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"
27 #include "xfs_trans.h"
28 #include "xfs_trans_priv.h"
32 #include "xfs_dmapi.h"
33 #include "xfs_mount.h"
34 #include "xfs_bmap_btree.h"
35 #include "xfs_alloc_btree.h"
36 #include "xfs_ialloc_btree.h"
37 #include "xfs_dir2_sf.h"
38 #include "xfs_attr_sf.h"
39 #include "xfs_dinode.h"
40 #include "xfs_inode.h"
41 #include "xfs_buf_item.h"
42 #include "xfs_inode_item.h"
43 #include "xfs_btree.h"
44 #include "xfs_alloc.h"
45 #include "xfs_ialloc.h"
48 #include "xfs_error.h"
49 #include "xfs_utils.h"
50 #include "xfs_dir2_trace.h"
51 #include "xfs_quota.h"
53 #include "xfs_filestream.h"
54 #include "xfs_vnodeops.h"
56 kmem_zone_t *xfs_ifork_zone;
57 kmem_zone_t *xfs_inode_zone;
60 * Used in xfs_itruncate(). This is the maximum number of extents
61 * freed from a file in a single transaction.
63 #define XFS_ITRUNC_MAX_EXTENTS 2
65 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
66 STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
67 STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
68 STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
72 * Make sure that the extents in the given memory buffer
82 xfs_bmbt_rec_host_t rec;
85 for (i = 0; i < nrecs; i++) {
86 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
87 rec.l0 = get_unaligned(&ep->l0);
88 rec.l1 = get_unaligned(&ep->l1);
89 xfs_bmbt_get_all(&rec, &irec);
90 if (fmt == XFS_EXTFMT_NOSTATE)
91 ASSERT(irec.br_state == XFS_EXT_NORM);
95 #define xfs_validate_extents(ifp, nrecs, fmt)
99 * Check that none of the inode's in the buffer have a next
100 * unlinked field of 0.
112 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
114 for (i = 0; i < j; i++) {
115 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
116 i * mp->m_sb.sb_inodesize);
117 if (!dip->di_next_unlinked) {
118 xfs_fs_cmn_err(CE_ALERT, mp,
119 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
121 ASSERT(dip->di_next_unlinked);
128 * Find the buffer associated with the given inode map
129 * We do basic validation checks on the buffer once it has been
130 * retrieved from disk.
146 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno,
147 (int)imap->im_len, buf_flags, &bp);
149 if (error != EAGAIN) {
151 "xfs_imap_to_bp: xfs_trans_read_buf()returned "
152 "an error %d on %s. Returning error.",
153 error, mp->m_fsname);
155 ASSERT(buf_flags & XFS_BUF_TRYLOCK);
161 * Validate the magic number and version of every inode in the buffer
162 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
165 ni = BBTOB(imap->im_len) >> mp->m_sb.sb_inodelog;
166 #else /* usual case */
170 for (i = 0; i < ni; i++) {
174 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
175 (i << mp->m_sb.sb_inodelog));
176 di_ok = be16_to_cpu(dip->di_core.di_magic) == XFS_DINODE_MAGIC &&
177 XFS_DINODE_GOOD_VERSION(dip->di_core.di_version);
178 if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
179 XFS_ERRTAG_ITOBP_INOTOBP,
180 XFS_RANDOM_ITOBP_INOTOBP))) {
181 if (imap_flags & XFS_IMAP_BULKSTAT) {
182 xfs_trans_brelse(tp, bp);
183 return XFS_ERROR(EINVAL);
185 XFS_CORRUPTION_ERROR("xfs_imap_to_bp",
186 XFS_ERRLEVEL_HIGH, mp, dip);
189 "Device %s - bad inode magic/vsn "
190 "daddr %lld #%d (magic=%x)",
191 XFS_BUFTARG_NAME(mp->m_ddev_targp),
192 (unsigned long long)imap->im_blkno, i,
193 be16_to_cpu(dip->di_core.di_magic));
195 xfs_trans_brelse(tp, bp);
196 return XFS_ERROR(EFSCORRUPTED);
200 xfs_inobp_check(mp, bp);
203 * Mark the buffer as an inode buffer now that it looks good
205 XFS_BUF_SET_VTYPE(bp, B_FS_INO);
212 * This routine is called to map an inode number within a file
213 * system to the buffer containing the on-disk version of the
214 * inode. It returns a pointer to the buffer containing the
215 * on-disk inode in the bpp parameter, and in the dip parameter
216 * it returns a pointer to the on-disk inode within that buffer.
218 * If a non-zero error is returned, then the contents of bpp and
219 * dipp are undefined.
221 * Use xfs_imap() to determine the size and location of the
222 * buffer to read from disk.
238 error = xfs_imap(mp, tp, ino, &imap, XFS_IMAP_LOOKUP);
242 error = xfs_imap_to_bp(mp, tp, &imap, &bp, XFS_BUF_LOCK, 0);
246 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
248 *offset = imap.im_boffset;
254 * This routine is called to map an inode to the buffer containing
255 * the on-disk version of the inode. It returns a pointer to the
256 * buffer containing the on-disk inode in the bpp parameter, and in
257 * the dip parameter it returns a pointer to the on-disk inode within
260 * If a non-zero error is returned, then the contents of bpp and
261 * dipp are undefined.
263 * If the inode is new and has not yet been initialized, use xfs_imap()
264 * to determine the size and location of the buffer to read from disk.
265 * If the inode has already been mapped to its buffer and read in once,
266 * then use the mapping information stored in the inode rather than
267 * calling xfs_imap(). This allows us to avoid the overhead of looking
268 * at the inode btree for small block file systems (see xfs_dilocate()).
269 * We can tell whether the inode has been mapped in before by comparing
270 * its disk block address to 0. Only uninitialized inodes will have
271 * 0 for the disk block address.
288 if (ip->i_blkno == (xfs_daddr_t)0) {
290 error = xfs_imap(mp, tp, ip->i_ino, &imap,
291 XFS_IMAP_LOOKUP | imap_flags);
296 * Fill in the fields in the inode that will be used to
297 * map the inode to its buffer from now on.
299 ip->i_blkno = imap.im_blkno;
300 ip->i_len = imap.im_len;
301 ip->i_boffset = imap.im_boffset;
304 * We've already mapped the inode once, so just use the
305 * mapping that we saved the first time.
307 imap.im_blkno = ip->i_blkno;
308 imap.im_len = ip->i_len;
309 imap.im_boffset = ip->i_boffset;
311 ASSERT(bno == 0 || bno == imap.im_blkno);
313 error = xfs_imap_to_bp(mp, tp, &imap, &bp, buf_flags, imap_flags);
318 ASSERT(buf_flags & XFS_BUF_TRYLOCK);
324 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
330 * Move inode type and inode format specific information from the
331 * on-disk inode to the in-core inode. For fifos, devs, and sockets
332 * this means set if_rdev to the proper value. For files, directories,
333 * and symlinks this means to bring in the in-line data or extent
334 * pointers. For a file in B-tree format, only the root is immediately
335 * brought in-core. The rest will be in-lined in if_extents when it
336 * is first referenced (see xfs_iread_extents()).
343 xfs_attr_shortform_t *atp;
347 ip->i_df.if_ext_max =
348 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
351 if (unlikely(be32_to_cpu(dip->di_core.di_nextents) +
352 be16_to_cpu(dip->di_core.di_anextents) >
353 be64_to_cpu(dip->di_core.di_nblocks))) {
354 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
355 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
356 (unsigned long long)ip->i_ino,
357 (int)(be32_to_cpu(dip->di_core.di_nextents) +
358 be16_to_cpu(dip->di_core.di_anextents)),
360 be64_to_cpu(dip->di_core.di_nblocks));
361 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
363 return XFS_ERROR(EFSCORRUPTED);
366 if (unlikely(dip->di_core.di_forkoff > ip->i_mount->m_sb.sb_inodesize)) {
367 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
368 "corrupt dinode %Lu, forkoff = 0x%x.",
369 (unsigned long long)ip->i_ino,
370 dip->di_core.di_forkoff);
371 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
373 return XFS_ERROR(EFSCORRUPTED);
376 switch (ip->i_d.di_mode & S_IFMT) {
381 if (unlikely(dip->di_core.di_format != XFS_DINODE_FMT_DEV)) {
382 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
384 return XFS_ERROR(EFSCORRUPTED);
388 ip->i_df.if_u2.if_rdev = be32_to_cpu(dip->di_u.di_dev);
394 switch (dip->di_core.di_format) {
395 case XFS_DINODE_FMT_LOCAL:
397 * no local regular files yet
399 if (unlikely((be16_to_cpu(dip->di_core.di_mode) & S_IFMT) == S_IFREG)) {
400 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
402 "(local format for regular file).",
403 (unsigned long long) ip->i_ino);
404 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
407 return XFS_ERROR(EFSCORRUPTED);
410 di_size = be64_to_cpu(dip->di_core.di_size);
411 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
412 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
414 "(bad size %Ld for local inode).",
415 (unsigned long long) ip->i_ino,
416 (long long) di_size);
417 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
420 return XFS_ERROR(EFSCORRUPTED);
424 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
426 case XFS_DINODE_FMT_EXTENTS:
427 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
429 case XFS_DINODE_FMT_BTREE:
430 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
433 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
435 return XFS_ERROR(EFSCORRUPTED);
440 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
441 return XFS_ERROR(EFSCORRUPTED);
446 if (!XFS_DFORK_Q(dip))
448 ASSERT(ip->i_afp == NULL);
449 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
450 ip->i_afp->if_ext_max =
451 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
452 switch (dip->di_core.di_aformat) {
453 case XFS_DINODE_FMT_LOCAL:
454 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
455 size = be16_to_cpu(atp->hdr.totsize);
456 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
458 case XFS_DINODE_FMT_EXTENTS:
459 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
461 case XFS_DINODE_FMT_BTREE:
462 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
465 error = XFS_ERROR(EFSCORRUPTED);
469 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
471 xfs_idestroy_fork(ip, XFS_DATA_FORK);
477 * The file is in-lined in the on-disk inode.
478 * If it fits into if_inline_data, then copy
479 * it there, otherwise allocate a buffer for it
480 * and copy the data there. Either way, set
481 * if_data to point at the data.
482 * If we allocate a buffer for the data, make
483 * sure that its size is a multiple of 4 and
484 * record the real size in i_real_bytes.
497 * If the size is unreasonable, then something
498 * is wrong and we just bail out rather than crash in
499 * kmem_alloc() or memcpy() below.
501 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
502 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
504 "(bad size %d for local fork, size = %d).",
505 (unsigned long long) ip->i_ino, size,
506 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
507 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
509 return XFS_ERROR(EFSCORRUPTED);
511 ifp = XFS_IFORK_PTR(ip, whichfork);
514 ifp->if_u1.if_data = NULL;
515 else if (size <= sizeof(ifp->if_u2.if_inline_data))
516 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
518 real_size = roundup(size, 4);
519 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
521 ifp->if_bytes = size;
522 ifp->if_real_bytes = real_size;
524 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
525 ifp->if_flags &= ~XFS_IFEXTENTS;
526 ifp->if_flags |= XFS_IFINLINE;
531 * The file consists of a set of extents all
532 * of which fit into the on-disk inode.
533 * If there are few enough extents to fit into
534 * the if_inline_ext, then copy them there.
535 * Otherwise allocate a buffer for them and copy
536 * them into it. Either way, set if_extents
537 * to point at the extents.
551 ifp = XFS_IFORK_PTR(ip, whichfork);
552 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
553 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
556 * If the number of extents is unreasonable, then something
557 * is wrong and we just bail out rather than crash in
558 * kmem_alloc() or memcpy() below.
560 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
561 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
562 "corrupt inode %Lu ((a)extents = %d).",
563 (unsigned long long) ip->i_ino, nex);
564 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
566 return XFS_ERROR(EFSCORRUPTED);
569 ifp->if_real_bytes = 0;
571 ifp->if_u1.if_extents = NULL;
572 else if (nex <= XFS_INLINE_EXTS)
573 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
575 xfs_iext_add(ifp, 0, nex);
577 ifp->if_bytes = size;
579 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
580 xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip));
581 for (i = 0; i < nex; i++, dp++) {
582 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
583 ep->l0 = be64_to_cpu(get_unaligned(&dp->l0));
584 ep->l1 = be64_to_cpu(get_unaligned(&dp->l1));
586 XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork);
587 if (whichfork != XFS_DATA_FORK ||
588 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
589 if (unlikely(xfs_check_nostate_extents(
591 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
594 return XFS_ERROR(EFSCORRUPTED);
597 ifp->if_flags |= XFS_IFEXTENTS;
602 * The file has too many extents to fit into
603 * the inode, so they are in B-tree format.
604 * Allocate a buffer for the root of the B-tree
605 * and copy the root into it. The i_extents
606 * field will remain NULL until all of the
607 * extents are read in (when they are needed).
615 xfs_bmdr_block_t *dfp;
621 ifp = XFS_IFORK_PTR(ip, whichfork);
622 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
623 size = XFS_BMAP_BROOT_SPACE(dfp);
624 nrecs = XFS_BMAP_BROOT_NUMRECS(dfp);
627 * blow out if -- fork has less extents than can fit in
628 * fork (fork shouldn't be a btree format), root btree
629 * block has more records than can fit into the fork,
630 * or the number of extents is greater than the number of
633 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
634 || XFS_BMDR_SPACE_CALC(nrecs) >
635 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
636 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
637 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
638 "corrupt inode %Lu (btree).",
639 (unsigned long long) ip->i_ino);
640 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
642 return XFS_ERROR(EFSCORRUPTED);
645 ifp->if_broot_bytes = size;
646 ifp->if_broot = kmem_alloc(size, KM_SLEEP);
647 ASSERT(ifp->if_broot != NULL);
649 * Copy and convert from the on-disk structure
650 * to the in-memory structure.
652 xfs_bmdr_to_bmbt(dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
653 ifp->if_broot, size);
654 ifp->if_flags &= ~XFS_IFEXTENTS;
655 ifp->if_flags |= XFS_IFBROOT;
661 xfs_dinode_from_disk(
663 xfs_dinode_core_t *from)
665 to->di_magic = be16_to_cpu(from->di_magic);
666 to->di_mode = be16_to_cpu(from->di_mode);
667 to->di_version = from ->di_version;
668 to->di_format = from->di_format;
669 to->di_onlink = be16_to_cpu(from->di_onlink);
670 to->di_uid = be32_to_cpu(from->di_uid);
671 to->di_gid = be32_to_cpu(from->di_gid);
672 to->di_nlink = be32_to_cpu(from->di_nlink);
673 to->di_projid = be16_to_cpu(from->di_projid);
674 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
675 to->di_flushiter = be16_to_cpu(from->di_flushiter);
676 to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec);
677 to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec);
678 to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec);
679 to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec);
680 to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec);
681 to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec);
682 to->di_size = be64_to_cpu(from->di_size);
683 to->di_nblocks = be64_to_cpu(from->di_nblocks);
684 to->di_extsize = be32_to_cpu(from->di_extsize);
685 to->di_nextents = be32_to_cpu(from->di_nextents);
686 to->di_anextents = be16_to_cpu(from->di_anextents);
687 to->di_forkoff = from->di_forkoff;
688 to->di_aformat = from->di_aformat;
689 to->di_dmevmask = be32_to_cpu(from->di_dmevmask);
690 to->di_dmstate = be16_to_cpu(from->di_dmstate);
691 to->di_flags = be16_to_cpu(from->di_flags);
692 to->di_gen = be32_to_cpu(from->di_gen);
697 xfs_dinode_core_t *to,
698 xfs_icdinode_t *from)
700 to->di_magic = cpu_to_be16(from->di_magic);
701 to->di_mode = cpu_to_be16(from->di_mode);
702 to->di_version = from ->di_version;
703 to->di_format = from->di_format;
704 to->di_onlink = cpu_to_be16(from->di_onlink);
705 to->di_uid = cpu_to_be32(from->di_uid);
706 to->di_gid = cpu_to_be32(from->di_gid);
707 to->di_nlink = cpu_to_be32(from->di_nlink);
708 to->di_projid = cpu_to_be16(from->di_projid);
709 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
710 to->di_flushiter = cpu_to_be16(from->di_flushiter);
711 to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec);
712 to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec);
713 to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec);
714 to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec);
715 to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec);
716 to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec);
717 to->di_size = cpu_to_be64(from->di_size);
718 to->di_nblocks = cpu_to_be64(from->di_nblocks);
719 to->di_extsize = cpu_to_be32(from->di_extsize);
720 to->di_nextents = cpu_to_be32(from->di_nextents);
721 to->di_anextents = cpu_to_be16(from->di_anextents);
722 to->di_forkoff = from->di_forkoff;
723 to->di_aformat = from->di_aformat;
724 to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
725 to->di_dmstate = cpu_to_be16(from->di_dmstate);
726 to->di_flags = cpu_to_be16(from->di_flags);
727 to->di_gen = cpu_to_be32(from->di_gen);
736 if (di_flags & XFS_DIFLAG_ANY) {
737 if (di_flags & XFS_DIFLAG_REALTIME)
738 flags |= XFS_XFLAG_REALTIME;
739 if (di_flags & XFS_DIFLAG_PREALLOC)
740 flags |= XFS_XFLAG_PREALLOC;
741 if (di_flags & XFS_DIFLAG_IMMUTABLE)
742 flags |= XFS_XFLAG_IMMUTABLE;
743 if (di_flags & XFS_DIFLAG_APPEND)
744 flags |= XFS_XFLAG_APPEND;
745 if (di_flags & XFS_DIFLAG_SYNC)
746 flags |= XFS_XFLAG_SYNC;
747 if (di_flags & XFS_DIFLAG_NOATIME)
748 flags |= XFS_XFLAG_NOATIME;
749 if (di_flags & XFS_DIFLAG_NODUMP)
750 flags |= XFS_XFLAG_NODUMP;
751 if (di_flags & XFS_DIFLAG_RTINHERIT)
752 flags |= XFS_XFLAG_RTINHERIT;
753 if (di_flags & XFS_DIFLAG_PROJINHERIT)
754 flags |= XFS_XFLAG_PROJINHERIT;
755 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
756 flags |= XFS_XFLAG_NOSYMLINKS;
757 if (di_flags & XFS_DIFLAG_EXTSIZE)
758 flags |= XFS_XFLAG_EXTSIZE;
759 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
760 flags |= XFS_XFLAG_EXTSZINHERIT;
761 if (di_flags & XFS_DIFLAG_NODEFRAG)
762 flags |= XFS_XFLAG_NODEFRAG;
763 if (di_flags & XFS_DIFLAG_FILESTREAM)
764 flags |= XFS_XFLAG_FILESTREAM;
774 xfs_icdinode_t *dic = &ip->i_d;
776 return _xfs_dic2xflags(dic->di_flags) |
777 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
784 xfs_dinode_core_t *dic = &dip->di_core;
786 return _xfs_dic2xflags(be16_to_cpu(dic->di_flags)) |
787 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
791 * Given a mount structure and an inode number, return a pointer
792 * to a newly allocated in-core inode corresponding to the given
795 * Initialize the inode's attributes and extent pointers if it
796 * already has them (it will not if the inode has no links).
812 ASSERT(xfs_inode_zone != NULL);
814 ip = kmem_zone_zalloc(xfs_inode_zone, KM_SLEEP);
817 atomic_set(&ip->i_iocount, 0);
818 spin_lock_init(&ip->i_flags_lock);
821 * Get pointer's to the on-disk inode and the buffer containing it.
822 * If the inode number refers to a block outside the file system
823 * then xfs_itobp() will return NULL. In this case we should
824 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
825 * know that this is a new incore inode.
827 error = xfs_itobp(mp, tp, ip, &dip, &bp, bno, imap_flags, XFS_BUF_LOCK);
829 kmem_zone_free(xfs_inode_zone, ip);
834 * Initialize inode's trace buffers.
835 * Do this before xfs_iformat in case it adds entries.
837 #ifdef XFS_INODE_TRACE
838 ip->i_trace = ktrace_alloc(INODE_TRACE_SIZE, KM_SLEEP);
840 #ifdef XFS_BMAP_TRACE
841 ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_SLEEP);
843 #ifdef XFS_BMBT_TRACE
844 ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_SLEEP);
847 ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_SLEEP);
849 #ifdef XFS_ILOCK_TRACE
850 ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_SLEEP);
852 #ifdef XFS_DIR2_TRACE
853 ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_SLEEP);
857 * If we got something that isn't an inode it means someone
858 * (nfs or dmi) has a stale handle.
860 if (be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC) {
861 kmem_zone_free(xfs_inode_zone, ip);
862 xfs_trans_brelse(tp, bp);
864 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
865 "dip->di_core.di_magic (0x%x) != "
866 "XFS_DINODE_MAGIC (0x%x)",
867 be16_to_cpu(dip->di_core.di_magic),
870 return XFS_ERROR(EINVAL);
874 * If the on-disk inode is already linked to a directory
875 * entry, copy all of the inode into the in-core inode.
876 * xfs_iformat() handles copying in the inode format
877 * specific information.
878 * Otherwise, just get the truly permanent information.
880 if (dip->di_core.di_mode) {
881 xfs_dinode_from_disk(&ip->i_d, &dip->di_core);
882 error = xfs_iformat(ip, dip);
884 kmem_zone_free(xfs_inode_zone, ip);
885 xfs_trans_brelse(tp, bp);
887 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
888 "xfs_iformat() returned error %d",
894 ip->i_d.di_magic = be16_to_cpu(dip->di_core.di_magic);
895 ip->i_d.di_version = dip->di_core.di_version;
896 ip->i_d.di_gen = be32_to_cpu(dip->di_core.di_gen);
897 ip->i_d.di_flushiter = be16_to_cpu(dip->di_core.di_flushiter);
899 * Make sure to pull in the mode here as well in
900 * case the inode is released without being used.
901 * This ensures that xfs_inactive() will see that
902 * the inode is already free and not try to mess
903 * with the uninitialized part of it.
907 * Initialize the per-fork minima and maxima for a new
908 * inode here. xfs_iformat will do it for old inodes.
910 ip->i_df.if_ext_max =
911 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
914 INIT_LIST_HEAD(&ip->i_reclaim);
917 * The inode format changed when we moved the link count and
918 * made it 32 bits long. If this is an old format inode,
919 * convert it in memory to look like a new one. If it gets
920 * flushed to disk we will convert back before flushing or
921 * logging it. We zero out the new projid field and the old link
922 * count field. We'll handle clearing the pad field (the remains
923 * of the old uuid field) when we actually convert the inode to
924 * the new format. We don't change the version number so that we
925 * can distinguish this from a real new format inode.
927 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
928 ip->i_d.di_nlink = ip->i_d.di_onlink;
929 ip->i_d.di_onlink = 0;
930 ip->i_d.di_projid = 0;
933 ip->i_delayed_blks = 0;
934 ip->i_size = ip->i_d.di_size;
937 * Mark the buffer containing the inode as something to keep
938 * around for a while. This helps to keep recently accessed
939 * meta-data in-core longer.
941 XFS_BUF_SET_REF(bp, XFS_INO_REF);
944 * Use xfs_trans_brelse() to release the buffer containing the
945 * on-disk inode, because it was acquired with xfs_trans_read_buf()
946 * in xfs_itobp() above. If tp is NULL, this is just a normal
947 * brelse(). If we're within a transaction, then xfs_trans_brelse()
948 * will only release the buffer if it is not dirty within the
949 * transaction. It will be OK to release the buffer in this case,
950 * because inodes on disk are never destroyed and we will be
951 * locking the new in-core inode before putting it in the hash
952 * table where other processes can find it. Thus we don't have
953 * to worry about the inode being changed just because we released
956 xfs_trans_brelse(tp, bp);
962 * Read in extents from a btree-format inode.
963 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
973 xfs_extnum_t nextents;
976 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
977 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
979 return XFS_ERROR(EFSCORRUPTED);
981 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
982 size = nextents * sizeof(xfs_bmbt_rec_t);
983 ifp = XFS_IFORK_PTR(ip, whichfork);
986 * We know that the size is valid (it's checked in iformat_btree)
988 ifp->if_lastex = NULLEXTNUM;
989 ifp->if_bytes = ifp->if_real_bytes = 0;
990 ifp->if_flags |= XFS_IFEXTENTS;
991 xfs_iext_add(ifp, 0, nextents);
992 error = xfs_bmap_read_extents(tp, ip, whichfork);
994 xfs_iext_destroy(ifp);
995 ifp->if_flags &= ~XFS_IFEXTENTS;
998 xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip));
1003 * Allocate an inode on disk and return a copy of its in-core version.
1004 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1005 * appropriately within the inode. The uid and gid for the inode are
1006 * set according to the contents of the given cred structure.
1008 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1009 * has a free inode available, call xfs_iget()
1010 * to obtain the in-core version of the allocated inode. Finally,
1011 * fill in the inode and log its initial contents. In this case,
1012 * ialloc_context would be set to NULL and call_again set to false.
1014 * If xfs_dialloc() does not have an available inode,
1015 * it will replenish its supply by doing an allocation. Since we can
1016 * only do one allocation within a transaction without deadlocks, we
1017 * must commit the current transaction before returning the inode itself.
1018 * In this case, therefore, we will set call_again to true and return.
1019 * The caller should then commit the current transaction, start a new
1020 * transaction, and call xfs_ialloc() again to actually get the inode.
1022 * To ensure that some other process does not grab the inode that
1023 * was allocated during the first call to xfs_ialloc(), this routine
1024 * also returns the [locked] bp pointing to the head of the freelist
1025 * as ialloc_context. The caller should hold this buffer across
1026 * the commit and pass it back into this routine on the second call.
1028 * If we are allocating quota inodes, we do not have a parent inode
1029 * to attach to or associate with (i.e. pip == NULL) because they
1030 * are not linked into the directory structure - they are attached
1031 * directly to the superblock - and so have no parent.
1043 xfs_buf_t **ialloc_context,
1044 boolean_t *call_again,
1054 * Call the space management code to pick
1055 * the on-disk inode to be allocated.
1057 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
1058 ialloc_context, call_again, &ino);
1062 if (*call_again || ino == NULLFSINO) {
1066 ASSERT(*ialloc_context == NULL);
1069 * Get the in-core inode with the lock held exclusively.
1070 * This is because we're setting fields here we need
1071 * to prevent others from looking at until we're done.
1073 error = xfs_trans_iget(tp->t_mountp, tp, ino,
1074 XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
1081 ip->i_d.di_mode = (__uint16_t)mode;
1082 ip->i_d.di_onlink = 0;
1083 ip->i_d.di_nlink = nlink;
1084 ASSERT(ip->i_d.di_nlink == nlink);
1085 ip->i_d.di_uid = current_fsuid(cr);
1086 ip->i_d.di_gid = current_fsgid(cr);
1087 ip->i_d.di_projid = prid;
1088 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1091 * If the superblock version is up to where we support new format
1092 * inodes and this is currently an old format inode, then change
1093 * the inode version number now. This way we only do the conversion
1094 * here rather than here and in the flush/logging code.
1096 if (xfs_sb_version_hasnlink(&tp->t_mountp->m_sb) &&
1097 ip->i_d.di_version == XFS_DINODE_VERSION_1) {
1098 ip->i_d.di_version = XFS_DINODE_VERSION_2;
1100 * We've already zeroed the old link count, the projid field,
1101 * and the pad field.
1106 * Project ids won't be stored on disk if we are using a version 1 inode.
1108 if ((prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1))
1109 xfs_bump_ino_vers2(tp, ip);
1111 if (pip && XFS_INHERIT_GID(pip)) {
1112 ip->i_d.di_gid = pip->i_d.di_gid;
1113 if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1114 ip->i_d.di_mode |= S_ISGID;
1119 * If the group ID of the new file does not match the effective group
1120 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1121 * (and only if the irix_sgid_inherit compatibility variable is set).
1123 if ((irix_sgid_inherit) &&
1124 (ip->i_d.di_mode & S_ISGID) &&
1125 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1126 ip->i_d.di_mode &= ~S_ISGID;
1129 ip->i_d.di_size = 0;
1131 ip->i_d.di_nextents = 0;
1132 ASSERT(ip->i_d.di_nblocks == 0);
1133 xfs_ichgtime(ip, XFS_ICHGTIME_CHG|XFS_ICHGTIME_ACC|XFS_ICHGTIME_MOD);
1135 * di_gen will have been taken care of in xfs_iread.
1137 ip->i_d.di_extsize = 0;
1138 ip->i_d.di_dmevmask = 0;
1139 ip->i_d.di_dmstate = 0;
1140 ip->i_d.di_flags = 0;
1141 flags = XFS_ILOG_CORE;
1142 switch (mode & S_IFMT) {
1147 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1148 ip->i_df.if_u2.if_rdev = rdev;
1149 ip->i_df.if_flags = 0;
1150 flags |= XFS_ILOG_DEV;
1153 if (pip && xfs_inode_is_filestream(pip)) {
1154 error = xfs_filestream_associate(pip, ip);
1158 xfs_iflags_set(ip, XFS_IFILESTREAM);
1162 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
1165 if ((mode & S_IFMT) == S_IFDIR) {
1166 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1167 di_flags |= XFS_DIFLAG_RTINHERIT;
1168 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1169 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1170 ip->i_d.di_extsize = pip->i_d.di_extsize;
1172 } else if ((mode & S_IFMT) == S_IFREG) {
1173 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1174 di_flags |= XFS_DIFLAG_REALTIME;
1175 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1176 di_flags |= XFS_DIFLAG_EXTSIZE;
1177 ip->i_d.di_extsize = pip->i_d.di_extsize;
1180 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1181 xfs_inherit_noatime)
1182 di_flags |= XFS_DIFLAG_NOATIME;
1183 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1185 di_flags |= XFS_DIFLAG_NODUMP;
1186 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1188 di_flags |= XFS_DIFLAG_SYNC;
1189 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1190 xfs_inherit_nosymlinks)
1191 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1192 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1193 di_flags |= XFS_DIFLAG_PROJINHERIT;
1194 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1195 xfs_inherit_nodefrag)
1196 di_flags |= XFS_DIFLAG_NODEFRAG;
1197 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
1198 di_flags |= XFS_DIFLAG_FILESTREAM;
1199 ip->i_d.di_flags |= di_flags;
1203 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1204 ip->i_df.if_flags = XFS_IFEXTENTS;
1205 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1206 ip->i_df.if_u1.if_extents = NULL;
1212 * Attribute fork settings for new inode.
1214 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1215 ip->i_d.di_anextents = 0;
1218 * Log the new values stuffed into the inode.
1220 xfs_trans_log_inode(tp, ip, flags);
1222 /* now that we have an i_mode we can setup inode ops and unlock */
1223 xfs_initialize_vnode(tp->t_mountp, vp, ip);
1230 * Check to make sure that there are no blocks allocated to the
1231 * file beyond the size of the file. We don't check this for
1232 * files with fixed size extents or real time extents, but we
1233 * at least do it for regular files.
1242 xfs_fileoff_t map_first;
1244 xfs_bmbt_irec_t imaps[2];
1246 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1249 if (XFS_IS_REALTIME_INODE(ip))
1252 if (ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE)
1256 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1258 * The filesystem could be shutting down, so bmapi may return
1261 if (xfs_bmapi(NULL, ip, map_first,
1263 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1265 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1268 ASSERT(nimaps == 1);
1269 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1274 * Calculate the last possible buffered byte in a file. This must
1275 * include data that was buffered beyond the EOF by the write code.
1276 * This also needs to deal with overflowing the xfs_fsize_t type
1277 * which can happen for sizes near the limit.
1279 * We also need to take into account any blocks beyond the EOF. It
1280 * may be the case that they were buffered by a write which failed.
1281 * In that case the pages will still be in memory, but the inode size
1282 * will never have been updated.
1289 xfs_fsize_t last_byte;
1290 xfs_fileoff_t last_block;
1291 xfs_fileoff_t size_last_block;
1294 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE | MR_ACCESS));
1298 * Only check for blocks beyond the EOF if the extents have
1299 * been read in. This eliminates the need for the inode lock,
1300 * and it also saves us from looking when it really isn't
1303 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1304 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1312 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_size);
1313 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1315 last_byte = XFS_FSB_TO_B(mp, last_block);
1316 if (last_byte < 0) {
1317 return XFS_MAXIOFFSET(mp);
1319 last_byte += (1 << mp->m_writeio_log);
1320 if (last_byte < 0) {
1321 return XFS_MAXIOFFSET(mp);
1326 #if defined(XFS_RW_TRACE)
1332 xfs_fsize_t new_size,
1333 xfs_off_t toss_start,
1334 xfs_off_t toss_finish)
1336 if (ip->i_rwtrace == NULL) {
1340 ktrace_enter(ip->i_rwtrace,
1343 (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
1344 (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
1345 (void*)((long)flag),
1346 (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
1347 (void*)(unsigned long)(new_size & 0xffffffff),
1348 (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
1349 (void*)(unsigned long)(toss_start & 0xffffffff),
1350 (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
1351 (void*)(unsigned long)(toss_finish & 0xffffffff),
1352 (void*)(unsigned long)current_cpu(),
1353 (void*)(unsigned long)current_pid(),
1359 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1363 * Start the truncation of the file to new_size. The new size
1364 * must be smaller than the current size. This routine will
1365 * clear the buffer and page caches of file data in the removed
1366 * range, and xfs_itruncate_finish() will remove the underlying
1369 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1370 * must NOT have the inode lock held at all. This is because we're
1371 * calling into the buffer/page cache code and we can't hold the
1372 * inode lock when we do so.
1374 * We need to wait for any direct I/Os in flight to complete before we
1375 * proceed with the truncate. This is needed to prevent the extents
1376 * being read or written by the direct I/Os from being removed while the
1377 * I/O is in flight as there is no other method of synchronising
1378 * direct I/O with the truncate operation. Also, because we hold
1379 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1380 * started until the truncate completes and drops the lock. Essentially,
1381 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1382 * between direct I/Os and the truncate operation.
1384 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1385 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1386 * in the case that the caller is locking things out of order and
1387 * may not be able to call xfs_itruncate_finish() with the inode lock
1388 * held without dropping the I/O lock. If the caller must drop the
1389 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1390 * must be called again with all the same restrictions as the initial
1394 xfs_itruncate_start(
1397 xfs_fsize_t new_size)
1399 xfs_fsize_t last_byte;
1400 xfs_off_t toss_start;
1405 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1406 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1407 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1408 (flags == XFS_ITRUNC_MAYBE));
1413 /* wait for the completion of any pending DIOs */
1414 if (new_size < ip->i_size)
1418 * Call toss_pages or flushinval_pages to get rid of pages
1419 * overlapping the region being removed. We have to use
1420 * the less efficient flushinval_pages in the case that the
1421 * caller may not be able to finish the truncate without
1422 * dropping the inode's I/O lock. Make sure
1423 * to catch any pages brought in by buffers overlapping
1424 * the EOF by searching out beyond the isize by our
1425 * block size. We round new_size up to a block boundary
1426 * so that we don't toss things on the same block as
1427 * new_size but before it.
1429 * Before calling toss_page or flushinval_pages, make sure to
1430 * call remapf() over the same region if the file is mapped.
1431 * This frees up mapped file references to the pages in the
1432 * given range and for the flushinval_pages case it ensures
1433 * that we get the latest mapped changes flushed out.
1435 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1436 toss_start = XFS_FSB_TO_B(mp, toss_start);
1437 if (toss_start < 0) {
1439 * The place to start tossing is beyond our maximum
1440 * file size, so there is no way that the data extended
1445 last_byte = xfs_file_last_byte(ip);
1446 xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
1448 if (last_byte > toss_start) {
1449 if (flags & XFS_ITRUNC_DEFINITE) {
1450 xfs_tosspages(ip, toss_start,
1451 -1, FI_REMAPF_LOCKED);
1453 error = xfs_flushinval_pages(ip, toss_start,
1454 -1, FI_REMAPF_LOCKED);
1459 if (new_size == 0) {
1460 ASSERT(VN_CACHED(vp) == 0);
1467 * Shrink the file to the given new_size. The new
1468 * size must be smaller than the current size.
1469 * This will free up the underlying blocks
1470 * in the removed range after a call to xfs_itruncate_start()
1471 * or xfs_atruncate_start().
1473 * The transaction passed to this routine must have made
1474 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1475 * This routine may commit the given transaction and
1476 * start new ones, so make sure everything involved in
1477 * the transaction is tidy before calling here.
1478 * Some transaction will be returned to the caller to be
1479 * committed. The incoming transaction must already include
1480 * the inode, and both inode locks must be held exclusively.
1481 * The inode must also be "held" within the transaction. On
1482 * return the inode will be "held" within the returned transaction.
1483 * This routine does NOT require any disk space to be reserved
1484 * for it within the transaction.
1486 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1487 * and it indicates the fork which is to be truncated. For the
1488 * attribute fork we only support truncation to size 0.
1490 * We use the sync parameter to indicate whether or not the first
1491 * transaction we perform might have to be synchronous. For the attr fork,
1492 * it needs to be so if the unlink of the inode is not yet known to be
1493 * permanent in the log. This keeps us from freeing and reusing the
1494 * blocks of the attribute fork before the unlink of the inode becomes
1497 * For the data fork, we normally have to run synchronously if we're
1498 * being called out of the inactive path or we're being called
1499 * out of the create path where we're truncating an existing file.
1500 * Either way, the truncate needs to be sync so blocks don't reappear
1501 * in the file with altered data in case of a crash. wsync filesystems
1502 * can run the first case async because anything that shrinks the inode
1503 * has to run sync so by the time we're called here from inactive, the
1504 * inode size is permanently set to 0.
1506 * Calls from the truncate path always need to be sync unless we're
1507 * in a wsync filesystem and the file has already been unlinked.
1509 * The caller is responsible for correctly setting the sync parameter.
1510 * It gets too hard for us to guess here which path we're being called
1511 * out of just based on inode state.
1514 xfs_itruncate_finish(
1517 xfs_fsize_t new_size,
1521 xfs_fsblock_t first_block;
1522 xfs_fileoff_t first_unmap_block;
1523 xfs_fileoff_t last_block;
1524 xfs_filblks_t unmap_len=0;
1529 xfs_bmap_free_t free_list;
1532 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1533 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
1534 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1535 ASSERT(*tp != NULL);
1536 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1537 ASSERT(ip->i_transp == *tp);
1538 ASSERT(ip->i_itemp != NULL);
1539 ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
1543 mp = (ntp)->t_mountp;
1544 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1547 * We only support truncating the entire attribute fork.
1549 if (fork == XFS_ATTR_FORK) {
1552 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1553 xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
1555 * The first thing we do is set the size to new_size permanently
1556 * on disk. This way we don't have to worry about anyone ever
1557 * being able to look at the data being freed even in the face
1558 * of a crash. What we're getting around here is the case where
1559 * we free a block, it is allocated to another file, it is written
1560 * to, and then we crash. If the new data gets written to the
1561 * file but the log buffers containing the free and reallocation
1562 * don't, then we'd end up with garbage in the blocks being freed.
1563 * As long as we make the new_size permanent before actually
1564 * freeing any blocks it doesn't matter if they get writtten to.
1566 * The callers must signal into us whether or not the size
1567 * setting here must be synchronous. There are a few cases
1568 * where it doesn't have to be synchronous. Those cases
1569 * occur if the file is unlinked and we know the unlink is
1570 * permanent or if the blocks being truncated are guaranteed
1571 * to be beyond the inode eof (regardless of the link count)
1572 * and the eof value is permanent. Both of these cases occur
1573 * only on wsync-mounted filesystems. In those cases, we're
1574 * guaranteed that no user will ever see the data in the blocks
1575 * that are being truncated so the truncate can run async.
1576 * In the free beyond eof case, the file may wind up with
1577 * more blocks allocated to it than it needs if we crash
1578 * and that won't get fixed until the next time the file
1579 * is re-opened and closed but that's ok as that shouldn't
1580 * be too many blocks.
1582 * However, we can't just make all wsync xactions run async
1583 * because there's one call out of the create path that needs
1584 * to run sync where it's truncating an existing file to size
1585 * 0 whose size is > 0.
1587 * It's probably possible to come up with a test in this
1588 * routine that would correctly distinguish all the above
1589 * cases from the values of the function parameters and the
1590 * inode state but for sanity's sake, I've decided to let the
1591 * layers above just tell us. It's simpler to correctly figure
1592 * out in the layer above exactly under what conditions we
1593 * can run async and I think it's easier for others read and
1594 * follow the logic in case something has to be changed.
1595 * cscope is your friend -- rcc.
1597 * The attribute fork is much simpler.
1599 * For the attribute fork we allow the caller to tell us whether
1600 * the unlink of the inode that led to this call is yet permanent
1601 * in the on disk log. If it is not and we will be freeing extents
1602 * in this inode then we make the first transaction synchronous
1603 * to make sure that the unlink is permanent by the time we free
1606 if (fork == XFS_DATA_FORK) {
1607 if (ip->i_d.di_nextents > 0) {
1609 * If we are not changing the file size then do
1610 * not update the on-disk file size - we may be
1611 * called from xfs_inactive_free_eofblocks(). If we
1612 * update the on-disk file size and then the system
1613 * crashes before the contents of the file are
1614 * flushed to disk then the files may be full of
1615 * holes (ie NULL files bug).
1617 if (ip->i_size != new_size) {
1618 ip->i_d.di_size = new_size;
1619 ip->i_size = new_size;
1620 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1624 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1625 if (ip->i_d.di_anextents > 0)
1626 xfs_trans_set_sync(ntp);
1628 ASSERT(fork == XFS_DATA_FORK ||
1629 (fork == XFS_ATTR_FORK &&
1630 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1631 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1634 * Since it is possible for space to become allocated beyond
1635 * the end of the file (in a crash where the space is allocated
1636 * but the inode size is not yet updated), simply remove any
1637 * blocks which show up between the new EOF and the maximum
1638 * possible file size. If the first block to be removed is
1639 * beyond the maximum file size (ie it is the same as last_block),
1640 * then there is nothing to do.
1642 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1643 ASSERT(first_unmap_block <= last_block);
1645 if (last_block == first_unmap_block) {
1648 unmap_len = last_block - first_unmap_block + 1;
1652 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1653 * will tell us whether it freed the entire range or
1654 * not. If this is a synchronous mount (wsync),
1655 * then we can tell bunmapi to keep all the
1656 * transactions asynchronous since the unlink
1657 * transaction that made this inode inactive has
1658 * already hit the disk. There's no danger of
1659 * the freed blocks being reused, there being a
1660 * crash, and the reused blocks suddenly reappearing
1661 * in this file with garbage in them once recovery
1664 XFS_BMAP_INIT(&free_list, &first_block);
1665 error = xfs_bunmapi(ntp, ip,
1666 first_unmap_block, unmap_len,
1667 XFS_BMAPI_AFLAG(fork) |
1668 (sync ? 0 : XFS_BMAPI_ASYNC),
1669 XFS_ITRUNC_MAX_EXTENTS,
1670 &first_block, &free_list,
1674 * If the bunmapi call encounters an error,
1675 * return to the caller where the transaction
1676 * can be properly aborted. We just need to
1677 * make sure we're not holding any resources
1678 * that we were not when we came in.
1680 xfs_bmap_cancel(&free_list);
1685 * Duplicate the transaction that has the permanent
1686 * reservation and commit the old transaction.
1688 error = xfs_bmap_finish(tp, &free_list, &committed);
1692 * If the bmap finish call encounters an error,
1693 * return to the caller where the transaction
1694 * can be properly aborted. We just need to
1695 * make sure we're not holding any resources
1696 * that we were not when we came in.
1698 * Aborting from this point might lose some
1699 * blocks in the file system, but oh well.
1701 xfs_bmap_cancel(&free_list);
1704 * If the passed in transaction committed
1705 * in xfs_bmap_finish(), then we want to
1706 * add the inode to this one before returning.
1707 * This keeps things simple for the higher
1708 * level code, because it always knows that
1709 * the inode is locked and held in the
1710 * transaction that returns to it whether
1711 * errors occur or not. We don't mark the
1712 * inode dirty so that this transaction can
1713 * be easily aborted if possible.
1715 xfs_trans_ijoin(ntp, ip,
1716 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1717 xfs_trans_ihold(ntp, ip);
1724 * The first xact was committed,
1725 * so add the inode to the new one.
1726 * Mark it dirty so it will be logged
1727 * and moved forward in the log as
1728 * part of every commit.
1730 xfs_trans_ijoin(ntp, ip,
1731 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1732 xfs_trans_ihold(ntp, ip);
1733 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1735 ntp = xfs_trans_dup(ntp);
1736 (void) xfs_trans_commit(*tp, 0);
1738 error = xfs_trans_reserve(ntp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1739 XFS_TRANS_PERM_LOG_RES,
1740 XFS_ITRUNCATE_LOG_COUNT);
1742 * Add the inode being truncated to the next chained
1745 xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1746 xfs_trans_ihold(ntp, ip);
1751 * Only update the size in the case of the data fork, but
1752 * always re-log the inode so that our permanent transaction
1753 * can keep on rolling it forward in the log.
1755 if (fork == XFS_DATA_FORK) {
1756 xfs_isize_check(mp, ip, new_size);
1758 * If we are not changing the file size then do
1759 * not update the on-disk file size - we may be
1760 * called from xfs_inactive_free_eofblocks(). If we
1761 * update the on-disk file size and then the system
1762 * crashes before the contents of the file are
1763 * flushed to disk then the files may be full of
1764 * holes (ie NULL files bug).
1766 if (ip->i_size != new_size) {
1767 ip->i_d.di_size = new_size;
1768 ip->i_size = new_size;
1771 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1772 ASSERT((new_size != 0) ||
1773 (fork == XFS_ATTR_FORK) ||
1774 (ip->i_delayed_blks == 0));
1775 ASSERT((new_size != 0) ||
1776 (fork == XFS_ATTR_FORK) ||
1777 (ip->i_d.di_nextents == 0));
1778 xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
1786 * Do the first part of growing a file: zero any data in the last
1787 * block that is beyond the old EOF. We need to do this before
1788 * the inode is joined to the transaction to modify the i_size.
1789 * That way we can drop the inode lock and call into the buffer
1790 * cache to get the buffer mapping the EOF.
1795 xfs_fsize_t new_size,
1798 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1799 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1800 ASSERT(new_size > ip->i_size);
1803 * Zero any pages that may have been created by
1804 * xfs_write_file() beyond the end of the file
1805 * and any blocks between the old and new file sizes.
1807 return xfs_zero_eof(ip, new_size, ip->i_size);
1813 * This routine is called to extend the size of a file.
1814 * The inode must have both the iolock and the ilock locked
1815 * for update and it must be a part of the current transaction.
1816 * The xfs_igrow_start() function must have been called previously.
1817 * If the change_flag is not zero, the inode change timestamp will
1824 xfs_fsize_t new_size,
1827 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1828 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1829 ASSERT(ip->i_transp == tp);
1830 ASSERT(new_size > ip->i_size);
1833 * Update the file size. Update the inode change timestamp
1834 * if change_flag set.
1836 ip->i_d.di_size = new_size;
1837 ip->i_size = new_size;
1839 xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
1840 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1846 * This is called when the inode's link count goes to 0.
1847 * We place the on-disk inode on a list in the AGI. It
1848 * will be pulled from this list when the inode is freed.
1860 xfs_agnumber_t agno;
1861 xfs_daddr_t agdaddr;
1868 ASSERT(ip->i_d.di_nlink == 0);
1869 ASSERT(ip->i_d.di_mode != 0);
1870 ASSERT(ip->i_transp == tp);
1874 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1875 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1878 * Get the agi buffer first. It ensures lock ordering
1881 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1882 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1887 * Validate the magic number of the agi block.
1889 agi = XFS_BUF_TO_AGI(agibp);
1891 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1892 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
1893 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK,
1894 XFS_RANDOM_IUNLINK))) {
1895 XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi);
1896 xfs_trans_brelse(tp, agibp);
1897 return XFS_ERROR(EFSCORRUPTED);
1900 * Get the index into the agi hash table for the
1901 * list this inode will go on.
1903 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1905 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1906 ASSERT(agi->agi_unlinked[bucket_index]);
1907 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1909 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
1911 * There is already another inode in the bucket we need
1912 * to add ourselves to. Add us at the front of the list.
1913 * Here we put the head pointer into our next pointer,
1914 * and then we fall through to point the head at us.
1916 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
1920 ASSERT(be32_to_cpu(dip->di_next_unlinked) == NULLAGINO);
1921 /* both on-disk, don't endian flip twice */
1922 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1923 offset = ip->i_boffset +
1924 offsetof(xfs_dinode_t, di_next_unlinked);
1925 xfs_trans_inode_buf(tp, ibp);
1926 xfs_trans_log_buf(tp, ibp, offset,
1927 (offset + sizeof(xfs_agino_t) - 1));
1928 xfs_inobp_check(mp, ibp);
1932 * Point the bucket head pointer at the inode being inserted.
1935 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1936 offset = offsetof(xfs_agi_t, agi_unlinked) +
1937 (sizeof(xfs_agino_t) * bucket_index);
1938 xfs_trans_log_buf(tp, agibp, offset,
1939 (offset + sizeof(xfs_agino_t) - 1));
1944 * Pull the on-disk inode from the AGI unlinked list.
1957 xfs_agnumber_t agno;
1958 xfs_daddr_t agdaddr;
1960 xfs_agino_t next_agino;
1961 xfs_buf_t *last_ibp;
1962 xfs_dinode_t *last_dip = NULL;
1964 int offset, last_offset = 0;
1969 * First pull the on-disk inode from the AGI unlinked list.
1973 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1974 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1977 * Get the agi buffer first. It ensures lock ordering
1980 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1981 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1984 "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
1985 error, mp->m_fsname);
1989 * Validate the magic number of the agi block.
1991 agi = XFS_BUF_TO_AGI(agibp);
1993 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1994 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
1995 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE,
1996 XFS_RANDOM_IUNLINK_REMOVE))) {
1997 XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW,
1999 xfs_trans_brelse(tp, agibp);
2001 "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
2003 return XFS_ERROR(EFSCORRUPTED);
2006 * Get the index into the agi hash table for the
2007 * list this inode will go on.
2009 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2011 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2012 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
2013 ASSERT(agi->agi_unlinked[bucket_index]);
2015 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2017 * We're at the head of the list. Get the inode's
2018 * on-disk buffer to see if there is anyone after us
2019 * on the list. Only modify our next pointer if it
2020 * is not already NULLAGINO. This saves us the overhead
2021 * of dealing with the buffer when there is no need to
2024 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
2027 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2028 error, mp->m_fsname);
2031 next_agino = be32_to_cpu(dip->di_next_unlinked);
2032 ASSERT(next_agino != 0);
2033 if (next_agino != NULLAGINO) {
2034 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2035 offset = ip->i_boffset +
2036 offsetof(xfs_dinode_t, di_next_unlinked);
2037 xfs_trans_inode_buf(tp, ibp);
2038 xfs_trans_log_buf(tp, ibp, offset,
2039 (offset + sizeof(xfs_agino_t) - 1));
2040 xfs_inobp_check(mp, ibp);
2042 xfs_trans_brelse(tp, ibp);
2045 * Point the bucket head pointer at the next inode.
2047 ASSERT(next_agino != 0);
2048 ASSERT(next_agino != agino);
2049 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2050 offset = offsetof(xfs_agi_t, agi_unlinked) +
2051 (sizeof(xfs_agino_t) * bucket_index);
2052 xfs_trans_log_buf(tp, agibp, offset,
2053 (offset + sizeof(xfs_agino_t) - 1));
2056 * We need to search the list for the inode being freed.
2058 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2060 while (next_agino != agino) {
2062 * If the last inode wasn't the one pointing to
2063 * us, then release its buffer since we're not
2064 * going to do anything with it.
2066 if (last_ibp != NULL) {
2067 xfs_trans_brelse(tp, last_ibp);
2069 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2070 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
2071 &last_ibp, &last_offset);
2074 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
2075 error, mp->m_fsname);
2078 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2079 ASSERT(next_agino != NULLAGINO);
2080 ASSERT(next_agino != 0);
2083 * Now last_ibp points to the buffer previous to us on
2084 * the unlinked list. Pull us from the list.
2086 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
2089 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2090 error, mp->m_fsname);
2093 next_agino = be32_to_cpu(dip->di_next_unlinked);
2094 ASSERT(next_agino != 0);
2095 ASSERT(next_agino != agino);
2096 if (next_agino != NULLAGINO) {
2097 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2098 offset = ip->i_boffset +
2099 offsetof(xfs_dinode_t, di_next_unlinked);
2100 xfs_trans_inode_buf(tp, ibp);
2101 xfs_trans_log_buf(tp, ibp, offset,
2102 (offset + sizeof(xfs_agino_t) - 1));
2103 xfs_inobp_check(mp, ibp);
2105 xfs_trans_brelse(tp, ibp);
2108 * Point the previous inode on the list to the next inode.
2110 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2111 ASSERT(next_agino != 0);
2112 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2113 xfs_trans_inode_buf(tp, last_ibp);
2114 xfs_trans_log_buf(tp, last_ibp, offset,
2115 (offset + sizeof(xfs_agino_t) - 1));
2116 xfs_inobp_check(mp, last_ibp);
2123 xfs_inode_t *free_ip,
2127 xfs_mount_t *mp = free_ip->i_mount;
2128 int blks_per_cluster;
2131 int i, j, found, pre_flushed;
2134 xfs_inode_t *ip, **ip_found;
2135 xfs_inode_log_item_t *iip;
2136 xfs_log_item_t *lip;
2137 xfs_perag_t *pag = xfs_get_perag(mp, inum);
2139 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
2140 blks_per_cluster = 1;
2141 ninodes = mp->m_sb.sb_inopblock;
2142 nbufs = XFS_IALLOC_BLOCKS(mp);
2144 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
2145 mp->m_sb.sb_blocksize;
2146 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
2147 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
2150 ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
2152 for (j = 0; j < nbufs; j++, inum += ninodes) {
2153 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2154 XFS_INO_TO_AGBNO(mp, inum));
2158 * Look for each inode in memory and attempt to lock it,
2159 * we can be racing with flush and tail pushing here.
2160 * any inode we get the locks on, add to an array of
2161 * inode items to process later.
2163 * The get the buffer lock, we could beat a flush
2164 * or tail pushing thread to the lock here, in which
2165 * case they will go looking for the inode buffer
2166 * and fail, we need some other form of interlock
2170 for (i = 0; i < ninodes; i++) {
2171 read_lock(&pag->pag_ici_lock);
2172 ip = radix_tree_lookup(&pag->pag_ici_root,
2173 XFS_INO_TO_AGINO(mp, (inum + i)));
2175 /* Inode not in memory or we found it already,
2178 if (!ip || xfs_iflags_test(ip, XFS_ISTALE)) {
2179 read_unlock(&pag->pag_ici_lock);
2183 if (xfs_inode_clean(ip)) {
2184 read_unlock(&pag->pag_ici_lock);
2188 /* If we can get the locks then add it to the
2189 * list, otherwise by the time we get the bp lock
2190 * below it will already be attached to the
2194 /* This inode will already be locked - by us, lets
2198 if (ip == free_ip) {
2199 if (xfs_iflock_nowait(ip)) {
2200 xfs_iflags_set(ip, XFS_ISTALE);
2201 if (xfs_inode_clean(ip)) {
2204 ip_found[found++] = ip;
2207 read_unlock(&pag->pag_ici_lock);
2211 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2212 if (xfs_iflock_nowait(ip)) {
2213 xfs_iflags_set(ip, XFS_ISTALE);
2215 if (xfs_inode_clean(ip)) {
2217 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2219 ip_found[found++] = ip;
2222 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2225 read_unlock(&pag->pag_ici_lock);
2228 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2229 mp->m_bsize * blks_per_cluster,
2233 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
2235 if (lip->li_type == XFS_LI_INODE) {
2236 iip = (xfs_inode_log_item_t *)lip;
2237 ASSERT(iip->ili_logged == 1);
2238 lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
2239 spin_lock(&mp->m_ail_lock);
2240 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2241 spin_unlock(&mp->m_ail_lock);
2242 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2245 lip = lip->li_bio_list;
2248 for (i = 0; i < found; i++) {
2253 ip->i_update_core = 0;
2255 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2259 iip->ili_last_fields = iip->ili_format.ilf_fields;
2260 iip->ili_format.ilf_fields = 0;
2261 iip->ili_logged = 1;
2262 spin_lock(&mp->m_ail_lock);
2263 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2264 spin_unlock(&mp->m_ail_lock);
2266 xfs_buf_attach_iodone(bp,
2267 (void(*)(xfs_buf_t*,xfs_log_item_t*))
2268 xfs_istale_done, (xfs_log_item_t *)iip);
2269 if (ip != free_ip) {
2270 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2274 if (found || pre_flushed)
2275 xfs_trans_stale_inode_buf(tp, bp);
2276 xfs_trans_binval(tp, bp);
2279 kmem_free(ip_found, ninodes * sizeof(xfs_inode_t *));
2280 xfs_put_perag(mp, pag);
2284 * This is called to return an inode to the inode free list.
2285 * The inode should already be truncated to 0 length and have
2286 * no pages associated with it. This routine also assumes that
2287 * the inode is already a part of the transaction.
2289 * The on-disk copy of the inode will have been added to the list
2290 * of unlinked inodes in the AGI. We need to remove the inode from
2291 * that list atomically with respect to freeing it here.
2297 xfs_bmap_free_t *flist)
2301 xfs_ino_t first_ino;
2305 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2306 ASSERT(ip->i_transp == tp);
2307 ASSERT(ip->i_d.di_nlink == 0);
2308 ASSERT(ip->i_d.di_nextents == 0);
2309 ASSERT(ip->i_d.di_anextents == 0);
2310 ASSERT((ip->i_d.di_size == 0 && ip->i_size == 0) ||
2311 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2312 ASSERT(ip->i_d.di_nblocks == 0);
2315 * Pull the on-disk inode from the AGI unlinked list.
2317 error = xfs_iunlink_remove(tp, ip);
2322 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2326 ip->i_d.di_mode = 0; /* mark incore inode as free */
2327 ip->i_d.di_flags = 0;
2328 ip->i_d.di_dmevmask = 0;
2329 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2330 ip->i_df.if_ext_max =
2331 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2332 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2333 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2335 * Bump the generation count so no one will be confused
2336 * by reincarnations of this inode.
2340 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2342 error = xfs_itobp(ip->i_mount, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
2347 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2348 * from picking up this inode when it is reclaimed (its incore state
2349 * initialzed but not flushed to disk yet). The in-core di_mode is
2350 * already cleared and a corresponding transaction logged.
2351 * The hack here just synchronizes the in-core to on-disk
2352 * di_mode value in advance before the actual inode sync to disk.
2353 * This is OK because the inode is already unlinked and would never
2354 * change its di_mode again for this inode generation.
2355 * This is a temporary hack that would require a proper fix
2358 dip->di_core.di_mode = 0;
2361 xfs_ifree_cluster(ip, tp, first_ino);
2368 * Reallocate the space for if_broot based on the number of records
2369 * being added or deleted as indicated in rec_diff. Move the records
2370 * and pointers in if_broot to fit the new size. When shrinking this
2371 * will eliminate holes between the records and pointers created by
2372 * the caller. When growing this will create holes to be filled in
2375 * The caller must not request to add more records than would fit in
2376 * the on-disk inode root. If the if_broot is currently NULL, then
2377 * if we adding records one will be allocated. The caller must also
2378 * not request that the number of records go below zero, although
2379 * it can go to zero.
2381 * ip -- the inode whose if_broot area is changing
2382 * ext_diff -- the change in the number of records, positive or negative,
2383 * requested for the if_broot array.
2393 xfs_bmbt_block_t *new_broot;
2400 * Handle the degenerate case quietly.
2402 if (rec_diff == 0) {
2406 ifp = XFS_IFORK_PTR(ip, whichfork);
2409 * If there wasn't any memory allocated before, just
2410 * allocate it now and get out.
2412 if (ifp->if_broot_bytes == 0) {
2413 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
2414 ifp->if_broot = (xfs_bmbt_block_t*)kmem_alloc(new_size,
2416 ifp->if_broot_bytes = (int)new_size;
2421 * If there is already an existing if_broot, then we need
2422 * to realloc() it and shift the pointers to their new
2423 * location. The records don't change location because
2424 * they are kept butted up against the btree block header.
2426 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2427 new_max = cur_max + rec_diff;
2428 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2429 ifp->if_broot = (xfs_bmbt_block_t *)
2430 kmem_realloc(ifp->if_broot,
2432 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2434 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2435 ifp->if_broot_bytes);
2436 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2438 ifp->if_broot_bytes = (int)new_size;
2439 ASSERT(ifp->if_broot_bytes <=
2440 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2441 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2446 * rec_diff is less than 0. In this case, we are shrinking the
2447 * if_broot buffer. It must already exist. If we go to zero
2448 * records, just get rid of the root and clear the status bit.
2450 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
2451 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2452 new_max = cur_max + rec_diff;
2453 ASSERT(new_max >= 0);
2455 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2459 new_broot = (xfs_bmbt_block_t *)kmem_alloc(new_size, KM_SLEEP);
2461 * First copy over the btree block header.
2463 memcpy(new_broot, ifp->if_broot, sizeof(xfs_bmbt_block_t));
2466 ifp->if_flags &= ~XFS_IFBROOT;
2470 * Only copy the records and pointers if there are any.
2474 * First copy the records.
2476 op = (char *)XFS_BMAP_BROOT_REC_ADDR(ifp->if_broot, 1,
2477 ifp->if_broot_bytes);
2478 np = (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot, 1,
2480 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2483 * Then copy the pointers.
2485 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2486 ifp->if_broot_bytes);
2487 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot, 1,
2489 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2491 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2492 ifp->if_broot = new_broot;
2493 ifp->if_broot_bytes = (int)new_size;
2494 ASSERT(ifp->if_broot_bytes <=
2495 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2501 * This is called when the amount of space needed for if_data
2502 * is increased or decreased. The change in size is indicated by
2503 * the number of bytes that need to be added or deleted in the
2504 * byte_diff parameter.
2506 * If the amount of space needed has decreased below the size of the
2507 * inline buffer, then switch to using the inline buffer. Otherwise,
2508 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2509 * to what is needed.
2511 * ip -- the inode whose if_data area is changing
2512 * byte_diff -- the change in the number of bytes, positive or negative,
2513 * requested for the if_data array.
2525 if (byte_diff == 0) {
2529 ifp = XFS_IFORK_PTR(ip, whichfork);
2530 new_size = (int)ifp->if_bytes + byte_diff;
2531 ASSERT(new_size >= 0);
2533 if (new_size == 0) {
2534 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2535 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2537 ifp->if_u1.if_data = NULL;
2539 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2541 * If the valid extents/data can fit in if_inline_ext/data,
2542 * copy them from the malloc'd vector and free it.
2544 if (ifp->if_u1.if_data == NULL) {
2545 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2546 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2547 ASSERT(ifp->if_real_bytes != 0);
2548 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2550 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2551 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2556 * Stuck with malloc/realloc.
2557 * For inline data, the underlying buffer must be
2558 * a multiple of 4 bytes in size so that it can be
2559 * logged and stay on word boundaries. We enforce
2562 real_size = roundup(new_size, 4);
2563 if (ifp->if_u1.if_data == NULL) {
2564 ASSERT(ifp->if_real_bytes == 0);
2565 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2566 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2568 * Only do the realloc if the underlying size
2569 * is really changing.
2571 if (ifp->if_real_bytes != real_size) {
2572 ifp->if_u1.if_data =
2573 kmem_realloc(ifp->if_u1.if_data,
2579 ASSERT(ifp->if_real_bytes == 0);
2580 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2581 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2585 ifp->if_real_bytes = real_size;
2586 ifp->if_bytes = new_size;
2587 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2594 * Map inode to disk block and offset.
2596 * mp -- the mount point structure for the current file system
2597 * tp -- the current transaction
2598 * ino -- the inode number of the inode to be located
2599 * imap -- this structure is filled in with the information necessary
2600 * to retrieve the given inode from disk
2601 * flags -- flags to pass to xfs_dilocate indicating whether or not
2602 * lookups in the inode btree were OK or not
2612 xfs_fsblock_t fsbno;
2617 fsbno = imap->im_blkno ?
2618 XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK;
2619 error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags);
2623 imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno);
2624 imap->im_len = XFS_FSB_TO_BB(mp, len);
2625 imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno);
2626 imap->im_ioffset = (ushort)off;
2627 imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog);
2630 * If the inode number maps to a block outside the bounds
2631 * of the file system then return NULL rather than calling
2632 * read_buf and panicing when we get an error from the
2635 if ((imap->im_blkno + imap->im_len) >
2636 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
2637 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
2638 "(imap->im_blkno (0x%llx) + imap->im_len (0x%llx)) > "
2639 " XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks) (0x%llx)",
2640 (unsigned long long) imap->im_blkno,
2641 (unsigned long long) imap->im_len,
2642 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
2655 ifp = XFS_IFORK_PTR(ip, whichfork);
2656 if (ifp->if_broot != NULL) {
2657 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2658 ifp->if_broot = NULL;
2662 * If the format is local, then we can't have an extents
2663 * array so just look for an inline data array. If we're
2664 * not local then we may or may not have an extents list,
2665 * so check and free it up if we do.
2667 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2668 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2669 (ifp->if_u1.if_data != NULL)) {
2670 ASSERT(ifp->if_real_bytes != 0);
2671 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2672 ifp->if_u1.if_data = NULL;
2673 ifp->if_real_bytes = 0;
2675 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2676 ((ifp->if_flags & XFS_IFEXTIREC) ||
2677 ((ifp->if_u1.if_extents != NULL) &&
2678 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
2679 ASSERT(ifp->if_real_bytes != 0);
2680 xfs_iext_destroy(ifp);
2682 ASSERT(ifp->if_u1.if_extents == NULL ||
2683 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2684 ASSERT(ifp->if_real_bytes == 0);
2685 if (whichfork == XFS_ATTR_FORK) {
2686 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2692 * This is called free all the memory associated with an inode.
2693 * It must free the inode itself and any buffers allocated for
2694 * if_extents/if_data and if_broot. It must also free the lock
2695 * associated with the inode.
2701 switch (ip->i_d.di_mode & S_IFMT) {
2705 xfs_idestroy_fork(ip, XFS_DATA_FORK);
2709 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
2710 mrfree(&ip->i_lock);
2711 mrfree(&ip->i_iolock);
2712 freesema(&ip->i_flock);
2714 #ifdef XFS_INODE_TRACE
2715 ktrace_free(ip->i_trace);
2717 #ifdef XFS_BMAP_TRACE
2718 ktrace_free(ip->i_xtrace);
2720 #ifdef XFS_BMBT_TRACE
2721 ktrace_free(ip->i_btrace);
2724 ktrace_free(ip->i_rwtrace);
2726 #ifdef XFS_ILOCK_TRACE
2727 ktrace_free(ip->i_lock_trace);
2729 #ifdef XFS_DIR2_TRACE
2730 ktrace_free(ip->i_dir_trace);
2734 * Only if we are shutting down the fs will we see an
2735 * inode still in the AIL. If it is there, we should remove
2736 * it to prevent a use-after-free from occurring.
2738 xfs_mount_t *mp = ip->i_mount;
2739 xfs_log_item_t *lip = &ip->i_itemp->ili_item;
2741 ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) ||
2742 XFS_FORCED_SHUTDOWN(ip->i_mount));
2743 if (lip->li_flags & XFS_LI_IN_AIL) {
2744 spin_lock(&mp->m_ail_lock);
2745 if (lip->li_flags & XFS_LI_IN_AIL)
2746 xfs_trans_delete_ail(mp, lip);
2748 spin_unlock(&mp->m_ail_lock);
2750 xfs_inode_item_destroy(ip);
2752 kmem_zone_free(xfs_inode_zone, ip);
2757 * Increment the pin count of the given buffer.
2758 * This value is protected by ipinlock spinlock in the mount structure.
2764 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2766 atomic_inc(&ip->i_pincount);
2770 * Decrement the pin count of the given inode, and wake up
2771 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2772 * inode must have been previously pinned with a call to xfs_ipin().
2778 ASSERT(atomic_read(&ip->i_pincount) > 0);
2780 if (atomic_dec_and_test(&ip->i_pincount))
2781 wake_up(&ip->i_ipin_wait);
2785 * This is called to unpin an inode. It can be directed to wait or to return
2786 * immediately without waiting for the inode to be unpinned. The caller must
2787 * have the inode locked in at least shared mode so that the buffer cannot be
2788 * subsequently pinned once someone is waiting for it to be unpinned.
2795 xfs_inode_log_item_t *iip = ip->i_itemp;
2797 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE | MR_ACCESS));
2798 if (atomic_read(&ip->i_pincount) == 0)
2801 /* Give the log a push to start the unpinning I/O */
2802 xfs_log_force(ip->i_mount, (iip && iip->ili_last_lsn) ?
2803 iip->ili_last_lsn : 0, XFS_LOG_FORCE);
2805 wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
2812 __xfs_iunpin_wait(ip, 1);
2819 __xfs_iunpin_wait(ip, 0);
2824 * xfs_iextents_copy()
2826 * This is called to copy the REAL extents (as opposed to the delayed
2827 * allocation extents) from the inode into the given buffer. It
2828 * returns the number of bytes copied into the buffer.
2830 * If there are no delayed allocation extents, then we can just
2831 * memcpy() the extents into the buffer. Otherwise, we need to
2832 * examine each extent in turn and skip those which are delayed.
2844 xfs_fsblock_t start_block;
2846 ifp = XFS_IFORK_PTR(ip, whichfork);
2847 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
2848 ASSERT(ifp->if_bytes > 0);
2850 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2851 XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork);
2855 * There are some delayed allocation extents in the
2856 * inode, so copy the extents one at a time and skip
2857 * the delayed ones. There must be at least one
2858 * non-delayed extent.
2861 for (i = 0; i < nrecs; i++) {
2862 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
2863 start_block = xfs_bmbt_get_startblock(ep);
2864 if (ISNULLSTARTBLOCK(start_block)) {
2866 * It's a delayed allocation extent, so skip it.
2871 /* Translate to on disk format */
2872 put_unaligned(cpu_to_be64(ep->l0), &dp->l0);
2873 put_unaligned(cpu_to_be64(ep->l1), &dp->l1);
2877 ASSERT(copied != 0);
2878 xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip));
2880 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2884 * Each of the following cases stores data into the same region
2885 * of the on-disk inode, so only one of them can be valid at
2886 * any given time. While it is possible to have conflicting formats
2887 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2888 * in EXTENTS format, this can only happen when the fork has
2889 * changed formats after being modified but before being flushed.
2890 * In these cases, the format always takes precedence, because the
2891 * format indicates the current state of the fork.
2898 xfs_inode_log_item_t *iip,
2905 #ifdef XFS_TRANS_DEBUG
2908 static const short brootflag[2] =
2909 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2910 static const short dataflag[2] =
2911 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2912 static const short extflag[2] =
2913 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2917 ifp = XFS_IFORK_PTR(ip, whichfork);
2919 * This can happen if we gave up in iformat in an error path,
2920 * for the attribute fork.
2923 ASSERT(whichfork == XFS_ATTR_FORK);
2926 cp = XFS_DFORK_PTR(dip, whichfork);
2928 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2929 case XFS_DINODE_FMT_LOCAL:
2930 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2931 (ifp->if_bytes > 0)) {
2932 ASSERT(ifp->if_u1.if_data != NULL);
2933 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2934 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2938 case XFS_DINODE_FMT_EXTENTS:
2939 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2940 !(iip->ili_format.ilf_fields & extflag[whichfork]));
2941 ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
2942 (ifp->if_bytes == 0));
2943 ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
2944 (ifp->if_bytes > 0));
2945 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
2946 (ifp->if_bytes > 0)) {
2947 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2948 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2953 case XFS_DINODE_FMT_BTREE:
2954 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
2955 (ifp->if_broot_bytes > 0)) {
2956 ASSERT(ifp->if_broot != NULL);
2957 ASSERT(ifp->if_broot_bytes <=
2958 (XFS_IFORK_SIZE(ip, whichfork) +
2959 XFS_BROOT_SIZE_ADJ));
2960 xfs_bmbt_to_bmdr(ifp->if_broot, ifp->if_broot_bytes,
2961 (xfs_bmdr_block_t *)cp,
2962 XFS_DFORK_SIZE(dip, mp, whichfork));
2966 case XFS_DINODE_FMT_DEV:
2967 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
2968 ASSERT(whichfork == XFS_DATA_FORK);
2969 dip->di_u.di_dev = cpu_to_be32(ip->i_df.if_u2.if_rdev);
2973 case XFS_DINODE_FMT_UUID:
2974 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
2975 ASSERT(whichfork == XFS_DATA_FORK);
2976 memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid,
2994 xfs_mount_t *mp = ip->i_mount;
2995 xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino);
2996 unsigned long first_index, mask;
2998 xfs_inode_t **ilist;
3005 ASSERT(pag->pagi_inodeok);
3006 ASSERT(pag->pag_ici_init);
3008 ilist_size = XFS_INODE_CLUSTER_SIZE(mp) * sizeof(xfs_inode_t *);
3009 ilist = kmem_alloc(ilist_size, KM_MAYFAIL);
3013 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
3014 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
3015 read_lock(&pag->pag_ici_lock);
3016 /* really need a gang lookup range call here */
3017 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
3019 XFS_INODE_CLUSTER_SIZE(mp));
3023 for (i = 0; i < nr_found; i++) {
3027 /* if the inode lies outside this cluster, we're done. */
3028 if ((XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index)
3031 * Do an un-protected check to see if the inode is dirty and
3032 * is a candidate for flushing. These checks will be repeated
3033 * later after the appropriate locks are acquired.
3035 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
3039 * Try to get locks. If any are unavailable or it is pinned,
3040 * then this inode cannot be flushed and is skipped.
3043 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
3045 if (!xfs_iflock_nowait(iq)) {
3046 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3049 if (xfs_ipincount(iq)) {
3051 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3056 * arriving here means that this inode can be flushed. First
3057 * re-check that it's dirty before flushing.
3059 if (!xfs_inode_clean(iq)) {
3061 error = xfs_iflush_int(iq, bp);
3063 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3064 goto cluster_corrupt_out;
3070 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3074 XFS_STATS_INC(xs_icluster_flushcnt);
3075 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
3079 read_unlock(&pag->pag_ici_lock);
3080 kmem_free(ilist, ilist_size);
3084 cluster_corrupt_out:
3086 * Corruption detected in the clustering loop. Invalidate the
3087 * inode buffer and shut down the filesystem.
3089 read_unlock(&pag->pag_ici_lock);
3091 * Clean up the buffer. If it was B_DELWRI, just release it --
3092 * brelse can handle it with no problems. If not, shut down the
3093 * filesystem before releasing the buffer.
3095 bufwasdelwri = XFS_BUF_ISDELAYWRITE(bp);
3099 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3101 if (!bufwasdelwri) {
3103 * Just like incore_relse: if we have b_iodone functions,
3104 * mark the buffer as an error and call them. Otherwise
3105 * mark it as stale and brelse.
3107 if (XFS_BUF_IODONE_FUNC(bp)) {
3108 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
3112 XFS_BUF_ERROR(bp,EIO);
3121 * Unlocks the flush lock
3123 xfs_iflush_abort(iq);
3124 kmem_free(ilist, ilist_size);
3125 return XFS_ERROR(EFSCORRUPTED);
3129 * xfs_iflush() will write a modified inode's changes out to the
3130 * inode's on disk home. The caller must have the inode lock held
3131 * in at least shared mode and the inode flush semaphore must be
3132 * held as well. The inode lock will still be held upon return from
3133 * the call and the caller is free to unlock it.
3134 * The inode flush lock will be unlocked when the inode reaches the disk.
3135 * The flags indicate how the inode's buffer should be written out.
3142 xfs_inode_log_item_t *iip;
3147 int noblock = (flags == XFS_IFLUSH_ASYNC_NOBLOCK);
3148 enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
3150 XFS_STATS_INC(xs_iflush_count);
3152 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
3153 ASSERT(issemalocked(&(ip->i_flock)));
3154 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3155 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3161 * If the inode isn't dirty, then just release the inode
3162 * flush lock and do nothing.
3164 if (xfs_inode_clean(ip)) {
3165 ASSERT((iip != NULL) ?
3166 !(iip->ili_item.li_flags & XFS_LI_IN_AIL) : 1);
3172 * We can't flush the inode until it is unpinned, so wait for it if we
3173 * are allowed to block. We know noone new can pin it, because we are
3174 * holding the inode lock shared and you need to hold it exclusively to
3177 * If we are not allowed to block, force the log out asynchronously so
3178 * that when we come back the inode will be unpinned. If other inodes
3179 * in the same cluster are dirty, they will probably write the inode
3180 * out for us if they occur after the log force completes.
3182 if (noblock && xfs_ipincount(ip)) {
3183 xfs_iunpin_nowait(ip);
3187 xfs_iunpin_wait(ip);
3190 * This may have been unpinned because the filesystem is shutting
3191 * down forcibly. If that's the case we must not write this inode
3192 * to disk, because the log record didn't make it to disk!
3194 if (XFS_FORCED_SHUTDOWN(mp)) {
3195 ip->i_update_core = 0;
3197 iip->ili_format.ilf_fields = 0;
3199 return XFS_ERROR(EIO);
3203 * Decide how buffer will be flushed out. This is done before
3204 * the call to xfs_iflush_int because this field is zeroed by it.
3206 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3208 * Flush out the inode buffer according to the directions
3209 * of the caller. In the cases where the caller has given
3210 * us a choice choose the non-delwri case. This is because
3211 * the inode is in the AIL and we need to get it out soon.
3214 case XFS_IFLUSH_SYNC:
3215 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3218 case XFS_IFLUSH_ASYNC_NOBLOCK:
3219 case XFS_IFLUSH_ASYNC:
3220 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3223 case XFS_IFLUSH_DELWRI:
3233 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3234 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3235 case XFS_IFLUSH_DELWRI:
3238 case XFS_IFLUSH_ASYNC_NOBLOCK:
3239 case XFS_IFLUSH_ASYNC:
3242 case XFS_IFLUSH_SYNC:
3253 * Get the buffer containing the on-disk inode.
3255 error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0, 0,
3256 noblock ? XFS_BUF_TRYLOCK : XFS_BUF_LOCK);
3263 * First flush out the inode that xfs_iflush was called with.
3265 error = xfs_iflush_int(ip, bp);
3270 * If the buffer is pinned then push on the log now so we won't
3271 * get stuck waiting in the write for too long.
3273 if (XFS_BUF_ISPINNED(bp))
3274 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
3278 * see if other inodes can be gathered into this write
3280 error = xfs_iflush_cluster(ip, bp);
3282 goto cluster_corrupt_out;
3284 if (flags & INT_DELWRI) {
3285 xfs_bdwrite(mp, bp);
3286 } else if (flags & INT_ASYNC) {
3287 xfs_bawrite(mp, bp);
3289 error = xfs_bwrite(mp, bp);
3295 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3296 cluster_corrupt_out:
3298 * Unlocks the flush lock
3300 xfs_iflush_abort(ip);
3301 return XFS_ERROR(EFSCORRUPTED);
3310 xfs_inode_log_item_t *iip;
3313 #ifdef XFS_TRANS_DEBUG
3317 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
3318 ASSERT(issemalocked(&(ip->i_flock)));
3319 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3320 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3327 * If the inode isn't dirty, then just release the inode
3328 * flush lock and do nothing.
3330 if (xfs_inode_clean(ip)) {
3335 /* set *dip = inode's place in the buffer */
3336 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset);
3339 * Clear i_update_core before copying out the data.
3340 * This is for coordination with our timestamp updates
3341 * that don't hold the inode lock. They will always
3342 * update the timestamps BEFORE setting i_update_core,
3343 * so if we clear i_update_core after they set it we
3344 * are guaranteed to see their updates to the timestamps.
3345 * I believe that this depends on strongly ordered memory
3346 * semantics, but we have that. We use the SYNCHRONIZE
3347 * macro to make sure that the compiler does not reorder
3348 * the i_update_core access below the data copy below.
3350 ip->i_update_core = 0;
3354 * Make sure to get the latest atime from the Linux inode.
3356 xfs_synchronize_atime(ip);
3358 if (XFS_TEST_ERROR(be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC,
3359 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3360 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3361 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3362 ip->i_ino, be16_to_cpu(dip->di_core.di_magic), dip);
3365 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3366 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3367 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3368 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3369 ip->i_ino, ip, ip->i_d.di_magic);
3372 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3374 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3375 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3376 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3377 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3378 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3382 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
3384 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3385 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3386 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3387 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3388 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3389 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3394 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3395 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3396 XFS_RANDOM_IFLUSH_5)) {
3397 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3398 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3400 ip->i_d.di_nextents + ip->i_d.di_anextents,
3405 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3406 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3407 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3408 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3409 ip->i_ino, ip->i_d.di_forkoff, ip);
3413 * bump the flush iteration count, used to detect flushes which
3414 * postdate a log record during recovery.
3417 ip->i_d.di_flushiter++;
3420 * Copy the dirty parts of the inode into the on-disk
3421 * inode. We always copy out the core of the inode,
3422 * because if the inode is dirty at all the core must
3425 xfs_dinode_to_disk(&dip->di_core, &ip->i_d);
3427 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3428 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3429 ip->i_d.di_flushiter = 0;
3432 * If this is really an old format inode and the superblock version
3433 * has not been updated to support only new format inodes, then
3434 * convert back to the old inode format. If the superblock version
3435 * has been updated, then make the conversion permanent.
3437 ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 ||
3438 xfs_sb_version_hasnlink(&mp->m_sb));
3439 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
3440 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
3444 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3445 dip->di_core.di_onlink = cpu_to_be16(ip->i_d.di_nlink);
3448 * The superblock version has already been bumped,
3449 * so just make the conversion to the new inode
3452 ip->i_d.di_version = XFS_DINODE_VERSION_2;
3453 dip->di_core.di_version = XFS_DINODE_VERSION_2;
3454 ip->i_d.di_onlink = 0;
3455 dip->di_core.di_onlink = 0;
3456 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3457 memset(&(dip->di_core.di_pad[0]), 0,
3458 sizeof(dip->di_core.di_pad));
3459 ASSERT(ip->i_d.di_projid == 0);
3463 if (xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp) == EFSCORRUPTED) {
3467 if (XFS_IFORK_Q(ip)) {
3469 * The only error from xfs_iflush_fork is on the data fork.
3471 (void) xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3473 xfs_inobp_check(mp, bp);
3476 * We've recorded everything logged in the inode, so we'd
3477 * like to clear the ilf_fields bits so we don't log and
3478 * flush things unnecessarily. However, we can't stop
3479 * logging all this information until the data we've copied
3480 * into the disk buffer is written to disk. If we did we might
3481 * overwrite the copy of the inode in the log with all the
3482 * data after re-logging only part of it, and in the face of
3483 * a crash we wouldn't have all the data we need to recover.
3485 * What we do is move the bits to the ili_last_fields field.
3486 * When logging the inode, these bits are moved back to the
3487 * ilf_fields field. In the xfs_iflush_done() routine we
3488 * clear ili_last_fields, since we know that the information
3489 * those bits represent is permanently on disk. As long as
3490 * the flush completes before the inode is logged again, then
3491 * both ilf_fields and ili_last_fields will be cleared.
3493 * We can play with the ilf_fields bits here, because the inode
3494 * lock must be held exclusively in order to set bits there
3495 * and the flush lock protects the ili_last_fields bits.
3496 * Set ili_logged so the flush done
3497 * routine can tell whether or not to look in the AIL.
3498 * Also, store the current LSN of the inode so that we can tell
3499 * whether the item has moved in the AIL from xfs_iflush_done().
3500 * In order to read the lsn we need the AIL lock, because
3501 * it is a 64 bit value that cannot be read atomically.
3503 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3504 iip->ili_last_fields = iip->ili_format.ilf_fields;
3505 iip->ili_format.ilf_fields = 0;
3506 iip->ili_logged = 1;
3508 ASSERT(sizeof(xfs_lsn_t) == 8); /* don't lock if it shrinks */
3509 spin_lock(&mp->m_ail_lock);
3510 iip->ili_flush_lsn = iip->ili_item.li_lsn;
3511 spin_unlock(&mp->m_ail_lock);
3514 * Attach the function xfs_iflush_done to the inode's
3515 * buffer. This will remove the inode from the AIL
3516 * and unlock the inode's flush lock when the inode is
3517 * completely written to disk.
3519 xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
3520 xfs_iflush_done, (xfs_log_item_t *)iip);
3522 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3523 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3526 * We're flushing an inode which is not in the AIL and has
3527 * not been logged but has i_update_core set. For this
3528 * case we can use a B_DELWRI flush and immediately drop
3529 * the inode flush lock because we can avoid the whole
3530 * AIL state thing. It's OK to drop the flush lock now,
3531 * because we've already locked the buffer and to do anything
3532 * you really need both.
3535 ASSERT(iip->ili_logged == 0);
3536 ASSERT(iip->ili_last_fields == 0);
3537 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3545 return XFS_ERROR(EFSCORRUPTED);
3550 * Flush all inactive inodes in mp.
3560 XFS_MOUNT_ILOCK(mp);
3566 /* Make sure we skip markers inserted by sync */
3567 if (ip->i_mount == NULL) {
3572 vp = XFS_ITOV_NULL(ip);
3574 XFS_MOUNT_IUNLOCK(mp);
3575 xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC);
3579 ASSERT(vn_count(vp) == 0);
3582 } while (ip != mp->m_inodes);
3584 XFS_MOUNT_IUNLOCK(mp);
3587 #ifdef XFS_ILOCK_TRACE
3588 ktrace_t *xfs_ilock_trace_buf;
3591 xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
3593 ktrace_enter(ip->i_lock_trace,
3595 (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
3596 (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
3597 (void *)ra, /* caller of ilock */
3598 (void *)(unsigned long)current_cpu(),
3599 (void *)(unsigned long)current_pid(),
3600 NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL);
3605 * Return a pointer to the extent record at file index idx.
3607 xfs_bmbt_rec_host_t *
3609 xfs_ifork_t *ifp, /* inode fork pointer */
3610 xfs_extnum_t idx) /* index of target extent */
3613 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
3614 return ifp->if_u1.if_ext_irec->er_extbuf;
3615 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3616 xfs_ext_irec_t *erp; /* irec pointer */
3617 int erp_idx = 0; /* irec index */
3618 xfs_extnum_t page_idx = idx; /* ext index in target list */
3620 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3621 return &erp->er_extbuf[page_idx];
3622 } else if (ifp->if_bytes) {
3623 return &ifp->if_u1.if_extents[idx];
3630 * Insert new item(s) into the extent records for incore inode
3631 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3635 xfs_ifork_t *ifp, /* inode fork pointer */
3636 xfs_extnum_t idx, /* starting index of new items */
3637 xfs_extnum_t count, /* number of inserted items */
3638 xfs_bmbt_irec_t *new) /* items to insert */
3640 xfs_extnum_t i; /* extent record index */
3642 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3643 xfs_iext_add(ifp, idx, count);
3644 for (i = idx; i < idx + count; i++, new++)
3645 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new);
3649 * This is called when the amount of space required for incore file
3650 * extents needs to be increased. The ext_diff parameter stores the
3651 * number of new extents being added and the idx parameter contains
3652 * the extent index where the new extents will be added. If the new
3653 * extents are being appended, then we just need to (re)allocate and
3654 * initialize the space. Otherwise, if the new extents are being
3655 * inserted into the middle of the existing entries, a bit more work
3656 * is required to make room for the new extents to be inserted. The
3657 * caller is responsible for filling in the new extent entries upon
3662 xfs_ifork_t *ifp, /* inode fork pointer */
3663 xfs_extnum_t idx, /* index to begin adding exts */
3664 int ext_diff) /* number of extents to add */
3666 int byte_diff; /* new bytes being added */
3667 int new_size; /* size of extents after adding */
3668 xfs_extnum_t nextents; /* number of extents in file */
3670 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3671 ASSERT((idx >= 0) && (idx <= nextents));
3672 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
3673 new_size = ifp->if_bytes + byte_diff;
3675 * If the new number of extents (nextents + ext_diff)
3676 * fits inside the inode, then continue to use the inline
3679 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
3680 if (idx < nextents) {
3681 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
3682 &ifp->if_u2.if_inline_ext[idx],
3683 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3684 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
3686 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3687 ifp->if_real_bytes = 0;
3688 ifp->if_lastex = nextents + ext_diff;
3691 * Otherwise use a linear (direct) extent list.
3692 * If the extents are currently inside the inode,
3693 * xfs_iext_realloc_direct will switch us from
3694 * inline to direct extent allocation mode.
3696 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
3697 xfs_iext_realloc_direct(ifp, new_size);
3698 if (idx < nextents) {
3699 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
3700 &ifp->if_u1.if_extents[idx],
3701 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3702 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
3705 /* Indirection array */
3707 xfs_ext_irec_t *erp;
3711 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
3712 if (ifp->if_flags & XFS_IFEXTIREC) {
3713 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3715 xfs_iext_irec_init(ifp);
3716 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3717 erp = ifp->if_u1.if_ext_irec;
3719 /* Extents fit in target extent page */
3720 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3721 if (page_idx < erp->er_extcount) {
3722 memmove(&erp->er_extbuf[page_idx + ext_diff],
3723 &erp->er_extbuf[page_idx],
3724 (erp->er_extcount - page_idx) *
3725 sizeof(xfs_bmbt_rec_t));
3726 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3728 erp->er_extcount += ext_diff;
3729 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3731 /* Insert a new extent page */
3733 xfs_iext_add_indirect_multi(ifp,
3734 erp_idx, page_idx, ext_diff);
3737 * If extent(s) are being appended to the last page in
3738 * the indirection array and the new extent(s) don't fit
3739 * in the page, then erp is NULL and erp_idx is set to
3740 * the next index needed in the indirection array.
3743 int count = ext_diff;
3746 erp = xfs_iext_irec_new(ifp, erp_idx);
3747 erp->er_extcount = count;
3748 count -= MIN(count, (int)XFS_LINEAR_EXTS);
3755 ifp->if_bytes = new_size;
3759 * This is called when incore extents are being added to the indirection
3760 * array and the new extents do not fit in the target extent list. The
3761 * erp_idx parameter contains the irec index for the target extent list
3762 * in the indirection array, and the idx parameter contains the extent
3763 * index within the list. The number of extents being added is stored
3764 * in the count parameter.
3766 * |-------| |-------|
3767 * | | | | idx - number of extents before idx
3769 * | | | | count - number of extents being inserted at idx
3770 * |-------| |-------|
3771 * | count | | nex2 | nex2 - number of extents after idx + count
3772 * |-------| |-------|
3775 xfs_iext_add_indirect_multi(
3776 xfs_ifork_t *ifp, /* inode fork pointer */
3777 int erp_idx, /* target extent irec index */
3778 xfs_extnum_t idx, /* index within target list */
3779 int count) /* new extents being added */
3781 int byte_diff; /* new bytes being added */
3782 xfs_ext_irec_t *erp; /* pointer to irec entry */
3783 xfs_extnum_t ext_diff; /* number of extents to add */
3784 xfs_extnum_t ext_cnt; /* new extents still needed */
3785 xfs_extnum_t nex2; /* extents after idx + count */
3786 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3787 int nlists; /* number of irec's (lists) */
3789 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3790 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3791 nex2 = erp->er_extcount - idx;
3792 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3795 * Save second part of target extent list
3796 * (all extents past */
3798 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3799 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_SLEEP);
3800 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3801 erp->er_extcount -= nex2;
3802 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3803 memset(&erp->er_extbuf[idx], 0, byte_diff);
3807 * Add the new extents to the end of the target
3808 * list, then allocate new irec record(s) and
3809 * extent buffer(s) as needed to store the rest
3810 * of the new extents.
3813 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3815 erp->er_extcount += ext_diff;
3816 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3817 ext_cnt -= ext_diff;
3821 erp = xfs_iext_irec_new(ifp, erp_idx);
3822 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3823 erp->er_extcount = ext_diff;
3824 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3825 ext_cnt -= ext_diff;
3828 /* Add nex2 extents back to indirection array */
3830 xfs_extnum_t ext_avail;
3833 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3834 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3837 * If nex2 extents fit in the current page, append
3838 * nex2_ep after the new extents.
3840 if (nex2 <= ext_avail) {
3841 i = erp->er_extcount;
3844 * Otherwise, check if space is available in the
3847 else if ((erp_idx < nlists - 1) &&
3848 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3849 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3852 /* Create a hole for nex2 extents */
3853 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3854 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3857 * Final choice, create a new extent page for
3862 erp = xfs_iext_irec_new(ifp, erp_idx);
3864 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
3865 kmem_free(nex2_ep, byte_diff);
3866 erp->er_extcount += nex2;
3867 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3872 * This is called when the amount of space required for incore file
3873 * extents needs to be decreased. The ext_diff parameter stores the
3874 * number of extents to be removed and the idx parameter contains
3875 * the extent index where the extents will be removed from.
3877 * If the amount of space needed has decreased below the linear
3878 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3879 * extent array. Otherwise, use kmem_realloc() to adjust the
3880 * size to what is needed.
3884 xfs_ifork_t *ifp, /* inode fork pointer */
3885 xfs_extnum_t idx, /* index to begin removing exts */
3886 int ext_diff) /* number of extents to remove */
3888 xfs_extnum_t nextents; /* number of extents in file */
3889 int new_size; /* size of extents after removal */
3891 ASSERT(ext_diff > 0);
3892 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3893 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3895 if (new_size == 0) {
3896 xfs_iext_destroy(ifp);
3897 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3898 xfs_iext_remove_indirect(ifp, idx, ext_diff);
3899 } else if (ifp->if_real_bytes) {
3900 xfs_iext_remove_direct(ifp, idx, ext_diff);
3902 xfs_iext_remove_inline(ifp, idx, ext_diff);
3904 ifp->if_bytes = new_size;
3908 * This removes ext_diff extents from the inline buffer, beginning
3909 * at extent index idx.
3912 xfs_iext_remove_inline(
3913 xfs_ifork_t *ifp, /* inode fork pointer */
3914 xfs_extnum_t idx, /* index to begin removing exts */
3915 int ext_diff) /* number of extents to remove */
3917 int nextents; /* number of extents in file */
3919 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3920 ASSERT(idx < XFS_INLINE_EXTS);
3921 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3922 ASSERT(((nextents - ext_diff) > 0) &&
3923 (nextents - ext_diff) < XFS_INLINE_EXTS);
3925 if (idx + ext_diff < nextents) {
3926 memmove(&ifp->if_u2.if_inline_ext[idx],
3927 &ifp->if_u2.if_inline_ext[idx + ext_diff],
3928 (nextents - (idx + ext_diff)) *
3929 sizeof(xfs_bmbt_rec_t));
3930 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
3931 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3933 memset(&ifp->if_u2.if_inline_ext[idx], 0,
3934 ext_diff * sizeof(xfs_bmbt_rec_t));
3939 * This removes ext_diff extents from a linear (direct) extent list,
3940 * beginning at extent index idx. If the extents are being removed
3941 * from the end of the list (ie. truncate) then we just need to re-
3942 * allocate the list to remove the extra space. Otherwise, if the
3943 * extents are being removed from the middle of the existing extent
3944 * entries, then we first need to move the extent records beginning
3945 * at idx + ext_diff up in the list to overwrite the records being
3946 * removed, then remove the extra space via kmem_realloc.
3949 xfs_iext_remove_direct(
3950 xfs_ifork_t *ifp, /* inode fork pointer */
3951 xfs_extnum_t idx, /* index to begin removing exts */
3952 int ext_diff) /* number of extents to remove */
3954 xfs_extnum_t nextents; /* number of extents in file */
3955 int new_size; /* size of extents after removal */
3957 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3958 new_size = ifp->if_bytes -
3959 (ext_diff * sizeof(xfs_bmbt_rec_t));
3960 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3962 if (new_size == 0) {
3963 xfs_iext_destroy(ifp);
3966 /* Move extents up in the list (if needed) */
3967 if (idx + ext_diff < nextents) {
3968 memmove(&ifp->if_u1.if_extents[idx],
3969 &ifp->if_u1.if_extents[idx + ext_diff],
3970 (nextents - (idx + ext_diff)) *
3971 sizeof(xfs_bmbt_rec_t));
3973 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
3974 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3976 * Reallocate the direct extent list. If the extents
3977 * will fit inside the inode then xfs_iext_realloc_direct
3978 * will switch from direct to inline extent allocation
3981 xfs_iext_realloc_direct(ifp, new_size);
3982 ifp->if_bytes = new_size;
3986 * This is called when incore extents are being removed from the
3987 * indirection array and the extents being removed span multiple extent
3988 * buffers. The idx parameter contains the file extent index where we
3989 * want to begin removing extents, and the count parameter contains
3990 * how many extents need to be removed.
3992 * |-------| |-------|
3993 * | nex1 | | | nex1 - number of extents before idx
3994 * |-------| | count |
3995 * | | | | count - number of extents being removed at idx
3996 * | count | |-------|
3997 * | | | nex2 | nex2 - number of extents after idx + count
3998 * |-------| |-------|
4001 xfs_iext_remove_indirect(
4002 xfs_ifork_t *ifp, /* inode fork pointer */
4003 xfs_extnum_t idx, /* index to begin removing extents */
4004 int count) /* number of extents to remove */
4006 xfs_ext_irec_t *erp; /* indirection array pointer */
4007 int erp_idx = 0; /* indirection array index */
4008 xfs_extnum_t ext_cnt; /* extents left to remove */
4009 xfs_extnum_t ext_diff; /* extents to remove in current list */
4010 xfs_extnum_t nex1; /* number of extents before idx */
4011 xfs_extnum_t nex2; /* extents after idx + count */
4012 int nlists; /* entries in indirection array */
4013 int page_idx = idx; /* index in target extent list */
4015 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4016 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
4017 ASSERT(erp != NULL);
4018 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4022 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
4023 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
4025 * Check for deletion of entire list;
4026 * xfs_iext_irec_remove() updates extent offsets.
4028 if (ext_diff == erp->er_extcount) {
4029 xfs_iext_irec_remove(ifp, erp_idx);
4030 ext_cnt -= ext_diff;
4033 ASSERT(erp_idx < ifp->if_real_bytes /
4035 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4042 /* Move extents up (if needed) */
4044 memmove(&erp->er_extbuf[nex1],
4045 &erp->er_extbuf[nex1 + ext_diff],
4046 nex2 * sizeof(xfs_bmbt_rec_t));
4048 /* Zero out rest of page */
4049 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
4050 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
4051 /* Update remaining counters */
4052 erp->er_extcount -= ext_diff;
4053 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
4054 ext_cnt -= ext_diff;
4059 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
4060 xfs_iext_irec_compact(ifp);
4064 * Create, destroy, or resize a linear (direct) block of extents.
4067 xfs_iext_realloc_direct(
4068 xfs_ifork_t *ifp, /* inode fork pointer */
4069 int new_size) /* new size of extents */
4071 int rnew_size; /* real new size of extents */
4073 rnew_size = new_size;
4075 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
4076 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
4077 (new_size != ifp->if_real_bytes)));
4079 /* Free extent records */
4080 if (new_size == 0) {
4081 xfs_iext_destroy(ifp);
4083 /* Resize direct extent list and zero any new bytes */
4084 else if (ifp->if_real_bytes) {
4085 /* Check if extents will fit inside the inode */
4086 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
4087 xfs_iext_direct_to_inline(ifp, new_size /
4088 (uint)sizeof(xfs_bmbt_rec_t));
4089 ifp->if_bytes = new_size;
4092 if (!is_power_of_2(new_size)){
4093 rnew_size = roundup_pow_of_two(new_size);
4095 if (rnew_size != ifp->if_real_bytes) {
4096 ifp->if_u1.if_extents =
4097 kmem_realloc(ifp->if_u1.if_extents,
4102 if (rnew_size > ifp->if_real_bytes) {
4103 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
4104 (uint)sizeof(xfs_bmbt_rec_t)], 0,
4105 rnew_size - ifp->if_real_bytes);
4109 * Switch from the inline extent buffer to a direct
4110 * extent list. Be sure to include the inline extent
4111 * bytes in new_size.
4114 new_size += ifp->if_bytes;
4115 if (!is_power_of_2(new_size)) {
4116 rnew_size = roundup_pow_of_two(new_size);
4118 xfs_iext_inline_to_direct(ifp, rnew_size);
4120 ifp->if_real_bytes = rnew_size;
4121 ifp->if_bytes = new_size;
4125 * Switch from linear (direct) extent records to inline buffer.
4128 xfs_iext_direct_to_inline(
4129 xfs_ifork_t *ifp, /* inode fork pointer */
4130 xfs_extnum_t nextents) /* number of extents in file */
4132 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4133 ASSERT(nextents <= XFS_INLINE_EXTS);
4135 * The inline buffer was zeroed when we switched
4136 * from inline to direct extent allocation mode,
4137 * so we don't need to clear it here.
4139 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
4140 nextents * sizeof(xfs_bmbt_rec_t));
4141 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
4142 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
4143 ifp->if_real_bytes = 0;
4147 * Switch from inline buffer to linear (direct) extent records.
4148 * new_size should already be rounded up to the next power of 2
4149 * by the caller (when appropriate), so use new_size as it is.
4150 * However, since new_size may be rounded up, we can't update
4151 * if_bytes here. It is the caller's responsibility to update
4152 * if_bytes upon return.
4155 xfs_iext_inline_to_direct(
4156 xfs_ifork_t *ifp, /* inode fork pointer */
4157 int new_size) /* number of extents in file */
4159 ifp->if_u1.if_extents = kmem_alloc(new_size, KM_SLEEP);
4160 memset(ifp->if_u1.if_extents, 0, new_size);
4161 if (ifp->if_bytes) {
4162 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
4164 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4165 sizeof(xfs_bmbt_rec_t));
4167 ifp->if_real_bytes = new_size;
4171 * Resize an extent indirection array to new_size bytes.
4174 xfs_iext_realloc_indirect(
4175 xfs_ifork_t *ifp, /* inode fork pointer */
4176 int new_size) /* new indirection array size */
4178 int nlists; /* number of irec's (ex lists) */
4179 int size; /* current indirection array size */
4181 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4182 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4183 size = nlists * sizeof(xfs_ext_irec_t);
4184 ASSERT(ifp->if_real_bytes);
4185 ASSERT((new_size >= 0) && (new_size != size));
4186 if (new_size == 0) {
4187 xfs_iext_destroy(ifp);
4189 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
4190 kmem_realloc(ifp->if_u1.if_ext_irec,
4191 new_size, size, KM_SLEEP);
4196 * Switch from indirection array to linear (direct) extent allocations.
4199 xfs_iext_indirect_to_direct(
4200 xfs_ifork_t *ifp) /* inode fork pointer */
4202 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
4203 xfs_extnum_t nextents; /* number of extents in file */
4204 int size; /* size of file extents */
4206 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4207 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4208 ASSERT(nextents <= XFS_LINEAR_EXTS);
4209 size = nextents * sizeof(xfs_bmbt_rec_t);
4211 xfs_iext_irec_compact_full(ifp);
4212 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
4214 ep = ifp->if_u1.if_ext_irec->er_extbuf;
4215 kmem_free(ifp->if_u1.if_ext_irec, sizeof(xfs_ext_irec_t));
4216 ifp->if_flags &= ~XFS_IFEXTIREC;
4217 ifp->if_u1.if_extents = ep;
4218 ifp->if_bytes = size;
4219 if (nextents < XFS_LINEAR_EXTS) {
4220 xfs_iext_realloc_direct(ifp, size);
4225 * Free incore file extents.
4229 xfs_ifork_t *ifp) /* inode fork pointer */
4231 if (ifp->if_flags & XFS_IFEXTIREC) {
4235 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4236 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
4237 xfs_iext_irec_remove(ifp, erp_idx);
4239 ifp->if_flags &= ~XFS_IFEXTIREC;
4240 } else if (ifp->if_real_bytes) {
4241 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
4242 } else if (ifp->if_bytes) {
4243 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4244 sizeof(xfs_bmbt_rec_t));
4246 ifp->if_u1.if_extents = NULL;
4247 ifp->if_real_bytes = 0;
4252 * Return a pointer to the extent record for file system block bno.
4254 xfs_bmbt_rec_host_t * /* pointer to found extent record */
4255 xfs_iext_bno_to_ext(
4256 xfs_ifork_t *ifp, /* inode fork pointer */
4257 xfs_fileoff_t bno, /* block number to search for */
4258 xfs_extnum_t *idxp) /* index of target extent */
4260 xfs_bmbt_rec_host_t *base; /* pointer to first extent */
4261 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
4262 xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */
4263 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4264 int high; /* upper boundary in search */
4265 xfs_extnum_t idx = 0; /* index of target extent */
4266 int low; /* lower boundary in search */
4267 xfs_extnum_t nextents; /* number of file extents */
4268 xfs_fileoff_t startoff = 0; /* start offset of extent */
4270 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4271 if (nextents == 0) {
4276 if (ifp->if_flags & XFS_IFEXTIREC) {
4277 /* Find target extent list */
4279 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
4280 base = erp->er_extbuf;
4281 high = erp->er_extcount - 1;
4283 base = ifp->if_u1.if_extents;
4284 high = nextents - 1;
4286 /* Binary search extent records */
4287 while (low <= high) {
4288 idx = (low + high) >> 1;
4290 startoff = xfs_bmbt_get_startoff(ep);
4291 blockcount = xfs_bmbt_get_blockcount(ep);
4292 if (bno < startoff) {
4294 } else if (bno >= startoff + blockcount) {
4297 /* Convert back to file-based extent index */
4298 if (ifp->if_flags & XFS_IFEXTIREC) {
4299 idx += erp->er_extoff;
4305 /* Convert back to file-based extent index */
4306 if (ifp->if_flags & XFS_IFEXTIREC) {
4307 idx += erp->er_extoff;
4309 if (bno >= startoff + blockcount) {
4310 if (++idx == nextents) {
4313 ep = xfs_iext_get_ext(ifp, idx);
4321 * Return a pointer to the indirection array entry containing the
4322 * extent record for filesystem block bno. Store the index of the
4323 * target irec in *erp_idxp.
4325 xfs_ext_irec_t * /* pointer to found extent record */
4326 xfs_iext_bno_to_irec(
4327 xfs_ifork_t *ifp, /* inode fork pointer */
4328 xfs_fileoff_t bno, /* block number to search for */
4329 int *erp_idxp) /* irec index of target ext list */
4331 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4332 xfs_ext_irec_t *erp_next; /* next indirection array entry */
4333 int erp_idx; /* indirection array index */
4334 int nlists; /* number of extent irec's (lists) */
4335 int high; /* binary search upper limit */
4336 int low; /* binary search lower limit */
4338 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4339 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4343 while (low <= high) {
4344 erp_idx = (low + high) >> 1;
4345 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4346 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
4347 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
4349 } else if (erp_next && bno >=
4350 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
4356 *erp_idxp = erp_idx;
4361 * Return a pointer to the indirection array entry containing the
4362 * extent record at file extent index *idxp. Store the index of the
4363 * target irec in *erp_idxp and store the page index of the target
4364 * extent record in *idxp.
4367 xfs_iext_idx_to_irec(
4368 xfs_ifork_t *ifp, /* inode fork pointer */
4369 xfs_extnum_t *idxp, /* extent index (file -> page) */
4370 int *erp_idxp, /* pointer to target irec */
4371 int realloc) /* new bytes were just added */
4373 xfs_ext_irec_t *prev; /* pointer to previous irec */
4374 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
4375 int erp_idx; /* indirection array index */
4376 int nlists; /* number of irec's (ex lists) */
4377 int high; /* binary search upper limit */
4378 int low; /* binary search lower limit */
4379 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
4381 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4382 ASSERT(page_idx >= 0 && page_idx <=
4383 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
4384 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4389 /* Binary search extent irec's */
4390 while (low <= high) {
4391 erp_idx = (low + high) >> 1;
4392 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4393 prev = erp_idx > 0 ? erp - 1 : NULL;
4394 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
4395 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
4397 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
4398 (page_idx == erp->er_extoff + erp->er_extcount &&
4401 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
4402 erp->er_extcount == XFS_LINEAR_EXTS) {
4406 erp = erp_idx < nlists ? erp + 1 : NULL;
4409 page_idx -= erp->er_extoff;
4414 *erp_idxp = erp_idx;
4419 * Allocate and initialize an indirection array once the space needed
4420 * for incore extents increases above XFS_IEXT_BUFSZ.
4424 xfs_ifork_t *ifp) /* inode fork pointer */
4426 xfs_ext_irec_t *erp; /* indirection array pointer */
4427 xfs_extnum_t nextents; /* number of extents in file */
4429 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4430 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4431 ASSERT(nextents <= XFS_LINEAR_EXTS);
4433 erp = (xfs_ext_irec_t *)
4434 kmem_alloc(sizeof(xfs_ext_irec_t), KM_SLEEP);
4436 if (nextents == 0) {
4437 ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
4438 } else if (!ifp->if_real_bytes) {
4439 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
4440 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
4441 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
4443 erp->er_extbuf = ifp->if_u1.if_extents;
4444 erp->er_extcount = nextents;
4447 ifp->if_flags |= XFS_IFEXTIREC;
4448 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
4449 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
4450 ifp->if_u1.if_ext_irec = erp;
4456 * Allocate and initialize a new entry in the indirection array.
4460 xfs_ifork_t *ifp, /* inode fork pointer */
4461 int erp_idx) /* index for new irec */
4463 xfs_ext_irec_t *erp; /* indirection array pointer */
4464 int i; /* loop counter */
4465 int nlists; /* number of irec's (ex lists) */
4467 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4468 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4470 /* Resize indirection array */
4471 xfs_iext_realloc_indirect(ifp, ++nlists *
4472 sizeof(xfs_ext_irec_t));
4474 * Move records down in the array so the
4475 * new page can use erp_idx.
4477 erp = ifp->if_u1.if_ext_irec;
4478 for (i = nlists - 1; i > erp_idx; i--) {
4479 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
4481 ASSERT(i == erp_idx);
4483 /* Initialize new extent record */
4484 erp = ifp->if_u1.if_ext_irec;
4485 erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
4486 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4487 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
4488 erp[erp_idx].er_extcount = 0;
4489 erp[erp_idx].er_extoff = erp_idx > 0 ?
4490 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
4491 return (&erp[erp_idx]);
4495 * Remove a record from the indirection array.
4498 xfs_iext_irec_remove(
4499 xfs_ifork_t *ifp, /* inode fork pointer */
4500 int erp_idx) /* irec index to remove */
4502 xfs_ext_irec_t *erp; /* indirection array pointer */
4503 int i; /* loop counter */
4504 int nlists; /* number of irec's (ex lists) */
4506 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4507 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4508 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4509 if (erp->er_extbuf) {
4510 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
4512 kmem_free(erp->er_extbuf, XFS_IEXT_BUFSZ);
4514 /* Compact extent records */
4515 erp = ifp->if_u1.if_ext_irec;
4516 for (i = erp_idx; i < nlists - 1; i++) {
4517 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
4520 * Manually free the last extent record from the indirection
4521 * array. A call to xfs_iext_realloc_indirect() with a size
4522 * of zero would result in a call to xfs_iext_destroy() which
4523 * would in turn call this function again, creating a nasty
4527 xfs_iext_realloc_indirect(ifp,
4528 nlists * sizeof(xfs_ext_irec_t));
4530 kmem_free(ifp->if_u1.if_ext_irec,
4531 sizeof(xfs_ext_irec_t));
4533 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4537 * This is called to clean up large amounts of unused memory allocated
4538 * by the indirection array. Before compacting anything though, verify
4539 * that the indirection array is still needed and switch back to the
4540 * linear extent list (or even the inline buffer) if possible. The
4541 * compaction policy is as follows:
4543 * Full Compaction: Extents fit into a single page (or inline buffer)
4544 * Full Compaction: Extents occupy less than 10% of allocated space
4545 * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
4546 * No Compaction: Extents occupy at least 50% of allocated space
4549 xfs_iext_irec_compact(
4550 xfs_ifork_t *ifp) /* inode fork pointer */
4552 xfs_extnum_t nextents; /* number of extents in file */
4553 int nlists; /* number of irec's (ex lists) */
4555 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4556 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4557 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4559 if (nextents == 0) {
4560 xfs_iext_destroy(ifp);
4561 } else if (nextents <= XFS_INLINE_EXTS) {
4562 xfs_iext_indirect_to_direct(ifp);
4563 xfs_iext_direct_to_inline(ifp, nextents);
4564 } else if (nextents <= XFS_LINEAR_EXTS) {
4565 xfs_iext_indirect_to_direct(ifp);
4566 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 3) {
4567 xfs_iext_irec_compact_full(ifp);
4568 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
4569 xfs_iext_irec_compact_pages(ifp);
4574 * Combine extents from neighboring extent pages.
4577 xfs_iext_irec_compact_pages(
4578 xfs_ifork_t *ifp) /* inode fork pointer */
4580 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
4581 int erp_idx = 0; /* indirection array index */
4582 int nlists; /* number of irec's (ex lists) */
4584 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4585 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4586 while (erp_idx < nlists - 1) {
4587 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4589 if (erp_next->er_extcount <=
4590 (XFS_LINEAR_EXTS - erp->er_extcount)) {
4591 memmove(&erp->er_extbuf[erp->er_extcount],
4592 erp_next->er_extbuf, erp_next->er_extcount *
4593 sizeof(xfs_bmbt_rec_t));
4594 erp->er_extcount += erp_next->er_extcount;
4596 * Free page before removing extent record
4597 * so er_extoffs don't get modified in
4598 * xfs_iext_irec_remove.
4600 kmem_free(erp_next->er_extbuf, XFS_IEXT_BUFSZ);
4601 erp_next->er_extbuf = NULL;
4602 xfs_iext_irec_remove(ifp, erp_idx + 1);
4603 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4611 * Fully compact the extent records managed by the indirection array.
4614 xfs_iext_irec_compact_full(
4615 xfs_ifork_t *ifp) /* inode fork pointer */
4617 xfs_bmbt_rec_host_t *ep, *ep_next; /* extent record pointers */
4618 xfs_ext_irec_t *erp, *erp_next; /* extent irec pointers */
4619 int erp_idx = 0; /* extent irec index */
4620 int ext_avail; /* empty entries in ex list */
4621 int ext_diff; /* number of exts to add */
4622 int nlists; /* number of irec's (ex lists) */
4624 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4625 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4626 erp = ifp->if_u1.if_ext_irec;
4627 ep = &erp->er_extbuf[erp->er_extcount];
4629 ep_next = erp_next->er_extbuf;
4630 while (erp_idx < nlists - 1) {
4631 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
4632 ext_diff = MIN(ext_avail, erp_next->er_extcount);
4633 memcpy(ep, ep_next, ext_diff * sizeof(xfs_bmbt_rec_t));
4634 erp->er_extcount += ext_diff;
4635 erp_next->er_extcount -= ext_diff;
4636 /* Remove next page */
4637 if (erp_next->er_extcount == 0) {
4639 * Free page before removing extent record
4640 * so er_extoffs don't get modified in
4641 * xfs_iext_irec_remove.
4643 kmem_free(erp_next->er_extbuf,
4644 erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
4645 erp_next->er_extbuf = NULL;
4646 xfs_iext_irec_remove(ifp, erp_idx + 1);
4647 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4648 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4649 /* Update next page */
4651 /* Move rest of page up to become next new page */
4652 memmove(erp_next->er_extbuf, ep_next,
4653 erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
4654 ep_next = erp_next->er_extbuf;
4655 memset(&ep_next[erp_next->er_extcount], 0,
4656 (XFS_LINEAR_EXTS - erp_next->er_extcount) *
4657 sizeof(xfs_bmbt_rec_t));
4659 if (erp->er_extcount == XFS_LINEAR_EXTS) {
4661 if (erp_idx < nlists)
4662 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4666 ep = &erp->er_extbuf[erp->er_extcount];
4668 ep_next = erp_next->er_extbuf;
4673 * This is called to update the er_extoff field in the indirection
4674 * array when extents have been added or removed from one of the
4675 * extent lists. erp_idx contains the irec index to begin updating
4676 * at and ext_diff contains the number of extents that were added
4680 xfs_iext_irec_update_extoffs(
4681 xfs_ifork_t *ifp, /* inode fork pointer */
4682 int erp_idx, /* irec index to update */
4683 int ext_diff) /* number of new extents */
4685 int i; /* loop counter */
4686 int nlists; /* number of irec's (ex lists */
4688 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4689 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4690 for (i = erp_idx; i < nlists; i++) {
4691 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;