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(). 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) {
113 xfs_fs_cmn_err(CE_ALERT, mp,
114 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
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 "xfs_imap_to_bp: xfs_trans_read_buf()returned "
147 "an error %d on %s. Returning error.",
148 error, mp->m_fsname);
150 ASSERT(buf_flags & XBF_TRYLOCK);
156 * Validate the magic number and version of every inode in the buffer
157 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
160 ni = BBTOB(imap->im_len) >> mp->m_sb.sb_inodelog;
161 #else /* usual case */
165 for (i = 0; i < ni; i++) {
169 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
170 (i << mp->m_sb.sb_inodelog));
171 di_ok = be16_to_cpu(dip->di_magic) == XFS_DINODE_MAGIC &&
172 XFS_DINODE_GOOD_VERSION(dip->di_version);
173 if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
174 XFS_ERRTAG_ITOBP_INOTOBP,
175 XFS_RANDOM_ITOBP_INOTOBP))) {
176 if (iget_flags & XFS_IGET_UNTRUSTED) {
177 xfs_trans_brelse(tp, bp);
178 return XFS_ERROR(EINVAL);
180 XFS_CORRUPTION_ERROR("xfs_imap_to_bp",
181 XFS_ERRLEVEL_HIGH, mp, dip);
184 "Device %s - bad inode magic/vsn "
185 "daddr %lld #%d (magic=%x)",
186 XFS_BUFTARG_NAME(mp->m_ddev_targp),
187 (unsigned long long)imap->im_blkno, i,
188 be16_to_cpu(dip->di_magic));
190 xfs_trans_brelse(tp, bp);
191 return XFS_ERROR(EFSCORRUPTED);
195 xfs_inobp_check(mp, bp);
198 * Mark the buffer as an inode buffer now that it looks good
200 XFS_BUF_SET_VTYPE(bp, B_FS_INO);
207 * This routine is called to map an inode number within a file
208 * system to the buffer containing the on-disk version of the
209 * inode. It returns a pointer to the buffer containing the
210 * on-disk inode in the bpp parameter, and in the dip parameter
211 * it returns a pointer to the on-disk inode within that buffer.
213 * If a non-zero error is returned, then the contents of bpp and
214 * dipp are undefined.
216 * Use xfs_imap() to determine the size and location of the
217 * buffer to read from disk.
229 struct xfs_imap imap;
234 error = xfs_imap(mp, tp, ino, &imap, imap_flags);
238 error = xfs_imap_to_bp(mp, tp, &imap, &bp, XBF_LOCK, imap_flags);
242 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
244 *offset = imap.im_boffset;
250 * This routine is called to map an inode to the buffer containing
251 * the on-disk version of the inode. It returns a pointer to the
252 * buffer containing the on-disk inode in the bpp parameter, and in
253 * the dip parameter it returns a pointer to the on-disk inode within
256 * If a non-zero error is returned, then the contents of bpp and
257 * dipp are undefined.
259 * The inode is expected to already been mapped to its buffer and read
260 * in once, thus we can use the mapping information stored in the inode
261 * rather than calling xfs_imap(). This allows us to avoid the overhead
262 * of looking at the inode btree for small block file systems
277 ASSERT(ip->i_imap.im_blkno != 0);
279 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp, buf_flags, 0);
284 ASSERT(buf_flags & XBF_TRYLOCK);
290 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
296 * Move inode type and inode format specific information from the
297 * on-disk inode to the in-core inode. For fifos, devs, and sockets
298 * this means set if_rdev to the proper value. For files, directories,
299 * and symlinks this means to bring in the in-line data or extent
300 * pointers. For a file in B-tree format, only the root is immediately
301 * brought in-core. The rest will be in-lined in if_extents when it
302 * is first referenced (see xfs_iread_extents()).
309 xfs_attr_shortform_t *atp;
313 ip->i_df.if_ext_max =
314 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
317 if (unlikely(be32_to_cpu(dip->di_nextents) +
318 be16_to_cpu(dip->di_anextents) >
319 be64_to_cpu(dip->di_nblocks))) {
320 xfs_warn(ip->i_mount,
321 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
322 (unsigned long long)ip->i_ino,
323 (int)(be32_to_cpu(dip->di_nextents) +
324 be16_to_cpu(dip->di_anextents)),
326 be64_to_cpu(dip->di_nblocks));
327 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
329 return XFS_ERROR(EFSCORRUPTED);
332 if (unlikely(dip->di_forkoff > ip->i_mount->m_sb.sb_inodesize)) {
333 xfs_warn(ip->i_mount, "corrupt dinode %Lu, forkoff = 0x%x.",
334 (unsigned long long)ip->i_ino,
336 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
338 return XFS_ERROR(EFSCORRUPTED);
341 if (unlikely((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) &&
342 !ip->i_mount->m_rtdev_targp)) {
343 xfs_warn(ip->i_mount,
344 "corrupt dinode %Lu, has realtime flag set.",
346 XFS_CORRUPTION_ERROR("xfs_iformat(realtime)",
347 XFS_ERRLEVEL_LOW, ip->i_mount, dip);
348 return XFS_ERROR(EFSCORRUPTED);
351 switch (ip->i_d.di_mode & S_IFMT) {
356 if (unlikely(dip->di_format != XFS_DINODE_FMT_DEV)) {
357 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
359 return XFS_ERROR(EFSCORRUPTED);
363 ip->i_df.if_u2.if_rdev = xfs_dinode_get_rdev(dip);
369 switch (dip->di_format) {
370 case XFS_DINODE_FMT_LOCAL:
372 * no local regular files yet
374 if (unlikely((be16_to_cpu(dip->di_mode) & S_IFMT) == S_IFREG)) {
375 xfs_warn(ip->i_mount,
376 "corrupt inode %Lu (local format for regular file).",
377 (unsigned long long) ip->i_ino);
378 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
381 return XFS_ERROR(EFSCORRUPTED);
384 di_size = be64_to_cpu(dip->di_size);
385 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
386 xfs_warn(ip->i_mount,
387 "corrupt inode %Lu (bad size %Ld for local inode).",
388 (unsigned long long) ip->i_ino,
389 (long long) di_size);
390 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
393 return XFS_ERROR(EFSCORRUPTED);
397 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
399 case XFS_DINODE_FMT_EXTENTS:
400 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
402 case XFS_DINODE_FMT_BTREE:
403 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
406 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
408 return XFS_ERROR(EFSCORRUPTED);
413 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
414 return XFS_ERROR(EFSCORRUPTED);
419 if (!XFS_DFORK_Q(dip))
421 ASSERT(ip->i_afp == NULL);
422 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP | KM_NOFS);
423 ip->i_afp->if_ext_max =
424 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
425 switch (dip->di_aformat) {
426 case XFS_DINODE_FMT_LOCAL:
427 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
428 size = be16_to_cpu(atp->hdr.totsize);
430 if (unlikely(size < sizeof(struct xfs_attr_sf_hdr))) {
431 xfs_warn(ip->i_mount,
432 "corrupt inode %Lu (bad attr fork size %Ld).",
433 (unsigned long long) ip->i_ino,
435 XFS_CORRUPTION_ERROR("xfs_iformat(8)",
438 return XFS_ERROR(EFSCORRUPTED);
441 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
443 case XFS_DINODE_FMT_EXTENTS:
444 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
446 case XFS_DINODE_FMT_BTREE:
447 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
450 error = XFS_ERROR(EFSCORRUPTED);
454 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
456 xfs_idestroy_fork(ip, XFS_DATA_FORK);
462 * The file is in-lined in the on-disk inode.
463 * If it fits into if_inline_data, then copy
464 * it there, otherwise allocate a buffer for it
465 * and copy the data there. Either way, set
466 * if_data to point at the data.
467 * If we allocate a buffer for the data, make
468 * sure that its size is a multiple of 4 and
469 * record the real size in i_real_bytes.
482 * If the size is unreasonable, then something
483 * is wrong and we just bail out rather than crash in
484 * kmem_alloc() or memcpy() below.
486 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
487 xfs_warn(ip->i_mount,
488 "corrupt inode %Lu (bad size %d for local fork, size = %d).",
489 (unsigned long long) ip->i_ino, size,
490 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
491 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
493 return XFS_ERROR(EFSCORRUPTED);
495 ifp = XFS_IFORK_PTR(ip, whichfork);
498 ifp->if_u1.if_data = NULL;
499 else if (size <= sizeof(ifp->if_u2.if_inline_data))
500 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
502 real_size = roundup(size, 4);
503 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP | KM_NOFS);
505 ifp->if_bytes = size;
506 ifp->if_real_bytes = real_size;
508 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
509 ifp->if_flags &= ~XFS_IFEXTENTS;
510 ifp->if_flags |= XFS_IFINLINE;
515 * The file consists of a set of extents all
516 * of which fit into the on-disk inode.
517 * If there are few enough extents to fit into
518 * the if_inline_ext, then copy them there.
519 * Otherwise allocate a buffer for them and copy
520 * them into it. Either way, set if_extents
521 * to point at the extents.
535 ifp = XFS_IFORK_PTR(ip, whichfork);
536 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
537 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
540 * If the number of extents is unreasonable, then something
541 * is wrong and we just bail out rather than crash in
542 * kmem_alloc() or memcpy() below.
544 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
545 xfs_warn(ip->i_mount, "corrupt inode %Lu ((a)extents = %d).",
546 (unsigned long long) ip->i_ino, nex);
547 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
549 return XFS_ERROR(EFSCORRUPTED);
552 ifp->if_real_bytes = 0;
554 ifp->if_u1.if_extents = NULL;
555 else if (nex <= XFS_INLINE_EXTS)
556 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
558 xfs_iext_add(ifp, 0, nex);
560 ifp->if_bytes = size;
562 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
563 xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip));
564 for (i = 0; i < nex; i++, dp++) {
565 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
566 ep->l0 = get_unaligned_be64(&dp->l0);
567 ep->l1 = get_unaligned_be64(&dp->l1);
569 XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork);
570 if (whichfork != XFS_DATA_FORK ||
571 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
572 if (unlikely(xfs_check_nostate_extents(
574 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
577 return XFS_ERROR(EFSCORRUPTED);
580 ifp->if_flags |= XFS_IFEXTENTS;
585 * The file has too many extents to fit into
586 * the inode, so they are in B-tree format.
587 * Allocate a buffer for the root of the B-tree
588 * and copy the root into it. The i_extents
589 * field will remain NULL until all of the
590 * extents are read in (when they are needed).
598 xfs_bmdr_block_t *dfp;
604 ifp = XFS_IFORK_PTR(ip, whichfork);
605 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
606 size = XFS_BMAP_BROOT_SPACE(dfp);
607 nrecs = be16_to_cpu(dfp->bb_numrecs);
610 * blow out if -- fork has less extents than can fit in
611 * fork (fork shouldn't be a btree format), root btree
612 * block has more records than can fit into the fork,
613 * or the number of extents is greater than the number of
616 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
617 || XFS_BMDR_SPACE_CALC(nrecs) >
618 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
619 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
620 xfs_warn(ip->i_mount, "corrupt inode %Lu (btree).",
621 (unsigned long long) ip->i_ino);
622 XFS_CORRUPTION_ERROR("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
624 return XFS_ERROR(EFSCORRUPTED);
627 ifp->if_broot_bytes = size;
628 ifp->if_broot = kmem_alloc(size, KM_SLEEP | KM_NOFS);
629 ASSERT(ifp->if_broot != NULL);
631 * Copy and convert from the on-disk structure
632 * to the in-memory structure.
634 xfs_bmdr_to_bmbt(ip->i_mount, dfp,
635 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
636 ifp->if_broot, size);
637 ifp->if_flags &= ~XFS_IFEXTENTS;
638 ifp->if_flags |= XFS_IFBROOT;
644 xfs_dinode_from_disk(
648 to->di_magic = be16_to_cpu(from->di_magic);
649 to->di_mode = be16_to_cpu(from->di_mode);
650 to->di_version = from ->di_version;
651 to->di_format = from->di_format;
652 to->di_onlink = be16_to_cpu(from->di_onlink);
653 to->di_uid = be32_to_cpu(from->di_uid);
654 to->di_gid = be32_to_cpu(from->di_gid);
655 to->di_nlink = be32_to_cpu(from->di_nlink);
656 to->di_projid_lo = be16_to_cpu(from->di_projid_lo);
657 to->di_projid_hi = be16_to_cpu(from->di_projid_hi);
658 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
659 to->di_flushiter = be16_to_cpu(from->di_flushiter);
660 to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec);
661 to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec);
662 to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec);
663 to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec);
664 to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec);
665 to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec);
666 to->di_size = be64_to_cpu(from->di_size);
667 to->di_nblocks = be64_to_cpu(from->di_nblocks);
668 to->di_extsize = be32_to_cpu(from->di_extsize);
669 to->di_nextents = be32_to_cpu(from->di_nextents);
670 to->di_anextents = be16_to_cpu(from->di_anextents);
671 to->di_forkoff = from->di_forkoff;
672 to->di_aformat = from->di_aformat;
673 to->di_dmevmask = be32_to_cpu(from->di_dmevmask);
674 to->di_dmstate = be16_to_cpu(from->di_dmstate);
675 to->di_flags = be16_to_cpu(from->di_flags);
676 to->di_gen = be32_to_cpu(from->di_gen);
682 xfs_icdinode_t *from)
684 to->di_magic = cpu_to_be16(from->di_magic);
685 to->di_mode = cpu_to_be16(from->di_mode);
686 to->di_version = from ->di_version;
687 to->di_format = from->di_format;
688 to->di_onlink = cpu_to_be16(from->di_onlink);
689 to->di_uid = cpu_to_be32(from->di_uid);
690 to->di_gid = cpu_to_be32(from->di_gid);
691 to->di_nlink = cpu_to_be32(from->di_nlink);
692 to->di_projid_lo = cpu_to_be16(from->di_projid_lo);
693 to->di_projid_hi = cpu_to_be16(from->di_projid_hi);
694 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
695 to->di_flushiter = cpu_to_be16(from->di_flushiter);
696 to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec);
697 to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec);
698 to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec);
699 to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec);
700 to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec);
701 to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec);
702 to->di_size = cpu_to_be64(from->di_size);
703 to->di_nblocks = cpu_to_be64(from->di_nblocks);
704 to->di_extsize = cpu_to_be32(from->di_extsize);
705 to->di_nextents = cpu_to_be32(from->di_nextents);
706 to->di_anextents = cpu_to_be16(from->di_anextents);
707 to->di_forkoff = from->di_forkoff;
708 to->di_aformat = from->di_aformat;
709 to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
710 to->di_dmstate = cpu_to_be16(from->di_dmstate);
711 to->di_flags = cpu_to_be16(from->di_flags);
712 to->di_gen = cpu_to_be32(from->di_gen);
721 if (di_flags & XFS_DIFLAG_ANY) {
722 if (di_flags & XFS_DIFLAG_REALTIME)
723 flags |= XFS_XFLAG_REALTIME;
724 if (di_flags & XFS_DIFLAG_PREALLOC)
725 flags |= XFS_XFLAG_PREALLOC;
726 if (di_flags & XFS_DIFLAG_IMMUTABLE)
727 flags |= XFS_XFLAG_IMMUTABLE;
728 if (di_flags & XFS_DIFLAG_APPEND)
729 flags |= XFS_XFLAG_APPEND;
730 if (di_flags & XFS_DIFLAG_SYNC)
731 flags |= XFS_XFLAG_SYNC;
732 if (di_flags & XFS_DIFLAG_NOATIME)
733 flags |= XFS_XFLAG_NOATIME;
734 if (di_flags & XFS_DIFLAG_NODUMP)
735 flags |= XFS_XFLAG_NODUMP;
736 if (di_flags & XFS_DIFLAG_RTINHERIT)
737 flags |= XFS_XFLAG_RTINHERIT;
738 if (di_flags & XFS_DIFLAG_PROJINHERIT)
739 flags |= XFS_XFLAG_PROJINHERIT;
740 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
741 flags |= XFS_XFLAG_NOSYMLINKS;
742 if (di_flags & XFS_DIFLAG_EXTSIZE)
743 flags |= XFS_XFLAG_EXTSIZE;
744 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
745 flags |= XFS_XFLAG_EXTSZINHERIT;
746 if (di_flags & XFS_DIFLAG_NODEFRAG)
747 flags |= XFS_XFLAG_NODEFRAG;
748 if (di_flags & XFS_DIFLAG_FILESTREAM)
749 flags |= XFS_XFLAG_FILESTREAM;
759 xfs_icdinode_t *dic = &ip->i_d;
761 return _xfs_dic2xflags(dic->di_flags) |
762 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
769 return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) |
770 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
774 * Read the disk inode attributes into the in-core inode structure.
788 * Fill in the location information in the in-core inode.
790 error = xfs_imap(mp, tp, ip->i_ino, &ip->i_imap, iget_flags);
795 * Get pointers to the on-disk inode and the buffer containing it.
797 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp,
798 XBF_LOCK, iget_flags);
801 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
804 * If we got something that isn't an inode it means someone
805 * (nfs or dmi) has a stale handle.
807 if (be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC) {
809 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
810 "dip->di_magic (0x%x) != "
811 "XFS_DINODE_MAGIC (0x%x)",
812 be16_to_cpu(dip->di_magic),
815 error = XFS_ERROR(EINVAL);
820 * If the on-disk inode is already linked to a directory
821 * entry, copy all of the inode into the in-core inode.
822 * xfs_iformat() handles copying in the inode format
823 * specific information.
824 * Otherwise, just get the truly permanent information.
827 xfs_dinode_from_disk(&ip->i_d, dip);
828 error = xfs_iformat(ip, dip);
831 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
832 "xfs_iformat() returned error %d",
838 ip->i_d.di_magic = be16_to_cpu(dip->di_magic);
839 ip->i_d.di_version = dip->di_version;
840 ip->i_d.di_gen = be32_to_cpu(dip->di_gen);
841 ip->i_d.di_flushiter = be16_to_cpu(dip->di_flushiter);
843 * Make sure to pull in the mode here as well in
844 * case the inode is released without being used.
845 * This ensures that xfs_inactive() will see that
846 * the inode is already free and not try to mess
847 * with the uninitialized part of it.
851 * Initialize the per-fork minima and maxima for a new
852 * inode here. xfs_iformat will do it for old inodes.
854 ip->i_df.if_ext_max =
855 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
859 * The inode format changed when we moved the link count and
860 * made it 32 bits long. If this is an old format inode,
861 * convert it in memory to look like a new one. If it gets
862 * flushed to disk we will convert back before flushing or
863 * logging it. We zero out the new projid field and the old link
864 * count field. We'll handle clearing the pad field (the remains
865 * of the old uuid field) when we actually convert the inode to
866 * the new format. We don't change the version number so that we
867 * can distinguish this from a real new format inode.
869 if (ip->i_d.di_version == 1) {
870 ip->i_d.di_nlink = ip->i_d.di_onlink;
871 ip->i_d.di_onlink = 0;
872 xfs_set_projid(ip, 0);
875 ip->i_delayed_blks = 0;
876 ip->i_size = ip->i_d.di_size;
879 * Mark the buffer containing the inode as something to keep
880 * around for a while. This helps to keep recently accessed
881 * meta-data in-core longer.
883 xfs_buf_set_ref(bp, XFS_INO_REF);
886 * Use xfs_trans_brelse() to release the buffer containing the
887 * on-disk inode, because it was acquired with xfs_trans_read_buf()
888 * in xfs_itobp() above. If tp is NULL, this is just a normal
889 * brelse(). If we're within a transaction, then xfs_trans_brelse()
890 * will only release the buffer if it is not dirty within the
891 * transaction. It will be OK to release the buffer in this case,
892 * because inodes on disk are never destroyed and we will be
893 * locking the new in-core inode before putting it in the hash
894 * table where other processes can find it. Thus we don't have
895 * to worry about the inode being changed just because we released
899 xfs_trans_brelse(tp, bp);
904 * Read in extents from a btree-format inode.
905 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
915 xfs_extnum_t nextents;
917 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
918 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
920 return XFS_ERROR(EFSCORRUPTED);
922 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
923 ifp = XFS_IFORK_PTR(ip, whichfork);
926 * We know that the size is valid (it's checked in iformat_btree)
928 ifp->if_lastex = NULLEXTNUM;
929 ifp->if_bytes = ifp->if_real_bytes = 0;
930 ifp->if_flags |= XFS_IFEXTENTS;
931 xfs_iext_add(ifp, 0, nextents);
932 error = xfs_bmap_read_extents(tp, ip, whichfork);
934 xfs_iext_destroy(ifp);
935 ifp->if_flags &= ~XFS_IFEXTENTS;
938 xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip));
943 * Allocate an inode on disk and return a copy of its in-core version.
944 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
945 * appropriately within the inode. The uid and gid for the inode are
946 * set according to the contents of the given cred structure.
948 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
949 * has a free inode available, call xfs_iget()
950 * to obtain the in-core version of the allocated inode. Finally,
951 * fill in the inode and log its initial contents. In this case,
952 * ialloc_context would be set to NULL and call_again set to false.
954 * If xfs_dialloc() does not have an available inode,
955 * it will replenish its supply by doing an allocation. Since we can
956 * only do one allocation within a transaction without deadlocks, we
957 * must commit the current transaction before returning the inode itself.
958 * In this case, therefore, we will set call_again to true and return.
959 * The caller should then commit the current transaction, start a new
960 * transaction, and call xfs_ialloc() again to actually get the inode.
962 * To ensure that some other process does not grab the inode that
963 * was allocated during the first call to xfs_ialloc(), this routine
964 * also returns the [locked] bp pointing to the head of the freelist
965 * as ialloc_context. The caller should hold this buffer across
966 * the commit and pass it back into this routine on the second call.
968 * If we are allocating quota inodes, we do not have a parent inode
969 * to attach to or associate with (i.e. pip == NULL) because they
970 * are not linked into the directory structure - they are attached
971 * directly to the superblock - and so have no parent.
982 xfs_buf_t **ialloc_context,
983 boolean_t *call_again,
994 * Call the space management code to pick
995 * the on-disk inode to be allocated.
997 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
998 ialloc_context, call_again, &ino);
1001 if (*call_again || ino == NULLFSINO) {
1005 ASSERT(*ialloc_context == NULL);
1008 * Get the in-core inode with the lock held exclusively.
1009 * This is because we're setting fields here we need
1010 * to prevent others from looking at until we're done.
1012 error = xfs_iget(tp->t_mountp, tp, ino, XFS_IGET_CREATE,
1013 XFS_ILOCK_EXCL, &ip);
1018 ip->i_d.di_mode = (__uint16_t)mode;
1019 ip->i_d.di_onlink = 0;
1020 ip->i_d.di_nlink = nlink;
1021 ASSERT(ip->i_d.di_nlink == nlink);
1022 ip->i_d.di_uid = current_fsuid();
1023 ip->i_d.di_gid = current_fsgid();
1024 xfs_set_projid(ip, prid);
1025 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1028 * If the superblock version is up to where we support new format
1029 * inodes and this is currently an old format inode, then change
1030 * the inode version number now. This way we only do the conversion
1031 * here rather than here and in the flush/logging code.
1033 if (xfs_sb_version_hasnlink(&tp->t_mountp->m_sb) &&
1034 ip->i_d.di_version == 1) {
1035 ip->i_d.di_version = 2;
1037 * We've already zeroed the old link count, the projid field,
1038 * and the pad field.
1043 * Project ids won't be stored on disk if we are using a version 1 inode.
1045 if ((prid != 0) && (ip->i_d.di_version == 1))
1046 xfs_bump_ino_vers2(tp, ip);
1048 if (pip && XFS_INHERIT_GID(pip)) {
1049 ip->i_d.di_gid = pip->i_d.di_gid;
1050 if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1051 ip->i_d.di_mode |= S_ISGID;
1056 * If the group ID of the new file does not match the effective group
1057 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1058 * (and only if the irix_sgid_inherit compatibility variable is set).
1060 if ((irix_sgid_inherit) &&
1061 (ip->i_d.di_mode & S_ISGID) &&
1062 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1063 ip->i_d.di_mode &= ~S_ISGID;
1066 ip->i_d.di_size = 0;
1068 ip->i_d.di_nextents = 0;
1069 ASSERT(ip->i_d.di_nblocks == 0);
1072 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
1073 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
1074 ip->i_d.di_atime = ip->i_d.di_mtime;
1075 ip->i_d.di_ctime = ip->i_d.di_mtime;
1078 * di_gen will have been taken care of in xfs_iread.
1080 ip->i_d.di_extsize = 0;
1081 ip->i_d.di_dmevmask = 0;
1082 ip->i_d.di_dmstate = 0;
1083 ip->i_d.di_flags = 0;
1084 flags = XFS_ILOG_CORE;
1085 switch (mode & S_IFMT) {
1090 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1091 ip->i_df.if_u2.if_rdev = rdev;
1092 ip->i_df.if_flags = 0;
1093 flags |= XFS_ILOG_DEV;
1097 * we can't set up filestreams until after the VFS inode
1098 * is set up properly.
1100 if (pip && xfs_inode_is_filestream(pip))
1104 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
1107 if ((mode & S_IFMT) == S_IFDIR) {
1108 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1109 di_flags |= XFS_DIFLAG_RTINHERIT;
1110 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1111 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1112 ip->i_d.di_extsize = pip->i_d.di_extsize;
1114 } else if ((mode & S_IFMT) == S_IFREG) {
1115 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1116 di_flags |= XFS_DIFLAG_REALTIME;
1117 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1118 di_flags |= XFS_DIFLAG_EXTSIZE;
1119 ip->i_d.di_extsize = pip->i_d.di_extsize;
1122 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1123 xfs_inherit_noatime)
1124 di_flags |= XFS_DIFLAG_NOATIME;
1125 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1127 di_flags |= XFS_DIFLAG_NODUMP;
1128 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1130 di_flags |= XFS_DIFLAG_SYNC;
1131 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1132 xfs_inherit_nosymlinks)
1133 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1134 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1135 di_flags |= XFS_DIFLAG_PROJINHERIT;
1136 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1137 xfs_inherit_nodefrag)
1138 di_flags |= XFS_DIFLAG_NODEFRAG;
1139 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
1140 di_flags |= XFS_DIFLAG_FILESTREAM;
1141 ip->i_d.di_flags |= di_flags;
1145 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1146 ip->i_df.if_flags = XFS_IFEXTENTS;
1147 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1148 ip->i_df.if_u1.if_extents = NULL;
1154 * Attribute fork settings for new inode.
1156 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1157 ip->i_d.di_anextents = 0;
1160 * Log the new values stuffed into the inode.
1162 xfs_trans_ijoin_ref(tp, ip, XFS_ILOCK_EXCL);
1163 xfs_trans_log_inode(tp, ip, flags);
1165 /* now that we have an i_mode we can setup inode ops and unlock */
1166 xfs_setup_inode(ip);
1168 /* now we have set up the vfs inode we can associate the filestream */
1170 error = xfs_filestream_associate(pip, ip);
1174 xfs_iflags_set(ip, XFS_IFILESTREAM);
1182 * Check to make sure that there are no blocks allocated to the
1183 * file beyond the size of the file. We don't check this for
1184 * files with fixed size extents or real time extents, but we
1185 * at least do it for regular files.
1194 xfs_fileoff_t map_first;
1196 xfs_bmbt_irec_t imaps[2];
1198 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1201 if (XFS_IS_REALTIME_INODE(ip))
1204 if (ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE)
1208 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1210 * The filesystem could be shutting down, so bmapi may return
1213 if (xfs_bmapi(NULL, ip, map_first,
1215 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1217 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1220 ASSERT(nimaps == 1);
1221 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1226 * Calculate the last possible buffered byte in a file. This must
1227 * include data that was buffered beyond the EOF by the write code.
1228 * This also needs to deal with overflowing the xfs_fsize_t type
1229 * which can happen for sizes near the limit.
1231 * We also need to take into account any blocks beyond the EOF. It
1232 * may be the case that they were buffered by a write which failed.
1233 * In that case the pages will still be in memory, but the inode size
1234 * will never have been updated.
1241 xfs_fsize_t last_byte;
1242 xfs_fileoff_t last_block;
1243 xfs_fileoff_t size_last_block;
1246 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED));
1250 * Only check for blocks beyond the EOF if the extents have
1251 * been read in. This eliminates the need for the inode lock,
1252 * and it also saves us from looking when it really isn't
1255 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1256 xfs_ilock(ip, XFS_ILOCK_SHARED);
1257 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1259 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1266 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_size);
1267 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1269 last_byte = XFS_FSB_TO_B(mp, last_block);
1270 if (last_byte < 0) {
1271 return XFS_MAXIOFFSET(mp);
1273 last_byte += (1 << mp->m_writeio_log);
1274 if (last_byte < 0) {
1275 return XFS_MAXIOFFSET(mp);
1281 * Start the truncation of the file to new_size. The new size
1282 * must be smaller than the current size. This routine will
1283 * clear the buffer and page caches of file data in the removed
1284 * range, and xfs_itruncate_finish() will remove the underlying
1287 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1288 * must NOT have the inode lock held at all. This is because we're
1289 * calling into the buffer/page cache code and we can't hold the
1290 * inode lock when we do so.
1292 * We need to wait for any direct I/Os in flight to complete before we
1293 * proceed with the truncate. This is needed to prevent the extents
1294 * being read or written by the direct I/Os from being removed while the
1295 * I/O is in flight as there is no other method of synchronising
1296 * direct I/O with the truncate operation. Also, because we hold
1297 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1298 * started until the truncate completes and drops the lock. Essentially,
1299 * the xfs_ioend_wait() call forms an I/O barrier that provides strict
1300 * ordering between direct I/Os and the truncate operation.
1302 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1303 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1304 * in the case that the caller is locking things out of order and
1305 * may not be able to call xfs_itruncate_finish() with the inode lock
1306 * held without dropping the I/O lock. If the caller must drop the
1307 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1308 * must be called again with all the same restrictions as the initial
1312 xfs_itruncate_start(
1315 xfs_fsize_t new_size)
1317 xfs_fsize_t last_byte;
1318 xfs_off_t toss_start;
1322 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1323 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1324 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1325 (flags == XFS_ITRUNC_MAYBE));
1329 /* wait for the completion of any pending DIOs */
1330 if (new_size == 0 || new_size < ip->i_size)
1334 * Call toss_pages or flushinval_pages to get rid of pages
1335 * overlapping the region being removed. We have to use
1336 * the less efficient flushinval_pages in the case that the
1337 * caller may not be able to finish the truncate without
1338 * dropping the inode's I/O lock. Make sure
1339 * to catch any pages brought in by buffers overlapping
1340 * the EOF by searching out beyond the isize by our
1341 * block size. We round new_size up to a block boundary
1342 * so that we don't toss things on the same block as
1343 * new_size but before it.
1345 * Before calling toss_page or flushinval_pages, make sure to
1346 * call remapf() over the same region if the file is mapped.
1347 * This frees up mapped file references to the pages in the
1348 * given range and for the flushinval_pages case it ensures
1349 * that we get the latest mapped changes flushed out.
1351 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1352 toss_start = XFS_FSB_TO_B(mp, toss_start);
1353 if (toss_start < 0) {
1355 * The place to start tossing is beyond our maximum
1356 * file size, so there is no way that the data extended
1361 last_byte = xfs_file_last_byte(ip);
1362 trace_xfs_itruncate_start(ip, flags, new_size, toss_start, last_byte);
1363 if (last_byte > toss_start) {
1364 if (flags & XFS_ITRUNC_DEFINITE) {
1365 xfs_tosspages(ip, toss_start,
1366 -1, FI_REMAPF_LOCKED);
1368 error = xfs_flushinval_pages(ip, toss_start,
1369 -1, FI_REMAPF_LOCKED);
1374 if (new_size == 0) {
1375 ASSERT(VN_CACHED(VFS_I(ip)) == 0);
1382 * Shrink the file to the given new_size. The new size must be smaller than
1383 * the current size. This will free up the underlying blocks in the removed
1384 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
1386 * The transaction passed to this routine must have made a permanent log
1387 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1388 * given transaction and start new ones, so make sure everything involved in
1389 * the transaction is tidy before calling here. Some transaction will be
1390 * returned to the caller to be committed. The incoming transaction must
1391 * already include the inode, and both inode locks must be held exclusively.
1392 * The inode must also be "held" within the transaction. On return the inode
1393 * will be "held" within the returned transaction. This routine does NOT
1394 * require any disk space to be reserved for it within the transaction.
1396 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1397 * indicates the fork which is to be truncated. For the attribute fork we only
1398 * support truncation to size 0.
1400 * We use the sync parameter to indicate whether or not the first transaction
1401 * we perform might have to be synchronous. For the attr fork, it needs to be
1402 * so if the unlink of the inode is not yet known to be permanent in the log.
1403 * This keeps us from freeing and reusing the blocks of the attribute fork
1404 * before the unlink of the inode becomes permanent.
1406 * For the data fork, we normally have to run synchronously if we're being
1407 * called out of the inactive path or we're being called out of the create path
1408 * where we're truncating an existing file. Either way, the truncate needs to
1409 * be sync so blocks don't reappear in the file with altered data in case of a
1410 * crash. wsync filesystems can run the first case async because anything that
1411 * shrinks the inode has to run sync so by the time we're called here from
1412 * inactive, the inode size is permanently set to 0.
1414 * Calls from the truncate path always need to be sync unless we're in a wsync
1415 * filesystem and the file has already been unlinked.
1417 * The caller is responsible for correctly setting the sync parameter. It gets
1418 * too hard for us to guess here which path we're being called out of just
1419 * based on inode state.
1421 * If we get an error, we must return with the inode locked and linked into the
1422 * current transaction. This keeps things simple for the higher level code,
1423 * because it always knows that the inode is locked and held in the transaction
1424 * that returns to it whether errors occur or not. We don't mark the inode
1425 * dirty on error so that transactions can be easily aborted if possible.
1428 xfs_itruncate_finish(
1431 xfs_fsize_t new_size,
1435 xfs_fsblock_t first_block;
1436 xfs_fileoff_t first_unmap_block;
1437 xfs_fileoff_t last_block;
1438 xfs_filblks_t unmap_len=0;
1443 xfs_bmap_free_t free_list;
1446 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
1447 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1448 ASSERT(*tp != NULL);
1449 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1450 ASSERT(ip->i_transp == *tp);
1451 ASSERT(ip->i_itemp != NULL);
1452 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1456 mp = (ntp)->t_mountp;
1457 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1460 * We only support truncating the entire attribute fork.
1462 if (fork == XFS_ATTR_FORK) {
1465 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1466 trace_xfs_itruncate_finish_start(ip, new_size);
1469 * The first thing we do is set the size to new_size permanently
1470 * on disk. This way we don't have to worry about anyone ever
1471 * being able to look at the data being freed even in the face
1472 * of a crash. What we're getting around here is the case where
1473 * we free a block, it is allocated to another file, it is written
1474 * to, and then we crash. If the new data gets written to the
1475 * file but the log buffers containing the free and reallocation
1476 * don't, then we'd end up with garbage in the blocks being freed.
1477 * As long as we make the new_size permanent before actually
1478 * freeing any blocks it doesn't matter if they get writtten to.
1480 * The callers must signal into us whether or not the size
1481 * setting here must be synchronous. There are a few cases
1482 * where it doesn't have to be synchronous. Those cases
1483 * occur if the file is unlinked and we know the unlink is
1484 * permanent or if the blocks being truncated are guaranteed
1485 * to be beyond the inode eof (regardless of the link count)
1486 * and the eof value is permanent. Both of these cases occur
1487 * only on wsync-mounted filesystems. In those cases, we're
1488 * guaranteed that no user will ever see the data in the blocks
1489 * that are being truncated so the truncate can run async.
1490 * In the free beyond eof case, the file may wind up with
1491 * more blocks allocated to it than it needs if we crash
1492 * and that won't get fixed until the next time the file
1493 * is re-opened and closed but that's ok as that shouldn't
1494 * be too many blocks.
1496 * However, we can't just make all wsync xactions run async
1497 * because there's one call out of the create path that needs
1498 * to run sync where it's truncating an existing file to size
1499 * 0 whose size is > 0.
1501 * It's probably possible to come up with a test in this
1502 * routine that would correctly distinguish all the above
1503 * cases from the values of the function parameters and the
1504 * inode state but for sanity's sake, I've decided to let the
1505 * layers above just tell us. It's simpler to correctly figure
1506 * out in the layer above exactly under what conditions we
1507 * can run async and I think it's easier for others read and
1508 * follow the logic in case something has to be changed.
1509 * cscope is your friend -- rcc.
1511 * The attribute fork is much simpler.
1513 * For the attribute fork we allow the caller to tell us whether
1514 * the unlink of the inode that led to this call is yet permanent
1515 * in the on disk log. If it is not and we will be freeing extents
1516 * in this inode then we make the first transaction synchronous
1517 * to make sure that the unlink is permanent by the time we free
1520 if (fork == XFS_DATA_FORK) {
1521 if (ip->i_d.di_nextents > 0) {
1523 * If we are not changing the file size then do
1524 * not update the on-disk file size - we may be
1525 * called from xfs_inactive_free_eofblocks(). If we
1526 * update the on-disk file size and then the system
1527 * crashes before the contents of the file are
1528 * flushed to disk then the files may be full of
1529 * holes (ie NULL files bug).
1531 if (ip->i_size != new_size) {
1532 ip->i_d.di_size = new_size;
1533 ip->i_size = new_size;
1534 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1538 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1539 if (ip->i_d.di_anextents > 0)
1540 xfs_trans_set_sync(ntp);
1542 ASSERT(fork == XFS_DATA_FORK ||
1543 (fork == XFS_ATTR_FORK &&
1544 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1545 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1548 * Since it is possible for space to become allocated beyond
1549 * the end of the file (in a crash where the space is allocated
1550 * but the inode size is not yet updated), simply remove any
1551 * blocks which show up between the new EOF and the maximum
1552 * possible file size. If the first block to be removed is
1553 * beyond the maximum file size (ie it is the same as last_block),
1554 * then there is nothing to do.
1556 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1557 ASSERT(first_unmap_block <= last_block);
1559 if (last_block == first_unmap_block) {
1562 unmap_len = last_block - first_unmap_block + 1;
1566 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1567 * will tell us whether it freed the entire range or
1568 * not. If this is a synchronous mount (wsync),
1569 * then we can tell bunmapi to keep all the
1570 * transactions asynchronous since the unlink
1571 * transaction that made this inode inactive has
1572 * already hit the disk. There's no danger of
1573 * the freed blocks being reused, there being a
1574 * crash, and the reused blocks suddenly reappearing
1575 * in this file with garbage in them once recovery
1578 xfs_bmap_init(&free_list, &first_block);
1579 error = xfs_bunmapi(ntp, ip,
1580 first_unmap_block, unmap_len,
1581 xfs_bmapi_aflag(fork),
1582 XFS_ITRUNC_MAX_EXTENTS,
1583 &first_block, &free_list,
1587 * If the bunmapi call encounters an error,
1588 * return to the caller where the transaction
1589 * can be properly aborted. We just need to
1590 * make sure we're not holding any resources
1591 * that we were not when we came in.
1593 xfs_bmap_cancel(&free_list);
1598 * Duplicate the transaction that has the permanent
1599 * reservation and commit the old transaction.
1601 error = xfs_bmap_finish(tp, &free_list, &committed);
1604 xfs_trans_ijoin(ntp, ip);
1608 * If the bmap finish call encounters an error, return
1609 * to the caller where the transaction can be properly
1610 * aborted. We just need to make sure we're not
1611 * holding any resources that we were not when we came
1614 * Aborting from this point might lose some blocks in
1615 * the file system, but oh well.
1617 xfs_bmap_cancel(&free_list);
1623 * Mark the inode dirty so it will be logged and
1624 * moved forward in the log as part of every commit.
1626 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1629 ntp = xfs_trans_dup(ntp);
1630 error = xfs_trans_commit(*tp, 0);
1633 xfs_trans_ijoin(ntp, ip);
1638 * transaction commit worked ok so we can drop the extra ticket
1639 * reference that we gained in xfs_trans_dup()
1641 xfs_log_ticket_put(ntp->t_ticket);
1642 error = xfs_trans_reserve(ntp, 0,
1643 XFS_ITRUNCATE_LOG_RES(mp), 0,
1644 XFS_TRANS_PERM_LOG_RES,
1645 XFS_ITRUNCATE_LOG_COUNT);
1650 * Only update the size in the case of the data fork, but
1651 * always re-log the inode so that our permanent transaction
1652 * can keep on rolling it forward in the log.
1654 if (fork == XFS_DATA_FORK) {
1655 xfs_isize_check(mp, ip, new_size);
1657 * If we are not changing the file size then do
1658 * not update the on-disk file size - we may be
1659 * called from xfs_inactive_free_eofblocks(). If we
1660 * update the on-disk file size and then the system
1661 * crashes before the contents of the file are
1662 * flushed to disk then the files may be full of
1663 * holes (ie NULL files bug).
1665 if (ip->i_size != new_size) {
1666 ip->i_d.di_size = new_size;
1667 ip->i_size = new_size;
1670 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1671 ASSERT((new_size != 0) ||
1672 (fork == XFS_ATTR_FORK) ||
1673 (ip->i_delayed_blks == 0));
1674 ASSERT((new_size != 0) ||
1675 (fork == XFS_ATTR_FORK) ||
1676 (ip->i_d.di_nextents == 0));
1677 trace_xfs_itruncate_finish_end(ip, new_size);
1682 * This is called when the inode's link count goes to 0.
1683 * We place the on-disk inode on a list in the AGI. It
1684 * will be pulled from this list when the inode is freed.
1701 ASSERT(ip->i_d.di_nlink == 0);
1702 ASSERT(ip->i_d.di_mode != 0);
1703 ASSERT(ip->i_transp == tp);
1708 * Get the agi buffer first. It ensures lock ordering
1711 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1714 agi = XFS_BUF_TO_AGI(agibp);
1717 * Get the index into the agi hash table for the
1718 * list this inode will go on.
1720 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1722 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1723 ASSERT(agi->agi_unlinked[bucket_index]);
1724 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1726 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
1728 * There is already another inode in the bucket we need
1729 * to add ourselves to. Add us at the front of the list.
1730 * Here we put the head pointer into our next pointer,
1731 * and then we fall through to point the head at us.
1733 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XBF_LOCK);
1737 ASSERT(be32_to_cpu(dip->di_next_unlinked) == NULLAGINO);
1738 /* both on-disk, don't endian flip twice */
1739 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1740 offset = ip->i_imap.im_boffset +
1741 offsetof(xfs_dinode_t, di_next_unlinked);
1742 xfs_trans_inode_buf(tp, ibp);
1743 xfs_trans_log_buf(tp, ibp, offset,
1744 (offset + sizeof(xfs_agino_t) - 1));
1745 xfs_inobp_check(mp, ibp);
1749 * Point the bucket head pointer at the inode being inserted.
1752 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1753 offset = offsetof(xfs_agi_t, agi_unlinked) +
1754 (sizeof(xfs_agino_t) * bucket_index);
1755 xfs_trans_log_buf(tp, agibp, offset,
1756 (offset + sizeof(xfs_agino_t) - 1));
1761 * Pull the on-disk inode from the AGI unlinked list.
1774 xfs_agnumber_t agno;
1776 xfs_agino_t next_agino;
1777 xfs_buf_t *last_ibp;
1778 xfs_dinode_t *last_dip = NULL;
1780 int offset, last_offset = 0;
1784 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1787 * Get the agi buffer first. It ensures lock ordering
1790 error = xfs_read_agi(mp, tp, agno, &agibp);
1794 agi = XFS_BUF_TO_AGI(agibp);
1797 * Get the index into the agi hash table for the
1798 * list this inode will go on.
1800 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1802 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1803 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
1804 ASSERT(agi->agi_unlinked[bucket_index]);
1806 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
1808 * We're at the head of the list. Get the inode's
1809 * on-disk buffer to see if there is anyone after us
1810 * on the list. Only modify our next pointer if it
1811 * is not already NULLAGINO. This saves us the overhead
1812 * of dealing with the buffer when there is no need to
1815 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XBF_LOCK);
1818 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
1819 error, mp->m_fsname);
1822 next_agino = be32_to_cpu(dip->di_next_unlinked);
1823 ASSERT(next_agino != 0);
1824 if (next_agino != NULLAGINO) {
1825 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
1826 offset = ip->i_imap.im_boffset +
1827 offsetof(xfs_dinode_t, di_next_unlinked);
1828 xfs_trans_inode_buf(tp, ibp);
1829 xfs_trans_log_buf(tp, ibp, offset,
1830 (offset + sizeof(xfs_agino_t) - 1));
1831 xfs_inobp_check(mp, ibp);
1833 xfs_trans_brelse(tp, ibp);
1836 * Point the bucket head pointer at the next inode.
1838 ASSERT(next_agino != 0);
1839 ASSERT(next_agino != agino);
1840 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
1841 offset = offsetof(xfs_agi_t, agi_unlinked) +
1842 (sizeof(xfs_agino_t) * bucket_index);
1843 xfs_trans_log_buf(tp, agibp, offset,
1844 (offset + sizeof(xfs_agino_t) - 1));
1847 * We need to search the list for the inode being freed.
1849 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
1851 while (next_agino != agino) {
1853 * If the last inode wasn't the one pointing to
1854 * us, then release its buffer since we're not
1855 * going to do anything with it.
1857 if (last_ibp != NULL) {
1858 xfs_trans_brelse(tp, last_ibp);
1860 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
1861 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
1862 &last_ibp, &last_offset, 0);
1865 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
1866 error, mp->m_fsname);
1869 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
1870 ASSERT(next_agino != NULLAGINO);
1871 ASSERT(next_agino != 0);
1874 * Now last_ibp points to the buffer previous to us on
1875 * the unlinked list. Pull us from the list.
1877 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XBF_LOCK);
1880 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
1881 error, mp->m_fsname);
1884 next_agino = be32_to_cpu(dip->di_next_unlinked);
1885 ASSERT(next_agino != 0);
1886 ASSERT(next_agino != agino);
1887 if (next_agino != NULLAGINO) {
1888 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
1889 offset = ip->i_imap.im_boffset +
1890 offsetof(xfs_dinode_t, di_next_unlinked);
1891 xfs_trans_inode_buf(tp, ibp);
1892 xfs_trans_log_buf(tp, ibp, offset,
1893 (offset + sizeof(xfs_agino_t) - 1));
1894 xfs_inobp_check(mp, ibp);
1896 xfs_trans_brelse(tp, ibp);
1899 * Point the previous inode on the list to the next inode.
1901 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
1902 ASSERT(next_agino != 0);
1903 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
1904 xfs_trans_inode_buf(tp, last_ibp);
1905 xfs_trans_log_buf(tp, last_ibp, offset,
1906 (offset + sizeof(xfs_agino_t) - 1));
1907 xfs_inobp_check(mp, last_ibp);
1913 * A big issue when freeing the inode cluster is is that we _cannot_ skip any
1914 * inodes that are in memory - they all must be marked stale and attached to
1915 * the cluster buffer.
1919 xfs_inode_t *free_ip,
1923 xfs_mount_t *mp = free_ip->i_mount;
1924 int blks_per_cluster;
1931 xfs_inode_log_item_t *iip;
1932 xfs_log_item_t *lip;
1933 struct xfs_perag *pag;
1935 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
1936 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
1937 blks_per_cluster = 1;
1938 ninodes = mp->m_sb.sb_inopblock;
1939 nbufs = XFS_IALLOC_BLOCKS(mp);
1941 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
1942 mp->m_sb.sb_blocksize;
1943 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
1944 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
1947 for (j = 0; j < nbufs; j++, inum += ninodes) {
1948 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
1949 XFS_INO_TO_AGBNO(mp, inum));
1952 * We obtain and lock the backing buffer first in the process
1953 * here, as we have to ensure that any dirty inode that we
1954 * can't get the flush lock on is attached to the buffer.
1955 * If we scan the in-memory inodes first, then buffer IO can
1956 * complete before we get a lock on it, and hence we may fail
1957 * to mark all the active inodes on the buffer stale.
1959 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
1960 mp->m_bsize * blks_per_cluster,
1964 * Walk the inodes already attached to the buffer and mark them
1965 * stale. These will all have the flush locks held, so an
1966 * in-memory inode walk can't lock them. By marking them all
1967 * stale first, we will not attempt to lock them in the loop
1968 * below as the XFS_ISTALE flag will be set.
1970 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
1972 if (lip->li_type == XFS_LI_INODE) {
1973 iip = (xfs_inode_log_item_t *)lip;
1974 ASSERT(iip->ili_logged == 1);
1975 lip->li_cb = xfs_istale_done;
1976 xfs_trans_ail_copy_lsn(mp->m_ail,
1977 &iip->ili_flush_lsn,
1978 &iip->ili_item.li_lsn);
1979 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
1981 lip = lip->li_bio_list;
1986 * For each inode in memory attempt to add it to the inode
1987 * buffer and set it up for being staled on buffer IO
1988 * completion. This is safe as we've locked out tail pushing
1989 * and flushing by locking the buffer.
1991 * We have already marked every inode that was part of a
1992 * transaction stale above, which means there is no point in
1993 * even trying to lock them.
1995 for (i = 0; i < ninodes; i++) {
1998 ip = radix_tree_lookup(&pag->pag_ici_root,
1999 XFS_INO_TO_AGINO(mp, (inum + i)));
2001 /* Inode not in memory, nothing to do */
2008 * because this is an RCU protected lookup, we could
2009 * find a recently freed or even reallocated inode
2010 * during the lookup. We need to check under the
2011 * i_flags_lock for a valid inode here. Skip it if it
2012 * is not valid, the wrong inode or stale.
2014 spin_lock(&ip->i_flags_lock);
2015 if (ip->i_ino != inum + i ||
2016 __xfs_iflags_test(ip, XFS_ISTALE)) {
2017 spin_unlock(&ip->i_flags_lock);
2021 spin_unlock(&ip->i_flags_lock);
2024 * Don't try to lock/unlock the current inode, but we
2025 * _cannot_ skip the other inodes that we did not find
2026 * in the list attached to the buffer and are not
2027 * already marked stale. If we can't lock it, back off
2030 if (ip != free_ip &&
2031 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2039 xfs_iflags_set(ip, XFS_ISTALE);
2042 * we don't need to attach clean inodes or those only
2043 * with unlogged changes (which we throw away, anyway).
2046 if (!iip || xfs_inode_clean(ip)) {
2047 ASSERT(ip != free_ip);
2048 ip->i_update_core = 0;
2050 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2054 iip->ili_last_fields = iip->ili_format.ilf_fields;
2055 iip->ili_format.ilf_fields = 0;
2056 iip->ili_logged = 1;
2057 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2058 &iip->ili_item.li_lsn);
2060 xfs_buf_attach_iodone(bp, xfs_istale_done,
2064 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2067 xfs_trans_stale_inode_buf(tp, bp);
2068 xfs_trans_binval(tp, bp);
2075 * This is called to return an inode to the inode free list.
2076 * The inode should already be truncated to 0 length and have
2077 * no pages associated with it. This routine also assumes that
2078 * the inode is already a part of the transaction.
2080 * The on-disk copy of the inode will have been added to the list
2081 * of unlinked inodes in the AGI. We need to remove the inode from
2082 * that list atomically with respect to freeing it here.
2088 xfs_bmap_free_t *flist)
2092 xfs_ino_t first_ino;
2096 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2097 ASSERT(ip->i_transp == tp);
2098 ASSERT(ip->i_d.di_nlink == 0);
2099 ASSERT(ip->i_d.di_nextents == 0);
2100 ASSERT(ip->i_d.di_anextents == 0);
2101 ASSERT((ip->i_d.di_size == 0 && ip->i_size == 0) ||
2102 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2103 ASSERT(ip->i_d.di_nblocks == 0);
2106 * Pull the on-disk inode from the AGI unlinked list.
2108 error = xfs_iunlink_remove(tp, ip);
2113 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2117 ip->i_d.di_mode = 0; /* mark incore inode as free */
2118 ip->i_d.di_flags = 0;
2119 ip->i_d.di_dmevmask = 0;
2120 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2121 ip->i_df.if_ext_max =
2122 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2123 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2124 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2126 * Bump the generation count so no one will be confused
2127 * by reincarnations of this inode.
2131 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2133 error = xfs_itobp(ip->i_mount, tp, ip, &dip, &ibp, XBF_LOCK);
2138 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2139 * from picking up this inode when it is reclaimed (its incore state
2140 * initialzed but not flushed to disk yet). The in-core di_mode is
2141 * already cleared and a corresponding transaction logged.
2142 * The hack here just synchronizes the in-core to on-disk
2143 * di_mode value in advance before the actual inode sync to disk.
2144 * This is OK because the inode is already unlinked and would never
2145 * change its di_mode again for this inode generation.
2146 * This is a temporary hack that would require a proper fix
2152 xfs_ifree_cluster(ip, tp, first_ino);
2159 * Reallocate the space for if_broot based on the number of records
2160 * being added or deleted as indicated in rec_diff. Move the records
2161 * and pointers in if_broot to fit the new size. When shrinking this
2162 * will eliminate holes between the records and pointers created by
2163 * the caller. When growing this will create holes to be filled in
2166 * The caller must not request to add more records than would fit in
2167 * the on-disk inode root. If the if_broot is currently NULL, then
2168 * if we adding records one will be allocated. The caller must also
2169 * not request that the number of records go below zero, although
2170 * it can go to zero.
2172 * ip -- the inode whose if_broot area is changing
2173 * ext_diff -- the change in the number of records, positive or negative,
2174 * requested for the if_broot array.
2182 struct xfs_mount *mp = ip->i_mount;
2185 struct xfs_btree_block *new_broot;
2192 * Handle the degenerate case quietly.
2194 if (rec_diff == 0) {
2198 ifp = XFS_IFORK_PTR(ip, whichfork);
2201 * If there wasn't any memory allocated before, just
2202 * allocate it now and get out.
2204 if (ifp->if_broot_bytes == 0) {
2205 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
2206 ifp->if_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
2207 ifp->if_broot_bytes = (int)new_size;
2212 * If there is already an existing if_broot, then we need
2213 * to realloc() it and shift the pointers to their new
2214 * location. The records don't change location because
2215 * they are kept butted up against the btree block header.
2217 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
2218 new_max = cur_max + rec_diff;
2219 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2220 ifp->if_broot = kmem_realloc(ifp->if_broot, new_size,
2221 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2222 KM_SLEEP | KM_NOFS);
2223 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2224 ifp->if_broot_bytes);
2225 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2227 ifp->if_broot_bytes = (int)new_size;
2228 ASSERT(ifp->if_broot_bytes <=
2229 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2230 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2235 * rec_diff is less than 0. In this case, we are shrinking the
2236 * if_broot buffer. It must already exist. If we go to zero
2237 * records, just get rid of the root and clear the status bit.
2239 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
2240 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
2241 new_max = cur_max + rec_diff;
2242 ASSERT(new_max >= 0);
2244 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2248 new_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
2250 * First copy over the btree block header.
2252 memcpy(new_broot, ifp->if_broot, XFS_BTREE_LBLOCK_LEN);
2255 ifp->if_flags &= ~XFS_IFBROOT;
2259 * Only copy the records and pointers if there are any.
2263 * First copy the records.
2265 op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1);
2266 np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1);
2267 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2270 * Then copy the pointers.
2272 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2273 ifp->if_broot_bytes);
2274 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1,
2276 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2278 kmem_free(ifp->if_broot);
2279 ifp->if_broot = new_broot;
2280 ifp->if_broot_bytes = (int)new_size;
2281 ASSERT(ifp->if_broot_bytes <=
2282 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2288 * This is called when the amount of space needed for if_data
2289 * is increased or decreased. The change in size is indicated by
2290 * the number of bytes that need to be added or deleted in the
2291 * byte_diff parameter.
2293 * If the amount of space needed has decreased below the size of the
2294 * inline buffer, then switch to using the inline buffer. Otherwise,
2295 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2296 * to what is needed.
2298 * ip -- the inode whose if_data area is changing
2299 * byte_diff -- the change in the number of bytes, positive or negative,
2300 * requested for the if_data array.
2312 if (byte_diff == 0) {
2316 ifp = XFS_IFORK_PTR(ip, whichfork);
2317 new_size = (int)ifp->if_bytes + byte_diff;
2318 ASSERT(new_size >= 0);
2320 if (new_size == 0) {
2321 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2322 kmem_free(ifp->if_u1.if_data);
2324 ifp->if_u1.if_data = NULL;
2326 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2328 * If the valid extents/data can fit in if_inline_ext/data,
2329 * copy them from the malloc'd vector and free it.
2331 if (ifp->if_u1.if_data == NULL) {
2332 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2333 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2334 ASSERT(ifp->if_real_bytes != 0);
2335 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2337 kmem_free(ifp->if_u1.if_data);
2338 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2343 * Stuck with malloc/realloc.
2344 * For inline data, the underlying buffer must be
2345 * a multiple of 4 bytes in size so that it can be
2346 * logged and stay on word boundaries. We enforce
2349 real_size = roundup(new_size, 4);
2350 if (ifp->if_u1.if_data == NULL) {
2351 ASSERT(ifp->if_real_bytes == 0);
2352 ifp->if_u1.if_data = kmem_alloc(real_size,
2353 KM_SLEEP | KM_NOFS);
2354 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2356 * Only do the realloc if the underlying size
2357 * is really changing.
2359 if (ifp->if_real_bytes != real_size) {
2360 ifp->if_u1.if_data =
2361 kmem_realloc(ifp->if_u1.if_data,
2364 KM_SLEEP | KM_NOFS);
2367 ASSERT(ifp->if_real_bytes == 0);
2368 ifp->if_u1.if_data = kmem_alloc(real_size,
2369 KM_SLEEP | KM_NOFS);
2370 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2374 ifp->if_real_bytes = real_size;
2375 ifp->if_bytes = new_size;
2376 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2386 ifp = XFS_IFORK_PTR(ip, whichfork);
2387 if (ifp->if_broot != NULL) {
2388 kmem_free(ifp->if_broot);
2389 ifp->if_broot = NULL;
2393 * If the format is local, then we can't have an extents
2394 * array so just look for an inline data array. If we're
2395 * not local then we may or may not have an extents list,
2396 * so check and free it up if we do.
2398 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2399 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2400 (ifp->if_u1.if_data != NULL)) {
2401 ASSERT(ifp->if_real_bytes != 0);
2402 kmem_free(ifp->if_u1.if_data);
2403 ifp->if_u1.if_data = NULL;
2404 ifp->if_real_bytes = 0;
2406 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2407 ((ifp->if_flags & XFS_IFEXTIREC) ||
2408 ((ifp->if_u1.if_extents != NULL) &&
2409 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
2410 ASSERT(ifp->if_real_bytes != 0);
2411 xfs_iext_destroy(ifp);
2413 ASSERT(ifp->if_u1.if_extents == NULL ||
2414 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2415 ASSERT(ifp->if_real_bytes == 0);
2416 if (whichfork == XFS_ATTR_FORK) {
2417 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2423 * This is called to unpin an inode. The caller must have the inode locked
2424 * in at least shared mode so that the buffer cannot be subsequently pinned
2425 * once someone is waiting for it to be unpinned.
2429 struct xfs_inode *ip)
2431 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2433 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2435 /* Give the log a push to start the unpinning I/O */
2436 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2442 struct xfs_inode *ip)
2444 if (xfs_ipincount(ip)) {
2445 xfs_iunpin_nowait(ip);
2446 wait_event(ip->i_ipin_wait, (xfs_ipincount(ip) == 0));
2451 * xfs_iextents_copy()
2453 * This is called to copy the REAL extents (as opposed to the delayed
2454 * allocation extents) from the inode into the given buffer. It
2455 * returns the number of bytes copied into the buffer.
2457 * If there are no delayed allocation extents, then we can just
2458 * memcpy() the extents into the buffer. Otherwise, we need to
2459 * examine each extent in turn and skip those which are delayed.
2471 xfs_fsblock_t start_block;
2473 ifp = XFS_IFORK_PTR(ip, whichfork);
2474 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2475 ASSERT(ifp->if_bytes > 0);
2477 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2478 XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork);
2482 * There are some delayed allocation extents in the
2483 * inode, so copy the extents one at a time and skip
2484 * the delayed ones. There must be at least one
2485 * non-delayed extent.
2488 for (i = 0; i < nrecs; i++) {
2489 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
2490 start_block = xfs_bmbt_get_startblock(ep);
2491 if (isnullstartblock(start_block)) {
2493 * It's a delayed allocation extent, so skip it.
2498 /* Translate to on disk format */
2499 put_unaligned(cpu_to_be64(ep->l0), &dp->l0);
2500 put_unaligned(cpu_to_be64(ep->l1), &dp->l1);
2504 ASSERT(copied != 0);
2505 xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip));
2507 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2511 * Each of the following cases stores data into the same region
2512 * of the on-disk inode, so only one of them can be valid at
2513 * any given time. While it is possible to have conflicting formats
2514 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2515 * in EXTENTS format, this can only happen when the fork has
2516 * changed formats after being modified but before being flushed.
2517 * In these cases, the format always takes precedence, because the
2518 * format indicates the current state of the fork.
2525 xfs_inode_log_item_t *iip,
2532 #ifdef XFS_TRANS_DEBUG
2535 static const short brootflag[2] =
2536 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2537 static const short dataflag[2] =
2538 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2539 static const short extflag[2] =
2540 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2544 ifp = XFS_IFORK_PTR(ip, whichfork);
2546 * This can happen if we gave up in iformat in an error path,
2547 * for the attribute fork.
2550 ASSERT(whichfork == XFS_ATTR_FORK);
2553 cp = XFS_DFORK_PTR(dip, whichfork);
2555 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2556 case XFS_DINODE_FMT_LOCAL:
2557 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2558 (ifp->if_bytes > 0)) {
2559 ASSERT(ifp->if_u1.if_data != NULL);
2560 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2561 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2565 case XFS_DINODE_FMT_EXTENTS:
2566 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2567 !(iip->ili_format.ilf_fields & extflag[whichfork]));
2568 ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
2569 (ifp->if_bytes == 0));
2570 ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
2571 (ifp->if_bytes > 0));
2572 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
2573 (ifp->if_bytes > 0)) {
2574 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2575 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2580 case XFS_DINODE_FMT_BTREE:
2581 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
2582 (ifp->if_broot_bytes > 0)) {
2583 ASSERT(ifp->if_broot != NULL);
2584 ASSERT(ifp->if_broot_bytes <=
2585 (XFS_IFORK_SIZE(ip, whichfork) +
2586 XFS_BROOT_SIZE_ADJ));
2587 xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes,
2588 (xfs_bmdr_block_t *)cp,
2589 XFS_DFORK_SIZE(dip, mp, whichfork));
2593 case XFS_DINODE_FMT_DEV:
2594 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
2595 ASSERT(whichfork == XFS_DATA_FORK);
2596 xfs_dinode_put_rdev(dip, ip->i_df.if_u2.if_rdev);
2600 case XFS_DINODE_FMT_UUID:
2601 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
2602 ASSERT(whichfork == XFS_DATA_FORK);
2603 memcpy(XFS_DFORK_DPTR(dip),
2604 &ip->i_df.if_u2.if_uuid,
2620 xfs_mount_t *mp = ip->i_mount;
2621 struct xfs_perag *pag;
2622 unsigned long first_index, mask;
2623 unsigned long inodes_per_cluster;
2625 xfs_inode_t **ilist;
2632 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2634 inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog;
2635 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
2636 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
2640 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
2641 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
2643 /* really need a gang lookup range call here */
2644 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
2645 first_index, inodes_per_cluster);
2649 for (i = 0; i < nr_found; i++) {
2655 * because this is an RCU protected lookup, we could find a
2656 * recently freed or even reallocated inode during the lookup.
2657 * We need to check under the i_flags_lock for a valid inode
2658 * here. Skip it if it is not valid or the wrong inode.
2660 spin_lock(&ip->i_flags_lock);
2662 (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) {
2663 spin_unlock(&ip->i_flags_lock);
2666 spin_unlock(&ip->i_flags_lock);
2669 * Do an un-protected check to see if the inode is dirty and
2670 * is a candidate for flushing. These checks will be repeated
2671 * later after the appropriate locks are acquired.
2673 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
2677 * Try to get locks. If any are unavailable or it is pinned,
2678 * then this inode cannot be flushed and is skipped.
2681 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
2683 if (!xfs_iflock_nowait(iq)) {
2684 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2687 if (xfs_ipincount(iq)) {
2689 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2694 * arriving here means that this inode can be flushed. First
2695 * re-check that it's dirty before flushing.
2697 if (!xfs_inode_clean(iq)) {
2699 error = xfs_iflush_int(iq, bp);
2701 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2702 goto cluster_corrupt_out;
2708 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2712 XFS_STATS_INC(xs_icluster_flushcnt);
2713 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
2724 cluster_corrupt_out:
2726 * Corruption detected in the clustering loop. Invalidate the
2727 * inode buffer and shut down the filesystem.
2731 * Clean up the buffer. If it was B_DELWRI, just release it --
2732 * brelse can handle it with no problems. If not, shut down the
2733 * filesystem before releasing the buffer.
2735 bufwasdelwri = XFS_BUF_ISDELAYWRITE(bp);
2739 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2741 if (!bufwasdelwri) {
2743 * Just like incore_relse: if we have b_iodone functions,
2744 * mark the buffer as an error and call them. Otherwise
2745 * mark it as stale and brelse.
2747 if (XFS_BUF_IODONE_FUNC(bp)) {
2750 XFS_BUF_ERROR(bp,EIO);
2751 xfs_buf_ioend(bp, 0);
2759 * Unlocks the flush lock
2761 xfs_iflush_abort(iq);
2764 return XFS_ERROR(EFSCORRUPTED);
2768 * xfs_iflush() will write a modified inode's changes out to the
2769 * inode's on disk home. The caller must have the inode lock held
2770 * in at least shared mode and the inode flush completion must be
2771 * active as well. The inode lock will still be held upon return from
2772 * the call and the caller is free to unlock it.
2773 * The inode flush will be completed when the inode reaches the disk.
2774 * The flags indicate how the inode's buffer should be written out.
2781 xfs_inode_log_item_t *iip;
2787 XFS_STATS_INC(xs_iflush_count);
2789 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2790 ASSERT(!completion_done(&ip->i_flush));
2791 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2792 ip->i_d.di_nextents > ip->i_df.if_ext_max);
2798 * We can't flush the inode until it is unpinned, so wait for it if we
2799 * are allowed to block. We know noone new can pin it, because we are
2800 * holding the inode lock shared and you need to hold it exclusively to
2803 * If we are not allowed to block, force the log out asynchronously so
2804 * that when we come back the inode will be unpinned. If other inodes
2805 * in the same cluster are dirty, they will probably write the inode
2806 * out for us if they occur after the log force completes.
2808 if (!(flags & SYNC_WAIT) && xfs_ipincount(ip)) {
2809 xfs_iunpin_nowait(ip);
2813 xfs_iunpin_wait(ip);
2816 * For stale inodes we cannot rely on the backing buffer remaining
2817 * stale in cache for the remaining life of the stale inode and so
2818 * xfs_itobp() below may give us a buffer that no longer contains
2819 * inodes below. We have to check this after ensuring the inode is
2820 * unpinned so that it is safe to reclaim the stale inode after the
2823 if (xfs_iflags_test(ip, XFS_ISTALE)) {
2829 * This may have been unpinned because the filesystem is shutting
2830 * down forcibly. If that's the case we must not write this inode
2831 * to disk, because the log record didn't make it to disk!
2833 if (XFS_FORCED_SHUTDOWN(mp)) {
2834 ip->i_update_core = 0;
2836 iip->ili_format.ilf_fields = 0;
2838 return XFS_ERROR(EIO);
2842 * Get the buffer containing the on-disk inode.
2844 error = xfs_itobp(mp, NULL, ip, &dip, &bp,
2845 (flags & SYNC_WAIT) ? XBF_LOCK : XBF_TRYLOCK);
2852 * First flush out the inode that xfs_iflush was called with.
2854 error = xfs_iflush_int(ip, bp);
2859 * If the buffer is pinned then push on the log now so we won't
2860 * get stuck waiting in the write for too long.
2862 if (XFS_BUF_ISPINNED(bp))
2863 xfs_log_force(mp, 0);
2867 * see if other inodes can be gathered into this write
2869 error = xfs_iflush_cluster(ip, bp);
2871 goto cluster_corrupt_out;
2873 if (flags & SYNC_WAIT)
2874 error = xfs_bwrite(mp, bp);
2876 xfs_bdwrite(mp, bp);
2881 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2882 cluster_corrupt_out:
2884 * Unlocks the flush lock
2886 xfs_iflush_abort(ip);
2887 return XFS_ERROR(EFSCORRUPTED);
2896 xfs_inode_log_item_t *iip;
2899 #ifdef XFS_TRANS_DEBUG
2903 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2904 ASSERT(!completion_done(&ip->i_flush));
2905 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2906 ip->i_d.di_nextents > ip->i_df.if_ext_max);
2911 /* set *dip = inode's place in the buffer */
2912 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
2915 * Clear i_update_core before copying out the data.
2916 * This is for coordination with our timestamp updates
2917 * that don't hold the inode lock. They will always
2918 * update the timestamps BEFORE setting i_update_core,
2919 * so if we clear i_update_core after they set it we
2920 * are guaranteed to see their updates to the timestamps.
2921 * I believe that this depends on strongly ordered memory
2922 * semantics, but we have that. We use the SYNCHRONIZE
2923 * macro to make sure that the compiler does not reorder
2924 * the i_update_core access below the data copy below.
2926 ip->i_update_core = 0;
2930 * Make sure to get the latest timestamps from the Linux inode.
2932 xfs_synchronize_times(ip);
2934 if (XFS_TEST_ERROR(be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC,
2935 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
2936 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2937 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
2938 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
2941 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
2942 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
2943 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2944 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
2945 __func__, ip->i_ino, ip, ip->i_d.di_magic);
2948 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
2950 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
2951 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
2952 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
2953 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2954 "%s: Bad regular inode %Lu, ptr 0x%p",
2955 __func__, ip->i_ino, ip);
2958 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
2960 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
2961 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
2962 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
2963 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
2964 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2965 "%s: Bad directory inode %Lu, ptr 0x%p",
2966 __func__, ip->i_ino, ip);
2970 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
2971 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
2972 XFS_RANDOM_IFLUSH_5)) {
2973 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2974 "%s: detected corrupt incore inode %Lu, "
2975 "total extents = %d, nblocks = %Ld, ptr 0x%p",
2976 __func__, ip->i_ino,
2977 ip->i_d.di_nextents + ip->i_d.di_anextents,
2978 ip->i_d.di_nblocks, ip);
2981 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
2982 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
2983 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2984 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
2985 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
2989 * bump the flush iteration count, used to detect flushes which
2990 * postdate a log record during recovery.
2993 ip->i_d.di_flushiter++;
2996 * Copy the dirty parts of the inode into the on-disk
2997 * inode. We always copy out the core of the inode,
2998 * because if the inode is dirty at all the core must
3001 xfs_dinode_to_disk(dip, &ip->i_d);
3003 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3004 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3005 ip->i_d.di_flushiter = 0;
3008 * If this is really an old format inode and the superblock version
3009 * has not been updated to support only new format inodes, then
3010 * convert back to the old inode format. If the superblock version
3011 * has been updated, then make the conversion permanent.
3013 ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
3014 if (ip->i_d.di_version == 1) {
3015 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
3019 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3020 dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink);
3023 * The superblock version has already been bumped,
3024 * so just make the conversion to the new inode
3027 ip->i_d.di_version = 2;
3028 dip->di_version = 2;
3029 ip->i_d.di_onlink = 0;
3031 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3032 memset(&(dip->di_pad[0]), 0,
3033 sizeof(dip->di_pad));
3034 ASSERT(xfs_get_projid(ip) == 0);
3038 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
3039 if (XFS_IFORK_Q(ip))
3040 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3041 xfs_inobp_check(mp, bp);
3044 * We've recorded everything logged in the inode, so we'd
3045 * like to clear the ilf_fields bits so we don't log and
3046 * flush things unnecessarily. However, we can't stop
3047 * logging all this information until the data we've copied
3048 * into the disk buffer is written to disk. If we did we might
3049 * overwrite the copy of the inode in the log with all the
3050 * data after re-logging only part of it, and in the face of
3051 * a crash we wouldn't have all the data we need to recover.
3053 * What we do is move the bits to the ili_last_fields field.
3054 * When logging the inode, these bits are moved back to the
3055 * ilf_fields field. In the xfs_iflush_done() routine we
3056 * clear ili_last_fields, since we know that the information
3057 * those bits represent is permanently on disk. As long as
3058 * the flush completes before the inode is logged again, then
3059 * both ilf_fields and ili_last_fields will be cleared.
3061 * We can play with the ilf_fields bits here, because the inode
3062 * lock must be held exclusively in order to set bits there
3063 * and the flush lock protects the ili_last_fields bits.
3064 * Set ili_logged so the flush done
3065 * routine can tell whether or not to look in the AIL.
3066 * Also, store the current LSN of the inode so that we can tell
3067 * whether the item has moved in the AIL from xfs_iflush_done().
3068 * In order to read the lsn we need the AIL lock, because
3069 * it is a 64 bit value that cannot be read atomically.
3071 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3072 iip->ili_last_fields = iip->ili_format.ilf_fields;
3073 iip->ili_format.ilf_fields = 0;
3074 iip->ili_logged = 1;
3076 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3077 &iip->ili_item.li_lsn);
3080 * Attach the function xfs_iflush_done to the inode's
3081 * buffer. This will remove the inode from the AIL
3082 * and unlock the inode's flush lock when the inode is
3083 * completely written to disk.
3085 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3087 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3088 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3091 * We're flushing an inode which is not in the AIL and has
3092 * not been logged but has i_update_core set. For this
3093 * case we can use a B_DELWRI flush and immediately drop
3094 * the inode flush lock because we can avoid the whole
3095 * AIL state thing. It's OK to drop the flush lock now,
3096 * because we've already locked the buffer and to do anything
3097 * you really need both.
3100 ASSERT(iip->ili_logged == 0);
3101 ASSERT(iip->ili_last_fields == 0);
3102 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3110 return XFS_ERROR(EFSCORRUPTED);
3114 * Return a pointer to the extent record at file index idx.
3116 xfs_bmbt_rec_host_t *
3118 xfs_ifork_t *ifp, /* inode fork pointer */
3119 xfs_extnum_t idx) /* index of target extent */
3122 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
3123 return ifp->if_u1.if_ext_irec->er_extbuf;
3124 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3125 xfs_ext_irec_t *erp; /* irec pointer */
3126 int erp_idx = 0; /* irec index */
3127 xfs_extnum_t page_idx = idx; /* ext index in target list */
3129 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3130 return &erp->er_extbuf[page_idx];
3131 } else if (ifp->if_bytes) {
3132 return &ifp->if_u1.if_extents[idx];
3139 * Insert new item(s) into the extent records for incore inode
3140 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3144 xfs_inode_t *ip, /* incore inode pointer */
3145 xfs_extnum_t idx, /* starting index of new items */
3146 xfs_extnum_t count, /* number of inserted items */
3147 xfs_bmbt_irec_t *new, /* items to insert */
3148 int state) /* type of extent conversion */
3150 xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
3151 xfs_extnum_t i; /* extent record index */
3153 trace_xfs_iext_insert(ip, idx, new, state, _RET_IP_);
3155 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3156 xfs_iext_add(ifp, idx, count);
3157 for (i = idx; i < idx + count; i++, new++)
3158 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new);
3162 * This is called when the amount of space required for incore file
3163 * extents needs to be increased. The ext_diff parameter stores the
3164 * number of new extents being added and the idx parameter contains
3165 * the extent index where the new extents will be added. If the new
3166 * extents are being appended, then we just need to (re)allocate and
3167 * initialize the space. Otherwise, if the new extents are being
3168 * inserted into the middle of the existing entries, a bit more work
3169 * is required to make room for the new extents to be inserted. The
3170 * caller is responsible for filling in the new extent entries upon
3175 xfs_ifork_t *ifp, /* inode fork pointer */
3176 xfs_extnum_t idx, /* index to begin adding exts */
3177 int ext_diff) /* number of extents to add */
3179 int byte_diff; /* new bytes being added */
3180 int new_size; /* size of extents after adding */
3181 xfs_extnum_t nextents; /* number of extents in file */
3183 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3184 ASSERT((idx >= 0) && (idx <= nextents));
3185 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
3186 new_size = ifp->if_bytes + byte_diff;
3188 * If the new number of extents (nextents + ext_diff)
3189 * fits inside the inode, then continue to use the inline
3192 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
3193 if (idx < nextents) {
3194 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
3195 &ifp->if_u2.if_inline_ext[idx],
3196 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3197 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
3199 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3200 ifp->if_real_bytes = 0;
3201 ifp->if_lastex = nextents + ext_diff;
3204 * Otherwise use a linear (direct) extent list.
3205 * If the extents are currently inside the inode,
3206 * xfs_iext_realloc_direct will switch us from
3207 * inline to direct extent allocation mode.
3209 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
3210 xfs_iext_realloc_direct(ifp, new_size);
3211 if (idx < nextents) {
3212 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
3213 &ifp->if_u1.if_extents[idx],
3214 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3215 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
3218 /* Indirection array */
3220 xfs_ext_irec_t *erp;
3224 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
3225 if (ifp->if_flags & XFS_IFEXTIREC) {
3226 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3228 xfs_iext_irec_init(ifp);
3229 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3230 erp = ifp->if_u1.if_ext_irec;
3232 /* Extents fit in target extent page */
3233 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3234 if (page_idx < erp->er_extcount) {
3235 memmove(&erp->er_extbuf[page_idx + ext_diff],
3236 &erp->er_extbuf[page_idx],
3237 (erp->er_extcount - page_idx) *
3238 sizeof(xfs_bmbt_rec_t));
3239 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3241 erp->er_extcount += ext_diff;
3242 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3244 /* Insert a new extent page */
3246 xfs_iext_add_indirect_multi(ifp,
3247 erp_idx, page_idx, ext_diff);
3250 * If extent(s) are being appended to the last page in
3251 * the indirection array and the new extent(s) don't fit
3252 * in the page, then erp is NULL and erp_idx is set to
3253 * the next index needed in the indirection array.
3256 int count = ext_diff;
3259 erp = xfs_iext_irec_new(ifp, erp_idx);
3260 erp->er_extcount = count;
3261 count -= MIN(count, (int)XFS_LINEAR_EXTS);
3268 ifp->if_bytes = new_size;
3272 * This is called when incore extents are being added to the indirection
3273 * array and the new extents do not fit in the target extent list. The
3274 * erp_idx parameter contains the irec index for the target extent list
3275 * in the indirection array, and the idx parameter contains the extent
3276 * index within the list. The number of extents being added is stored
3277 * in the count parameter.
3279 * |-------| |-------|
3280 * | | | | idx - number of extents before idx
3282 * | | | | count - number of extents being inserted at idx
3283 * |-------| |-------|
3284 * | count | | nex2 | nex2 - number of extents after idx + count
3285 * |-------| |-------|
3288 xfs_iext_add_indirect_multi(
3289 xfs_ifork_t *ifp, /* inode fork pointer */
3290 int erp_idx, /* target extent irec index */
3291 xfs_extnum_t idx, /* index within target list */
3292 int count) /* new extents being added */
3294 int byte_diff; /* new bytes being added */
3295 xfs_ext_irec_t *erp; /* pointer to irec entry */
3296 xfs_extnum_t ext_diff; /* number of extents to add */
3297 xfs_extnum_t ext_cnt; /* new extents still needed */
3298 xfs_extnum_t nex2; /* extents after idx + count */
3299 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3300 int nlists; /* number of irec's (lists) */
3302 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3303 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3304 nex2 = erp->er_extcount - idx;
3305 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3308 * Save second part of target extent list
3309 * (all extents past */
3311 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3312 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS);
3313 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3314 erp->er_extcount -= nex2;
3315 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3316 memset(&erp->er_extbuf[idx], 0, byte_diff);
3320 * Add the new extents to the end of the target
3321 * list, then allocate new irec record(s) and
3322 * extent buffer(s) as needed to store the rest
3323 * of the new extents.
3326 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3328 erp->er_extcount += ext_diff;
3329 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3330 ext_cnt -= ext_diff;
3334 erp = xfs_iext_irec_new(ifp, erp_idx);
3335 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3336 erp->er_extcount = ext_diff;
3337 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3338 ext_cnt -= ext_diff;
3341 /* Add nex2 extents back to indirection array */
3343 xfs_extnum_t ext_avail;
3346 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3347 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3350 * If nex2 extents fit in the current page, append
3351 * nex2_ep after the new extents.
3353 if (nex2 <= ext_avail) {
3354 i = erp->er_extcount;
3357 * Otherwise, check if space is available in the
3360 else if ((erp_idx < nlists - 1) &&
3361 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3362 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3365 /* Create a hole for nex2 extents */
3366 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3367 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3370 * Final choice, create a new extent page for
3375 erp = xfs_iext_irec_new(ifp, erp_idx);
3377 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
3379 erp->er_extcount += nex2;
3380 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3385 * This is called when the amount of space required for incore file
3386 * extents needs to be decreased. The ext_diff parameter stores the
3387 * number of extents to be removed and the idx parameter contains
3388 * the extent index where the extents will be removed from.
3390 * If the amount of space needed has decreased below the linear
3391 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3392 * extent array. Otherwise, use kmem_realloc() to adjust the
3393 * size to what is needed.
3397 xfs_inode_t *ip, /* incore inode pointer */
3398 xfs_extnum_t idx, /* index to begin removing exts */
3399 int ext_diff, /* number of extents to remove */
3400 int state) /* type of extent conversion */
3402 xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
3403 xfs_extnum_t nextents; /* number of extents in file */
3404 int new_size; /* size of extents after removal */
3406 trace_xfs_iext_remove(ip, idx, state, _RET_IP_);
3408 ASSERT(ext_diff > 0);
3409 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3410 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3412 if (new_size == 0) {
3413 xfs_iext_destroy(ifp);
3414 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3415 xfs_iext_remove_indirect(ifp, idx, ext_diff);
3416 } else if (ifp->if_real_bytes) {
3417 xfs_iext_remove_direct(ifp, idx, ext_diff);
3419 xfs_iext_remove_inline(ifp, idx, ext_diff);
3421 ifp->if_bytes = new_size;
3425 * This removes ext_diff extents from the inline buffer, beginning
3426 * at extent index idx.
3429 xfs_iext_remove_inline(
3430 xfs_ifork_t *ifp, /* inode fork pointer */
3431 xfs_extnum_t idx, /* index to begin removing exts */
3432 int ext_diff) /* number of extents to remove */
3434 int nextents; /* number of extents in file */
3436 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3437 ASSERT(idx < XFS_INLINE_EXTS);
3438 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3439 ASSERT(((nextents - ext_diff) > 0) &&
3440 (nextents - ext_diff) < XFS_INLINE_EXTS);
3442 if (idx + ext_diff < nextents) {
3443 memmove(&ifp->if_u2.if_inline_ext[idx],
3444 &ifp->if_u2.if_inline_ext[idx + ext_diff],
3445 (nextents - (idx + ext_diff)) *
3446 sizeof(xfs_bmbt_rec_t));
3447 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
3448 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3450 memset(&ifp->if_u2.if_inline_ext[idx], 0,
3451 ext_diff * sizeof(xfs_bmbt_rec_t));
3456 * This removes ext_diff extents from a linear (direct) extent list,
3457 * beginning at extent index idx. If the extents are being removed
3458 * from the end of the list (ie. truncate) then we just need to re-
3459 * allocate the list to remove the extra space. Otherwise, if the
3460 * extents are being removed from the middle of the existing extent
3461 * entries, then we first need to move the extent records beginning
3462 * at idx + ext_diff up in the list to overwrite the records being
3463 * removed, then remove the extra space via kmem_realloc.
3466 xfs_iext_remove_direct(
3467 xfs_ifork_t *ifp, /* inode fork pointer */
3468 xfs_extnum_t idx, /* index to begin removing exts */
3469 int ext_diff) /* number of extents to remove */
3471 xfs_extnum_t nextents; /* number of extents in file */
3472 int new_size; /* size of extents after removal */
3474 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3475 new_size = ifp->if_bytes -
3476 (ext_diff * sizeof(xfs_bmbt_rec_t));
3477 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3479 if (new_size == 0) {
3480 xfs_iext_destroy(ifp);
3483 /* Move extents up in the list (if needed) */
3484 if (idx + ext_diff < nextents) {
3485 memmove(&ifp->if_u1.if_extents[idx],
3486 &ifp->if_u1.if_extents[idx + ext_diff],
3487 (nextents - (idx + ext_diff)) *
3488 sizeof(xfs_bmbt_rec_t));
3490 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
3491 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3493 * Reallocate the direct extent list. If the extents
3494 * will fit inside the inode then xfs_iext_realloc_direct
3495 * will switch from direct to inline extent allocation
3498 xfs_iext_realloc_direct(ifp, new_size);
3499 ifp->if_bytes = new_size;
3503 * This is called when incore extents are being removed from the
3504 * indirection array and the extents being removed span multiple extent
3505 * buffers. The idx parameter contains the file extent index where we
3506 * want to begin removing extents, and the count parameter contains
3507 * how many extents need to be removed.
3509 * |-------| |-------|
3510 * | nex1 | | | nex1 - number of extents before idx
3511 * |-------| | count |
3512 * | | | | count - number of extents being removed at idx
3513 * | count | |-------|
3514 * | | | nex2 | nex2 - number of extents after idx + count
3515 * |-------| |-------|
3518 xfs_iext_remove_indirect(
3519 xfs_ifork_t *ifp, /* inode fork pointer */
3520 xfs_extnum_t idx, /* index to begin removing extents */
3521 int count) /* number of extents to remove */
3523 xfs_ext_irec_t *erp; /* indirection array pointer */
3524 int erp_idx = 0; /* indirection array index */
3525 xfs_extnum_t ext_cnt; /* extents left to remove */
3526 xfs_extnum_t ext_diff; /* extents to remove in current list */
3527 xfs_extnum_t nex1; /* number of extents before idx */
3528 xfs_extnum_t nex2; /* extents after idx + count */
3529 int page_idx = idx; /* index in target extent list */
3531 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3532 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3533 ASSERT(erp != NULL);
3537 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
3538 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
3540 * Check for deletion of entire list;
3541 * xfs_iext_irec_remove() updates extent offsets.
3543 if (ext_diff == erp->er_extcount) {
3544 xfs_iext_irec_remove(ifp, erp_idx);
3545 ext_cnt -= ext_diff;
3548 ASSERT(erp_idx < ifp->if_real_bytes /
3550 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3557 /* Move extents up (if needed) */
3559 memmove(&erp->er_extbuf[nex1],
3560 &erp->er_extbuf[nex1 + ext_diff],
3561 nex2 * sizeof(xfs_bmbt_rec_t));
3563 /* Zero out rest of page */
3564 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
3565 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
3566 /* Update remaining counters */
3567 erp->er_extcount -= ext_diff;
3568 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
3569 ext_cnt -= ext_diff;
3574 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
3575 xfs_iext_irec_compact(ifp);
3579 * Create, destroy, or resize a linear (direct) block of extents.
3582 xfs_iext_realloc_direct(
3583 xfs_ifork_t *ifp, /* inode fork pointer */
3584 int new_size) /* new size of extents */
3586 int rnew_size; /* real new size of extents */
3588 rnew_size = new_size;
3590 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
3591 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
3592 (new_size != ifp->if_real_bytes)));
3594 /* Free extent records */
3595 if (new_size == 0) {
3596 xfs_iext_destroy(ifp);
3598 /* Resize direct extent list and zero any new bytes */
3599 else if (ifp->if_real_bytes) {
3600 /* Check if extents will fit inside the inode */
3601 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
3602 xfs_iext_direct_to_inline(ifp, new_size /
3603 (uint)sizeof(xfs_bmbt_rec_t));
3604 ifp->if_bytes = new_size;
3607 if (!is_power_of_2(new_size)){
3608 rnew_size = roundup_pow_of_two(new_size);
3610 if (rnew_size != ifp->if_real_bytes) {
3611 ifp->if_u1.if_extents =
3612 kmem_realloc(ifp->if_u1.if_extents,
3614 ifp->if_real_bytes, KM_NOFS);
3616 if (rnew_size > ifp->if_real_bytes) {
3617 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
3618 (uint)sizeof(xfs_bmbt_rec_t)], 0,
3619 rnew_size - ifp->if_real_bytes);
3623 * Switch from the inline extent buffer to a direct
3624 * extent list. Be sure to include the inline extent
3625 * bytes in new_size.
3628 new_size += ifp->if_bytes;
3629 if (!is_power_of_2(new_size)) {
3630 rnew_size = roundup_pow_of_two(new_size);
3632 xfs_iext_inline_to_direct(ifp, rnew_size);
3634 ifp->if_real_bytes = rnew_size;
3635 ifp->if_bytes = new_size;
3639 * Switch from linear (direct) extent records to inline buffer.
3642 xfs_iext_direct_to_inline(
3643 xfs_ifork_t *ifp, /* inode fork pointer */
3644 xfs_extnum_t nextents) /* number of extents in file */
3646 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3647 ASSERT(nextents <= XFS_INLINE_EXTS);
3649 * The inline buffer was zeroed when we switched
3650 * from inline to direct extent allocation mode,
3651 * so we don't need to clear it here.
3653 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
3654 nextents * sizeof(xfs_bmbt_rec_t));
3655 kmem_free(ifp->if_u1.if_extents);
3656 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3657 ifp->if_real_bytes = 0;
3661 * Switch from inline buffer to linear (direct) extent records.
3662 * new_size should already be rounded up to the next power of 2
3663 * by the caller (when appropriate), so use new_size as it is.
3664 * However, since new_size may be rounded up, we can't update
3665 * if_bytes here. It is the caller's responsibility to update
3666 * if_bytes upon return.
3669 xfs_iext_inline_to_direct(
3670 xfs_ifork_t *ifp, /* inode fork pointer */
3671 int new_size) /* number of extents in file */
3673 ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS);
3674 memset(ifp->if_u1.if_extents, 0, new_size);
3675 if (ifp->if_bytes) {
3676 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
3678 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3679 sizeof(xfs_bmbt_rec_t));
3681 ifp->if_real_bytes = new_size;
3685 * Resize an extent indirection array to new_size bytes.
3688 xfs_iext_realloc_indirect(
3689 xfs_ifork_t *ifp, /* inode fork pointer */
3690 int new_size) /* new indirection array size */
3692 int nlists; /* number of irec's (ex lists) */
3693 int size; /* current indirection array size */
3695 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3696 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3697 size = nlists * sizeof(xfs_ext_irec_t);
3698 ASSERT(ifp->if_real_bytes);
3699 ASSERT((new_size >= 0) && (new_size != size));
3700 if (new_size == 0) {
3701 xfs_iext_destroy(ifp);
3703 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
3704 kmem_realloc(ifp->if_u1.if_ext_irec,
3705 new_size, size, KM_NOFS);
3710 * Switch from indirection array to linear (direct) extent allocations.
3713 xfs_iext_indirect_to_direct(
3714 xfs_ifork_t *ifp) /* inode fork pointer */
3716 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
3717 xfs_extnum_t nextents; /* number of extents in file */
3718 int size; /* size of file extents */
3720 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3721 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3722 ASSERT(nextents <= XFS_LINEAR_EXTS);
3723 size = nextents * sizeof(xfs_bmbt_rec_t);
3725 xfs_iext_irec_compact_pages(ifp);
3726 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
3728 ep = ifp->if_u1.if_ext_irec->er_extbuf;
3729 kmem_free(ifp->if_u1.if_ext_irec);
3730 ifp->if_flags &= ~XFS_IFEXTIREC;
3731 ifp->if_u1.if_extents = ep;
3732 ifp->if_bytes = size;
3733 if (nextents < XFS_LINEAR_EXTS) {
3734 xfs_iext_realloc_direct(ifp, size);
3739 * Free incore file extents.
3743 xfs_ifork_t *ifp) /* inode fork pointer */
3745 if (ifp->if_flags & XFS_IFEXTIREC) {
3749 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3750 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
3751 xfs_iext_irec_remove(ifp, erp_idx);
3753 ifp->if_flags &= ~XFS_IFEXTIREC;
3754 } else if (ifp->if_real_bytes) {
3755 kmem_free(ifp->if_u1.if_extents);
3756 } else if (ifp->if_bytes) {
3757 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3758 sizeof(xfs_bmbt_rec_t));
3760 ifp->if_u1.if_extents = NULL;
3761 ifp->if_real_bytes = 0;
3766 * Return a pointer to the extent record for file system block bno.
3768 xfs_bmbt_rec_host_t * /* pointer to found extent record */
3769 xfs_iext_bno_to_ext(
3770 xfs_ifork_t *ifp, /* inode fork pointer */
3771 xfs_fileoff_t bno, /* block number to search for */
3772 xfs_extnum_t *idxp) /* index of target extent */
3774 xfs_bmbt_rec_host_t *base; /* pointer to first extent */
3775 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
3776 xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */
3777 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
3778 int high; /* upper boundary in search */
3779 xfs_extnum_t idx = 0; /* index of target extent */
3780 int low; /* lower boundary in search */
3781 xfs_extnum_t nextents; /* number of file extents */
3782 xfs_fileoff_t startoff = 0; /* start offset of extent */
3784 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3785 if (nextents == 0) {
3790 if (ifp->if_flags & XFS_IFEXTIREC) {
3791 /* Find target extent list */
3793 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
3794 base = erp->er_extbuf;
3795 high = erp->er_extcount - 1;
3797 base = ifp->if_u1.if_extents;
3798 high = nextents - 1;
3800 /* Binary search extent records */
3801 while (low <= high) {
3802 idx = (low + high) >> 1;
3804 startoff = xfs_bmbt_get_startoff(ep);
3805 blockcount = xfs_bmbt_get_blockcount(ep);
3806 if (bno < startoff) {
3808 } else if (bno >= startoff + blockcount) {
3811 /* Convert back to file-based extent index */
3812 if (ifp->if_flags & XFS_IFEXTIREC) {
3813 idx += erp->er_extoff;
3819 /* Convert back to file-based extent index */
3820 if (ifp->if_flags & XFS_IFEXTIREC) {
3821 idx += erp->er_extoff;
3823 if (bno >= startoff + blockcount) {
3824 if (++idx == nextents) {
3827 ep = xfs_iext_get_ext(ifp, idx);
3835 * Return a pointer to the indirection array entry containing the
3836 * extent record for filesystem block bno. Store the index of the
3837 * target irec in *erp_idxp.
3839 xfs_ext_irec_t * /* pointer to found extent record */
3840 xfs_iext_bno_to_irec(
3841 xfs_ifork_t *ifp, /* inode fork pointer */
3842 xfs_fileoff_t bno, /* block number to search for */
3843 int *erp_idxp) /* irec index of target ext list */
3845 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
3846 xfs_ext_irec_t *erp_next; /* next indirection array entry */
3847 int erp_idx; /* indirection array index */
3848 int nlists; /* number of extent irec's (lists) */
3849 int high; /* binary search upper limit */
3850 int low; /* binary search lower limit */
3852 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3853 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3857 while (low <= high) {
3858 erp_idx = (low + high) >> 1;
3859 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3860 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
3861 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
3863 } else if (erp_next && bno >=
3864 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
3870 *erp_idxp = erp_idx;
3875 * Return a pointer to the indirection array entry containing the
3876 * extent record at file extent index *idxp. Store the index of the
3877 * target irec in *erp_idxp and store the page index of the target
3878 * extent record in *idxp.
3881 xfs_iext_idx_to_irec(
3882 xfs_ifork_t *ifp, /* inode fork pointer */
3883 xfs_extnum_t *idxp, /* extent index (file -> page) */
3884 int *erp_idxp, /* pointer to target irec */
3885 int realloc) /* new bytes were just added */
3887 xfs_ext_irec_t *prev; /* pointer to previous irec */
3888 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
3889 int erp_idx; /* indirection array index */
3890 int nlists; /* number of irec's (ex lists) */
3891 int high; /* binary search upper limit */
3892 int low; /* binary search lower limit */
3893 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
3895 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3896 ASSERT(page_idx >= 0 && page_idx <=
3897 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
3898 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3903 /* Binary search extent irec's */
3904 while (low <= high) {
3905 erp_idx = (low + high) >> 1;
3906 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3907 prev = erp_idx > 0 ? erp - 1 : NULL;
3908 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
3909 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
3911 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
3912 (page_idx == erp->er_extoff + erp->er_extcount &&
3915 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
3916 erp->er_extcount == XFS_LINEAR_EXTS) {
3920 erp = erp_idx < nlists ? erp + 1 : NULL;
3923 page_idx -= erp->er_extoff;
3928 *erp_idxp = erp_idx;
3933 * Allocate and initialize an indirection array once the space needed
3934 * for incore extents increases above XFS_IEXT_BUFSZ.
3938 xfs_ifork_t *ifp) /* inode fork pointer */
3940 xfs_ext_irec_t *erp; /* indirection array pointer */
3941 xfs_extnum_t nextents; /* number of extents in file */
3943 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3944 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3945 ASSERT(nextents <= XFS_LINEAR_EXTS);
3947 erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS);
3949 if (nextents == 0) {
3950 ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
3951 } else if (!ifp->if_real_bytes) {
3952 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
3953 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
3954 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
3956 erp->er_extbuf = ifp->if_u1.if_extents;
3957 erp->er_extcount = nextents;
3960 ifp->if_flags |= XFS_IFEXTIREC;
3961 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
3962 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
3963 ifp->if_u1.if_ext_irec = erp;
3969 * Allocate and initialize a new entry in the indirection array.
3973 xfs_ifork_t *ifp, /* inode fork pointer */
3974 int erp_idx) /* index for new irec */
3976 xfs_ext_irec_t *erp; /* indirection array pointer */
3977 int i; /* loop counter */
3978 int nlists; /* number of irec's (ex lists) */
3980 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3981 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3983 /* Resize indirection array */
3984 xfs_iext_realloc_indirect(ifp, ++nlists *
3985 sizeof(xfs_ext_irec_t));
3987 * Move records down in the array so the
3988 * new page can use erp_idx.
3990 erp = ifp->if_u1.if_ext_irec;
3991 for (i = nlists - 1; i > erp_idx; i--) {
3992 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
3994 ASSERT(i == erp_idx);
3996 /* Initialize new extent record */
3997 erp = ifp->if_u1.if_ext_irec;
3998 erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
3999 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4000 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
4001 erp[erp_idx].er_extcount = 0;
4002 erp[erp_idx].er_extoff = erp_idx > 0 ?
4003 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
4004 return (&erp[erp_idx]);
4008 * Remove a record from the indirection array.
4011 xfs_iext_irec_remove(
4012 xfs_ifork_t *ifp, /* inode fork pointer */
4013 int erp_idx) /* irec index to remove */
4015 xfs_ext_irec_t *erp; /* indirection array pointer */
4016 int i; /* loop counter */
4017 int nlists; /* number of irec's (ex lists) */
4019 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4020 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4021 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4022 if (erp->er_extbuf) {
4023 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
4025 kmem_free(erp->er_extbuf);
4027 /* Compact extent records */
4028 erp = ifp->if_u1.if_ext_irec;
4029 for (i = erp_idx; i < nlists - 1; i++) {
4030 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
4033 * Manually free the last extent record from the indirection
4034 * array. A call to xfs_iext_realloc_indirect() with a size
4035 * of zero would result in a call to xfs_iext_destroy() which
4036 * would in turn call this function again, creating a nasty
4040 xfs_iext_realloc_indirect(ifp,
4041 nlists * sizeof(xfs_ext_irec_t));
4043 kmem_free(ifp->if_u1.if_ext_irec);
4045 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4049 * This is called to clean up large amounts of unused memory allocated
4050 * by the indirection array. Before compacting anything though, verify
4051 * that the indirection array is still needed and switch back to the
4052 * linear extent list (or even the inline buffer) if possible. The
4053 * compaction policy is as follows:
4055 * Full Compaction: Extents fit into a single page (or inline buffer)
4056 * Partial Compaction: Extents occupy less than 50% of allocated space
4057 * No Compaction: Extents occupy at least 50% of allocated space
4060 xfs_iext_irec_compact(
4061 xfs_ifork_t *ifp) /* inode fork pointer */
4063 xfs_extnum_t nextents; /* number of extents in file */
4064 int nlists; /* number of irec's (ex lists) */
4066 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4067 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4068 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4070 if (nextents == 0) {
4071 xfs_iext_destroy(ifp);
4072 } else if (nextents <= XFS_INLINE_EXTS) {
4073 xfs_iext_indirect_to_direct(ifp);
4074 xfs_iext_direct_to_inline(ifp, nextents);
4075 } else if (nextents <= XFS_LINEAR_EXTS) {
4076 xfs_iext_indirect_to_direct(ifp);
4077 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
4078 xfs_iext_irec_compact_pages(ifp);
4083 * Combine extents from neighboring extent pages.
4086 xfs_iext_irec_compact_pages(
4087 xfs_ifork_t *ifp) /* inode fork pointer */
4089 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
4090 int erp_idx = 0; /* indirection array index */
4091 int nlists; /* number of irec's (ex lists) */
4093 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4094 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4095 while (erp_idx < nlists - 1) {
4096 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4098 if (erp_next->er_extcount <=
4099 (XFS_LINEAR_EXTS - erp->er_extcount)) {
4100 memcpy(&erp->er_extbuf[erp->er_extcount],
4101 erp_next->er_extbuf, erp_next->er_extcount *
4102 sizeof(xfs_bmbt_rec_t));
4103 erp->er_extcount += erp_next->er_extcount;
4105 * Free page before removing extent record
4106 * so er_extoffs don't get modified in
4107 * xfs_iext_irec_remove.
4109 kmem_free(erp_next->er_extbuf);
4110 erp_next->er_extbuf = NULL;
4111 xfs_iext_irec_remove(ifp, erp_idx + 1);
4112 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4120 * This is called to update the er_extoff field in the indirection
4121 * array when extents have been added or removed from one of the
4122 * extent lists. erp_idx contains the irec index to begin updating
4123 * at and ext_diff contains the number of extents that were added
4127 xfs_iext_irec_update_extoffs(
4128 xfs_ifork_t *ifp, /* inode fork pointer */
4129 int erp_idx, /* irec index to update */
4130 int ext_diff) /* number of new extents */
4132 int i; /* loop counter */
4133 int nlists; /* number of irec's (ex lists */
4135 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4136 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4137 for (i = erp_idx; i < nlists; i++) {
4138 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;