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_lo = be16_to_cpu(from->di_projid_lo);
664 to->di_projid_hi = be16_to_cpu(from->di_projid_hi);
665 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
666 to->di_flushiter = be16_to_cpu(from->di_flushiter);
667 to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec);
668 to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec);
669 to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec);
670 to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec);
671 to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec);
672 to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec);
673 to->di_size = be64_to_cpu(from->di_size);
674 to->di_nblocks = be64_to_cpu(from->di_nblocks);
675 to->di_extsize = be32_to_cpu(from->di_extsize);
676 to->di_nextents = be32_to_cpu(from->di_nextents);
677 to->di_anextents = be16_to_cpu(from->di_anextents);
678 to->di_forkoff = from->di_forkoff;
679 to->di_aformat = from->di_aformat;
680 to->di_dmevmask = be32_to_cpu(from->di_dmevmask);
681 to->di_dmstate = be16_to_cpu(from->di_dmstate);
682 to->di_flags = be16_to_cpu(from->di_flags);
683 to->di_gen = be32_to_cpu(from->di_gen);
689 xfs_icdinode_t *from)
691 to->di_magic = cpu_to_be16(from->di_magic);
692 to->di_mode = cpu_to_be16(from->di_mode);
693 to->di_version = from ->di_version;
694 to->di_format = from->di_format;
695 to->di_onlink = cpu_to_be16(from->di_onlink);
696 to->di_uid = cpu_to_be32(from->di_uid);
697 to->di_gid = cpu_to_be32(from->di_gid);
698 to->di_nlink = cpu_to_be32(from->di_nlink);
699 to->di_projid_lo = cpu_to_be16(from->di_projid_lo);
700 to->di_projid_hi = cpu_to_be16(from->di_projid_hi);
701 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
702 to->di_flushiter = cpu_to_be16(from->di_flushiter);
703 to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec);
704 to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec);
705 to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec);
706 to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec);
707 to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec);
708 to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec);
709 to->di_size = cpu_to_be64(from->di_size);
710 to->di_nblocks = cpu_to_be64(from->di_nblocks);
711 to->di_extsize = cpu_to_be32(from->di_extsize);
712 to->di_nextents = cpu_to_be32(from->di_nextents);
713 to->di_anextents = cpu_to_be16(from->di_anextents);
714 to->di_forkoff = from->di_forkoff;
715 to->di_aformat = from->di_aformat;
716 to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
717 to->di_dmstate = cpu_to_be16(from->di_dmstate);
718 to->di_flags = cpu_to_be16(from->di_flags);
719 to->di_gen = cpu_to_be32(from->di_gen);
728 if (di_flags & XFS_DIFLAG_ANY) {
729 if (di_flags & XFS_DIFLAG_REALTIME)
730 flags |= XFS_XFLAG_REALTIME;
731 if (di_flags & XFS_DIFLAG_PREALLOC)
732 flags |= XFS_XFLAG_PREALLOC;
733 if (di_flags & XFS_DIFLAG_IMMUTABLE)
734 flags |= XFS_XFLAG_IMMUTABLE;
735 if (di_flags & XFS_DIFLAG_APPEND)
736 flags |= XFS_XFLAG_APPEND;
737 if (di_flags & XFS_DIFLAG_SYNC)
738 flags |= XFS_XFLAG_SYNC;
739 if (di_flags & XFS_DIFLAG_NOATIME)
740 flags |= XFS_XFLAG_NOATIME;
741 if (di_flags & XFS_DIFLAG_NODUMP)
742 flags |= XFS_XFLAG_NODUMP;
743 if (di_flags & XFS_DIFLAG_RTINHERIT)
744 flags |= XFS_XFLAG_RTINHERIT;
745 if (di_flags & XFS_DIFLAG_PROJINHERIT)
746 flags |= XFS_XFLAG_PROJINHERIT;
747 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
748 flags |= XFS_XFLAG_NOSYMLINKS;
749 if (di_flags & XFS_DIFLAG_EXTSIZE)
750 flags |= XFS_XFLAG_EXTSIZE;
751 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
752 flags |= XFS_XFLAG_EXTSZINHERIT;
753 if (di_flags & XFS_DIFLAG_NODEFRAG)
754 flags |= XFS_XFLAG_NODEFRAG;
755 if (di_flags & XFS_DIFLAG_FILESTREAM)
756 flags |= XFS_XFLAG_FILESTREAM;
766 xfs_icdinode_t *dic = &ip->i_d;
768 return _xfs_dic2xflags(dic->di_flags) |
769 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
776 return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) |
777 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
781 * Read the disk inode attributes into the in-core inode structure.
795 * Fill in the location information in the in-core inode.
797 error = xfs_imap(mp, tp, ip->i_ino, &ip->i_imap, iget_flags);
802 * Get pointers to the on-disk inode and the buffer containing it.
804 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp,
805 XBF_LOCK, iget_flags);
808 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
811 * If we got something that isn't an inode it means someone
812 * (nfs or dmi) has a stale handle.
814 if (be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC) {
816 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
817 "dip->di_magic (0x%x) != "
818 "XFS_DINODE_MAGIC (0x%x)",
819 be16_to_cpu(dip->di_magic),
822 error = XFS_ERROR(EINVAL);
827 * If the on-disk inode is already linked to a directory
828 * entry, copy all of the inode into the in-core inode.
829 * xfs_iformat() handles copying in the inode format
830 * specific information.
831 * Otherwise, just get the truly permanent information.
834 xfs_dinode_from_disk(&ip->i_d, dip);
835 error = xfs_iformat(ip, dip);
838 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
839 "xfs_iformat() returned error %d",
845 ip->i_d.di_magic = be16_to_cpu(dip->di_magic);
846 ip->i_d.di_version = dip->di_version;
847 ip->i_d.di_gen = be32_to_cpu(dip->di_gen);
848 ip->i_d.di_flushiter = be16_to_cpu(dip->di_flushiter);
850 * Make sure to pull in the mode here as well in
851 * case the inode is released without being used.
852 * This ensures that xfs_inactive() will see that
853 * the inode is already free and not try to mess
854 * with the uninitialized part of it.
858 * Initialize the per-fork minima and maxima for a new
859 * inode here. xfs_iformat will do it for old inodes.
861 ip->i_df.if_ext_max =
862 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
866 * The inode format changed when we moved the link count and
867 * made it 32 bits long. If this is an old format inode,
868 * convert it in memory to look like a new one. If it gets
869 * flushed to disk we will convert back before flushing or
870 * logging it. We zero out the new projid field and the old link
871 * count field. We'll handle clearing the pad field (the remains
872 * of the old uuid field) when we actually convert the inode to
873 * the new format. We don't change the version number so that we
874 * can distinguish this from a real new format inode.
876 if (ip->i_d.di_version == 1) {
877 ip->i_d.di_nlink = ip->i_d.di_onlink;
878 ip->i_d.di_onlink = 0;
879 xfs_set_projid(ip, 0);
882 ip->i_delayed_blks = 0;
883 ip->i_size = ip->i_d.di_size;
886 * Mark the buffer containing the inode as something to keep
887 * around for a while. This helps to keep recently accessed
888 * meta-data in-core longer.
890 xfs_buf_set_ref(bp, XFS_INO_REF);
893 * Use xfs_trans_brelse() to release the buffer containing the
894 * on-disk inode, because it was acquired with xfs_trans_read_buf()
895 * in xfs_itobp() above. If tp is NULL, this is just a normal
896 * brelse(). If we're within a transaction, then xfs_trans_brelse()
897 * will only release the buffer if it is not dirty within the
898 * transaction. It will be OK to release the buffer in this case,
899 * because inodes on disk are never destroyed and we will be
900 * locking the new in-core inode before putting it in the hash
901 * table where other processes can find it. Thus we don't have
902 * to worry about the inode being changed just because we released
906 xfs_trans_brelse(tp, bp);
911 * Read in extents from a btree-format inode.
912 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
922 xfs_extnum_t nextents;
924 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
925 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
927 return XFS_ERROR(EFSCORRUPTED);
929 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
930 ifp = XFS_IFORK_PTR(ip, whichfork);
933 * We know that the size is valid (it's checked in iformat_btree)
935 ifp->if_lastex = NULLEXTNUM;
936 ifp->if_bytes = ifp->if_real_bytes = 0;
937 ifp->if_flags |= XFS_IFEXTENTS;
938 xfs_iext_add(ifp, 0, nextents);
939 error = xfs_bmap_read_extents(tp, ip, whichfork);
941 xfs_iext_destroy(ifp);
942 ifp->if_flags &= ~XFS_IFEXTENTS;
945 xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip));
950 * Allocate an inode on disk and return a copy of its in-core version.
951 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
952 * appropriately within the inode. The uid and gid for the inode are
953 * set according to the contents of the given cred structure.
955 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
956 * has a free inode available, call xfs_iget()
957 * to obtain the in-core version of the allocated inode. Finally,
958 * fill in the inode and log its initial contents. In this case,
959 * ialloc_context would be set to NULL and call_again set to false.
961 * If xfs_dialloc() does not have an available inode,
962 * it will replenish its supply by doing an allocation. Since we can
963 * only do one allocation within a transaction without deadlocks, we
964 * must commit the current transaction before returning the inode itself.
965 * In this case, therefore, we will set call_again to true and return.
966 * The caller should then commit the current transaction, start a new
967 * transaction, and call xfs_ialloc() again to actually get the inode.
969 * To ensure that some other process does not grab the inode that
970 * was allocated during the first call to xfs_ialloc(), this routine
971 * also returns the [locked] bp pointing to the head of the freelist
972 * as ialloc_context. The caller should hold this buffer across
973 * the commit and pass it back into this routine on the second call.
975 * If we are allocating quota inodes, we do not have a parent inode
976 * to attach to or associate with (i.e. pip == NULL) because they
977 * are not linked into the directory structure - they are attached
978 * directly to the superblock - and so have no parent.
989 xfs_buf_t **ialloc_context,
990 boolean_t *call_again,
1001 * Call the space management code to pick
1002 * the on-disk inode to be allocated.
1004 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
1005 ialloc_context, call_again, &ino);
1008 if (*call_again || ino == NULLFSINO) {
1012 ASSERT(*ialloc_context == NULL);
1015 * Get the in-core inode with the lock held exclusively.
1016 * This is because we're setting fields here we need
1017 * to prevent others from looking at until we're done.
1019 error = xfs_iget(tp->t_mountp, tp, ino, XFS_IGET_CREATE,
1020 XFS_ILOCK_EXCL, &ip);
1025 ip->i_d.di_mode = (__uint16_t)mode;
1026 ip->i_d.di_onlink = 0;
1027 ip->i_d.di_nlink = nlink;
1028 ASSERT(ip->i_d.di_nlink == nlink);
1029 ip->i_d.di_uid = current_fsuid();
1030 ip->i_d.di_gid = current_fsgid();
1031 xfs_set_projid(ip, prid);
1032 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1035 * If the superblock version is up to where we support new format
1036 * inodes and this is currently an old format inode, then change
1037 * the inode version number now. This way we only do the conversion
1038 * here rather than here and in the flush/logging code.
1040 if (xfs_sb_version_hasnlink(&tp->t_mountp->m_sb) &&
1041 ip->i_d.di_version == 1) {
1042 ip->i_d.di_version = 2;
1044 * We've already zeroed the old link count, the projid field,
1045 * and the pad field.
1050 * Project ids won't be stored on disk if we are using a version 1 inode.
1052 if ((prid != 0) && (ip->i_d.di_version == 1))
1053 xfs_bump_ino_vers2(tp, ip);
1055 if (pip && XFS_INHERIT_GID(pip)) {
1056 ip->i_d.di_gid = pip->i_d.di_gid;
1057 if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1058 ip->i_d.di_mode |= S_ISGID;
1063 * If the group ID of the new file does not match the effective group
1064 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1065 * (and only if the irix_sgid_inherit compatibility variable is set).
1067 if ((irix_sgid_inherit) &&
1068 (ip->i_d.di_mode & S_ISGID) &&
1069 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1070 ip->i_d.di_mode &= ~S_ISGID;
1073 ip->i_d.di_size = 0;
1075 ip->i_d.di_nextents = 0;
1076 ASSERT(ip->i_d.di_nblocks == 0);
1079 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
1080 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
1081 ip->i_d.di_atime = ip->i_d.di_mtime;
1082 ip->i_d.di_ctime = ip->i_d.di_mtime;
1085 * di_gen will have been taken care of in xfs_iread.
1087 ip->i_d.di_extsize = 0;
1088 ip->i_d.di_dmevmask = 0;
1089 ip->i_d.di_dmstate = 0;
1090 ip->i_d.di_flags = 0;
1091 flags = XFS_ILOG_CORE;
1092 switch (mode & S_IFMT) {
1097 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1098 ip->i_df.if_u2.if_rdev = rdev;
1099 ip->i_df.if_flags = 0;
1100 flags |= XFS_ILOG_DEV;
1104 * we can't set up filestreams until after the VFS inode
1105 * is set up properly.
1107 if (pip && xfs_inode_is_filestream(pip))
1111 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
1114 if ((mode & S_IFMT) == S_IFDIR) {
1115 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1116 di_flags |= XFS_DIFLAG_RTINHERIT;
1117 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1118 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1119 ip->i_d.di_extsize = pip->i_d.di_extsize;
1121 } else if ((mode & S_IFMT) == S_IFREG) {
1122 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1123 di_flags |= XFS_DIFLAG_REALTIME;
1124 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1125 di_flags |= XFS_DIFLAG_EXTSIZE;
1126 ip->i_d.di_extsize = pip->i_d.di_extsize;
1129 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1130 xfs_inherit_noatime)
1131 di_flags |= XFS_DIFLAG_NOATIME;
1132 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1134 di_flags |= XFS_DIFLAG_NODUMP;
1135 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1137 di_flags |= XFS_DIFLAG_SYNC;
1138 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1139 xfs_inherit_nosymlinks)
1140 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1141 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1142 di_flags |= XFS_DIFLAG_PROJINHERIT;
1143 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1144 xfs_inherit_nodefrag)
1145 di_flags |= XFS_DIFLAG_NODEFRAG;
1146 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
1147 di_flags |= XFS_DIFLAG_FILESTREAM;
1148 ip->i_d.di_flags |= di_flags;
1152 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1153 ip->i_df.if_flags = XFS_IFEXTENTS;
1154 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1155 ip->i_df.if_u1.if_extents = NULL;
1161 * Attribute fork settings for new inode.
1163 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1164 ip->i_d.di_anextents = 0;
1167 * Log the new values stuffed into the inode.
1169 xfs_trans_ijoin_ref(tp, ip, XFS_ILOCK_EXCL);
1170 xfs_trans_log_inode(tp, ip, flags);
1172 /* now that we have an i_mode we can setup inode ops and unlock */
1173 xfs_setup_inode(ip);
1175 /* now we have set up the vfs inode we can associate the filestream */
1177 error = xfs_filestream_associate(pip, ip);
1181 xfs_iflags_set(ip, XFS_IFILESTREAM);
1189 * Check to make sure that there are no blocks allocated to the
1190 * file beyond the size of the file. We don't check this for
1191 * files with fixed size extents or real time extents, but we
1192 * at least do it for regular files.
1201 xfs_fileoff_t map_first;
1203 xfs_bmbt_irec_t imaps[2];
1205 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1208 if (XFS_IS_REALTIME_INODE(ip))
1211 if (ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE)
1215 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1217 * The filesystem could be shutting down, so bmapi may return
1220 if (xfs_bmapi(NULL, ip, map_first,
1222 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1224 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1227 ASSERT(nimaps == 1);
1228 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1233 * Calculate the last possible buffered byte in a file. This must
1234 * include data that was buffered beyond the EOF by the write code.
1235 * This also needs to deal with overflowing the xfs_fsize_t type
1236 * which can happen for sizes near the limit.
1238 * We also need to take into account any blocks beyond the EOF. It
1239 * may be the case that they were buffered by a write which failed.
1240 * In that case the pages will still be in memory, but the inode size
1241 * will never have been updated.
1248 xfs_fsize_t last_byte;
1249 xfs_fileoff_t last_block;
1250 xfs_fileoff_t size_last_block;
1253 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED));
1257 * Only check for blocks beyond the EOF if the extents have
1258 * been read in. This eliminates the need for the inode lock,
1259 * and it also saves us from looking when it really isn't
1262 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1263 xfs_ilock(ip, XFS_ILOCK_SHARED);
1264 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1266 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1273 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_size);
1274 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1276 last_byte = XFS_FSB_TO_B(mp, last_block);
1277 if (last_byte < 0) {
1278 return XFS_MAXIOFFSET(mp);
1280 last_byte += (1 << mp->m_writeio_log);
1281 if (last_byte < 0) {
1282 return XFS_MAXIOFFSET(mp);
1288 * Start the truncation of the file to new_size. The new size
1289 * must be smaller than the current size. This routine will
1290 * clear the buffer and page caches of file data in the removed
1291 * range, and xfs_itruncate_finish() will remove the underlying
1294 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1295 * must NOT have the inode lock held at all. This is because we're
1296 * calling into the buffer/page cache code and we can't hold the
1297 * inode lock when we do so.
1299 * We need to wait for any direct I/Os in flight to complete before we
1300 * proceed with the truncate. This is needed to prevent the extents
1301 * being read or written by the direct I/Os from being removed while the
1302 * I/O is in flight as there is no other method of synchronising
1303 * direct I/O with the truncate operation. Also, because we hold
1304 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1305 * started until the truncate completes and drops the lock. Essentially,
1306 * the xfs_ioend_wait() call forms an I/O barrier that provides strict
1307 * ordering between direct I/Os and the truncate operation.
1309 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1310 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1311 * in the case that the caller is locking things out of order and
1312 * may not be able to call xfs_itruncate_finish() with the inode lock
1313 * held without dropping the I/O lock. If the caller must drop the
1314 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1315 * must be called again with all the same restrictions as the initial
1319 xfs_itruncate_start(
1322 xfs_fsize_t new_size)
1324 xfs_fsize_t last_byte;
1325 xfs_off_t toss_start;
1329 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1330 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1331 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1332 (flags == XFS_ITRUNC_MAYBE));
1336 /* wait for the completion of any pending DIOs */
1337 if (new_size == 0 || new_size < ip->i_size)
1341 * Call toss_pages or flushinval_pages to get rid of pages
1342 * overlapping the region being removed. We have to use
1343 * the less efficient flushinval_pages in the case that the
1344 * caller may not be able to finish the truncate without
1345 * dropping the inode's I/O lock. Make sure
1346 * to catch any pages brought in by buffers overlapping
1347 * the EOF by searching out beyond the isize by our
1348 * block size. We round new_size up to a block boundary
1349 * so that we don't toss things on the same block as
1350 * new_size but before it.
1352 * Before calling toss_page or flushinval_pages, make sure to
1353 * call remapf() over the same region if the file is mapped.
1354 * This frees up mapped file references to the pages in the
1355 * given range and for the flushinval_pages case it ensures
1356 * that we get the latest mapped changes flushed out.
1358 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1359 toss_start = XFS_FSB_TO_B(mp, toss_start);
1360 if (toss_start < 0) {
1362 * The place to start tossing is beyond our maximum
1363 * file size, so there is no way that the data extended
1368 last_byte = xfs_file_last_byte(ip);
1369 trace_xfs_itruncate_start(ip, flags, new_size, toss_start, last_byte);
1370 if (last_byte > toss_start) {
1371 if (flags & XFS_ITRUNC_DEFINITE) {
1372 xfs_tosspages(ip, toss_start,
1373 -1, FI_REMAPF_LOCKED);
1375 error = xfs_flushinval_pages(ip, toss_start,
1376 -1, FI_REMAPF_LOCKED);
1381 if (new_size == 0) {
1382 ASSERT(VN_CACHED(VFS_I(ip)) == 0);
1389 * Shrink the file to the given new_size. The new size must be smaller than
1390 * the current size. This will free up the underlying blocks in the removed
1391 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
1393 * The transaction passed to this routine must have made a permanent log
1394 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1395 * given transaction and start new ones, so make sure everything involved in
1396 * the transaction is tidy before calling here. Some transaction will be
1397 * returned to the caller to be committed. The incoming transaction must
1398 * already include the inode, and both inode locks must be held exclusively.
1399 * The inode must also be "held" within the transaction. On return the inode
1400 * will be "held" within the returned transaction. This routine does NOT
1401 * require any disk space to be reserved for it within the transaction.
1403 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1404 * indicates the fork which is to be truncated. For the attribute fork we only
1405 * support truncation to size 0.
1407 * We use the sync parameter to indicate whether or not the first transaction
1408 * we perform might have to be synchronous. For the attr fork, it needs to be
1409 * so if the unlink of the inode is not yet known to be permanent in the log.
1410 * This keeps us from freeing and reusing the blocks of the attribute fork
1411 * before the unlink of the inode becomes permanent.
1413 * For the data fork, we normally have to run synchronously if we're being
1414 * called out of the inactive path or we're being called out of the create path
1415 * where we're truncating an existing file. Either way, the truncate needs to
1416 * be sync so blocks don't reappear in the file with altered data in case of a
1417 * crash. wsync filesystems can run the first case async because anything that
1418 * shrinks the inode has to run sync so by the time we're called here from
1419 * inactive, the inode size is permanently set to 0.
1421 * Calls from the truncate path always need to be sync unless we're in a wsync
1422 * filesystem and the file has already been unlinked.
1424 * The caller is responsible for correctly setting the sync parameter. It gets
1425 * too hard for us to guess here which path we're being called out of just
1426 * based on inode state.
1428 * If we get an error, we must return with the inode locked and linked into the
1429 * current transaction. This keeps things simple for the higher level code,
1430 * because it always knows that the inode is locked and held in the transaction
1431 * that returns to it whether errors occur or not. We don't mark the inode
1432 * dirty on error so that transactions can be easily aborted if possible.
1435 xfs_itruncate_finish(
1438 xfs_fsize_t new_size,
1442 xfs_fsblock_t first_block;
1443 xfs_fileoff_t first_unmap_block;
1444 xfs_fileoff_t last_block;
1445 xfs_filblks_t unmap_len=0;
1450 xfs_bmap_free_t free_list;
1453 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
1454 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1455 ASSERT(*tp != NULL);
1456 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1457 ASSERT(ip->i_transp == *tp);
1458 ASSERT(ip->i_itemp != NULL);
1459 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1463 mp = (ntp)->t_mountp;
1464 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1467 * We only support truncating the entire attribute fork.
1469 if (fork == XFS_ATTR_FORK) {
1472 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1473 trace_xfs_itruncate_finish_start(ip, new_size);
1476 * The first thing we do is set the size to new_size permanently
1477 * on disk. This way we don't have to worry about anyone ever
1478 * being able to look at the data being freed even in the face
1479 * of a crash. What we're getting around here is the case where
1480 * we free a block, it is allocated to another file, it is written
1481 * to, and then we crash. If the new data gets written to the
1482 * file but the log buffers containing the free and reallocation
1483 * don't, then we'd end up with garbage in the blocks being freed.
1484 * As long as we make the new_size permanent before actually
1485 * freeing any blocks it doesn't matter if they get writtten to.
1487 * The callers must signal into us whether or not the size
1488 * setting here must be synchronous. There are a few cases
1489 * where it doesn't have to be synchronous. Those cases
1490 * occur if the file is unlinked and we know the unlink is
1491 * permanent or if the blocks being truncated are guaranteed
1492 * to be beyond the inode eof (regardless of the link count)
1493 * and the eof value is permanent. Both of these cases occur
1494 * only on wsync-mounted filesystems. In those cases, we're
1495 * guaranteed that no user will ever see the data in the blocks
1496 * that are being truncated so the truncate can run async.
1497 * In the free beyond eof case, the file may wind up with
1498 * more blocks allocated to it than it needs if we crash
1499 * and that won't get fixed until the next time the file
1500 * is re-opened and closed but that's ok as that shouldn't
1501 * be too many blocks.
1503 * However, we can't just make all wsync xactions run async
1504 * because there's one call out of the create path that needs
1505 * to run sync where it's truncating an existing file to size
1506 * 0 whose size is > 0.
1508 * It's probably possible to come up with a test in this
1509 * routine that would correctly distinguish all the above
1510 * cases from the values of the function parameters and the
1511 * inode state but for sanity's sake, I've decided to let the
1512 * layers above just tell us. It's simpler to correctly figure
1513 * out in the layer above exactly under what conditions we
1514 * can run async and I think it's easier for others read and
1515 * follow the logic in case something has to be changed.
1516 * cscope is your friend -- rcc.
1518 * The attribute fork is much simpler.
1520 * For the attribute fork we allow the caller to tell us whether
1521 * the unlink of the inode that led to this call is yet permanent
1522 * in the on disk log. If it is not and we will be freeing extents
1523 * in this inode then we make the first transaction synchronous
1524 * to make sure that the unlink is permanent by the time we free
1527 if (fork == XFS_DATA_FORK) {
1528 if (ip->i_d.di_nextents > 0) {
1530 * If we are not changing the file size then do
1531 * not update the on-disk file size - we may be
1532 * called from xfs_inactive_free_eofblocks(). If we
1533 * update the on-disk file size and then the system
1534 * crashes before the contents of the file are
1535 * flushed to disk then the files may be full of
1536 * holes (ie NULL files bug).
1538 if (ip->i_size != new_size) {
1539 ip->i_d.di_size = new_size;
1540 ip->i_size = new_size;
1541 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1545 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1546 if (ip->i_d.di_anextents > 0)
1547 xfs_trans_set_sync(ntp);
1549 ASSERT(fork == XFS_DATA_FORK ||
1550 (fork == XFS_ATTR_FORK &&
1551 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1552 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1555 * Since it is possible for space to become allocated beyond
1556 * the end of the file (in a crash where the space is allocated
1557 * but the inode size is not yet updated), simply remove any
1558 * blocks which show up between the new EOF and the maximum
1559 * possible file size. If the first block to be removed is
1560 * beyond the maximum file size (ie it is the same as last_block),
1561 * then there is nothing to do.
1563 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1564 ASSERT(first_unmap_block <= last_block);
1566 if (last_block == first_unmap_block) {
1569 unmap_len = last_block - first_unmap_block + 1;
1573 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1574 * will tell us whether it freed the entire range or
1575 * not. If this is a synchronous mount (wsync),
1576 * then we can tell bunmapi to keep all the
1577 * transactions asynchronous since the unlink
1578 * transaction that made this inode inactive has
1579 * already hit the disk. There's no danger of
1580 * the freed blocks being reused, there being a
1581 * crash, and the reused blocks suddenly reappearing
1582 * in this file with garbage in them once recovery
1585 xfs_bmap_init(&free_list, &first_block);
1586 error = xfs_bunmapi(ntp, ip,
1587 first_unmap_block, unmap_len,
1588 xfs_bmapi_aflag(fork),
1589 XFS_ITRUNC_MAX_EXTENTS,
1590 &first_block, &free_list,
1594 * If the bunmapi call encounters an error,
1595 * return to the caller where the transaction
1596 * can be properly aborted. We just need to
1597 * make sure we're not holding any resources
1598 * that we were not when we came in.
1600 xfs_bmap_cancel(&free_list);
1605 * Duplicate the transaction that has the permanent
1606 * reservation and commit the old transaction.
1608 error = xfs_bmap_finish(tp, &free_list, &committed);
1611 xfs_trans_ijoin(ntp, ip);
1615 * If the bmap finish call encounters an error, return
1616 * to the caller where the transaction can be properly
1617 * aborted. We just need to make sure we're not
1618 * holding any resources that we were not when we came
1621 * Aborting from this point might lose some blocks in
1622 * the file system, but oh well.
1624 xfs_bmap_cancel(&free_list);
1630 * Mark the inode dirty so it will be logged and
1631 * moved forward in the log as part of every commit.
1633 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1636 ntp = xfs_trans_dup(ntp);
1637 error = xfs_trans_commit(*tp, 0);
1640 xfs_trans_ijoin(ntp, ip);
1645 * transaction commit worked ok so we can drop the extra ticket
1646 * reference that we gained in xfs_trans_dup()
1648 xfs_log_ticket_put(ntp->t_ticket);
1649 error = xfs_trans_reserve(ntp, 0,
1650 XFS_ITRUNCATE_LOG_RES(mp), 0,
1651 XFS_TRANS_PERM_LOG_RES,
1652 XFS_ITRUNCATE_LOG_COUNT);
1657 * Only update the size in the case of the data fork, but
1658 * always re-log the inode so that our permanent transaction
1659 * can keep on rolling it forward in the log.
1661 if (fork == XFS_DATA_FORK) {
1662 xfs_isize_check(mp, ip, new_size);
1664 * If we are not changing the file size then do
1665 * not update the on-disk file size - we may be
1666 * called from xfs_inactive_free_eofblocks(). If we
1667 * update the on-disk file size and then the system
1668 * crashes before the contents of the file are
1669 * flushed to disk then the files may be full of
1670 * holes (ie NULL files bug).
1672 if (ip->i_size != new_size) {
1673 ip->i_d.di_size = new_size;
1674 ip->i_size = new_size;
1677 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1678 ASSERT((new_size != 0) ||
1679 (fork == XFS_ATTR_FORK) ||
1680 (ip->i_delayed_blks == 0));
1681 ASSERT((new_size != 0) ||
1682 (fork == XFS_ATTR_FORK) ||
1683 (ip->i_d.di_nextents == 0));
1684 trace_xfs_itruncate_finish_end(ip, new_size);
1689 * This is called when the inode's link count goes to 0.
1690 * We place the on-disk inode on a list in the AGI. It
1691 * will be pulled from this list when the inode is freed.
1708 ASSERT(ip->i_d.di_nlink == 0);
1709 ASSERT(ip->i_d.di_mode != 0);
1710 ASSERT(ip->i_transp == tp);
1715 * Get the agi buffer first. It ensures lock ordering
1718 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1721 agi = XFS_BUF_TO_AGI(agibp);
1724 * Get the index into the agi hash table for the
1725 * list this inode will go on.
1727 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1729 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1730 ASSERT(agi->agi_unlinked[bucket_index]);
1731 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1733 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
1735 * There is already another inode in the bucket we need
1736 * to add ourselves to. Add us at the front of the list.
1737 * Here we put the head pointer into our next pointer,
1738 * and then we fall through to point the head at us.
1740 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XBF_LOCK);
1744 ASSERT(be32_to_cpu(dip->di_next_unlinked) == NULLAGINO);
1745 /* both on-disk, don't endian flip twice */
1746 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1747 offset = ip->i_imap.im_boffset +
1748 offsetof(xfs_dinode_t, di_next_unlinked);
1749 xfs_trans_inode_buf(tp, ibp);
1750 xfs_trans_log_buf(tp, ibp, offset,
1751 (offset + sizeof(xfs_agino_t) - 1));
1752 xfs_inobp_check(mp, ibp);
1756 * Point the bucket head pointer at the inode being inserted.
1759 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1760 offset = offsetof(xfs_agi_t, agi_unlinked) +
1761 (sizeof(xfs_agino_t) * bucket_index);
1762 xfs_trans_log_buf(tp, agibp, offset,
1763 (offset + sizeof(xfs_agino_t) - 1));
1768 * Pull the on-disk inode from the AGI unlinked list.
1781 xfs_agnumber_t agno;
1783 xfs_agino_t next_agino;
1784 xfs_buf_t *last_ibp;
1785 xfs_dinode_t *last_dip = NULL;
1787 int offset, last_offset = 0;
1791 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1794 * Get the agi buffer first. It ensures lock ordering
1797 error = xfs_read_agi(mp, tp, agno, &agibp);
1801 agi = XFS_BUF_TO_AGI(agibp);
1804 * Get the index into the agi hash table for the
1805 * list this inode will go on.
1807 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1809 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1810 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
1811 ASSERT(agi->agi_unlinked[bucket_index]);
1813 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
1815 * We're at the head of the list. Get the inode's
1816 * on-disk buffer to see if there is anyone after us
1817 * on the list. Only modify our next pointer if it
1818 * is not already NULLAGINO. This saves us the overhead
1819 * of dealing with the buffer when there is no need to
1822 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XBF_LOCK);
1825 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
1826 error, mp->m_fsname);
1829 next_agino = be32_to_cpu(dip->di_next_unlinked);
1830 ASSERT(next_agino != 0);
1831 if (next_agino != NULLAGINO) {
1832 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
1833 offset = ip->i_imap.im_boffset +
1834 offsetof(xfs_dinode_t, di_next_unlinked);
1835 xfs_trans_inode_buf(tp, ibp);
1836 xfs_trans_log_buf(tp, ibp, offset,
1837 (offset + sizeof(xfs_agino_t) - 1));
1838 xfs_inobp_check(mp, ibp);
1840 xfs_trans_brelse(tp, ibp);
1843 * Point the bucket head pointer at the next inode.
1845 ASSERT(next_agino != 0);
1846 ASSERT(next_agino != agino);
1847 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
1848 offset = offsetof(xfs_agi_t, agi_unlinked) +
1849 (sizeof(xfs_agino_t) * bucket_index);
1850 xfs_trans_log_buf(tp, agibp, offset,
1851 (offset + sizeof(xfs_agino_t) - 1));
1854 * We need to search the list for the inode being freed.
1856 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
1858 while (next_agino != agino) {
1860 * If the last inode wasn't the one pointing to
1861 * us, then release its buffer since we're not
1862 * going to do anything with it.
1864 if (last_ibp != NULL) {
1865 xfs_trans_brelse(tp, last_ibp);
1867 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
1868 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
1869 &last_ibp, &last_offset, 0);
1872 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
1873 error, mp->m_fsname);
1876 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
1877 ASSERT(next_agino != NULLAGINO);
1878 ASSERT(next_agino != 0);
1881 * Now last_ibp points to the buffer previous to us on
1882 * the unlinked list. Pull us from the list.
1884 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XBF_LOCK);
1887 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
1888 error, mp->m_fsname);
1891 next_agino = be32_to_cpu(dip->di_next_unlinked);
1892 ASSERT(next_agino != 0);
1893 ASSERT(next_agino != agino);
1894 if (next_agino != NULLAGINO) {
1895 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
1896 offset = ip->i_imap.im_boffset +
1897 offsetof(xfs_dinode_t, di_next_unlinked);
1898 xfs_trans_inode_buf(tp, ibp);
1899 xfs_trans_log_buf(tp, ibp, offset,
1900 (offset + sizeof(xfs_agino_t) - 1));
1901 xfs_inobp_check(mp, ibp);
1903 xfs_trans_brelse(tp, ibp);
1906 * Point the previous inode on the list to the next inode.
1908 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
1909 ASSERT(next_agino != 0);
1910 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
1911 xfs_trans_inode_buf(tp, last_ibp);
1912 xfs_trans_log_buf(tp, last_ibp, offset,
1913 (offset + sizeof(xfs_agino_t) - 1));
1914 xfs_inobp_check(mp, last_ibp);
1920 * A big issue when freeing the inode cluster is is that we _cannot_ skip any
1921 * inodes that are in memory - they all must be marked stale and attached to
1922 * the cluster buffer.
1926 xfs_inode_t *free_ip,
1930 xfs_mount_t *mp = free_ip->i_mount;
1931 int blks_per_cluster;
1938 xfs_inode_log_item_t *iip;
1939 xfs_log_item_t *lip;
1940 struct xfs_perag *pag;
1942 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
1943 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
1944 blks_per_cluster = 1;
1945 ninodes = mp->m_sb.sb_inopblock;
1946 nbufs = XFS_IALLOC_BLOCKS(mp);
1948 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
1949 mp->m_sb.sb_blocksize;
1950 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
1951 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
1954 for (j = 0; j < nbufs; j++, inum += ninodes) {
1955 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
1956 XFS_INO_TO_AGBNO(mp, inum));
1959 * We obtain and lock the backing buffer first in the process
1960 * here, as we have to ensure that any dirty inode that we
1961 * can't get the flush lock on is attached to the buffer.
1962 * If we scan the in-memory inodes first, then buffer IO can
1963 * complete before we get a lock on it, and hence we may fail
1964 * to mark all the active inodes on the buffer stale.
1966 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
1967 mp->m_bsize * blks_per_cluster,
1971 * Walk the inodes already attached to the buffer and mark them
1972 * stale. These will all have the flush locks held, so an
1973 * in-memory inode walk can't lock them. By marking them all
1974 * stale first, we will not attempt to lock them in the loop
1975 * below as the XFS_ISTALE flag will be set.
1977 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
1979 if (lip->li_type == XFS_LI_INODE) {
1980 iip = (xfs_inode_log_item_t *)lip;
1981 ASSERT(iip->ili_logged == 1);
1982 lip->li_cb = xfs_istale_done;
1983 xfs_trans_ail_copy_lsn(mp->m_ail,
1984 &iip->ili_flush_lsn,
1985 &iip->ili_item.li_lsn);
1986 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
1988 lip = lip->li_bio_list;
1993 * For each inode in memory attempt to add it to the inode
1994 * buffer and set it up for being staled on buffer IO
1995 * completion. This is safe as we've locked out tail pushing
1996 * and flushing by locking the buffer.
1998 * We have already marked every inode that was part of a
1999 * transaction stale above, which means there is no point in
2000 * even trying to lock them.
2002 for (i = 0; i < ninodes; i++) {
2005 ip = radix_tree_lookup(&pag->pag_ici_root,
2006 XFS_INO_TO_AGINO(mp, (inum + i)));
2008 /* Inode not in memory, nothing to do */
2015 * because this is an RCU protected lookup, we could
2016 * find a recently freed or even reallocated inode
2017 * during the lookup. We need to check under the
2018 * i_flags_lock for a valid inode here. Skip it if it
2019 * is not valid, the wrong inode or stale.
2021 spin_lock(&ip->i_flags_lock);
2022 if (ip->i_ino != inum + i ||
2023 __xfs_iflags_test(ip, XFS_ISTALE)) {
2024 spin_unlock(&ip->i_flags_lock);
2028 spin_unlock(&ip->i_flags_lock);
2031 * Don't try to lock/unlock the current inode, but we
2032 * _cannot_ skip the other inodes that we did not find
2033 * in the list attached to the buffer and are not
2034 * already marked stale. If we can't lock it, back off
2037 if (ip != free_ip &&
2038 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2046 xfs_iflags_set(ip, XFS_ISTALE);
2049 * we don't need to attach clean inodes or those only
2050 * with unlogged changes (which we throw away, anyway).
2053 if (!iip || xfs_inode_clean(ip)) {
2054 ASSERT(ip != free_ip);
2055 ip->i_update_core = 0;
2057 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2061 iip->ili_last_fields = iip->ili_format.ilf_fields;
2062 iip->ili_format.ilf_fields = 0;
2063 iip->ili_logged = 1;
2064 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2065 &iip->ili_item.li_lsn);
2067 xfs_buf_attach_iodone(bp, xfs_istale_done,
2071 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2074 xfs_trans_stale_inode_buf(tp, bp);
2075 xfs_trans_binval(tp, bp);
2082 * This is called to return an inode to the inode free list.
2083 * The inode should already be truncated to 0 length and have
2084 * no pages associated with it. This routine also assumes that
2085 * the inode is already a part of the transaction.
2087 * The on-disk copy of the inode will have been added to the list
2088 * of unlinked inodes in the AGI. We need to remove the inode from
2089 * that list atomically with respect to freeing it here.
2095 xfs_bmap_free_t *flist)
2099 xfs_ino_t first_ino;
2103 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2104 ASSERT(ip->i_transp == tp);
2105 ASSERT(ip->i_d.di_nlink == 0);
2106 ASSERT(ip->i_d.di_nextents == 0);
2107 ASSERT(ip->i_d.di_anextents == 0);
2108 ASSERT((ip->i_d.di_size == 0 && ip->i_size == 0) ||
2109 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2110 ASSERT(ip->i_d.di_nblocks == 0);
2113 * Pull the on-disk inode from the AGI unlinked list.
2115 error = xfs_iunlink_remove(tp, ip);
2120 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2124 ip->i_d.di_mode = 0; /* mark incore inode as free */
2125 ip->i_d.di_flags = 0;
2126 ip->i_d.di_dmevmask = 0;
2127 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2128 ip->i_df.if_ext_max =
2129 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2130 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2131 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2133 * Bump the generation count so no one will be confused
2134 * by reincarnations of this inode.
2138 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2140 error = xfs_itobp(ip->i_mount, tp, ip, &dip, &ibp, XBF_LOCK);
2145 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2146 * from picking up this inode when it is reclaimed (its incore state
2147 * initialzed but not flushed to disk yet). The in-core di_mode is
2148 * already cleared and a corresponding transaction logged.
2149 * The hack here just synchronizes the in-core to on-disk
2150 * di_mode value in advance before the actual inode sync to disk.
2151 * This is OK because the inode is already unlinked and would never
2152 * change its di_mode again for this inode generation.
2153 * This is a temporary hack that would require a proper fix
2159 xfs_ifree_cluster(ip, tp, first_ino);
2166 * Reallocate the space for if_broot based on the number of records
2167 * being added or deleted as indicated in rec_diff. Move the records
2168 * and pointers in if_broot to fit the new size. When shrinking this
2169 * will eliminate holes between the records and pointers created by
2170 * the caller. When growing this will create holes to be filled in
2173 * The caller must not request to add more records than would fit in
2174 * the on-disk inode root. If the if_broot is currently NULL, then
2175 * if we adding records one will be allocated. The caller must also
2176 * not request that the number of records go below zero, although
2177 * it can go to zero.
2179 * ip -- the inode whose if_broot area is changing
2180 * ext_diff -- the change in the number of records, positive or negative,
2181 * requested for the if_broot array.
2189 struct xfs_mount *mp = ip->i_mount;
2192 struct xfs_btree_block *new_broot;
2199 * Handle the degenerate case quietly.
2201 if (rec_diff == 0) {
2205 ifp = XFS_IFORK_PTR(ip, whichfork);
2208 * If there wasn't any memory allocated before, just
2209 * allocate it now and get out.
2211 if (ifp->if_broot_bytes == 0) {
2212 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
2213 ifp->if_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
2214 ifp->if_broot_bytes = (int)new_size;
2219 * If there is already an existing if_broot, then we need
2220 * to realloc() it and shift the pointers to their new
2221 * location. The records don't change location because
2222 * they are kept butted up against the btree block header.
2224 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
2225 new_max = cur_max + rec_diff;
2226 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2227 ifp->if_broot = kmem_realloc(ifp->if_broot, new_size,
2228 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2229 KM_SLEEP | KM_NOFS);
2230 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2231 ifp->if_broot_bytes);
2232 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2234 ifp->if_broot_bytes = (int)new_size;
2235 ASSERT(ifp->if_broot_bytes <=
2236 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2237 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2242 * rec_diff is less than 0. In this case, we are shrinking the
2243 * if_broot buffer. It must already exist. If we go to zero
2244 * records, just get rid of the root and clear the status bit.
2246 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
2247 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
2248 new_max = cur_max + rec_diff;
2249 ASSERT(new_max >= 0);
2251 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2255 new_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
2257 * First copy over the btree block header.
2259 memcpy(new_broot, ifp->if_broot, XFS_BTREE_LBLOCK_LEN);
2262 ifp->if_flags &= ~XFS_IFBROOT;
2266 * Only copy the records and pointers if there are any.
2270 * First copy the records.
2272 op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1);
2273 np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1);
2274 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2277 * Then copy the pointers.
2279 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2280 ifp->if_broot_bytes);
2281 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1,
2283 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2285 kmem_free(ifp->if_broot);
2286 ifp->if_broot = new_broot;
2287 ifp->if_broot_bytes = (int)new_size;
2288 ASSERT(ifp->if_broot_bytes <=
2289 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2295 * This is called when the amount of space needed for if_data
2296 * is increased or decreased. The change in size is indicated by
2297 * the number of bytes that need to be added or deleted in the
2298 * byte_diff parameter.
2300 * If the amount of space needed has decreased below the size of the
2301 * inline buffer, then switch to using the inline buffer. Otherwise,
2302 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2303 * to what is needed.
2305 * ip -- the inode whose if_data area is changing
2306 * byte_diff -- the change in the number of bytes, positive or negative,
2307 * requested for the if_data array.
2319 if (byte_diff == 0) {
2323 ifp = XFS_IFORK_PTR(ip, whichfork);
2324 new_size = (int)ifp->if_bytes + byte_diff;
2325 ASSERT(new_size >= 0);
2327 if (new_size == 0) {
2328 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2329 kmem_free(ifp->if_u1.if_data);
2331 ifp->if_u1.if_data = NULL;
2333 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2335 * If the valid extents/data can fit in if_inline_ext/data,
2336 * copy them from the malloc'd vector and free it.
2338 if (ifp->if_u1.if_data == NULL) {
2339 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2340 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2341 ASSERT(ifp->if_real_bytes != 0);
2342 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2344 kmem_free(ifp->if_u1.if_data);
2345 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2350 * Stuck with malloc/realloc.
2351 * For inline data, the underlying buffer must be
2352 * a multiple of 4 bytes in size so that it can be
2353 * logged and stay on word boundaries. We enforce
2356 real_size = roundup(new_size, 4);
2357 if (ifp->if_u1.if_data == NULL) {
2358 ASSERT(ifp->if_real_bytes == 0);
2359 ifp->if_u1.if_data = kmem_alloc(real_size,
2360 KM_SLEEP | KM_NOFS);
2361 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2363 * Only do the realloc if the underlying size
2364 * is really changing.
2366 if (ifp->if_real_bytes != real_size) {
2367 ifp->if_u1.if_data =
2368 kmem_realloc(ifp->if_u1.if_data,
2371 KM_SLEEP | KM_NOFS);
2374 ASSERT(ifp->if_real_bytes == 0);
2375 ifp->if_u1.if_data = kmem_alloc(real_size,
2376 KM_SLEEP | KM_NOFS);
2377 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2381 ifp->if_real_bytes = real_size;
2382 ifp->if_bytes = new_size;
2383 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2393 ifp = XFS_IFORK_PTR(ip, whichfork);
2394 if (ifp->if_broot != NULL) {
2395 kmem_free(ifp->if_broot);
2396 ifp->if_broot = NULL;
2400 * If the format is local, then we can't have an extents
2401 * array so just look for an inline data array. If we're
2402 * not local then we may or may not have an extents list,
2403 * so check and free it up if we do.
2405 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2406 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2407 (ifp->if_u1.if_data != NULL)) {
2408 ASSERT(ifp->if_real_bytes != 0);
2409 kmem_free(ifp->if_u1.if_data);
2410 ifp->if_u1.if_data = NULL;
2411 ifp->if_real_bytes = 0;
2413 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2414 ((ifp->if_flags & XFS_IFEXTIREC) ||
2415 ((ifp->if_u1.if_extents != NULL) &&
2416 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
2417 ASSERT(ifp->if_real_bytes != 0);
2418 xfs_iext_destroy(ifp);
2420 ASSERT(ifp->if_u1.if_extents == NULL ||
2421 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2422 ASSERT(ifp->if_real_bytes == 0);
2423 if (whichfork == XFS_ATTR_FORK) {
2424 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2430 * This is called to unpin an inode. The caller must have the inode locked
2431 * in at least shared mode so that the buffer cannot be subsequently pinned
2432 * once someone is waiting for it to be unpinned.
2436 struct xfs_inode *ip)
2438 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2440 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2442 /* Give the log a push to start the unpinning I/O */
2443 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2449 struct xfs_inode *ip)
2451 if (xfs_ipincount(ip)) {
2452 xfs_iunpin_nowait(ip);
2453 wait_event(ip->i_ipin_wait, (xfs_ipincount(ip) == 0));
2458 * xfs_iextents_copy()
2460 * This is called to copy the REAL extents (as opposed to the delayed
2461 * allocation extents) from the inode into the given buffer. It
2462 * returns the number of bytes copied into the buffer.
2464 * If there are no delayed allocation extents, then we can just
2465 * memcpy() the extents into the buffer. Otherwise, we need to
2466 * examine each extent in turn and skip those which are delayed.
2478 xfs_fsblock_t start_block;
2480 ifp = XFS_IFORK_PTR(ip, whichfork);
2481 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2482 ASSERT(ifp->if_bytes > 0);
2484 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2485 XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork);
2489 * There are some delayed allocation extents in the
2490 * inode, so copy the extents one at a time and skip
2491 * the delayed ones. There must be at least one
2492 * non-delayed extent.
2495 for (i = 0; i < nrecs; i++) {
2496 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
2497 start_block = xfs_bmbt_get_startblock(ep);
2498 if (isnullstartblock(start_block)) {
2500 * It's a delayed allocation extent, so skip it.
2505 /* Translate to on disk format */
2506 put_unaligned(cpu_to_be64(ep->l0), &dp->l0);
2507 put_unaligned(cpu_to_be64(ep->l1), &dp->l1);
2511 ASSERT(copied != 0);
2512 xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip));
2514 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2518 * Each of the following cases stores data into the same region
2519 * of the on-disk inode, so only one of them can be valid at
2520 * any given time. While it is possible to have conflicting formats
2521 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2522 * in EXTENTS format, this can only happen when the fork has
2523 * changed formats after being modified but before being flushed.
2524 * In these cases, the format always takes precedence, because the
2525 * format indicates the current state of the fork.
2532 xfs_inode_log_item_t *iip,
2539 #ifdef XFS_TRANS_DEBUG
2542 static const short brootflag[2] =
2543 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2544 static const short dataflag[2] =
2545 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2546 static const short extflag[2] =
2547 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2551 ifp = XFS_IFORK_PTR(ip, whichfork);
2553 * This can happen if we gave up in iformat in an error path,
2554 * for the attribute fork.
2557 ASSERT(whichfork == XFS_ATTR_FORK);
2560 cp = XFS_DFORK_PTR(dip, whichfork);
2562 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2563 case XFS_DINODE_FMT_LOCAL:
2564 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2565 (ifp->if_bytes > 0)) {
2566 ASSERT(ifp->if_u1.if_data != NULL);
2567 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2568 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2572 case XFS_DINODE_FMT_EXTENTS:
2573 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2574 !(iip->ili_format.ilf_fields & extflag[whichfork]));
2575 ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
2576 (ifp->if_bytes == 0));
2577 ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
2578 (ifp->if_bytes > 0));
2579 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
2580 (ifp->if_bytes > 0)) {
2581 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2582 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2587 case XFS_DINODE_FMT_BTREE:
2588 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
2589 (ifp->if_broot_bytes > 0)) {
2590 ASSERT(ifp->if_broot != NULL);
2591 ASSERT(ifp->if_broot_bytes <=
2592 (XFS_IFORK_SIZE(ip, whichfork) +
2593 XFS_BROOT_SIZE_ADJ));
2594 xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes,
2595 (xfs_bmdr_block_t *)cp,
2596 XFS_DFORK_SIZE(dip, mp, whichfork));
2600 case XFS_DINODE_FMT_DEV:
2601 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
2602 ASSERT(whichfork == XFS_DATA_FORK);
2603 xfs_dinode_put_rdev(dip, ip->i_df.if_u2.if_rdev);
2607 case XFS_DINODE_FMT_UUID:
2608 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
2609 ASSERT(whichfork == XFS_DATA_FORK);
2610 memcpy(XFS_DFORK_DPTR(dip),
2611 &ip->i_df.if_u2.if_uuid,
2627 xfs_mount_t *mp = ip->i_mount;
2628 struct xfs_perag *pag;
2629 unsigned long first_index, mask;
2630 unsigned long inodes_per_cluster;
2632 xfs_inode_t **ilist;
2639 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2641 inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog;
2642 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
2643 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
2647 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
2648 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
2650 /* really need a gang lookup range call here */
2651 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
2652 first_index, inodes_per_cluster);
2656 for (i = 0; i < nr_found; i++) {
2662 * because this is an RCU protected lookup, we could find a
2663 * recently freed or even reallocated inode during the lookup.
2664 * We need to check under the i_flags_lock for a valid inode
2665 * here. Skip it if it is not valid or the wrong inode.
2667 spin_lock(&ip->i_flags_lock);
2669 (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) {
2670 spin_unlock(&ip->i_flags_lock);
2673 spin_unlock(&ip->i_flags_lock);
2676 * Do an un-protected check to see if the inode is dirty and
2677 * is a candidate for flushing. These checks will be repeated
2678 * later after the appropriate locks are acquired.
2680 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
2684 * Try to get locks. If any are unavailable or it is pinned,
2685 * then this inode cannot be flushed and is skipped.
2688 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
2690 if (!xfs_iflock_nowait(iq)) {
2691 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2694 if (xfs_ipincount(iq)) {
2696 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2701 * arriving here means that this inode can be flushed. First
2702 * re-check that it's dirty before flushing.
2704 if (!xfs_inode_clean(iq)) {
2706 error = xfs_iflush_int(iq, bp);
2708 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2709 goto cluster_corrupt_out;
2715 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2719 XFS_STATS_INC(xs_icluster_flushcnt);
2720 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
2731 cluster_corrupt_out:
2733 * Corruption detected in the clustering loop. Invalidate the
2734 * inode buffer and shut down the filesystem.
2738 * Clean up the buffer. If it was B_DELWRI, just release it --
2739 * brelse can handle it with no problems. If not, shut down the
2740 * filesystem before releasing the buffer.
2742 bufwasdelwri = XFS_BUF_ISDELAYWRITE(bp);
2746 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2748 if (!bufwasdelwri) {
2750 * Just like incore_relse: if we have b_iodone functions,
2751 * mark the buffer as an error and call them. Otherwise
2752 * mark it as stale and brelse.
2754 if (XFS_BUF_IODONE_FUNC(bp)) {
2757 XFS_BUF_ERROR(bp,EIO);
2758 xfs_buf_ioend(bp, 0);
2766 * Unlocks the flush lock
2768 xfs_iflush_abort(iq);
2771 return XFS_ERROR(EFSCORRUPTED);
2775 * xfs_iflush() will write a modified inode's changes out to the
2776 * inode's on disk home. The caller must have the inode lock held
2777 * in at least shared mode and the inode flush completion must be
2778 * active as well. The inode lock will still be held upon return from
2779 * the call and the caller is free to unlock it.
2780 * The inode flush will be completed when the inode reaches the disk.
2781 * The flags indicate how the inode's buffer should be written out.
2788 xfs_inode_log_item_t *iip;
2794 XFS_STATS_INC(xs_iflush_count);
2796 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2797 ASSERT(!completion_done(&ip->i_flush));
2798 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2799 ip->i_d.di_nextents > ip->i_df.if_ext_max);
2805 * We can't flush the inode until it is unpinned, so wait for it if we
2806 * are allowed to block. We know noone new can pin it, because we are
2807 * holding the inode lock shared and you need to hold it exclusively to
2810 * If we are not allowed to block, force the log out asynchronously so
2811 * that when we come back the inode will be unpinned. If other inodes
2812 * in the same cluster are dirty, they will probably write the inode
2813 * out for us if they occur after the log force completes.
2815 if (!(flags & SYNC_WAIT) && xfs_ipincount(ip)) {
2816 xfs_iunpin_nowait(ip);
2820 xfs_iunpin_wait(ip);
2823 * For stale inodes we cannot rely on the backing buffer remaining
2824 * stale in cache for the remaining life of the stale inode and so
2825 * xfs_itobp() below may give us a buffer that no longer contains
2826 * inodes below. We have to check this after ensuring the inode is
2827 * unpinned so that it is safe to reclaim the stale inode after the
2830 if (xfs_iflags_test(ip, XFS_ISTALE)) {
2836 * This may have been unpinned because the filesystem is shutting
2837 * down forcibly. If that's the case we must not write this inode
2838 * to disk, because the log record didn't make it to disk!
2840 if (XFS_FORCED_SHUTDOWN(mp)) {
2841 ip->i_update_core = 0;
2843 iip->ili_format.ilf_fields = 0;
2845 return XFS_ERROR(EIO);
2849 * Get the buffer containing the on-disk inode.
2851 error = xfs_itobp(mp, NULL, ip, &dip, &bp,
2852 (flags & SYNC_WAIT) ? XBF_LOCK : XBF_TRYLOCK);
2859 * First flush out the inode that xfs_iflush was called with.
2861 error = xfs_iflush_int(ip, bp);
2866 * If the buffer is pinned then push on the log now so we won't
2867 * get stuck waiting in the write for too long.
2869 if (XFS_BUF_ISPINNED(bp))
2870 xfs_log_force(mp, 0);
2874 * see if other inodes can be gathered into this write
2876 error = xfs_iflush_cluster(ip, bp);
2878 goto cluster_corrupt_out;
2880 if (flags & SYNC_WAIT)
2881 error = xfs_bwrite(mp, bp);
2883 xfs_bdwrite(mp, bp);
2888 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2889 cluster_corrupt_out:
2891 * Unlocks the flush lock
2893 xfs_iflush_abort(ip);
2894 return XFS_ERROR(EFSCORRUPTED);
2903 xfs_inode_log_item_t *iip;
2906 #ifdef XFS_TRANS_DEBUG
2910 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2911 ASSERT(!completion_done(&ip->i_flush));
2912 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2913 ip->i_d.di_nextents > ip->i_df.if_ext_max);
2918 /* set *dip = inode's place in the buffer */
2919 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
2922 * Clear i_update_core before copying out the data.
2923 * This is for coordination with our timestamp updates
2924 * that don't hold the inode lock. They will always
2925 * update the timestamps BEFORE setting i_update_core,
2926 * so if we clear i_update_core after they set it we
2927 * are guaranteed to see their updates to the timestamps.
2928 * I believe that this depends on strongly ordered memory
2929 * semantics, but we have that. We use the SYNCHRONIZE
2930 * macro to make sure that the compiler does not reorder
2931 * the i_update_core access below the data copy below.
2933 ip->i_update_core = 0;
2937 * Make sure to get the latest timestamps from the Linux inode.
2939 xfs_synchronize_times(ip);
2941 if (XFS_TEST_ERROR(be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC,
2942 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
2943 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
2944 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
2945 ip->i_ino, be16_to_cpu(dip->di_magic), dip);
2948 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
2949 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
2950 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
2951 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
2952 ip->i_ino, ip, ip->i_d.di_magic);
2955 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
2957 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
2958 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
2959 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
2960 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
2961 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
2965 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
2967 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
2968 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
2969 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
2970 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
2971 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
2972 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
2977 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
2978 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
2979 XFS_RANDOM_IFLUSH_5)) {
2980 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
2981 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
2983 ip->i_d.di_nextents + ip->i_d.di_anextents,
2988 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
2989 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
2990 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
2991 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
2992 ip->i_ino, ip->i_d.di_forkoff, ip);
2996 * bump the flush iteration count, used to detect flushes which
2997 * postdate a log record during recovery.
3000 ip->i_d.di_flushiter++;
3003 * Copy the dirty parts of the inode into the on-disk
3004 * inode. We always copy out the core of the inode,
3005 * because if the inode is dirty at all the core must
3008 xfs_dinode_to_disk(dip, &ip->i_d);
3010 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3011 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3012 ip->i_d.di_flushiter = 0;
3015 * If this is really an old format inode and the superblock version
3016 * has not been updated to support only new format inodes, then
3017 * convert back to the old inode format. If the superblock version
3018 * has been updated, then make the conversion permanent.
3020 ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
3021 if (ip->i_d.di_version == 1) {
3022 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
3026 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3027 dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink);
3030 * The superblock version has already been bumped,
3031 * so just make the conversion to the new inode
3034 ip->i_d.di_version = 2;
3035 dip->di_version = 2;
3036 ip->i_d.di_onlink = 0;
3038 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3039 memset(&(dip->di_pad[0]), 0,
3040 sizeof(dip->di_pad));
3041 ASSERT(xfs_get_projid(ip) == 0);
3045 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
3046 if (XFS_IFORK_Q(ip))
3047 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3048 xfs_inobp_check(mp, bp);
3051 * We've recorded everything logged in the inode, so we'd
3052 * like to clear the ilf_fields bits so we don't log and
3053 * flush things unnecessarily. However, we can't stop
3054 * logging all this information until the data we've copied
3055 * into the disk buffer is written to disk. If we did we might
3056 * overwrite the copy of the inode in the log with all the
3057 * data after re-logging only part of it, and in the face of
3058 * a crash we wouldn't have all the data we need to recover.
3060 * What we do is move the bits to the ili_last_fields field.
3061 * When logging the inode, these bits are moved back to the
3062 * ilf_fields field. In the xfs_iflush_done() routine we
3063 * clear ili_last_fields, since we know that the information
3064 * those bits represent is permanently on disk. As long as
3065 * the flush completes before the inode is logged again, then
3066 * both ilf_fields and ili_last_fields will be cleared.
3068 * We can play with the ilf_fields bits here, because the inode
3069 * lock must be held exclusively in order to set bits there
3070 * and the flush lock protects the ili_last_fields bits.
3071 * Set ili_logged so the flush done
3072 * routine can tell whether or not to look in the AIL.
3073 * Also, store the current LSN of the inode so that we can tell
3074 * whether the item has moved in the AIL from xfs_iflush_done().
3075 * In order to read the lsn we need the AIL lock, because
3076 * it is a 64 bit value that cannot be read atomically.
3078 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3079 iip->ili_last_fields = iip->ili_format.ilf_fields;
3080 iip->ili_format.ilf_fields = 0;
3081 iip->ili_logged = 1;
3083 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3084 &iip->ili_item.li_lsn);
3087 * Attach the function xfs_iflush_done to the inode's
3088 * buffer. This will remove the inode from the AIL
3089 * and unlock the inode's flush lock when the inode is
3090 * completely written to disk.
3092 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3094 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3095 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3098 * We're flushing an inode which is not in the AIL and has
3099 * not been logged but has i_update_core set. For this
3100 * case we can use a B_DELWRI flush and immediately drop
3101 * the inode flush lock because we can avoid the whole
3102 * AIL state thing. It's OK to drop the flush lock now,
3103 * because we've already locked the buffer and to do anything
3104 * you really need both.
3107 ASSERT(iip->ili_logged == 0);
3108 ASSERT(iip->ili_last_fields == 0);
3109 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3117 return XFS_ERROR(EFSCORRUPTED);
3121 * Return a pointer to the extent record at file index idx.
3123 xfs_bmbt_rec_host_t *
3125 xfs_ifork_t *ifp, /* inode fork pointer */
3126 xfs_extnum_t idx) /* index of target extent */
3129 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
3130 return ifp->if_u1.if_ext_irec->er_extbuf;
3131 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3132 xfs_ext_irec_t *erp; /* irec pointer */
3133 int erp_idx = 0; /* irec index */
3134 xfs_extnum_t page_idx = idx; /* ext index in target list */
3136 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3137 return &erp->er_extbuf[page_idx];
3138 } else if (ifp->if_bytes) {
3139 return &ifp->if_u1.if_extents[idx];
3146 * Insert new item(s) into the extent records for incore inode
3147 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3151 xfs_inode_t *ip, /* incore inode pointer */
3152 xfs_extnum_t idx, /* starting index of new items */
3153 xfs_extnum_t count, /* number of inserted items */
3154 xfs_bmbt_irec_t *new, /* items to insert */
3155 int state) /* type of extent conversion */
3157 xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
3158 xfs_extnum_t i; /* extent record index */
3160 trace_xfs_iext_insert(ip, idx, new, state, _RET_IP_);
3162 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3163 xfs_iext_add(ifp, idx, count);
3164 for (i = idx; i < idx + count; i++, new++)
3165 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new);
3169 * This is called when the amount of space required for incore file
3170 * extents needs to be increased. The ext_diff parameter stores the
3171 * number of new extents being added and the idx parameter contains
3172 * the extent index where the new extents will be added. If the new
3173 * extents are being appended, then we just need to (re)allocate and
3174 * initialize the space. Otherwise, if the new extents are being
3175 * inserted into the middle of the existing entries, a bit more work
3176 * is required to make room for the new extents to be inserted. The
3177 * caller is responsible for filling in the new extent entries upon
3182 xfs_ifork_t *ifp, /* inode fork pointer */
3183 xfs_extnum_t idx, /* index to begin adding exts */
3184 int ext_diff) /* number of extents to add */
3186 int byte_diff; /* new bytes being added */
3187 int new_size; /* size of extents after adding */
3188 xfs_extnum_t nextents; /* number of extents in file */
3190 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3191 ASSERT((idx >= 0) && (idx <= nextents));
3192 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
3193 new_size = ifp->if_bytes + byte_diff;
3195 * If the new number of extents (nextents + ext_diff)
3196 * fits inside the inode, then continue to use the inline
3199 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
3200 if (idx < nextents) {
3201 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
3202 &ifp->if_u2.if_inline_ext[idx],
3203 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3204 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
3206 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3207 ifp->if_real_bytes = 0;
3208 ifp->if_lastex = nextents + ext_diff;
3211 * Otherwise use a linear (direct) extent list.
3212 * If the extents are currently inside the inode,
3213 * xfs_iext_realloc_direct will switch us from
3214 * inline to direct extent allocation mode.
3216 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
3217 xfs_iext_realloc_direct(ifp, new_size);
3218 if (idx < nextents) {
3219 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
3220 &ifp->if_u1.if_extents[idx],
3221 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3222 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
3225 /* Indirection array */
3227 xfs_ext_irec_t *erp;
3231 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
3232 if (ifp->if_flags & XFS_IFEXTIREC) {
3233 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3235 xfs_iext_irec_init(ifp);
3236 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3237 erp = ifp->if_u1.if_ext_irec;
3239 /* Extents fit in target extent page */
3240 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3241 if (page_idx < erp->er_extcount) {
3242 memmove(&erp->er_extbuf[page_idx + ext_diff],
3243 &erp->er_extbuf[page_idx],
3244 (erp->er_extcount - page_idx) *
3245 sizeof(xfs_bmbt_rec_t));
3246 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3248 erp->er_extcount += ext_diff;
3249 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3251 /* Insert a new extent page */
3253 xfs_iext_add_indirect_multi(ifp,
3254 erp_idx, page_idx, ext_diff);
3257 * If extent(s) are being appended to the last page in
3258 * the indirection array and the new extent(s) don't fit
3259 * in the page, then erp is NULL and erp_idx is set to
3260 * the next index needed in the indirection array.
3263 int count = ext_diff;
3266 erp = xfs_iext_irec_new(ifp, erp_idx);
3267 erp->er_extcount = count;
3268 count -= MIN(count, (int)XFS_LINEAR_EXTS);
3275 ifp->if_bytes = new_size;
3279 * This is called when incore extents are being added to the indirection
3280 * array and the new extents do not fit in the target extent list. The
3281 * erp_idx parameter contains the irec index for the target extent list
3282 * in the indirection array, and the idx parameter contains the extent
3283 * index within the list. The number of extents being added is stored
3284 * in the count parameter.
3286 * |-------| |-------|
3287 * | | | | idx - number of extents before idx
3289 * | | | | count - number of extents being inserted at idx
3290 * |-------| |-------|
3291 * | count | | nex2 | nex2 - number of extents after idx + count
3292 * |-------| |-------|
3295 xfs_iext_add_indirect_multi(
3296 xfs_ifork_t *ifp, /* inode fork pointer */
3297 int erp_idx, /* target extent irec index */
3298 xfs_extnum_t idx, /* index within target list */
3299 int count) /* new extents being added */
3301 int byte_diff; /* new bytes being added */
3302 xfs_ext_irec_t *erp; /* pointer to irec entry */
3303 xfs_extnum_t ext_diff; /* number of extents to add */
3304 xfs_extnum_t ext_cnt; /* new extents still needed */
3305 xfs_extnum_t nex2; /* extents after idx + count */
3306 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3307 int nlists; /* number of irec's (lists) */
3309 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3310 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3311 nex2 = erp->er_extcount - idx;
3312 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3315 * Save second part of target extent list
3316 * (all extents past */
3318 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3319 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS);
3320 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3321 erp->er_extcount -= nex2;
3322 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3323 memset(&erp->er_extbuf[idx], 0, byte_diff);
3327 * Add the new extents to the end of the target
3328 * list, then allocate new irec record(s) and
3329 * extent buffer(s) as needed to store the rest
3330 * of the new extents.
3333 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3335 erp->er_extcount += ext_diff;
3336 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3337 ext_cnt -= ext_diff;
3341 erp = xfs_iext_irec_new(ifp, erp_idx);
3342 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3343 erp->er_extcount = ext_diff;
3344 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3345 ext_cnt -= ext_diff;
3348 /* Add nex2 extents back to indirection array */
3350 xfs_extnum_t ext_avail;
3353 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3354 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3357 * If nex2 extents fit in the current page, append
3358 * nex2_ep after the new extents.
3360 if (nex2 <= ext_avail) {
3361 i = erp->er_extcount;
3364 * Otherwise, check if space is available in the
3367 else if ((erp_idx < nlists - 1) &&
3368 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3369 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3372 /* Create a hole for nex2 extents */
3373 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3374 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3377 * Final choice, create a new extent page for
3382 erp = xfs_iext_irec_new(ifp, erp_idx);
3384 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
3386 erp->er_extcount += nex2;
3387 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3392 * This is called when the amount of space required for incore file
3393 * extents needs to be decreased. The ext_diff parameter stores the
3394 * number of extents to be removed and the idx parameter contains
3395 * the extent index where the extents will be removed from.
3397 * If the amount of space needed has decreased below the linear
3398 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3399 * extent array. Otherwise, use kmem_realloc() to adjust the
3400 * size to what is needed.
3404 xfs_inode_t *ip, /* incore inode pointer */
3405 xfs_extnum_t idx, /* index to begin removing exts */
3406 int ext_diff, /* number of extents to remove */
3407 int state) /* type of extent conversion */
3409 xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
3410 xfs_extnum_t nextents; /* number of extents in file */
3411 int new_size; /* size of extents after removal */
3413 trace_xfs_iext_remove(ip, idx, state, _RET_IP_);
3415 ASSERT(ext_diff > 0);
3416 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3417 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3419 if (new_size == 0) {
3420 xfs_iext_destroy(ifp);
3421 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3422 xfs_iext_remove_indirect(ifp, idx, ext_diff);
3423 } else if (ifp->if_real_bytes) {
3424 xfs_iext_remove_direct(ifp, idx, ext_diff);
3426 xfs_iext_remove_inline(ifp, idx, ext_diff);
3428 ifp->if_bytes = new_size;
3432 * This removes ext_diff extents from the inline buffer, beginning
3433 * at extent index idx.
3436 xfs_iext_remove_inline(
3437 xfs_ifork_t *ifp, /* inode fork pointer */
3438 xfs_extnum_t idx, /* index to begin removing exts */
3439 int ext_diff) /* number of extents to remove */
3441 int nextents; /* number of extents in file */
3443 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3444 ASSERT(idx < XFS_INLINE_EXTS);
3445 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3446 ASSERT(((nextents - ext_diff) > 0) &&
3447 (nextents - ext_diff) < XFS_INLINE_EXTS);
3449 if (idx + ext_diff < nextents) {
3450 memmove(&ifp->if_u2.if_inline_ext[idx],
3451 &ifp->if_u2.if_inline_ext[idx + ext_diff],
3452 (nextents - (idx + ext_diff)) *
3453 sizeof(xfs_bmbt_rec_t));
3454 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
3455 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3457 memset(&ifp->if_u2.if_inline_ext[idx], 0,
3458 ext_diff * sizeof(xfs_bmbt_rec_t));
3463 * This removes ext_diff extents from a linear (direct) extent list,
3464 * beginning at extent index idx. If the extents are being removed
3465 * from the end of the list (ie. truncate) then we just need to re-
3466 * allocate the list to remove the extra space. Otherwise, if the
3467 * extents are being removed from the middle of the existing extent
3468 * entries, then we first need to move the extent records beginning
3469 * at idx + ext_diff up in the list to overwrite the records being
3470 * removed, then remove the extra space via kmem_realloc.
3473 xfs_iext_remove_direct(
3474 xfs_ifork_t *ifp, /* inode fork pointer */
3475 xfs_extnum_t idx, /* index to begin removing exts */
3476 int ext_diff) /* number of extents to remove */
3478 xfs_extnum_t nextents; /* number of extents in file */
3479 int new_size; /* size of extents after removal */
3481 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3482 new_size = ifp->if_bytes -
3483 (ext_diff * sizeof(xfs_bmbt_rec_t));
3484 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3486 if (new_size == 0) {
3487 xfs_iext_destroy(ifp);
3490 /* Move extents up in the list (if needed) */
3491 if (idx + ext_diff < nextents) {
3492 memmove(&ifp->if_u1.if_extents[idx],
3493 &ifp->if_u1.if_extents[idx + ext_diff],
3494 (nextents - (idx + ext_diff)) *
3495 sizeof(xfs_bmbt_rec_t));
3497 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
3498 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3500 * Reallocate the direct extent list. If the extents
3501 * will fit inside the inode then xfs_iext_realloc_direct
3502 * will switch from direct to inline extent allocation
3505 xfs_iext_realloc_direct(ifp, new_size);
3506 ifp->if_bytes = new_size;
3510 * This is called when incore extents are being removed from the
3511 * indirection array and the extents being removed span multiple extent
3512 * buffers. The idx parameter contains the file extent index where we
3513 * want to begin removing extents, and the count parameter contains
3514 * how many extents need to be removed.
3516 * |-------| |-------|
3517 * | nex1 | | | nex1 - number of extents before idx
3518 * |-------| | count |
3519 * | | | | count - number of extents being removed at idx
3520 * | count | |-------|
3521 * | | | nex2 | nex2 - number of extents after idx + count
3522 * |-------| |-------|
3525 xfs_iext_remove_indirect(
3526 xfs_ifork_t *ifp, /* inode fork pointer */
3527 xfs_extnum_t idx, /* index to begin removing extents */
3528 int count) /* number of extents to remove */
3530 xfs_ext_irec_t *erp; /* indirection array pointer */
3531 int erp_idx = 0; /* indirection array index */
3532 xfs_extnum_t ext_cnt; /* extents left to remove */
3533 xfs_extnum_t ext_diff; /* extents to remove in current list */
3534 xfs_extnum_t nex1; /* number of extents before idx */
3535 xfs_extnum_t nex2; /* extents after idx + count */
3536 int page_idx = idx; /* index in target extent list */
3538 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3539 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3540 ASSERT(erp != NULL);
3544 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
3545 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
3547 * Check for deletion of entire list;
3548 * xfs_iext_irec_remove() updates extent offsets.
3550 if (ext_diff == erp->er_extcount) {
3551 xfs_iext_irec_remove(ifp, erp_idx);
3552 ext_cnt -= ext_diff;
3555 ASSERT(erp_idx < ifp->if_real_bytes /
3557 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3564 /* Move extents up (if needed) */
3566 memmove(&erp->er_extbuf[nex1],
3567 &erp->er_extbuf[nex1 + ext_diff],
3568 nex2 * sizeof(xfs_bmbt_rec_t));
3570 /* Zero out rest of page */
3571 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
3572 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
3573 /* Update remaining counters */
3574 erp->er_extcount -= ext_diff;
3575 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
3576 ext_cnt -= ext_diff;
3581 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
3582 xfs_iext_irec_compact(ifp);
3586 * Create, destroy, or resize a linear (direct) block of extents.
3589 xfs_iext_realloc_direct(
3590 xfs_ifork_t *ifp, /* inode fork pointer */
3591 int new_size) /* new size of extents */
3593 int rnew_size; /* real new size of extents */
3595 rnew_size = new_size;
3597 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
3598 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
3599 (new_size != ifp->if_real_bytes)));
3601 /* Free extent records */
3602 if (new_size == 0) {
3603 xfs_iext_destroy(ifp);
3605 /* Resize direct extent list and zero any new bytes */
3606 else if (ifp->if_real_bytes) {
3607 /* Check if extents will fit inside the inode */
3608 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
3609 xfs_iext_direct_to_inline(ifp, new_size /
3610 (uint)sizeof(xfs_bmbt_rec_t));
3611 ifp->if_bytes = new_size;
3614 if (!is_power_of_2(new_size)){
3615 rnew_size = roundup_pow_of_two(new_size);
3617 if (rnew_size != ifp->if_real_bytes) {
3618 ifp->if_u1.if_extents =
3619 kmem_realloc(ifp->if_u1.if_extents,
3621 ifp->if_real_bytes, KM_NOFS);
3623 if (rnew_size > ifp->if_real_bytes) {
3624 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
3625 (uint)sizeof(xfs_bmbt_rec_t)], 0,
3626 rnew_size - ifp->if_real_bytes);
3630 * Switch from the inline extent buffer to a direct
3631 * extent list. Be sure to include the inline extent
3632 * bytes in new_size.
3635 new_size += ifp->if_bytes;
3636 if (!is_power_of_2(new_size)) {
3637 rnew_size = roundup_pow_of_two(new_size);
3639 xfs_iext_inline_to_direct(ifp, rnew_size);
3641 ifp->if_real_bytes = rnew_size;
3642 ifp->if_bytes = new_size;
3646 * Switch from linear (direct) extent records to inline buffer.
3649 xfs_iext_direct_to_inline(
3650 xfs_ifork_t *ifp, /* inode fork pointer */
3651 xfs_extnum_t nextents) /* number of extents in file */
3653 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3654 ASSERT(nextents <= XFS_INLINE_EXTS);
3656 * The inline buffer was zeroed when we switched
3657 * from inline to direct extent allocation mode,
3658 * so we don't need to clear it here.
3660 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
3661 nextents * sizeof(xfs_bmbt_rec_t));
3662 kmem_free(ifp->if_u1.if_extents);
3663 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3664 ifp->if_real_bytes = 0;
3668 * Switch from inline buffer to linear (direct) extent records.
3669 * new_size should already be rounded up to the next power of 2
3670 * by the caller (when appropriate), so use new_size as it is.
3671 * However, since new_size may be rounded up, we can't update
3672 * if_bytes here. It is the caller's responsibility to update
3673 * if_bytes upon return.
3676 xfs_iext_inline_to_direct(
3677 xfs_ifork_t *ifp, /* inode fork pointer */
3678 int new_size) /* number of extents in file */
3680 ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS);
3681 memset(ifp->if_u1.if_extents, 0, new_size);
3682 if (ifp->if_bytes) {
3683 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
3685 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3686 sizeof(xfs_bmbt_rec_t));
3688 ifp->if_real_bytes = new_size;
3692 * Resize an extent indirection array to new_size bytes.
3695 xfs_iext_realloc_indirect(
3696 xfs_ifork_t *ifp, /* inode fork pointer */
3697 int new_size) /* new indirection array size */
3699 int nlists; /* number of irec's (ex lists) */
3700 int size; /* current indirection array size */
3702 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3703 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3704 size = nlists * sizeof(xfs_ext_irec_t);
3705 ASSERT(ifp->if_real_bytes);
3706 ASSERT((new_size >= 0) && (new_size != size));
3707 if (new_size == 0) {
3708 xfs_iext_destroy(ifp);
3710 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
3711 kmem_realloc(ifp->if_u1.if_ext_irec,
3712 new_size, size, KM_NOFS);
3717 * Switch from indirection array to linear (direct) extent allocations.
3720 xfs_iext_indirect_to_direct(
3721 xfs_ifork_t *ifp) /* inode fork pointer */
3723 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
3724 xfs_extnum_t nextents; /* number of extents in file */
3725 int size; /* size of file extents */
3727 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3728 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3729 ASSERT(nextents <= XFS_LINEAR_EXTS);
3730 size = nextents * sizeof(xfs_bmbt_rec_t);
3732 xfs_iext_irec_compact_pages(ifp);
3733 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
3735 ep = ifp->if_u1.if_ext_irec->er_extbuf;
3736 kmem_free(ifp->if_u1.if_ext_irec);
3737 ifp->if_flags &= ~XFS_IFEXTIREC;
3738 ifp->if_u1.if_extents = ep;
3739 ifp->if_bytes = size;
3740 if (nextents < XFS_LINEAR_EXTS) {
3741 xfs_iext_realloc_direct(ifp, size);
3746 * Free incore file extents.
3750 xfs_ifork_t *ifp) /* inode fork pointer */
3752 if (ifp->if_flags & XFS_IFEXTIREC) {
3756 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3757 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
3758 xfs_iext_irec_remove(ifp, erp_idx);
3760 ifp->if_flags &= ~XFS_IFEXTIREC;
3761 } else if (ifp->if_real_bytes) {
3762 kmem_free(ifp->if_u1.if_extents);
3763 } else if (ifp->if_bytes) {
3764 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3765 sizeof(xfs_bmbt_rec_t));
3767 ifp->if_u1.if_extents = NULL;
3768 ifp->if_real_bytes = 0;
3773 * Return a pointer to the extent record for file system block bno.
3775 xfs_bmbt_rec_host_t * /* pointer to found extent record */
3776 xfs_iext_bno_to_ext(
3777 xfs_ifork_t *ifp, /* inode fork pointer */
3778 xfs_fileoff_t bno, /* block number to search for */
3779 xfs_extnum_t *idxp) /* index of target extent */
3781 xfs_bmbt_rec_host_t *base; /* pointer to first extent */
3782 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
3783 xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */
3784 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
3785 int high; /* upper boundary in search */
3786 xfs_extnum_t idx = 0; /* index of target extent */
3787 int low; /* lower boundary in search */
3788 xfs_extnum_t nextents; /* number of file extents */
3789 xfs_fileoff_t startoff = 0; /* start offset of extent */
3791 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3792 if (nextents == 0) {
3797 if (ifp->if_flags & XFS_IFEXTIREC) {
3798 /* Find target extent list */
3800 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
3801 base = erp->er_extbuf;
3802 high = erp->er_extcount - 1;
3804 base = ifp->if_u1.if_extents;
3805 high = nextents - 1;
3807 /* Binary search extent records */
3808 while (low <= high) {
3809 idx = (low + high) >> 1;
3811 startoff = xfs_bmbt_get_startoff(ep);
3812 blockcount = xfs_bmbt_get_blockcount(ep);
3813 if (bno < startoff) {
3815 } else if (bno >= startoff + blockcount) {
3818 /* Convert back to file-based extent index */
3819 if (ifp->if_flags & XFS_IFEXTIREC) {
3820 idx += erp->er_extoff;
3826 /* Convert back to file-based extent index */
3827 if (ifp->if_flags & XFS_IFEXTIREC) {
3828 idx += erp->er_extoff;
3830 if (bno >= startoff + blockcount) {
3831 if (++idx == nextents) {
3834 ep = xfs_iext_get_ext(ifp, idx);
3842 * Return a pointer to the indirection array entry containing the
3843 * extent record for filesystem block bno. Store the index of the
3844 * target irec in *erp_idxp.
3846 xfs_ext_irec_t * /* pointer to found extent record */
3847 xfs_iext_bno_to_irec(
3848 xfs_ifork_t *ifp, /* inode fork pointer */
3849 xfs_fileoff_t bno, /* block number to search for */
3850 int *erp_idxp) /* irec index of target ext list */
3852 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
3853 xfs_ext_irec_t *erp_next; /* next indirection array entry */
3854 int erp_idx; /* indirection array index */
3855 int nlists; /* number of extent irec's (lists) */
3856 int high; /* binary search upper limit */
3857 int low; /* binary search lower limit */
3859 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3860 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3864 while (low <= high) {
3865 erp_idx = (low + high) >> 1;
3866 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3867 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
3868 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
3870 } else if (erp_next && bno >=
3871 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
3877 *erp_idxp = erp_idx;
3882 * Return a pointer to the indirection array entry containing the
3883 * extent record at file extent index *idxp. Store the index of the
3884 * target irec in *erp_idxp and store the page index of the target
3885 * extent record in *idxp.
3888 xfs_iext_idx_to_irec(
3889 xfs_ifork_t *ifp, /* inode fork pointer */
3890 xfs_extnum_t *idxp, /* extent index (file -> page) */
3891 int *erp_idxp, /* pointer to target irec */
3892 int realloc) /* new bytes were just added */
3894 xfs_ext_irec_t *prev; /* pointer to previous irec */
3895 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
3896 int erp_idx; /* indirection array index */
3897 int nlists; /* number of irec's (ex lists) */
3898 int high; /* binary search upper limit */
3899 int low; /* binary search lower limit */
3900 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
3902 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3903 ASSERT(page_idx >= 0 && page_idx <=
3904 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
3905 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3910 /* Binary search extent irec's */
3911 while (low <= high) {
3912 erp_idx = (low + high) >> 1;
3913 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3914 prev = erp_idx > 0 ? erp - 1 : NULL;
3915 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
3916 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
3918 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
3919 (page_idx == erp->er_extoff + erp->er_extcount &&
3922 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
3923 erp->er_extcount == XFS_LINEAR_EXTS) {
3927 erp = erp_idx < nlists ? erp + 1 : NULL;
3930 page_idx -= erp->er_extoff;
3935 *erp_idxp = erp_idx;
3940 * Allocate and initialize an indirection array once the space needed
3941 * for incore extents increases above XFS_IEXT_BUFSZ.
3945 xfs_ifork_t *ifp) /* inode fork pointer */
3947 xfs_ext_irec_t *erp; /* indirection array pointer */
3948 xfs_extnum_t nextents; /* number of extents in file */
3950 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3951 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3952 ASSERT(nextents <= XFS_LINEAR_EXTS);
3954 erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS);
3956 if (nextents == 0) {
3957 ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
3958 } else if (!ifp->if_real_bytes) {
3959 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
3960 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
3961 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
3963 erp->er_extbuf = ifp->if_u1.if_extents;
3964 erp->er_extcount = nextents;
3967 ifp->if_flags |= XFS_IFEXTIREC;
3968 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
3969 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
3970 ifp->if_u1.if_ext_irec = erp;
3976 * Allocate and initialize a new entry in the indirection array.
3980 xfs_ifork_t *ifp, /* inode fork pointer */
3981 int erp_idx) /* index for new irec */
3983 xfs_ext_irec_t *erp; /* indirection array pointer */
3984 int i; /* loop counter */
3985 int nlists; /* number of irec's (ex lists) */
3987 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3988 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3990 /* Resize indirection array */
3991 xfs_iext_realloc_indirect(ifp, ++nlists *
3992 sizeof(xfs_ext_irec_t));
3994 * Move records down in the array so the
3995 * new page can use erp_idx.
3997 erp = ifp->if_u1.if_ext_irec;
3998 for (i = nlists - 1; i > erp_idx; i--) {
3999 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
4001 ASSERT(i == erp_idx);
4003 /* Initialize new extent record */
4004 erp = ifp->if_u1.if_ext_irec;
4005 erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
4006 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4007 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
4008 erp[erp_idx].er_extcount = 0;
4009 erp[erp_idx].er_extoff = erp_idx > 0 ?
4010 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
4011 return (&erp[erp_idx]);
4015 * Remove a record from the indirection array.
4018 xfs_iext_irec_remove(
4019 xfs_ifork_t *ifp, /* inode fork pointer */
4020 int erp_idx) /* irec index to remove */
4022 xfs_ext_irec_t *erp; /* indirection array pointer */
4023 int i; /* loop counter */
4024 int nlists; /* number of irec's (ex lists) */
4026 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4027 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4028 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4029 if (erp->er_extbuf) {
4030 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
4032 kmem_free(erp->er_extbuf);
4034 /* Compact extent records */
4035 erp = ifp->if_u1.if_ext_irec;
4036 for (i = erp_idx; i < nlists - 1; i++) {
4037 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
4040 * Manually free the last extent record from the indirection
4041 * array. A call to xfs_iext_realloc_indirect() with a size
4042 * of zero would result in a call to xfs_iext_destroy() which
4043 * would in turn call this function again, creating a nasty
4047 xfs_iext_realloc_indirect(ifp,
4048 nlists * sizeof(xfs_ext_irec_t));
4050 kmem_free(ifp->if_u1.if_ext_irec);
4052 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4056 * This is called to clean up large amounts of unused memory allocated
4057 * by the indirection array. Before compacting anything though, verify
4058 * that the indirection array is still needed and switch back to the
4059 * linear extent list (or even the inline buffer) if possible. The
4060 * compaction policy is as follows:
4062 * Full Compaction: Extents fit into a single page (or inline buffer)
4063 * Partial Compaction: Extents occupy less than 50% of allocated space
4064 * No Compaction: Extents occupy at least 50% of allocated space
4067 xfs_iext_irec_compact(
4068 xfs_ifork_t *ifp) /* inode fork pointer */
4070 xfs_extnum_t nextents; /* number of extents in file */
4071 int nlists; /* number of irec's (ex lists) */
4073 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4074 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4075 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4077 if (nextents == 0) {
4078 xfs_iext_destroy(ifp);
4079 } else if (nextents <= XFS_INLINE_EXTS) {
4080 xfs_iext_indirect_to_direct(ifp);
4081 xfs_iext_direct_to_inline(ifp, nextents);
4082 } else if (nextents <= XFS_LINEAR_EXTS) {
4083 xfs_iext_indirect_to_direct(ifp);
4084 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
4085 xfs_iext_irec_compact_pages(ifp);
4090 * Combine extents from neighboring extent pages.
4093 xfs_iext_irec_compact_pages(
4094 xfs_ifork_t *ifp) /* inode fork pointer */
4096 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
4097 int erp_idx = 0; /* indirection array index */
4098 int nlists; /* number of irec's (ex lists) */
4100 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4101 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4102 while (erp_idx < nlists - 1) {
4103 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4105 if (erp_next->er_extcount <=
4106 (XFS_LINEAR_EXTS - erp->er_extcount)) {
4107 memcpy(&erp->er_extbuf[erp->er_extcount],
4108 erp_next->er_extbuf, erp_next->er_extcount *
4109 sizeof(xfs_bmbt_rec_t));
4110 erp->er_extcount += erp_next->er_extcount;
4112 * Free page before removing extent record
4113 * so er_extoffs don't get modified in
4114 * xfs_iext_irec_remove.
4116 kmem_free(erp_next->er_extbuf);
4117 erp_next->er_extbuf = NULL;
4118 xfs_iext_irec_remove(ifp, erp_idx + 1);
4119 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4127 * This is called to update the er_extoff field in the indirection
4128 * array when extents have been added or removed from one of the
4129 * extent lists. erp_idx contains the irec index to begin updating
4130 * at and ext_diff contains the number of extents that were added
4134 xfs_iext_irec_update_extoffs(
4135 xfs_ifork_t *ifp, /* inode fork pointer */
4136 int erp_idx, /* irec index to update */
4137 int ext_diff) /* number of new extents */
4139 int i; /* loop counter */
4140 int nlists; /* number of irec's (ex lists */
4142 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4143 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4144 for (i = erp_idx; i < nlists; i++) {
4145 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;