs_ifork_t *ifp) /* inode fork pointer */
3670 xfs_ext_irec_t *erp; /* indirection array pointer */
3671 xfs_extnum_t nextents; /* number of extents in file */
3673 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3674 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3675 ASSERT(nextents <= XFS_LINEAR_EXTS);
3677 erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS);
3679 if (nextents == 0) {
3680 ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
3681 } else if (!ifp->if_real_bytes) {
3682 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
3683 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
3684 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
3686 erp->er_extbuf = ifp->if_u1.if_extents;
3687 erp->er_extcount = nextents;
3690 ifp->if_flags |= XFS_IFEXTIREC;
3691 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
3692 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
3693 ifp->if_u1.if_ext_irec = erp;
3699 * Allocate and initialize a new entry in the indirection array.
3703 xfs_ifork_t *ifp, /* inode fork pointer */
3704 int erp_idx) /* index for new irec */
3706 xfs_ext_irec_t *erp; /* indirection array pointer */
3707 int i; /* loop counter */
3708 int nlists; /* number of irec's (ex lists) */
3710 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3711 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3713 /* Resize indirection array */
3714 xfs_iext_realloc_indirect(ifp, ++nlists *
3715 sizeof(xfs_ext_irec_t));
3717 * Move records down in the array so the
3718 * new page can use erp_idx.
3720 erp = ifp->if_u1.if_ext_irec;
3721 for (i = nlists - 1; i > erp_idx; i--) {
3722 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
3724 ASSERT(i == erp_idx);
3726 /* Initialize new extent record */
3727 erp = ifp->if_u1.if_ext_irec;
3728 erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
3729 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
3730 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
3731 erp[erp_idx].er_extcount = 0;
3732 erp[erp_idx].er_extoff = erp_idx > 0 ?
3733 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
3734 return (&erp[erp_idx]);
3738 * Remove a record from the indirection array.
3741 xfs_iext_irec_remove(
3742 xfs_ifork_t *ifp, /* inode fork pointer */
3743 int erp_idx) /* irec index to remove */
3745 xfs_ext_irec_t *erp; /* indirection array pointer */
3746 int i; /* loop counter */
3747 int nlists; /* number of irec's (ex lists) */
3749 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3750 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3751 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3752 if (erp->er_extbuf) {
3753 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
3755 kmem_free(erp->er_extbuf);
3757 /* Compact extent records */
3758 erp = ifp->if_u1.if_ext_irec;
3759 for (i = erp_idx; i < nlists - 1; i++) {
3760 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
3763 * Manually free the last extent record from the indirection
3764 * array. A call to xfs_iext_realloc_indirect() with a size
3765 * of zero would result in a call to xfs_iext_destroy() which
3766 * would in turn call this function again, creating a nasty
3770 xfs_iext_realloc_indirect(ifp,
3771 nlists * sizeof(xfs_ext_irec_t));
3773 kmem_free(ifp->if_u1.if_ext_irec);
3775 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
3779 * This is called to clean up large amounts of unused memory allocated
3780 * by the indirection array. Before compacting anything though, verify
3781 * that the indirection array is still needed and switch back to the
3782 * linear extent list (or even the inline buffer) if possible. The
3783 * compaction policy is as follows:
3785 * Full Compaction: Extents fit into a single page (or inline buffer)
3786 * Partial Compaction: Extents occupy less than 50% of allocated space
3787 * No Compaction: Extents occupy at least 50% of allocated space
3790 xfs_iext_irec_compact(
3791 xfs_ifork_t *ifp) /* inode fork pointer */
3793 xfs_extnum_t nextents; /* number of extents in file */
3794 int nlists; /* number of irec's (ex lists) */
3796 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3797 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3798 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3800 if (nextents == 0) {
3801 xfs_iext_destroy(ifp);
3802 } else if (nextents <= XFS_INLINE_EXTS) {
3803 xfs_iext_indirect_to_direct(ifp);
3804 xfs_iext_direct_to_inline(ifp, nextents);
3805 } else if (nextents <= XFS_LINEAR_EXTS) {
3806 xfs_iext_indirect_to_direct(ifp);
3807 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
3808 xfs_iext_irec_compact_pages(ifp);
3813 * Combine extents from neighboring extent pages.
3816 xfs_iext_irec_compact_pages(
3817 xfs_ifork_t *ifp) /* inode fork pointer */
3819 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
3820 int erp_idx = 0; /* indirection array index */
3821 int nlists; /* number of irec's (ex lists) */
3823 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3824 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3825 while (erp_idx < nlists - 1) {
3826 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3828 if (erp_next->er_extcount <=
3829 (XFS_LINEAR_EXTS - erp->er_extcount)) {
3830 memcpy(&erp->er_extbuf[erp->er_extcount],
3831 erp_next->er_extbuf, erp_next->er_extcount *
3832 sizeof(xfs_bmbt_rec_t));
3833 erp->er_extcount += erp_next->er_extcount;
3835 * Free page before removing extent record
3836 * so er_extoffs don't get modified in
3837 * xfs_iext_irec_remove.
3839 kmem_free(erp_next->er_extbuf);
3840 erp_next->er_extbuf = NULL;
3841 xfs_iext_irec_remove(ifp, erp_idx + 1);
3842 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3850 * This is called to update the er_extoff field in the indirection
3851 * array when extents have been added or removed from one of the
3852 * extent lists. erp_idx contains the irec index to begin updating
3853 * at and ext_diff contains the number of extents that were added
3857 xfs_iext_irec_update_extoffs(
3858 xfs_ifork_t *ifp, /* inode fork pointer */
3859 int erp_idx, /* irec index to update */
3860 int ext_diff) /* number of new extents */
3862 int i; /* loop counter */
3863 int nlists; /* number of irec's (ex lists */
3865 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3866 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3867 for (i = erp_idx; i < nlists; i++) {
3868 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;