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_fs_repair_cmn_err(CE_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_fs_repair_cmn_err(CE_WARN, ip->i_mount,
334 "corrupt dinode %Lu, forkoff = 0x%x.",
335 (unsigned long long)ip->i_ino,
337 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
339 return XFS_ERROR(EFSCORRUPTED);
342 if (unlikely((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) &&
343 !ip->i_mount->m_rtdev_targp)) {
344 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
345 "corrupt dinode %Lu, has realtime flag set.",
347 XFS_CORRUPTION_ERROR("xfs_iformat(realtime)",
348 XFS_ERRLEVEL_LOW, ip->i_mount, dip);
349 return XFS_ERROR(EFSCORRUPTED);
352 switch (ip->i_d.di_mode & S_IFMT) {
357 if (unlikely(dip->di_format != XFS_DINODE_FMT_DEV)) {
358 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
360 return XFS_ERROR(EFSCORRUPTED);
364 ip->i_df.if_u2.if_rdev = xfs_dinode_get_rdev(dip);
370 switch (dip->di_format) {
371 case XFS_DINODE_FMT_LOCAL:
373 * no local regular files yet
375 if (unlikely((be16_to_cpu(dip->di_mode) & S_IFMT) == S_IFREG)) {
376 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
378 "(local format for regular file).",
379 (unsigned long long) ip->i_ino);
380 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
383 return XFS_ERROR(EFSCORRUPTED);
386 di_size = be64_to_cpu(dip->di_size);
387 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
388 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
390 "(bad size %Ld for local inode).",
391 (unsigned long long) ip->i_ino,
392 (long long) di_size);
393 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
396 return XFS_ERROR(EFSCORRUPTED);
400 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
402 case XFS_DINODE_FMT_EXTENTS:
403 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
405 case XFS_DINODE_FMT_BTREE:
406 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
409 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
411 return XFS_ERROR(EFSCORRUPTED);
416 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
417 return XFS_ERROR(EFSCORRUPTED);
422 if (!XFS_DFORK_Q(dip))
424 ASSERT(ip->i_afp == NULL);
425 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP | KM_NOFS);
426 ip->i_afp->if_ext_max =
427 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
428 switch (dip->di_aformat) {
429 case XFS_DINODE_FMT_LOCAL:
430 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
431 size = be16_to_cpu(atp->hdr.totsize);
433 if (unlikely(size < sizeof(struct xfs_attr_sf_hdr))) {
434 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
436 "(bad attr fork size %Ld).",
437 (unsigned long long) ip->i_ino,
439 XFS_CORRUPTION_ERROR("xfs_iformat(8)",
442 return XFS_ERROR(EFSCORRUPTED);
445 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
447 case XFS_DINODE_FMT_EXTENTS:
448 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
450 case XFS_DINODE_FMT_BTREE:
451 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
454 error = XFS_ERROR(EFSCORRUPTED);
458 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
460 xfs_idestroy_fork(ip, XFS_DATA_FORK);
466 * The file is in-lined in the on-disk inode.
467 * If it fits into if_inline_data, then copy
468 * it there, otherwise allocate a buffer for it
469 * and copy the data there. Either way, set
470 * if_data to point at the data.
471 * If we allocate a buffer for the data, make
472 * sure that its size is a multiple of 4 and
473 * record the real size in i_real_bytes.
486 * If the size is unreasonable, then something
487 * is wrong and we just bail out rather than crash in
488 * kmem_alloc() or memcpy() below.
490 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
491 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
493 "(bad size %d for local fork, size = %d).",
494 (unsigned long long) ip->i_ino, size,
495 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
496 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
498 return XFS_ERROR(EFSCORRUPTED);
500 ifp = XFS_IFORK_PTR(ip, whichfork);
503 ifp->if_u1.if_data = NULL;
504 else if (size <= sizeof(ifp->if_u2.if_inline_data))
505 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
507 real_size = roundup(size, 4);
508 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP | KM_NOFS);
510 ifp->if_bytes = size;
511 ifp->if_real_bytes = real_size;
513 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
514 ifp->if_flags &= ~XFS_IFEXTENTS;
515 ifp->if_flags |= XFS_IFINLINE;
520 * The file consists of a set of extents all
521 * of which fit into the on-disk inode.
522 * If there are few enough extents to fit into
523 * the if_inline_ext, then copy them there.
524 * Otherwise allocate a buffer for them and copy
525 * them into it. Either way, set if_extents
526 * to point at the extents.
540 ifp = XFS_IFORK_PTR(ip, whichfork);
541 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
542 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
545 * If the number of extents is unreasonable, then something
546 * is wrong and we just bail out rather than crash in
547 * kmem_alloc() or memcpy() below.
549 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
550 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
551 "corrupt inode %Lu ((a)extents = %d).",
552 (unsigned long long) ip->i_ino, nex);
553 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
555 return XFS_ERROR(EFSCORRUPTED);
558 ifp->if_real_bytes = 0;
560 ifp->if_u1.if_extents = NULL;
561 else if (nex <= XFS_INLINE_EXTS)
562 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
564 xfs_iext_add(ifp, 0, nex);
566 ifp->if_bytes = size;
568 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
569 xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip));
570 for (i = 0; i < nex; i++, dp++) {
571 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
572 ep->l0 = get_unaligned_be64(&dp->l0);
573 ep->l1 = get_unaligned_be64(&dp->l1);
575 XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork);
576 if (whichfork != XFS_DATA_FORK ||
577 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
578 if (unlikely(xfs_check_nostate_extents(
580 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
583 return XFS_ERROR(EFSCORRUPTED);
586 ifp->if_flags |= XFS_IFEXTENTS;
591 * The file has too many extents to fit into
592 * the inode, so they are in B-tree format.
593 * Allocate a buffer for the root of the B-tree
594 * and copy the root into it. The i_extents
595 * field will remain NULL until all of the
596 * extents are read in (when they are needed).
604 xfs_bmdr_block_t *dfp;
610 ifp = XFS_IFORK_PTR(ip, whichfork);
611 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
612 size = XFS_BMAP_BROOT_SPACE(dfp);
613 nrecs = be16_to_cpu(dfp->bb_numrecs);
616 * blow out if -- fork has less extents than can fit in
617 * fork (fork shouldn't be a btree format), root btree
618 * block has more records than can fit into the fork,
619 * or the number of extents is greater than the number of
622 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
623 || XFS_BMDR_SPACE_CALC(nrecs) >
624 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
625 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
626 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
627 "corrupt inode %Lu (btree).",
628 (unsigned long long) ip->i_ino);
629 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
631 return XFS_ERROR(EFSCORRUPTED);
634 ifp->if_broot_bytes = size;
635 ifp->if_broot = kmem_alloc(size, KM_SLEEP | KM_NOFS);
636 ASSERT(ifp->if_broot != NULL);
638 * Copy and convert from the on-disk structure
639 * to the in-memory structure.
641 xfs_bmdr_to_bmbt(ip->i_mount, dfp,
642 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
643 ifp->if_broot, size);
644 ifp->if_flags &= ~XFS_IFEXTENTS;
645 ifp->if_flags |= XFS_IFBROOT;
651 xfs_dinode_from_disk(
655 to->di_magic = be16_to_cpu(from->di_magic);
656 to->di_mode = be16_to_cpu(from->di_mode);
657 to->di_version = from ->di_version;
658 to->di_format = from->di_format;
659 to->di_onlink = be16_to_cpu(from->di_onlink);
660 to->di_uid = be32_to_cpu(from->di_uid);
661 to->di_gid = be32_to_cpu(from->di_gid);
662 to->di_nlink = be32_to_cpu(from->di_nlink);
663 to->di_projid = be16_to_cpu(from->di_projid);
664 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
665 to->di_flushiter = be16_to_cpu(from->di_flushiter);
666 to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec);
667 to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec);
668 to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec);
669 to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec);
670 to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec);
671 to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec);
672 to->di_size = be64_to_cpu(from->di_size);
673 to->di_nblocks = be64_to_cpu(from->di_nblocks);
674 to->di_extsize = be32_to_cpu(from->di_extsize);
675 to->di_nextents = be32_to_cpu(from->di_nextents);
676 to->di_anextents = be16_to_cpu(from->di_anextents);
677 to->di_forkoff = from->di_forkoff;
678 to->di_aformat = from->di_aformat;
679 to->di_dmevmask = be32_to_cpu(from->di_dmevmask);
680 to->di_dmstate = be16_to_cpu(from->di_dmstate);
681 to->di_flags = be16_to_cpu(from->di_flags);
682 to->di_gen = be32_to_cpu(from->di_gen);
688 xfs_icdinode_t *from)
690 to->di_magic = cpu_to_be16(from->di_magic);
691 to->di_mode = cpu_to_be16(from->di_mode);
692 to->di_version = from ->di_version;
693 to->di_format = from->di_format;
694 to->di_onlink = cpu_to_be16(from->di_onlink);
695 to->di_uid = cpu_to_be32(from->di_uid);
696 to->di_gid = cpu_to_be32(from->di_gid);
697 to->di_nlink = cpu_to_be32(from->di_nlink);
698 to->di_projid = cpu_to_be16(from->di_projid);
699 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
700 to->di_flushiter = cpu_to_be16(from->di_flushiter);
701 to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec);
702 to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec);
703 to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec);
704 to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec);
705 to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec);
706 to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec);
707 to->di_size = cpu_to_be64(from->di_size);
708 to->di_nblocks = cpu_to_be64(from->di_nblocks);
709 to->di_extsize = cpu_to_be32(from->di_extsize);
710 to->di_nextents = cpu_to_be32(from->di_nextents);
711 to->di_anextents = cpu_to_be16(from->di_anextents);
712 to->di_forkoff = from->di_forkoff;
713 to->di_aformat = from->di_aformat;
714 to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
715 to->di_dmstate = cpu_to_be16(from->di_dmstate);
716 to->di_flags = cpu_to_be16(from->di_flags);
717 to->di_gen = cpu_to_be32(from->di_gen);
726 if (di_flags & XFS_DIFLAG_ANY) {
727 if (di_flags & XFS_DIFLAG_REALTIME)
728 flags |= XFS_XFLAG_REALTIME;
729 if (di_flags & XFS_DIFLAG_PREALLOC)
730 flags |= XFS_XFLAG_PREALLOC;
731 if (di_flags & XFS_DIFLAG_IMMUTABLE)
732 flags |= XFS_XFLAG_IMMUTABLE;
733 if (di_flags & XFS_DIFLAG_APPEND)
734 flags |= XFS_XFLAG_APPEND;
735 if (di_flags & XFS_DIFLAG_SYNC)
736 flags |= XFS_XFLAG_SYNC;
737 if (di_flags & XFS_DIFLAG_NOATIME)
738 flags |= XFS_XFLAG_NOATIME;
739 if (di_flags & XFS_DIFLAG_NODUMP)
740 flags |= XFS_XFLAG_NODUMP;
741 if (di_flags & XFS_DIFLAG_RTINHERIT)
742 flags |= XFS_XFLAG_RTINHERIT;
743 if (di_flags & XFS_DIFLAG_PROJINHERIT)
744 flags |= XFS_XFLAG_PROJINHERIT;
745 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
746 flags |= XFS_XFLAG_NOSYMLINKS;
747 if (di_flags & XFS_DIFLAG_EXTSIZE)
748 flags |= XFS_XFLAG_EXTSIZE;
749 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
750 flags |= XFS_XFLAG_EXTSZINHERIT;
751 if (di_flags & XFS_DIFLAG_NODEFRAG)
752 flags |= XFS_XFLAG_NODEFRAG;
753 if (di_flags & XFS_DIFLAG_FILESTREAM)
754 flags |= XFS_XFLAG_FILESTREAM;
764 xfs_icdinode_t *dic = &ip->i_d;
766 return _xfs_dic2xflags(dic->di_flags) |
767 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
774 return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) |
775 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
779 * Read the disk inode attributes into the in-core inode structure.
793 * Fill in the location information in the in-core inode.
795 error = xfs_imap(mp, tp, ip->i_ino, &ip->i_imap, iget_flags);
800 * Get pointers to the on-disk inode and the buffer containing it.
802 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp,
803 XBF_LOCK, iget_flags);
806 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
809 * If we got something that isn't an inode it means someone
810 * (nfs or dmi) has a stale handle.
812 if (be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC) {
814 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
815 "dip->di_magic (0x%x) != "
816 "XFS_DINODE_MAGIC (0x%x)",
817 be16_to_cpu(dip->di_magic),
820 error = XFS_ERROR(EINVAL);
825 * If the on-disk inode is already linked to a directory
826 * entry, copy all of the inode into the in-core inode.
827 * xfs_iformat() handles copying in the inode format
828 * specific information.
829 * Otherwise, just get the truly permanent information.
832 xfs_dinode_from_disk(&ip->i_d, dip);
833 error = xfs_iformat(ip, dip);
836 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
837 "xfs_iformat() returned error %d",
843 ip->i_d.di_magic = be16_to_cpu(dip->di_magic);
844 ip->i_d.di_version = dip->di_version;
845 ip->i_d.di_gen = be32_to_cpu(dip->di_gen);
846 ip->i_d.di_flushiter = be16_to_cpu(dip->di_flushiter);
848 * Make sure to pull in the mode here as well in
849 * case the inode is released without being used.
850 * This ensures that xfs_inactive() will see that
851 * the inode is already free and not try to mess
852 * with the uninitialized part of it.
856 * Initialize the per-fork minima and maxima for a new
857 * inode here. xfs_iformat will do it for old inodes.
859 ip->i_df.if_ext_max =
860 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
864 * The inode format changed when we moved the link count and
865 * made it 32 bits long. If this is an old format inode,
866 * convert it in memory to look like a new one. If it gets
867 * flushed to disk we will convert back before flushing or
868 * logging it. We zero out the new projid field and the old link
869 * count field. We'll handle clearing the pad field (the remains
870 * of the old uuid field) when we actually convert the inode to
871 * the new format. We don't change the version number so that we
872 * can distinguish this from a real new format inode.
874 if (ip->i_d.di_version == 1) {
875 ip->i_d.di_nlink = ip->i_d.di_onlink;
876 ip->i_d.di_onlink = 0;
877 ip->i_d.di_projid = 0;
880 ip->i_delayed_blks = 0;
881 ip->i_size = ip->i_d.di_size;
884 * Mark the buffer containing the inode as something to keep
885 * around for a while. This helps to keep recently accessed
886 * meta-data in-core longer.
888 XFS_BUF_SET_REF(bp, XFS_INO_REF);
891 * Use xfs_trans_brelse() to release the buffer containing the
892 * on-disk inode, because it was acquired with xfs_trans_read_buf()
893 * in xfs_itobp() above. If tp is NULL, this is just a normal
894 * brelse(). If we're within a transaction, then xfs_trans_brelse()
895 * will only release the buffer if it is not dirty within the
896 * transaction. It will be OK to release the buffer in this case,
897 * because inodes on disk are never destroyed and we will be
898 * locking the new in-core inode before putting it in the hash
899 * table where other processes can find it. Thus we don't have
900 * to worry about the inode being changed just because we released
904 xfs_trans_brelse(tp, bp);
909 * Read in extents from a btree-format inode.
910 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
920 xfs_extnum_t nextents;
922 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
923 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
925 return XFS_ERROR(EFSCORRUPTED);
927 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
928 ifp = XFS_IFORK_PTR(ip, whichfork);
931 * We know that the size is valid (it's checked in iformat_btree)
933 ifp->if_lastex = NULLEXTNUM;
934 ifp->if_bytes = ifp->if_real_bytes = 0;
935 ifp->if_flags |= XFS_IFEXTENTS;
936 xfs_iext_add(ifp, 0, nextents);
937 error = xfs_bmap_read_extents(tp, ip, whichfork);
939 xfs_iext_destroy(ifp);
940 ifp->if_flags &= ~XFS_IFEXTENTS;
943 xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip));
948 * Allocate an inode on disk and return a copy of its in-core version.
949 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
950 * appropriately within the inode. The uid and gid for the inode are
951 * set according to the contents of the given cred structure.
953 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
954 * has a free inode available, call xfs_iget()
955 * to obtain the in-core version of the allocated inode. Finally,
956 * fill in the inode and log its initial contents. In this case,
957 * ialloc_context would be set to NULL and call_again set to false.
959 * If xfs_dialloc() does not have an available inode,
960 * it will replenish its supply by doing an allocation. Since we can
961 * only do one allocation within a transaction without deadlocks, we
962 * must commit the current transaction before returning the inode itself.
963 * In this case, therefore, we will set call_again to true and return.
964 * The caller should then commit the current transaction, start a new
965 * transaction, and call xfs_ialloc() again to actually get the inode.
967 * To ensure that some other process does not grab the inode that
968 * was allocated during the first call to xfs_ialloc(), this routine
969 * also returns the [locked] bp pointing to the head of the freelist
970 * as ialloc_context. The caller should hold this buffer across
971 * the commit and pass it back into this routine on the second call.
973 * If we are allocating quota inodes, we do not have a parent inode
974 * to attach to or associate with (i.e. pip == NULL) because they
975 * are not linked into the directory structure - they are attached
976 * directly to the superblock - and so have no parent.
987 xfs_buf_t **ialloc_context,
988 boolean_t *call_again,
999 * Call the space management code to pick
1000 * the on-disk inode to be allocated.
1002 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
1003 ialloc_context, call_again, &ino);
1006 if (*call_again || ino == NULLFSINO) {
1010 ASSERT(*ialloc_context == NULL);
1013 * Get the in-core inode with the lock held exclusively.
1014 * This is because we're setting fields here we need
1015 * to prevent others from looking at until we're done.
1017 error = xfs_trans_iget(tp->t_mountp, tp, ino,
1018 XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
1023 ip->i_d.di_mode = (__uint16_t)mode;
1024 ip->i_d.di_onlink = 0;
1025 ip->i_d.di_nlink = nlink;
1026 ASSERT(ip->i_d.di_nlink == nlink);
1027 ip->i_d.di_uid = current_fsuid();
1028 ip->i_d.di_gid = current_fsgid();
1029 ip->i_d.di_projid = prid;
1030 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1033 * If the superblock version is up to where we support new format
1034 * inodes and this is currently an old format inode, then change
1035 * the inode version number now. This way we only do the conversion
1036 * here rather than here and in the flush/logging code.
1038 if (xfs_sb_version_hasnlink(&tp->t_mountp->m_sb) &&
1039 ip->i_d.di_version == 1) {
1040 ip->i_d.di_version = 2;
1042 * We've already zeroed the old link count, the projid field,
1043 * and the pad field.
1048 * Project ids won't be stored on disk if we are using a version 1 inode.
1050 if ((prid != 0) && (ip->i_d.di_version == 1))
1051 xfs_bump_ino_vers2(tp, ip);
1053 if (pip && XFS_INHERIT_GID(pip)) {
1054 ip->i_d.di_gid = pip->i_d.di_gid;
1055 if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1056 ip->i_d.di_mode |= S_ISGID;
1061 * If the group ID of the new file does not match the effective group
1062 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1063 * (and only if the irix_sgid_inherit compatibility variable is set).
1065 if ((irix_sgid_inherit) &&
1066 (ip->i_d.di_mode & S_ISGID) &&
1067 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1068 ip->i_d.di_mode &= ~S_ISGID;
1071 ip->i_d.di_size = 0;
1073 ip->i_d.di_nextents = 0;
1074 ASSERT(ip->i_d.di_nblocks == 0);
1077 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
1078 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
1079 ip->i_d.di_atime = ip->i_d.di_mtime;
1080 ip->i_d.di_ctime = ip->i_d.di_mtime;
1083 * di_gen will have been taken care of in xfs_iread.
1085 ip->i_d.di_extsize = 0;
1086 ip->i_d.di_dmevmask = 0;
1087 ip->i_d.di_dmstate = 0;
1088 ip->i_d.di_flags = 0;
1089 flags = XFS_ILOG_CORE;
1090 switch (mode & S_IFMT) {
1095 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1096 ip->i_df.if_u2.if_rdev = rdev;
1097 ip->i_df.if_flags = 0;
1098 flags |= XFS_ILOG_DEV;
1102 * we can't set up filestreams until after the VFS inode
1103 * is set up properly.
1105 if (pip && xfs_inode_is_filestream(pip))
1109 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
1112 if ((mode & S_IFMT) == S_IFDIR) {
1113 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1114 di_flags |= XFS_DIFLAG_RTINHERIT;
1115 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1116 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1117 ip->i_d.di_extsize = pip->i_d.di_extsize;
1119 } else if ((mode & S_IFMT) == S_IFREG) {
1120 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1121 di_flags |= XFS_DIFLAG_REALTIME;
1122 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1123 di_flags |= XFS_DIFLAG_EXTSIZE;
1124 ip->i_d.di_extsize = pip->i_d.di_extsize;
1127 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1128 xfs_inherit_noatime)
1129 di_flags |= XFS_DIFLAG_NOATIME;
1130 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1132 di_flags |= XFS_DIFLAG_NODUMP;
1133 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1135 di_flags |= XFS_DIFLAG_SYNC;
1136 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1137 xfs_inherit_nosymlinks)
1138 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1139 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1140 di_flags |= XFS_DIFLAG_PROJINHERIT;
1141 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1142 xfs_inherit_nodefrag)
1143 di_flags |= XFS_DIFLAG_NODEFRAG;
1144 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
1145 di_flags |= XFS_DIFLAG_FILESTREAM;
1146 ip->i_d.di_flags |= di_flags;
1150 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1151 ip->i_df.if_flags = XFS_IFEXTENTS;
1152 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1153 ip->i_df.if_u1.if_extents = NULL;
1159 * Attribute fork settings for new inode.
1161 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1162 ip->i_d.di_anextents = 0;
1165 * Log the new values stuffed into the inode.
1167 xfs_trans_log_inode(tp, ip, flags);
1169 /* now that we have an i_mode we can setup inode ops and unlock */
1170 xfs_setup_inode(ip);
1172 /* now we have set up the vfs inode we can associate the filestream */
1174 error = xfs_filestream_associate(pip, ip);
1178 xfs_iflags_set(ip, XFS_IFILESTREAM);
1186 * Check to make sure that there are no blocks allocated to the
1187 * file beyond the size of the file. We don't check this for
1188 * files with fixed size extents or real time extents, but we
1189 * at least do it for regular files.
1198 xfs_fileoff_t map_first;
1200 xfs_bmbt_irec_t imaps[2];
1202 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1205 if (XFS_IS_REALTIME_INODE(ip))
1208 if (ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE)
1212 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1214 * The filesystem could be shutting down, so bmapi may return
1217 if (xfs_bmapi(NULL, ip, map_first,
1219 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1221 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1224 ASSERT(nimaps == 1);
1225 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1230 * Calculate the last possible buffered byte in a file. This must
1231 * include data that was buffered beyond the EOF by the write code.
1232 * This also needs to deal with overflowing the xfs_fsize_t type
1233 * which can happen for sizes near the limit.
1235 * We also need to take into account any blocks beyond the EOF. It
1236 * may be the case that they were buffered by a write which failed.
1237 * In that case the pages will still be in memory, but the inode size
1238 * will never have been updated.
1245 xfs_fsize_t last_byte;
1246 xfs_fileoff_t last_block;
1247 xfs_fileoff_t size_last_block;
1250 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED));
1254 * Only check for blocks beyond the EOF if the extents have
1255 * been read in. This eliminates the need for the inode lock,
1256 * and it also saves us from looking when it really isn't
1259 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1260 xfs_ilock(ip, XFS_ILOCK_SHARED);
1261 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1263 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1270 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_size);
1271 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1273 last_byte = XFS_FSB_TO_B(mp, last_block);
1274 if (last_byte < 0) {
1275 return XFS_MAXIOFFSET(mp);
1277 last_byte += (1 << mp->m_writeio_log);
1278 if (last_byte < 0) {
1279 return XFS_MAXIOFFSET(mp);
1285 * Start the truncation of the file to new_size. The new size
1286 * must be smaller than the current size. This routine will
1287 * clear the buffer and page caches of file data in the removed
1288 * range, and xfs_itruncate_finish() will remove the underlying
1291 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1292 * must NOT have the inode lock held at all. This is because we're
1293 * calling into the buffer/page cache code and we can't hold the
1294 * inode lock when we do so.
1296 * We need to wait for any direct I/Os in flight to complete before we
1297 * proceed with the truncate. This is needed to prevent the extents
1298 * being read or written by the direct I/Os from being removed while the
1299 * I/O is in flight as there is no other method of synchronising
1300 * direct I/O with the truncate operation. Also, because we hold
1301 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1302 * started until the truncate completes and drops the lock. Essentially,
1303 * the xfs_ioend_wait() call forms an I/O barrier that provides strict
1304 * ordering between direct I/Os and the truncate operation.
1306 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1307 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1308 * in the case that the caller is locking things out of order and
1309 * may not be able to call xfs_itruncate_finish() with the inode lock
1310 * held without dropping the I/O lock. If the caller must drop the
1311 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1312 * must be called again with all the same restrictions as the initial
1316 xfs_itruncate_start(
1319 xfs_fsize_t new_size)
1321 xfs_fsize_t last_byte;
1322 xfs_off_t toss_start;
1326 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1327 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1328 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1329 (flags == XFS_ITRUNC_MAYBE));
1333 /* wait for the completion of any pending DIOs */
1334 if (new_size == 0 || new_size < ip->i_size)
1338 * Call toss_pages or flushinval_pages to get rid of pages
1339 * overlapping the region being removed. We have to use
1340 * the less efficient flushinval_pages in the case that the
1341 * caller may not be able to finish the truncate without
1342 * dropping the inode's I/O lock. Make sure
1343 * to catch any pages brought in by buffers overlapping
1344 * the EOF by searching out beyond the isize by our
1345 * block size. We round new_size up to a block boundary
1346 * so that we don't toss things on the same block as
1347 * new_size but before it.
1349 * Before calling toss_page or flushinval_pages, make sure to
1350 * call remapf() over the same region if the file is mapped.
1351 * This frees up mapped file references to the pages in the
1352 * given range and for the flushinval_pages case it ensures
1353 * that we get the latest mapped changes flushed out.
1355 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1356 toss_start = XFS_FSB_TO_B(mp, toss_start);
1357 if (toss_start < 0) {
1359 * The place to start tossing is beyond our maximum
1360 * file size, so there is no way that the data extended
1365 last_byte = xfs_file_last_byte(ip);
1366 trace_xfs_itruncate_start(ip, flags, new_size, toss_start, last_byte);
1367 if (last_byte > toss_start) {
1368 if (flags & XFS_ITRUNC_DEFINITE) {
1369 xfs_tosspages(ip, toss_start,
1370 -1, FI_REMAPF_LOCKED);
1372 error = xfs_flushinval_pages(ip, toss_start,
1373 -1, FI_REMAPF_LOCKED);
1378 if (new_size == 0) {
1379 ASSERT(VN_CACHED(VFS_I(ip)) == 0);
1386 * Shrink the file to the given new_size. The new size must be smaller than
1387 * the current size. This will free up the underlying blocks in the removed
1388 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
1390 * The transaction passed to this routine must have made a permanent log
1391 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1392 * given transaction and start new ones, so make sure everything involved in
1393 * the transaction is tidy before calling here. Some transaction will be
1394 * returned to the caller to be committed. The incoming transaction must
1395 * already include the inode, and both inode locks must be held exclusively.
1396 * The inode must also be "held" within the transaction. On return the inode
1397 * will be "held" within the returned transaction. This routine does NOT
1398 * require any disk space to be reserved for it within the transaction.
1400 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1401 * indicates the fork which is to be truncated. For the attribute fork we only
1402 * support truncation to size 0.
1404 * We use the sync parameter to indicate whether or not the first transaction
1405 * we perform might have to be synchronous. For the attr fork, it needs to be
1406 * so if the unlink of the inode is not yet known to be permanent in the log.
1407 * This keeps us from freeing and reusing the blocks of the attribute fork
1408 * before the unlink of the inode becomes permanent.
1410 * For the data fork, we normally have to run synchronously if we're being
1411 * called out of the inactive path or we're being called out of the create path
1412 * where we're truncating an existing file. Either way, the truncate needs to
1413 * be sync so blocks don't reappear in the file with altered data in case of a
1414 * crash. wsync filesystems can run the first case async because anything that
1415 * shrinks the inode has to run sync so by the time we're called here from
1416 * inactive, the inode size is permanently set to 0.
1418 * Calls from the truncate path always need to be sync unless we're in a wsync
1419 * filesystem and the file has already been unlinked.
1421 * The caller is responsible for correctly setting the sync parameter. It gets
1422 * too hard for us to guess here which path we're being called out of just
1423 * based on inode state.
1425 * If we get an error, we must return with the inode locked and linked into the
1426 * current transaction. This keeps things simple for the higher level code,
1427 * because it always knows that the inode is locked and held in the transaction
1428 * that returns to it whether errors occur or not. We don't mark the inode
1429 * dirty on error so that transactions can be easily aborted if possible.
1432 xfs_itruncate_finish(
1435 xfs_fsize_t new_size,
1439 xfs_fsblock_t first_block;
1440 xfs_fileoff_t first_unmap_block;
1441 xfs_fileoff_t last_block;
1442 xfs_filblks_t unmap_len=0;
1447 xfs_bmap_free_t free_list;
1450 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
1451 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1452 ASSERT(*tp != NULL);
1453 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1454 ASSERT(ip->i_transp == *tp);
1455 ASSERT(ip->i_itemp != NULL);
1456 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1460 mp = (ntp)->t_mountp;
1461 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1464 * We only support truncating the entire attribute fork.
1466 if (fork == XFS_ATTR_FORK) {
1469 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1470 trace_xfs_itruncate_finish_start(ip, new_size);
1473 * The first thing we do is set the size to new_size permanently
1474 * on disk. This way we don't have to worry about anyone ever
1475 * being able to look at the data being freed even in the face
1476 * of a crash. What we're getting around here is the case where
1477 * we free a block, it is allocated to another file, it is written
1478 * to, and then we crash. If the new data gets written to the
1479 * file but the log buffers containing the free and reallocation
1480 * don't, then we'd end up with garbage in the blocks being freed.
1481 * As long as we make the new_size permanent before actually
1482 * freeing any blocks it doesn't matter if they get writtten to.
1484 * The callers must signal into us whether or not the size
1485 * setting here must be synchronous. There are a few cases
1486 * where it doesn't have to be synchronous. Those cases
1487 * occur if the file is unlinked and we know the unlink is
1488 * permanent or if the blocks being truncated are guaranteed
1489 * to be beyond the inode eof (regardless of the link count)
1490 * and the eof value is permanent. Both of these cases occur
1491 * only on wsync-mounted filesystems. In those cases, we're
1492 * guaranteed that no user will ever see the data in the blocks
1493 * that are being truncated so the truncate can run async.
1494 * In the free beyond eof case, the file may wind up with
1495 * more blocks allocated to it than it needs if we crash
1496 * and that won't get fixed until the next time the file
1497 * is re-opened and closed but that's ok as that shouldn't
1498 * be too many blocks.
1500 * However, we can't just make all wsync xactions run async
1501 * because there's one call out of the create path that needs
1502 * to run sync where it's truncating an existing file to size
1503 * 0 whose size is > 0.
1505 * It's probably possible to come up with a test in this
1506 * routine that would correctly distinguish all the above
1507 * cases from the values of the function parameters and the
1508 * inode state but for sanity's sake, I've decided to let the
1509 * layers above just tell us. It's simpler to correctly figure
1510 * out in the layer above exactly under what conditions we
1511 * can run async and I think it's easier for others read and
1512 * follow the logic in case something has to be changed.
1513 * cscope is your friend -- rcc.
1515 * The attribute fork is much simpler.
1517 * For the attribute fork we allow the caller to tell us whether
1518 * the unlink of the inode that led to this call is yet permanent
1519 * in the on disk log. If it is not and we will be freeing extents
1520 * in this inode then we make the first transaction synchronous
1521 * to make sure that the unlink is permanent by the time we free
1524 if (fork == XFS_DATA_FORK) {
1525 if (ip->i_d.di_nextents > 0) {
1527 * If we are not changing the file size then do
1528 * not update the on-disk file size - we may be
1529 * called from xfs_inactive_free_eofblocks(). If we
1530 * update the on-disk file size and then the system
1531 * crashes before the contents of the file are
1532 * flushed to disk then the files may be full of
1533 * holes (ie NULL files bug).
1535 if (ip->i_size != new_size) {
1536 ip->i_d.di_size = new_size;
1537 ip->i_size = new_size;
1538 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1542 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1543 if (ip->i_d.di_anextents > 0)
1544 xfs_trans_set_sync(ntp);
1546 ASSERT(fork == XFS_DATA_FORK ||
1547 (fork == XFS_ATTR_FORK &&
1548 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1549 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1552 * Since it is possible for space to become allocated beyond
1553 * the end of the file (in a crash where the space is allocated
1554 * but the inode size is not yet updated), simply remove any
1555 * blocks which show up between the new EOF and the maximum
1556 * possible file size. If the first block to be removed is
1557 * beyond the maximum file size (ie it is the same as last_block),
1558 * then there is nothing to do.
1560 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1561 ASSERT(first_unmap_block <= last_block);
1563 if (last_block == first_unmap_block) {
1566 unmap_len = last_block - first_unmap_block + 1;
1570 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1571 * will tell us whether it freed the entire range or
1572 * not. If this is a synchronous mount (wsync),
1573 * then we can tell bunmapi to keep all the
1574 * transactions asynchronous since the unlink
1575 * transaction that made this inode inactive has
1576 * already hit the disk. There's no danger of
1577 * the freed blocks being reused, there being a
1578 * crash, and the reused blocks suddenly reappearing
1579 * in this file with garbage in them once recovery
1582 xfs_bmap_init(&free_list, &first_block);
1583 error = xfs_bunmapi(ntp, ip,
1584 first_unmap_block, unmap_len,
1585 xfs_bmapi_aflag(fork),
1586 XFS_ITRUNC_MAX_EXTENTS,
1587 &first_block, &free_list,
1591 * If the bunmapi call encounters an error,
1592 * return to the caller where the transaction
1593 * can be properly aborted. We just need to
1594 * make sure we're not holding any resources
1595 * that we were not when we came in.
1597 xfs_bmap_cancel(&free_list);
1602 * Duplicate the transaction that has the permanent
1603 * reservation and commit the old transaction.
1605 error = xfs_bmap_finish(tp, &free_list, &committed);
1608 xfs_trans_ijoin(ntp, ip);
1612 * If the bmap finish call encounters an error, return
1613 * to the caller where the transaction can be properly
1614 * aborted. We just need to make sure we're not
1615 * holding any resources that we were not when we came
1618 * Aborting from this point might lose some blocks in
1619 * the file system, but oh well.
1621 xfs_bmap_cancel(&free_list);
1627 * Mark the inode dirty so it will be logged and
1628 * moved forward in the log as part of every commit.
1630 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1633 ntp = xfs_trans_dup(ntp);
1634 error = xfs_trans_commit(*tp, 0);
1637 xfs_trans_ijoin(ntp, ip);
1642 * transaction commit worked ok so we can drop the extra ticket
1643 * reference that we gained in xfs_trans_dup()
1645 xfs_log_ticket_put(ntp->t_ticket);
1646 error = xfs_trans_reserve(ntp, 0,
1647 XFS_ITRUNCATE_LOG_RES(mp), 0,
1648 XFS_TRANS_PERM_LOG_RES,
1649 XFS_ITRUNCATE_LOG_COUNT);
1654 * Only update the size in the case of the data fork, but
1655 * always re-log the inode so that our permanent transaction
1656 * can keep on rolling it forward in the log.
1658 if (fork == XFS_DATA_FORK) {
1659 xfs_isize_check(mp, ip, new_size);
1661 * If we are not changing the file size then do
1662 * not update the on-disk file size - we may be
1663 * called from xfs_inactive_free_eofblocks(). If we
1664 * update the on-disk file size and then the system
1665 * crashes before the contents of the file are
1666 * flushed to disk then the files may be full of
1667 * holes (ie NULL files bug).
1669 if (ip->i_size != new_size) {
1670 ip->i_d.di_size = new_size;
1671 ip->i_size = new_size;
1674 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1675 ASSERT((new_size != 0) ||
1676 (fork == XFS_ATTR_FORK) ||
1677 (ip->i_delayed_blks == 0));
1678 ASSERT((new_size != 0) ||
1679 (fork == XFS_ATTR_FORK) ||
1680 (ip->i_d.di_nextents == 0));
1681 trace_xfs_itruncate_finish_end(ip, new_size);
1686 * This is called when the inode's link count goes to 0.
1687 * We place the on-disk inode on a list in the AGI. It
1688 * will be pulled from this list when the inode is freed.
1705 ASSERT(ip->i_d.di_nlink == 0);
1706 ASSERT(ip->i_d.di_mode != 0);
1707 ASSERT(ip->i_transp == tp);
1712 * Get the agi buffer first. It ensures lock ordering
1715 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1718 agi = XFS_BUF_TO_AGI(agibp);
1721 * Get the index into the agi hash table for the
1722 * list this inode will go on.
1724 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1726 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1727 ASSERT(agi->agi_unlinked[bucket_index]);
1728 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1730 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
1732 * There is already another inode in the bucket we need
1733 * to add ourselves to. Add us at the front of the list.
1734 * Here we put the head pointer into our next pointer,
1735 * and then we fall through to point the head at us.
1737 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XBF_LOCK);
1741 ASSERT(be32_to_cpu(dip->di_next_unlinked) == NULLAGINO);
1742 /* both on-disk, don't endian flip twice */
1743 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1744 offset = ip->i_imap.im_boffset +
1745 offsetof(xfs_dinode_t, di_next_unlinked);
1746 xfs_trans_inode_buf(tp, ibp);
1747 xfs_trans_log_buf(tp, ibp, offset,
1748 (offset + sizeof(xfs_agino_t) - 1));
1749 xfs_inobp_check(mp, ibp);
1753 * Point the bucket head pointer at the inode being inserted.
1756 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1757 offset = offsetof(xfs_agi_t, agi_unlinked) +
1758 (sizeof(xfs_agino_t) * bucket_index);
1759 xfs_trans_log_buf(tp, agibp, offset,
1760 (offset + sizeof(xfs_agino_t) - 1));
1765 * Pull the on-disk inode from the AGI unlinked list.
1778 xfs_agnumber_t agno;
1780 xfs_agino_t next_agino;
1781 xfs_buf_t *last_ibp;
1782 xfs_dinode_t *last_dip = NULL;
1784 int offset, last_offset = 0;
1788 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1791 * Get the agi buffer first. It ensures lock ordering
1794 error = xfs_read_agi(mp, tp, agno, &agibp);
1798 agi = XFS_BUF_TO_AGI(agibp);
1801 * Get the index into the agi hash table for the
1802 * list this inode will go on.
1804 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1806 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1807 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
1808 ASSERT(agi->agi_unlinked[bucket_index]);
1810 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
1812 * We're at the head of the list. Get the inode's
1813 * on-disk buffer to see if there is anyone after us
1814 * on the list. Only modify our next pointer if it
1815 * is not already NULLAGINO. This saves us the overhead
1816 * of dealing with the buffer when there is no need to
1819 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XBF_LOCK);
1822 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
1823 error, mp->m_fsname);
1826 next_agino = be32_to_cpu(dip->di_next_unlinked);
1827 ASSERT(next_agino != 0);
1828 if (next_agino != NULLAGINO) {
1829 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
1830 offset = ip->i_imap.im_boffset +
1831 offsetof(xfs_dinode_t, di_next_unlinked);
1832 xfs_trans_inode_buf(tp, ibp);
1833 xfs_trans_log_buf(tp, ibp, offset,
1834 (offset + sizeof(xfs_agino_t) - 1));
1835 xfs_inobp_check(mp, ibp);
1837 xfs_trans_brelse(tp, ibp);
1840 * Point the bucket head pointer at the next inode.
1842 ASSERT(next_agino != 0);
1843 ASSERT(next_agino != agino);
1844 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
1845 offset = offsetof(xfs_agi_t, agi_unlinked) +
1846 (sizeof(xfs_agino_t) * bucket_index);
1847 xfs_trans_log_buf(tp, agibp, offset,
1848 (offset + sizeof(xfs_agino_t) - 1));
1851 * We need to search the list for the inode being freed.
1853 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
1855 while (next_agino != agino) {
1857 * If the last inode wasn't the one pointing to
1858 * us, then release its buffer since we're not
1859 * going to do anything with it.
1861 if (last_ibp != NULL) {
1862 xfs_trans_brelse(tp, last_ibp);
1864 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
1865 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
1866 &last_ibp, &last_offset, 0);
1869 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
1870 error, mp->m_fsname);
1873 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
1874 ASSERT(next_agino != NULLAGINO);
1875 ASSERT(next_agino != 0);
1878 * Now last_ibp points to the buffer previous to us on
1879 * the unlinked list. Pull us from the list.
1881 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XBF_LOCK);
1884 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
1885 error, mp->m_fsname);
1888 next_agino = be32_to_cpu(dip->di_next_unlinked);
1889 ASSERT(next_agino != 0);
1890 ASSERT(next_agino != agino);
1891 if (next_agino != NULLAGINO) {
1892 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
1893 offset = ip->i_imap.im_boffset +
1894 offsetof(xfs_dinode_t, di_next_unlinked);
1895 xfs_trans_inode_buf(tp, ibp);
1896 xfs_trans_log_buf(tp, ibp, offset,
1897 (offset + sizeof(xfs_agino_t) - 1));
1898 xfs_inobp_check(mp, ibp);
1900 xfs_trans_brelse(tp, ibp);
1903 * Point the previous inode on the list to the next inode.
1905 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
1906 ASSERT(next_agino != 0);
1907 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
1908 xfs_trans_inode_buf(tp, last_ibp);
1909 xfs_trans_log_buf(tp, last_ibp, offset,
1910 (offset + sizeof(xfs_agino_t) - 1));
1911 xfs_inobp_check(mp, last_ibp);
1917 * A big issue when freeing the inode cluster is is that we _cannot_ skip any
1918 * inodes that are in memory - they all must be marked stale and attached to
1919 * the cluster buffer.
1923 xfs_inode_t *free_ip,
1927 xfs_mount_t *mp = free_ip->i_mount;
1928 int blks_per_cluster;
1935 xfs_inode_log_item_t *iip;
1936 xfs_log_item_t *lip;
1937 struct xfs_perag *pag;
1939 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
1940 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
1941 blks_per_cluster = 1;
1942 ninodes = mp->m_sb.sb_inopblock;
1943 nbufs = XFS_IALLOC_BLOCKS(mp);
1945 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
1946 mp->m_sb.sb_blocksize;
1947 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
1948 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
1951 for (j = 0; j < nbufs; j++, inum += ninodes) {
1952 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
1953 XFS_INO_TO_AGBNO(mp, inum));
1956 * We obtain and lock the backing buffer first in the process
1957 * here, as we have to ensure that any dirty inode that we
1958 * can't get the flush lock on is attached to the buffer.
1959 * If we scan the in-memory inodes first, then buffer IO can
1960 * complete before we get a lock on it, and hence we may fail
1961 * to mark all the active inodes on the buffer stale.
1963 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
1964 mp->m_bsize * blks_per_cluster,
1968 * Walk the inodes already attached to the buffer and mark them
1969 * stale. These will all have the flush locks held, so an
1970 * in-memory inode walk can't lock them. By marking them all
1971 * stale first, we will not attempt to lock them in the loop
1972 * below as the XFS_ISTALE flag will be set.
1974 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
1976 if (lip->li_type == XFS_LI_INODE) {
1977 iip = (xfs_inode_log_item_t *)lip;
1978 ASSERT(iip->ili_logged == 1);
1979 lip->li_cb = xfs_istale_done;
1980 xfs_trans_ail_copy_lsn(mp->m_ail,
1981 &iip->ili_flush_lsn,
1982 &iip->ili_item.li_lsn);
1983 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
1985 lip = lip->li_bio_list;
1990 * For each inode in memory attempt to add it to the inode
1991 * buffer and set it up for being staled on buffer IO
1992 * completion. This is safe as we've locked out tail pushing
1993 * and flushing by locking the buffer.
1995 * We have already marked every inode that was part of a
1996 * transaction stale above, which means there is no point in
1997 * even trying to lock them.
1999 for (i = 0; i < ninodes; i++) {
2001 read_lock(&pag->pag_ici_lock);
2002 ip = radix_tree_lookup(&pag->pag_ici_root,
2003 XFS_INO_TO_AGINO(mp, (inum + i)));
2005 /* Inode not in memory or stale, nothing to do */
2006 if (!ip || xfs_iflags_test(ip, XFS_ISTALE)) {
2007 read_unlock(&pag->pag_ici_lock);
2012 * Don't try to lock/unlock the current inode, but we
2013 * _cannot_ skip the other inodes that we did not find
2014 * in the list attached to the buffer and are not
2015 * already marked stale. If we can't lock it, back off
2018 if (ip != free_ip &&
2019 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2020 read_unlock(&pag->pag_ici_lock);
2024 read_unlock(&pag->pag_ici_lock);
2027 xfs_iflags_set(ip, XFS_ISTALE);
2030 * we don't need to attach clean inodes or those only
2031 * with unlogged changes (which we throw away, anyway).
2034 if (!iip || xfs_inode_clean(ip)) {
2035 ASSERT(ip != free_ip);
2036 ip->i_update_core = 0;
2038 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2042 iip->ili_last_fields = iip->ili_format.ilf_fields;
2043 iip->ili_format.ilf_fields = 0;
2044 iip->ili_logged = 1;
2045 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2046 &iip->ili_item.li_lsn);
2048 xfs_buf_attach_iodone(bp, xfs_istale_done,
2052 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2055 xfs_trans_stale_inode_buf(tp, bp);
2056 xfs_trans_binval(tp, bp);
2063 * This is called to return an inode to the inode free list.
2064 * The inode should already be truncated to 0 length and have
2065 * no pages associated with it. This routine also assumes that
2066 * the inode is already a part of the transaction.
2068 * The on-disk copy of the inode will have been added to the list
2069 * of unlinked inodes in the AGI. We need to remove the inode from
2070 * that list atomically with respect to freeing it here.
2076 xfs_bmap_free_t *flist)
2080 xfs_ino_t first_ino;
2084 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2085 ASSERT(ip->i_transp == tp);
2086 ASSERT(ip->i_d.di_nlink == 0);
2087 ASSERT(ip->i_d.di_nextents == 0);
2088 ASSERT(ip->i_d.di_anextents == 0);
2089 ASSERT((ip->i_d.di_size == 0 && ip->i_size == 0) ||
2090 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2091 ASSERT(ip->i_d.di_nblocks == 0);
2094 * Pull the on-disk inode from the AGI unlinked list.
2096 error = xfs_iunlink_remove(tp, ip);
2101 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2105 ip->i_d.di_mode = 0; /* mark incore inode as free */
2106 ip->i_d.di_flags = 0;
2107 ip->i_d.di_dmevmask = 0;
2108 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2109 ip->i_df.if_ext_max =
2110 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2111 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2112 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2114 * Bump the generation count so no one will be confused
2115 * by reincarnations of this inode.
2119 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2121 error = xfs_itobp(ip->i_mount, tp, ip, &dip, &ibp, XBF_LOCK);
2126 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2127 * from picking up this inode when it is reclaimed (its incore state
2128 * initialzed but not flushed to disk yet). The in-core di_mode is
2129 * already cleared and a corresponding transaction logged.
2130 * The hack here just synchronizes the in-core to on-disk
2131 * di_mode value in advance before the actual inode sync to disk.
2132 * This is OK because the inode is already unlinked and would never
2133 * change its di_mode again for this inode generation.
2134 * This is a temporary hack that would require a proper fix
2140 xfs_ifree_cluster(ip, tp, first_ino);
2147 * Reallocate the space for if_broot based on the number of records
2148 * being added or deleted as indicated in rec_diff. Move the records
2149 * and pointers in if_broot to fit the new size. When shrinking this
2150 * will eliminate holes between the records and pointers created by
2151 * the caller. When growing this will create holes to be filled in
2154 * The caller must not request to add more records than would fit in
2155 * the on-disk inode root. If the if_broot is currently NULL, then
2156 * if we adding records one will be allocated. The caller must also
2157 * not request that the number of records go below zero, although
2158 * it can go to zero.
2160 * ip -- the inode whose if_broot area is changing
2161 * ext_diff -- the change in the number of records, positive or negative,
2162 * requested for the if_broot array.
2170 struct xfs_mount *mp = ip->i_mount;
2173 struct xfs_btree_block *new_broot;
2180 * Handle the degenerate case quietly.
2182 if (rec_diff == 0) {
2186 ifp = XFS_IFORK_PTR(ip, whichfork);
2189 * If there wasn't any memory allocated before, just
2190 * allocate it now and get out.
2192 if (ifp->if_broot_bytes == 0) {
2193 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
2194 ifp->if_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
2195 ifp->if_broot_bytes = (int)new_size;
2200 * If there is already an existing if_broot, then we need
2201 * to realloc() it and shift the pointers to their new
2202 * location. The records don't change location because
2203 * they are kept butted up against the btree block header.
2205 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
2206 new_max = cur_max + rec_diff;
2207 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2208 ifp->if_broot = kmem_realloc(ifp->if_broot, new_size,
2209 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2210 KM_SLEEP | KM_NOFS);
2211 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2212 ifp->if_broot_bytes);
2213 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2215 ifp->if_broot_bytes = (int)new_size;
2216 ASSERT(ifp->if_broot_bytes <=
2217 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2218 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2223 * rec_diff is less than 0. In this case, we are shrinking the
2224 * if_broot buffer. It must already exist. If we go to zero
2225 * records, just get rid of the root and clear the status bit.
2227 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
2228 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
2229 new_max = cur_max + rec_diff;
2230 ASSERT(new_max >= 0);
2232 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2236 new_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
2238 * First copy over the btree block header.
2240 memcpy(new_broot, ifp->if_broot, XFS_BTREE_LBLOCK_LEN);
2243 ifp->if_flags &= ~XFS_IFBROOT;
2247 * Only copy the records and pointers if there are any.
2251 * First copy the records.
2253 op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1);
2254 np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1);
2255 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2258 * Then copy the pointers.
2260 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2261 ifp->if_broot_bytes);
2262 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1,
2264 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2266 kmem_free(ifp->if_broot);
2267 ifp->if_broot = new_broot;
2268 ifp->if_broot_bytes = (int)new_size;
2269 ASSERT(ifp->if_broot_bytes <=
2270 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2276 * This is called when the amount of space needed for if_data
2277 * is increased or decreased. The change in size is indicated by
2278 * the number of bytes that need to be added or deleted in the
2279 * byte_diff parameter.
2281 * If the amount of space needed has decreased below the size of the
2282 * inline buffer, then switch to using the inline buffer. Otherwise,
2283 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2284 * to what is needed.
2286 * ip -- the inode whose if_data area is changing
2287 * byte_diff -- the change in the number of bytes, positive or negative,
2288 * requested for the if_data array.
2300 if (byte_diff == 0) {
2304 ifp = XFS_IFORK_PTR(ip, whichfork);
2305 new_size = (int)ifp->if_bytes + byte_diff;
2306 ASSERT(new_size >= 0);
2308 if (new_size == 0) {
2309 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2310 kmem_free(ifp->if_u1.if_data);
2312 ifp->if_u1.if_data = NULL;
2314 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2316 * If the valid extents/data can fit in if_inline_ext/data,
2317 * copy them from the malloc'd vector and free it.
2319 if (ifp->if_u1.if_data == NULL) {
2320 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2321 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2322 ASSERT(ifp->if_real_bytes != 0);
2323 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2325 kmem_free(ifp->if_u1.if_data);
2326 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2331 * Stuck with malloc/realloc.
2332 * For inline data, the underlying buffer must be
2333 * a multiple of 4 bytes in size so that it can be
2334 * logged and stay on word boundaries. We enforce
2337 real_size = roundup(new_size, 4);
2338 if (ifp->if_u1.if_data == NULL) {
2339 ASSERT(ifp->if_real_bytes == 0);
2340 ifp->if_u1.if_data = kmem_alloc(real_size,
2341 KM_SLEEP | KM_NOFS);
2342 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2344 * Only do the realloc if the underlying size
2345 * is really changing.
2347 if (ifp->if_real_bytes != real_size) {
2348 ifp->if_u1.if_data =
2349 kmem_realloc(ifp->if_u1.if_data,
2352 KM_SLEEP | KM_NOFS);
2355 ASSERT(ifp->if_real_bytes == 0);
2356 ifp->if_u1.if_data = kmem_alloc(real_size,
2357 KM_SLEEP | KM_NOFS);
2358 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2362 ifp->if_real_bytes = real_size;
2363 ifp->if_bytes = new_size;
2364 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2374 ifp = XFS_IFORK_PTR(ip, whichfork);
2375 if (ifp->if_broot != NULL) {
2376 kmem_free(ifp->if_broot);
2377 ifp->if_broot = NULL;
2381 * If the format is local, then we can't have an extents
2382 * array so just look for an inline data array. If we're
2383 * not local then we may or may not have an extents list,
2384 * so check and free it up if we do.
2386 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2387 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2388 (ifp->if_u1.if_data != NULL)) {
2389 ASSERT(ifp->if_real_bytes != 0);
2390 kmem_free(ifp->if_u1.if_data);
2391 ifp->if_u1.if_data = NULL;
2392 ifp->if_real_bytes = 0;
2394 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2395 ((ifp->if_flags & XFS_IFEXTIREC) ||
2396 ((ifp->if_u1.if_extents != NULL) &&
2397 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
2398 ASSERT(ifp->if_real_bytes != 0);
2399 xfs_iext_destroy(ifp);
2401 ASSERT(ifp->if_u1.if_extents == NULL ||
2402 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2403 ASSERT(ifp->if_real_bytes == 0);
2404 if (whichfork == XFS_ATTR_FORK) {
2405 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2411 * This is called to unpin an inode. The caller must have the inode locked
2412 * in at least shared mode so that the buffer cannot be subsequently pinned
2413 * once someone is waiting for it to be unpinned.
2417 struct xfs_inode *ip)
2419 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2421 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2423 /* Give the log a push to start the unpinning I/O */
2424 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2430 struct xfs_inode *ip)
2432 if (xfs_ipincount(ip)) {
2433 xfs_iunpin_nowait(ip);
2434 wait_event(ip->i_ipin_wait, (xfs_ipincount(ip) == 0));
2439 * xfs_iextents_copy()
2441 * This is called to copy the REAL extents (as opposed to the delayed
2442 * allocation extents) from the inode into the given buffer. It
2443 * returns the number of bytes copied into the buffer.
2445 * If there are no delayed allocation extents, then we can just
2446 * memcpy() the extents into the buffer. Otherwise, we need to
2447 * examine each extent in turn and skip those which are delayed.
2459 xfs_fsblock_t start_block;
2461 ifp = XFS_IFORK_PTR(ip, whichfork);
2462 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2463 ASSERT(ifp->if_bytes > 0);
2465 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2466 XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork);
2470 * There are some delayed allocation extents in the
2471 * inode, so copy the extents one at a time and skip
2472 * the delayed ones. There must be at least one
2473 * non-delayed extent.
2476 for (i = 0; i < nrecs; i++) {
2477 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
2478 start_block = xfs_bmbt_get_startblock(ep);
2479 if (isnullstartblock(start_block)) {
2481 * It's a delayed allocation extent, so skip it.
2486 /* Translate to on disk format */
2487 put_unaligned(cpu_to_be64(ep->l0), &dp->l0);
2488 put_unaligned(cpu_to_be64(ep->l1), &dp->l1);
2492 ASSERT(copied != 0);
2493 xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip));
2495 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2499 * Each of the following cases stores data into the same region
2500 * of the on-disk inode, so only one of them can be valid at
2501 * any given time. While it is possible to have conflicting formats
2502 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2503 * in EXTENTS format, this can only happen when the fork has
2504 * changed formats after being modified but before being flushed.
2505 * In these cases, the format always takes precedence, because the
2506 * format indicates the current state of the fork.
2513 xfs_inode_log_item_t *iip,
2520 #ifdef XFS_TRANS_DEBUG
2523 static const short brootflag[2] =
2524 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2525 static const short dataflag[2] =
2526 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2527 static const short extflag[2] =
2528 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2532 ifp = XFS_IFORK_PTR(ip, whichfork);
2534 * This can happen if we gave up in iformat in an error path,
2535 * for the attribute fork.
2538 ASSERT(whichfork == XFS_ATTR_FORK);
2541 cp = XFS_DFORK_PTR(dip, whichfork);
2543 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2544 case XFS_DINODE_FMT_LOCAL:
2545 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2546 (ifp->if_bytes > 0)) {
2547 ASSERT(ifp->if_u1.if_data != NULL);
2548 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2549 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2553 case XFS_DINODE_FMT_EXTENTS:
2554 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2555 !(iip->ili_format.ilf_fields & extflag[whichfork]));
2556 ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
2557 (ifp->if_bytes == 0));
2558 ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
2559 (ifp->if_bytes > 0));
2560 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
2561 (ifp->if_bytes > 0)) {
2562 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2563 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2568 case XFS_DINODE_FMT_BTREE:
2569 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
2570 (ifp->if_broot_bytes > 0)) {
2571 ASSERT(ifp->if_broot != NULL);
2572 ASSERT(ifp->if_broot_bytes <=
2573 (XFS_IFORK_SIZE(ip, whichfork) +
2574 XFS_BROOT_SIZE_ADJ));
2575 xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes,
2576 (xfs_bmdr_block_t *)cp,
2577 XFS_DFORK_SIZE(dip, mp, whichfork));
2581 case XFS_DINODE_FMT_DEV:
2582 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
2583 ASSERT(whichfork == XFS_DATA_FORK);
2584 xfs_dinode_put_rdev(dip, ip->i_df.if_u2.if_rdev);
2588 case XFS_DINODE_FMT_UUID:
2589 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
2590 ASSERT(whichfork == XFS_DATA_FORK);
2591 memcpy(XFS_DFORK_DPTR(dip),
2592 &ip->i_df.if_u2.if_uuid,
2608 xfs_mount_t *mp = ip->i_mount;
2609 struct xfs_perag *pag;
2610 unsigned long first_index, mask;
2611 unsigned long inodes_per_cluster;
2613 xfs_inode_t **ilist;
2620 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2622 inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog;
2623 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
2624 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
2628 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
2629 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
2630 read_lock(&pag->pag_ici_lock);
2631 /* really need a gang lookup range call here */
2632 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
2633 first_index, inodes_per_cluster);
2637 for (i = 0; i < nr_found; i++) {
2641 /* if the inode lies outside this cluster, we're done. */
2642 if ((XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index)
2645 * Do an un-protected check to see if the inode is dirty and
2646 * is a candidate for flushing. These checks will be repeated
2647 * later after the appropriate locks are acquired.
2649 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
2653 * Try to get locks. If any are unavailable or it is pinned,
2654 * then this inode cannot be flushed and is skipped.
2657 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
2659 if (!xfs_iflock_nowait(iq)) {
2660 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2663 if (xfs_ipincount(iq)) {
2665 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2670 * arriving here means that this inode can be flushed. First
2671 * re-check that it's dirty before flushing.
2673 if (!xfs_inode_clean(iq)) {
2675 error = xfs_iflush_int(iq, bp);
2677 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2678 goto cluster_corrupt_out;
2684 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2688 XFS_STATS_INC(xs_icluster_flushcnt);
2689 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
2693 read_unlock(&pag->pag_ici_lock);
2700 cluster_corrupt_out:
2702 * Corruption detected in the clustering loop. Invalidate the
2703 * inode buffer and shut down the filesystem.
2705 read_unlock(&pag->pag_ici_lock);
2707 * Clean up the buffer. If it was B_DELWRI, just release it --
2708 * brelse can handle it with no problems. If not, shut down the
2709 * filesystem before releasing the buffer.
2711 bufwasdelwri = XFS_BUF_ISDELAYWRITE(bp);
2715 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2717 if (!bufwasdelwri) {
2719 * Just like incore_relse: if we have b_iodone functions,
2720 * mark the buffer as an error and call them. Otherwise
2721 * mark it as stale and brelse.
2723 if (XFS_BUF_IODONE_FUNC(bp)) {
2726 XFS_BUF_ERROR(bp,EIO);
2735 * Unlocks the flush lock
2737 xfs_iflush_abort(iq);
2740 return XFS_ERROR(EFSCORRUPTED);
2744 * xfs_iflush() will write a modified inode's changes out to the
2745 * inode's on disk home. The caller must have the inode lock held
2746 * in at least shared mode and the inode flush completion must be
2747 * active as well. The inode lock will still be held upon return from
2748 * the call and the caller is free to unlock it.
2749 * The inode flush will be completed when the inode reaches the disk.
2750 * The flags indicate how the inode's buffer should be written out.
2757 xfs_inode_log_item_t *iip;
2763 XFS_STATS_INC(xs_iflush_count);
2765 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2766 ASSERT(!completion_done(&ip->i_flush));
2767 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2768 ip->i_d.di_nextents > ip->i_df.if_ext_max);
2774 * We can't flush the inode until it is unpinned, so wait for it if we
2775 * are allowed to block. We know noone new can pin it, because we are
2776 * holding the inode lock shared and you need to hold it exclusively to
2779 * If we are not allowed to block, force the log out asynchronously so
2780 * that when we come back the inode will be unpinned. If other inodes
2781 * in the same cluster are dirty, they will probably write the inode
2782 * out for us if they occur after the log force completes.
2784 if (!(flags & SYNC_WAIT) && xfs_ipincount(ip)) {
2785 xfs_iunpin_nowait(ip);
2789 xfs_iunpin_wait(ip);
2792 * For stale inodes we cannot rely on the backing buffer remaining
2793 * stale in cache for the remaining life of the stale inode and so
2794 * xfs_itobp() below may give us a buffer that no longer contains
2795 * inodes below. We have to check this after ensuring the inode is
2796 * unpinned so that it is safe to reclaim the stale inode after the
2799 if (xfs_iflags_test(ip, XFS_ISTALE)) {
2805 * This may have been unpinned because the filesystem is shutting
2806 * down forcibly. If that's the case we must not write this inode
2807 * to disk, because the log record didn't make it to disk!
2809 if (XFS_FORCED_SHUTDOWN(mp)) {
2810 ip->i_update_core = 0;
2812 iip->ili_format.ilf_fields = 0;
2814 return XFS_ERROR(EIO);
2818 * Get the buffer containing the on-disk inode.
2820 error = xfs_itobp(mp, NULL, ip, &dip, &bp,
2821 (flags & SYNC_WAIT) ? XBF_LOCK : XBF_TRYLOCK);
2828 * First flush out the inode that xfs_iflush was called with.
2830 error = xfs_iflush_int(ip, bp);
2835 * If the buffer is pinned then push on the log now so we won't
2836 * get stuck waiting in the write for too long.
2838 if (XFS_BUF_ISPINNED(bp))
2839 xfs_log_force(mp, 0);
2843 * see if other inodes can be gathered into this write
2845 error = xfs_iflush_cluster(ip, bp);
2847 goto cluster_corrupt_out;
2849 if (flags & SYNC_WAIT)
2850 error = xfs_bwrite(mp, bp);
2852 xfs_bdwrite(mp, bp);
2857 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2858 cluster_corrupt_out:
2860 * Unlocks the flush lock
2862 xfs_iflush_abort(ip);
2863 return XFS_ERROR(EFSCORRUPTED);
2872 xfs_inode_log_item_t *iip;
2875 #ifdef XFS_TRANS_DEBUG
2879 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2880 ASSERT(!completion_done(&ip->i_flush));
2881 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2882 ip->i_d.di_nextents > ip->i_df.if_ext_max);
2887 /* set *dip = inode's place in the buffer */
2888 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
2891 * Clear i_update_core before copying out the data.
2892 * This is for coordination with our timestamp updates
2893 * that don't hold the inode lock. They will always
2894 * update the timestamps BEFORE setting i_update_core,
2895 * so if we clear i_update_core after they set it we
2896 * are guaranteed to see their updates to the timestamps.
2897 * I believe that this depends on strongly ordered memory
2898 * semantics, but we have that. We use the SYNCHRONIZE
2899 * macro to make sure that the compiler does not reorder
2900 * the i_update_core access below the data copy below.
2902 ip->i_update_core = 0;
2906 * Make sure to get the latest timestamps from the Linux inode.
2908 xfs_synchronize_times(ip);
2910 if (XFS_TEST_ERROR(be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC,
2911 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
2912 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
2913 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
2914 ip->i_ino, be16_to_cpu(dip->di_magic), dip);
2917 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
2918 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
2919 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
2920 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
2921 ip->i_ino, ip, ip->i_d.di_magic);
2924 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
2926 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
2927 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
2928 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
2929 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
2930 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
2934 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
2936 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
2937 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
2938 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
2939 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
2940 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
2941 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
2946 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
2947 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
2948 XFS_RANDOM_IFLUSH_5)) {
2949 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
2950 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
2952 ip->i_d.di_nextents + ip->i_d.di_anextents,
2957 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
2958 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
2959 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
2960 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
2961 ip->i_ino, ip->i_d.di_forkoff, ip);
2965 * bump the flush iteration count, used to detect flushes which
2966 * postdate a log record during recovery.
2969 ip->i_d.di_flushiter++;
2972 * Copy the dirty parts of the inode into the on-disk
2973 * inode. We always copy out the core of the inode,
2974 * because if the inode is dirty at all the core must
2977 xfs_dinode_to_disk(dip, &ip->i_d);
2979 /* Wrap, we never let the log put out DI_MAX_FLUSH */
2980 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
2981 ip->i_d.di_flushiter = 0;
2984 * If this is really an old format inode and the superblock version
2985 * has not been updated to support only new format inodes, then
2986 * convert back to the old inode format. If the superblock version
2987 * has been updated, then make the conversion permanent.
2989 ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
2990 if (ip->i_d.di_version == 1) {
2991 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
2995 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
2996 dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink);
2999 * The superblock version has already been bumped,
3000 * so just make the conversion to the new inode
3003 ip->i_d.di_version = 2;
3004 dip->di_version = 2;
3005 ip->i_d.di_onlink = 0;
3007 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3008 memset(&(dip->di_pad[0]), 0,
3009 sizeof(dip->di_pad));
3010 ASSERT(ip->i_d.di_projid == 0);
3014 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
3015 if (XFS_IFORK_Q(ip))
3016 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3017 xfs_inobp_check(mp, bp);
3020 * We've recorded everything logged in the inode, so we'd
3021 * like to clear the ilf_fields bits so we don't log and
3022 * flush things unnecessarily. However, we can't stop
3023 * logging all this information until the data we've copied
3024 * into the disk buffer is written to disk. If we did we might
3025 * overwrite the copy of the inode in the log with all the
3026 * data after re-logging only part of it, and in the face of
3027 * a crash we wouldn't have all the data we need to recover.
3029 * What we do is move the bits to the ili_last_fields field.
3030 * When logging the inode, these bits are moved back to the
3031 * ilf_fields field. In the xfs_iflush_done() routine we
3032 * clear ili_last_fields, since we know that the information
3033 * those bits represent is permanently on disk. As long as
3034 * the flush completes before the inode is logged again, then
3035 * both ilf_fields and ili_last_fields will be cleared.
3037 * We can play with the ilf_fields bits here, because the inode
3038 * lock must be held exclusively in order to set bits there
3039 * and the flush lock protects the ili_last_fields bits.
3040 * Set ili_logged so the flush done
3041 * routine can tell whether or not to look in the AIL.
3042 * Also, store the current LSN of the inode so that we can tell
3043 * whether the item has moved in the AIL from xfs_iflush_done().
3044 * In order to read the lsn we need the AIL lock, because
3045 * it is a 64 bit value that cannot be read atomically.
3047 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3048 iip->ili_last_fields = iip->ili_format.ilf_fields;
3049 iip->ili_format.ilf_fields = 0;
3050 iip->ili_logged = 1;
3052 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3053 &iip->ili_item.li_lsn);
3056 * Attach the function xfs_iflush_done to the inode's
3057 * buffer. This will remove the inode from the AIL
3058 * and unlock the inode's flush lock when the inode is
3059 * completely written to disk.
3061 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3063 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3064 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3067 * We're flushing an inode which is not in the AIL and has
3068 * not been logged but has i_update_core set. For this
3069 * case we can use a B_DELWRI flush and immediately drop
3070 * the inode flush lock because we can avoid the whole
3071 * AIL state thing. It's OK to drop the flush lock now,
3072 * because we've already locked the buffer and to do anything
3073 * you really need both.
3076 ASSERT(iip->ili_logged == 0);
3077 ASSERT(iip->ili_last_fields == 0);
3078 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3086 return XFS_ERROR(EFSCORRUPTED);
3090 * Return a pointer to the extent record at file index idx.
3092 xfs_bmbt_rec_host_t *
3094 xfs_ifork_t *ifp, /* inode fork pointer */
3095 xfs_extnum_t idx) /* index of target extent */
3098 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
3099 return ifp->if_u1.if_ext_irec->er_extbuf;
3100 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3101 xfs_ext_irec_t *erp; /* irec pointer */
3102 int erp_idx = 0; /* irec index */
3103 xfs_extnum_t page_idx = idx; /* ext index in target list */
3105 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3106 return &erp->er_extbuf[page_idx];
3107 } else if (ifp->if_bytes) {
3108 return &ifp->if_u1.if_extents[idx];
3115 * Insert new item(s) into the extent records for incore inode
3116 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3120 xfs_inode_t *ip, /* incore inode pointer */
3121 xfs_extnum_t idx, /* starting index of new items */
3122 xfs_extnum_t count, /* number of inserted items */
3123 xfs_bmbt_irec_t *new, /* items to insert */
3124 int state) /* type of extent conversion */
3126 xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
3127 xfs_extnum_t i; /* extent record index */
3129 trace_xfs_iext_insert(ip, idx, new, state, _RET_IP_);
3131 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3132 xfs_iext_add(ifp, idx, count);
3133 for (i = idx; i < idx + count; i++, new++)
3134 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new);
3138 * This is called when the amount of space required for incore file
3139 * extents needs to be increased. The ext_diff parameter stores the
3140 * number of new extents being added and the idx parameter contains
3141 * the extent index where the new extents will be added. If the new
3142 * extents are being appended, then we just need to (re)allocate and
3143 * initialize the space. Otherwise, if the new extents are being
3144 * inserted into the middle of the existing entries, a bit more work
3145 * is required to make room for the new extents to be inserted. The
3146 * caller is responsible for filling in the new extent entries upon
3151 xfs_ifork_t *ifp, /* inode fork pointer */
3152 xfs_extnum_t idx, /* index to begin adding exts */
3153 int ext_diff) /* number of extents to add */
3155 int byte_diff; /* new bytes being added */
3156 int new_size; /* size of extents after adding */
3157 xfs_extnum_t nextents; /* number of extents in file */
3159 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3160 ASSERT((idx >= 0) && (idx <= nextents));
3161 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
3162 new_size = ifp->if_bytes + byte_diff;
3164 * If the new number of extents (nextents + ext_diff)
3165 * fits inside the inode, then continue to use the inline
3168 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
3169 if (idx < nextents) {
3170 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
3171 &ifp->if_u2.if_inline_ext[idx],
3172 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3173 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
3175 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3176 ifp->if_real_bytes = 0;
3177 ifp->if_lastex = nextents + ext_diff;
3180 * Otherwise use a linear (direct) extent list.
3181 * If the extents are currently inside the inode,
3182 * xfs_iext_realloc_direct will switch us from
3183 * inline to direct extent allocation mode.
3185 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
3186 xfs_iext_realloc_direct(ifp, new_size);
3187 if (idx < nextents) {
3188 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
3189 &ifp->if_u1.if_extents[idx],
3190 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3191 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
3194 /* Indirection array */
3196 xfs_ext_irec_t *erp;
3200 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
3201 if (ifp->if_flags & XFS_IFEXTIREC) {
3202 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3204 xfs_iext_irec_init(ifp);
3205 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3206 erp = ifp->if_u1.if_ext_irec;
3208 /* Extents fit in target extent page */
3209 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3210 if (page_idx < erp->er_extcount) {
3211 memmove(&erp->er_extbuf[page_idx + ext_diff],
3212 &erp->er_extbuf[page_idx],
3213 (erp->er_extcount - page_idx) *
3214 sizeof(xfs_bmbt_rec_t));
3215 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3217 erp->er_extcount += ext_diff;
3218 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3220 /* Insert a new extent page */
3222 xfs_iext_add_indirect_multi(ifp,
3223 erp_idx, page_idx, ext_diff);
3226 * If extent(s) are being appended to the last page in
3227 * the indirection array and the new extent(s) don't fit
3228 * in the page, then erp is NULL and erp_idx is set to
3229 * the next index needed in the indirection array.
3232 int count = ext_diff;
3235 erp = xfs_iext_irec_new(ifp, erp_idx);
3236 erp->er_extcount = count;
3237 count -= MIN(count, (int)XFS_LINEAR_EXTS);
3244 ifp->if_bytes = new_size;
3248 * This is called when incore extents are being added to the indirection
3249 * array and the new extents do not fit in the target extent list. The
3250 * erp_idx parameter contains the irec index for the target extent list
3251 * in the indirection array, and the idx parameter contains the extent
3252 * index within the list. The number of extents being added is stored
3253 * in the count parameter.
3255 * |-------| |-------|
3256 * | | | | idx - number of extents before idx
3258 * | | | | count - number of extents being inserted at idx
3259 * |-------| |-------|
3260 * | count | | nex2 | nex2 - number of extents after idx + count
3261 * |-------| |-------|
3264 xfs_iext_add_indirect_multi(
3265 xfs_ifork_t *ifp, /* inode fork pointer */
3266 int erp_idx, /* target extent irec index */
3267 xfs_extnum_t idx, /* index within target list */
3268 int count) /* new extents being added */
3270 int byte_diff; /* new bytes being added */
3271 xfs_ext_irec_t *erp; /* pointer to irec entry */
3272 xfs_extnum_t ext_diff; /* number of extents to add */
3273 xfs_extnum_t ext_cnt; /* new extents still needed */
3274 xfs_extnum_t nex2; /* extents after idx + count */
3275 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3276 int nlists; /* number of irec's (lists) */
3278 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3279 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3280 nex2 = erp->er_extcount - idx;
3281 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3284 * Save second part of target extent list
3285 * (all extents past */
3287 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3288 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS);
3289 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3290 erp->er_extcount -= nex2;
3291 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3292 memset(&erp->er_extbuf[idx], 0, byte_diff);
3296 * Add the new extents to the end of the target
3297 * list, then allocate new irec record(s) and
3298 * extent buffer(s) as needed to store the rest
3299 * of the new extents.
3302 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3304 erp->er_extcount += ext_diff;
3305 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3306 ext_cnt -= ext_diff;
3310 erp = xfs_iext_irec_new(ifp, erp_idx);
3311 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3312 erp->er_extcount = ext_diff;
3313 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3314 ext_cnt -= ext_diff;
3317 /* Add nex2 extents back to indirection array */
3319 xfs_extnum_t ext_avail;
3322 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3323 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3326 * If nex2 extents fit in the current page, append
3327 * nex2_ep after the new extents.
3329 if (nex2 <= ext_avail) {
3330 i = erp->er_extcount;
3333 * Otherwise, check if space is available in the
3336 else if ((erp_idx < nlists - 1) &&
3337 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3338 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3341 /* Create a hole for nex2 extents */
3342 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3343 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3346 * Final choice, create a new extent page for
3351 erp = xfs_iext_irec_new(ifp, erp_idx);
3353 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
3355 erp->er_extcount += nex2;
3356 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3361 * This is called when the amount of space required for incore file
3362 * extents needs to be decreased. The ext_diff parameter stores the
3363 * number of extents to be removed and the idx parameter contains
3364 * the extent index where the extents will be removed from.
3366 * If the amount of space needed has decreased below the linear
3367 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3368 * extent array. Otherwise, use kmem_realloc() to adjust the
3369 * size to what is needed.
3373 xfs_inode_t *ip, /* incore inode pointer */
3374 xfs_extnum_t idx, /* index to begin removing exts */
3375 int ext_diff, /* number of extents to remove */
3376 int state) /* type of extent conversion */
3378 xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
3379 xfs_extnum_t nextents; /* number of extents in file */
3380 int new_size; /* size of extents after removal */
3382 trace_xfs_iext_remove(ip, idx, state, _RET_IP_);
3384 ASSERT(ext_diff > 0);
3385 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3386 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3388 if (new_size == 0) {
3389 xfs_iext_destroy(ifp);
3390 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3391 xfs_iext_remove_indirect(ifp, idx, ext_diff);
3392 } else if (ifp->if_real_bytes) {
3393 xfs_iext_remove_direct(ifp, idx, ext_diff);
3395 xfs_iext_remove_inline(ifp, idx, ext_diff);
3397 ifp->if_bytes = new_size;
3401 * This removes ext_diff extents from the inline buffer, beginning
3402 * at extent index idx.
3405 xfs_iext_remove_inline(
3406 xfs_ifork_t *ifp, /* inode fork pointer */
3407 xfs_extnum_t idx, /* index to begin removing exts */
3408 int ext_diff) /* number of extents to remove */
3410 int nextents; /* number of extents in file */
3412 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3413 ASSERT(idx < XFS_INLINE_EXTS);
3414 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3415 ASSERT(((nextents - ext_diff) > 0) &&
3416 (nextents - ext_diff) < XFS_INLINE_EXTS);
3418 if (idx + ext_diff < nextents) {
3419 memmove(&ifp->if_u2.if_inline_ext[idx],
3420 &ifp->if_u2.if_inline_ext[idx + ext_diff],
3421 (nextents - (idx + ext_diff)) *
3422 sizeof(xfs_bmbt_rec_t));
3423 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
3424 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3426 memset(&ifp->if_u2.if_inline_ext[idx], 0,
3427 ext_diff * sizeof(xfs_bmbt_rec_t));
3432 * This removes ext_diff extents from a linear (direct) extent list,
3433 * beginning at extent index idx. If the extents are being removed
3434 * from the end of the list (ie. truncate) then we just need to re-
3435 * allocate the list to remove the extra space. Otherwise, if the
3436 * extents are being removed from the middle of the existing extent
3437 * entries, then we first need to move the extent records beginning
3438 * at idx + ext_diff up in the list to overwrite the records being
3439 * removed, then remove the extra space via kmem_realloc.
3442 xfs_iext_remove_direct(
3443 xfs_ifork_t *ifp, /* inode fork pointer */
3444 xfs_extnum_t idx, /* index to begin removing exts */
3445 int ext_diff) /* number of extents to remove */
3447 xfs_extnum_t nextents; /* number of extents in file */
3448 int new_size; /* size of extents after removal */
3450 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3451 new_size = ifp->if_bytes -
3452 (ext_diff * sizeof(xfs_bmbt_rec_t));
3453 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3455 if (new_size == 0) {
3456 xfs_iext_destroy(ifp);
3459 /* Move extents up in the list (if needed) */
3460 if (idx + ext_diff < nextents) {
3461 memmove(&ifp->if_u1.if_extents[idx],
3462 &ifp->if_u1.if_extents[idx + ext_diff],
3463 (nextents - (idx + ext_diff)) *
3464 sizeof(xfs_bmbt_rec_t));
3466 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
3467 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3469 * Reallocate the direct extent list. If the extents
3470 * will fit inside the inode then xfs_iext_realloc_direct
3471 * will switch from direct to inline extent allocation
3474 xfs_iext_realloc_direct(ifp, new_size);
3475 ifp->if_bytes = new_size;
3479 * This is called when incore extents are being removed from the
3480 * indirection array and the extents being removed span multiple extent
3481 * buffers. The idx parameter contains the file extent index where we
3482 * want to begin removing extents, and the count parameter contains
3483 * how many extents need to be removed.
3485 * |-------| |-------|
3486 * | nex1 | | | nex1 - number of extents before idx
3487 * |-------| | count |
3488 * | | | | count - number of extents being removed at idx
3489 * | count | |-------|
3490 * | | | nex2 | nex2 - number of extents after idx + count
3491 * |-------| |-------|
3494 xfs_iext_remove_indirect(
3495 xfs_ifork_t *ifp, /* inode fork pointer */
3496 xfs_extnum_t idx, /* index to begin removing extents */
3497 int count) /* number of extents to remove */
3499 xfs_ext_irec_t *erp; /* indirection array pointer */
3500 int erp_idx = 0; /* indirection array index */
3501 xfs_extnum_t ext_cnt; /* extents left to remove */
3502 xfs_extnum_t ext_diff; /* extents to remove in current list */
3503 xfs_extnum_t nex1; /* number of extents before idx */
3504 xfs_extnum_t nex2; /* extents after idx + count */
3505 int page_idx = idx; /* index in target extent list */
3507 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3508 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3509 ASSERT(erp != NULL);
3513 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
3514 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
3516 * Check for deletion of entire list;
3517 * xfs_iext_irec_remove() updates extent offsets.
3519 if (ext_diff == erp->er_extcount) {
3520 xfs_iext_irec_remove(ifp, erp_idx);
3521 ext_cnt -= ext_diff;
3524 ASSERT(erp_idx < ifp->if_real_bytes /
3526 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3533 /* Move extents up (if needed) */
3535 memmove(&erp->er_extbuf[nex1],
3536 &erp->er_extbuf[nex1 + ext_diff],
3537 nex2 * sizeof(xfs_bmbt_rec_t));
3539 /* Zero out rest of page */
3540 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
3541 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
3542 /* Update remaining counters */
3543 erp->er_extcount -= ext_diff;
3544 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
3545 ext_cnt -= ext_diff;
3550 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
3551 xfs_iext_irec_compact(ifp);
3555 * Create, destroy, or resize a linear (direct) block of extents.
3558 xfs_iext_realloc_direct(
3559 xfs_ifork_t *ifp, /* inode fork pointer */
3560 int new_size) /* new size of extents */
3562 int rnew_size; /* real new size of extents */
3564 rnew_size = new_size;
3566 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
3567 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
3568 (new_size != ifp->if_real_bytes)));
3570 /* Free extent records */
3571 if (new_size == 0) {
3572 xfs_iext_destroy(ifp);
3574 /* Resize direct extent list and zero any new bytes */
3575 else if (ifp->if_real_bytes) {
3576 /* Check if extents will fit inside the inode */
3577 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
3578 xfs_iext_direct_to_inline(ifp, new_size /
3579 (uint)sizeof(xfs_bmbt_rec_t));
3580 ifp->if_bytes = new_size;
3583 if (!is_power_of_2(new_size)){
3584 rnew_size = roundup_pow_of_two(new_size);
3586 if (rnew_size != ifp->if_real_bytes) {
3587 ifp->if_u1.if_extents =
3588 kmem_realloc(ifp->if_u1.if_extents,
3590 ifp->if_real_bytes, KM_NOFS);
3592 if (rnew_size > ifp->if_real_bytes) {
3593 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
3594 (uint)sizeof(xfs_bmbt_rec_t)], 0,
3595 rnew_size - ifp->if_real_bytes);
3599 * Switch from the inline extent buffer to a direct
3600 * extent list. Be sure to include the inline extent
3601 * bytes in new_size.
3604 new_size += ifp->if_bytes;
3605 if (!is_power_of_2(new_size)) {
3606 rnew_size = roundup_pow_of_two(new_size);
3608 xfs_iext_inline_to_direct(ifp, rnew_size);
3610 ifp->if_real_bytes = rnew_size;
3611 ifp->if_bytes = new_size;
3615 * Switch from linear (direct) extent records to inline buffer.
3618 xfs_iext_direct_to_inline(
3619 xfs_ifork_t *ifp, /* inode fork pointer */
3620 xfs_extnum_t nextents) /* number of extents in file */
3622 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3623 ASSERT(nextents <= XFS_INLINE_EXTS);
3625 * The inline buffer was zeroed when we switched
3626 * from inline to direct extent allocation mode,
3627 * so we don't need to clear it here.
3629 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
3630 nextents * sizeof(xfs_bmbt_rec_t));
3631 kmem_free(ifp->if_u1.if_extents);
3632 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3633 ifp->if_real_bytes = 0;
3637 * Switch from inline buffer to linear (direct) extent records.
3638 * new_size should already be rounded up to the next power of 2
3639 * by the caller (when appropriate), so use new_size as it is.
3640 * However, since new_size may be rounded up, we can't update
3641 * if_bytes here. It is the caller's responsibility to update
3642 * if_bytes upon return.
3645 xfs_iext_inline_to_direct(
3646 xfs_ifork_t *ifp, /* inode fork pointer */
3647 int new_size) /* number of extents in file */
3649 ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS);
3650 memset(ifp->if_u1.if_extents, 0, new_size);
3651 if (ifp->if_bytes) {
3652 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
3654 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3655 sizeof(xfs_bmbt_rec_t));
3657 ifp->if_real_bytes = new_size;
3661 * Resize an extent indirection array to new_size bytes.
3664 xfs_iext_realloc_indirect(
3665 xfs_ifork_t *ifp, /* inode fork pointer */
3666 int new_size) /* new indirection array size */
3668 int nlists; /* number of irec's (ex lists) */
3669 int size; /* current indirection array size */
3671 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3672 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3673 size = nlists * sizeof(xfs_ext_irec_t);
3674 ASSERT(ifp->if_real_bytes);
3675 ASSERT((new_size >= 0) && (new_size != size));
3676 if (new_size == 0) {
3677 xfs_iext_destroy(ifp);
3679 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
3680 kmem_realloc(ifp->if_u1.if_ext_irec,
3681 new_size, size, KM_NOFS);
3686 * Switch from indirection array to linear (direct) extent allocations.
3689 xfs_iext_indirect_to_direct(
3690 xfs_ifork_t *ifp) /* inode fork pointer */
3692 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
3693 xfs_extnum_t nextents; /* number of extents in file */
3694 int size; /* size of file extents */
3696 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3697 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3698 ASSERT(nextents <= XFS_LINEAR_EXTS);
3699 size = nextents * sizeof(xfs_bmbt_rec_t);
3701 xfs_iext_irec_compact_pages(ifp);
3702 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
3704 ep = ifp->if_u1.if_ext_irec->er_extbuf;
3705 kmem_free(ifp->if_u1.if_ext_irec);
3706 ifp->if_flags &= ~XFS_IFEXTIREC;
3707 ifp->if_u1.if_extents = ep;
3708 ifp->if_bytes = size;
3709 if (nextents < XFS_LINEAR_EXTS) {
3710 xfs_iext_realloc_direct(ifp, size);
3715 * Free incore file extents.
3719 xfs_ifork_t *ifp) /* inode fork pointer */
3721 if (ifp->if_flags & XFS_IFEXTIREC) {
3725 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3726 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
3727 xfs_iext_irec_remove(ifp, erp_idx);
3729 ifp->if_flags &= ~XFS_IFEXTIREC;
3730 } else if (ifp->if_real_bytes) {
3731 kmem_free(ifp->if_u1.if_extents);
3732 } else if (ifp->if_bytes) {
3733 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3734 sizeof(xfs_bmbt_rec_t));
3736 ifp->if_u1.if_extents = NULL;
3737 ifp->if_real_bytes = 0;
3742 * Return a pointer to the extent record for file system block bno.
3744 xfs_bmbt_rec_host_t * /* pointer to found extent record */
3745 xfs_iext_bno_to_ext(
3746 xfs_ifork_t *ifp, /* inode fork pointer */
3747 xfs_fileoff_t bno, /* block number to search for */
3748 xfs_extnum_t *idxp) /* index of target extent */
3750 xfs_bmbt_rec_host_t *base; /* pointer to first extent */
3751 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
3752 xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */
3753 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
3754 int high; /* upper boundary in search */
3755 xfs_extnum_t idx = 0; /* index of target extent */
3756 int low; /* lower boundary in search */
3757 xfs_extnum_t nextents; /* number of file extents */
3758 xfs_fileoff_t startoff = 0; /* start offset of extent */
3760 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3761 if (nextents == 0) {
3766 if (ifp->if_flags & XFS_IFEXTIREC) {
3767 /* Find target extent list */
3769 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
3770 base = erp->er_extbuf;
3771 high = erp->er_extcount - 1;
3773 base = ifp->if_u1.if_extents;
3774 high = nextents - 1;
3776 /* Binary search extent records */
3777 while (low <= high) {
3778 idx = (low + high) >> 1;
3780 startoff = xfs_bmbt_get_startoff(ep);
3781 blockcount = xfs_bmbt_get_blockcount(ep);
3782 if (bno < startoff) {
3784 } else if (bno >= startoff + blockcount) {
3787 /* Convert back to file-based extent index */
3788 if (ifp->if_flags & XFS_IFEXTIREC) {
3789 idx += erp->er_extoff;
3795 /* Convert back to file-based extent index */
3796 if (ifp->if_flags & XFS_IFEXTIREC) {
3797 idx += erp->er_extoff;
3799 if (bno >= startoff + blockcount) {
3800 if (++idx == nextents) {
3803 ep = xfs_iext_get_ext(ifp, idx);
3811 * Return a pointer to the indirection array entry containing the
3812 * extent record for filesystem block bno. Store the index of the
3813 * target irec in *erp_idxp.
3815 xfs_ext_irec_t * /* pointer to found extent record */
3816 xfs_iext_bno_to_irec(
3817 xfs_ifork_t *ifp, /* inode fork pointer */
3818 xfs_fileoff_t bno, /* block number to search for */
3819 int *erp_idxp) /* irec index of target ext list */
3821 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
3822 xfs_ext_irec_t *erp_next; /* next indirection array entry */
3823 int erp_idx; /* indirection array index */
3824 int nlists; /* number of extent irec's (lists) */
3825 int high; /* binary search upper limit */
3826 int low; /* binary search lower limit */
3828 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3829 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3833 while (low <= high) {
3834 erp_idx = (low + high) >> 1;
3835 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3836 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
3837 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
3839 } else if (erp_next && bno >=
3840 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
3846 *erp_idxp = erp_idx;
3851 * Return a pointer to the indirection array entry containing the
3852 * extent record at file extent index *idxp. Store the index of the
3853 * target irec in *erp_idxp and store the page index of the target
3854 * extent record in *idxp.
3857 xfs_iext_idx_to_irec(
3858 xfs_ifork_t *ifp, /* inode fork pointer */
3859 xfs_extnum_t *idxp, /* extent index (file -> page) */
3860 int *erp_idxp, /* pointer to target irec */
3861 int realloc) /* new bytes were just added */
3863 xfs_ext_irec_t *prev; /* pointer to previous irec */
3864 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
3865 int erp_idx; /* indirection array index */
3866 int nlists; /* number of irec's (ex lists) */
3867 int high; /* binary search upper limit */
3868 int low; /* binary search lower limit */
3869 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
3871 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3872 ASSERT(page_idx >= 0 && page_idx <=
3873 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
3874 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3879 /* Binary search extent irec's */
3880 while (low <= high) {
3881 erp_idx = (low + high) >> 1;
3882 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3883 prev = erp_idx > 0 ? erp - 1 : NULL;
3884 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
3885 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
3887 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
3888 (page_idx == erp->er_extoff + erp->er_extcount &&
3891 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
3892 erp->er_extcount == XFS_LINEAR_EXTS) {
3896 erp = erp_idx < nlists ? erp + 1 : NULL;
3899 page_idx -= erp->er_extoff;
3904 *erp_idxp = erp_idx;
3909 * Allocate and initialize an indirection array once the space needed
3910 * for incore extents increases above XFS_IEXT_BUFSZ.
3914 xfs_ifork_t *ifp) /* inode fork pointer */
3916 xfs_ext_irec_t *erp; /* indirection array pointer */
3917 xfs_extnum_t nextents; /* number of extents in file */
3919 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3920 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3921 ASSERT(nextents <= XFS_LINEAR_EXTS);
3923 erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS);
3925 if (nextents == 0) {
3926 ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
3927 } else if (!ifp->if_real_bytes) {
3928 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
3929 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
3930 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
3932 erp->er_extbuf = ifp->if_u1.if_extents;
3933 erp->er_extcount = nextents;
3936 ifp->if_flags |= XFS_IFEXTIREC;
3937 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
3938 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
3939 ifp->if_u1.if_ext_irec = erp;
3945 * Allocate and initialize a new entry in the indirection array.
3949 xfs_ifork_t *ifp, /* inode fork pointer */
3950 int erp_idx) /* index for new irec */
3952 xfs_ext_irec_t *erp; /* indirection array pointer */
3953 int i; /* loop counter */
3954 int nlists; /* number of irec's (ex lists) */
3956 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3957 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3959 /* Resize indirection array */
3960 xfs_iext_realloc_indirect(ifp, ++nlists *
3961 sizeof(xfs_ext_irec_t));
3963 * Move records down in the array so the
3964 * new page can use erp_idx.
3966 erp = ifp->if_u1.if_ext_irec;
3967 for (i = nlists - 1; i > erp_idx; i--) {
3968 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
3970 ASSERT(i == erp_idx);
3972 /* Initialize new extent record */
3973 erp = ifp->if_u1.if_ext_irec;
3974 erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
3975 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
3976 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
3977 erp[erp_idx].er_extcount = 0;
3978 erp[erp_idx].er_extoff = erp_idx > 0 ?
3979 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
3980 return (&erp[erp_idx]);
3984 * Remove a record from the indirection array.
3987 xfs_iext_irec_remove(
3988 xfs_ifork_t *ifp, /* inode fork pointer */
3989 int erp_idx) /* irec index to remove */
3991 xfs_ext_irec_t *erp; /* indirection array pointer */
3992 int i; /* loop counter */
3993 int nlists; /* number of irec's (ex lists) */
3995 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3996 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3997 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3998 if (erp->er_extbuf) {
3999 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
4001 kmem_free(erp->er_extbuf);
4003 /* Compact extent records */
4004 erp = ifp->if_u1.if_ext_irec;
4005 for (i = erp_idx; i < nlists - 1; i++) {
4006 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
4009 * Manually free the last extent record from the indirection
4010 * array. A call to xfs_iext_realloc_indirect() with a size
4011 * of zero would result in a call to xfs_iext_destroy() which
4012 * would in turn call this function again, creating a nasty
4016 xfs_iext_realloc_indirect(ifp,
4017 nlists * sizeof(xfs_ext_irec_t));
4019 kmem_free(ifp->if_u1.if_ext_irec);
4021 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4025 * This is called to clean up large amounts of unused memory allocated
4026 * by the indirection array. Before compacting anything though, verify
4027 * that the indirection array is still needed and switch back to the
4028 * linear extent list (or even the inline buffer) if possible. The
4029 * compaction policy is as follows:
4031 * Full Compaction: Extents fit into a single page (or inline buffer)
4032 * Partial Compaction: Extents occupy less than 50% of allocated space
4033 * No Compaction: Extents occupy at least 50% of allocated space
4036 xfs_iext_irec_compact(
4037 xfs_ifork_t *ifp) /* inode fork pointer */
4039 xfs_extnum_t nextents; /* number of extents in file */
4040 int nlists; /* number of irec's (ex lists) */
4042 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4043 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4044 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4046 if (nextents == 0) {
4047 xfs_iext_destroy(ifp);
4048 } else if (nextents <= XFS_INLINE_EXTS) {
4049 xfs_iext_indirect_to_direct(ifp);
4050 xfs_iext_direct_to_inline(ifp, nextents);
4051 } else if (nextents <= XFS_LINEAR_EXTS) {
4052 xfs_iext_indirect_to_direct(ifp);
4053 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
4054 xfs_iext_irec_compact_pages(ifp);
4059 * Combine extents from neighboring extent pages.
4062 xfs_iext_irec_compact_pages(
4063 xfs_ifork_t *ifp) /* inode fork pointer */
4065 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
4066 int erp_idx = 0; /* indirection array index */
4067 int nlists; /* number of irec's (ex lists) */
4069 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4070 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4071 while (erp_idx < nlists - 1) {
4072 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4074 if (erp_next->er_extcount <=
4075 (XFS_LINEAR_EXTS - erp->er_extcount)) {
4076 memcpy(&erp->er_extbuf[erp->er_extcount],
4077 erp_next->er_extbuf, erp_next->er_extcount *
4078 sizeof(xfs_bmbt_rec_t));
4079 erp->er_extcount += erp_next->er_extcount;
4081 * Free page before removing extent record
4082 * so er_extoffs don't get modified in
4083 * xfs_iext_irec_remove.
4085 kmem_free(erp_next->er_extbuf);
4086 erp_next->er_extbuf = NULL;
4087 xfs_iext_irec_remove(ifp, erp_idx + 1);
4088 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4096 * This is called to update the er_extoff field in the indirection
4097 * array when extents have been added or removed from one of the
4098 * extent lists. erp_idx contains the irec index to begin updating
4099 * at and ext_diff contains the number of extents that were added
4103 xfs_iext_irec_update_extoffs(
4104 xfs_ifork_t *ifp, /* inode fork pointer */
4105 int erp_idx, /* irec index to update */
4106 int ext_diff) /* number of new extents */
4108 int i; /* loop counter */
4109 int nlists; /* number of irec's (ex lists */
4111 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4112 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4113 for (i = erp_idx; i < nlists; i++) {
4114 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;