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_btree_trace.h"
45 #include "xfs_alloc.h"
46 #include "xfs_ialloc.h"
49 #include "xfs_error.h"
50 #include "xfs_utils.h"
51 #include "xfs_dir2_trace.h"
52 #include "xfs_quota.h"
54 #include "xfs_filestream.h"
55 #include "xfs_vnodeops.h"
57 kmem_zone_t *xfs_ifork_zone;
58 kmem_zone_t *xfs_inode_zone;
61 * Used in xfs_itruncate(). This is the maximum number of extents
62 * freed from a file in a single transaction.
64 #define XFS_ITRUNC_MAX_EXTENTS 2
66 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
67 STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
68 STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
69 STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
73 * Make sure that the extents in the given memory buffer
83 xfs_bmbt_rec_host_t rec;
86 for (i = 0; i < nrecs; i++) {
87 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
88 rec.l0 = get_unaligned(&ep->l0);
89 rec.l1 = get_unaligned(&ep->l1);
90 xfs_bmbt_get_all(&rec, &irec);
91 if (fmt == XFS_EXTFMT_NOSTATE)
92 ASSERT(irec.br_state == XFS_EXT_NORM);
96 #define xfs_validate_extents(ifp, nrecs, fmt)
100 * Check that none of the inode's in the buffer have a next
101 * unlinked field of 0.
113 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
115 for (i = 0; i < j; i++) {
116 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
117 i * mp->m_sb.sb_inodesize);
118 if (!dip->di_next_unlinked) {
119 xfs_fs_cmn_err(CE_ALERT, mp,
120 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
122 ASSERT(dip->di_next_unlinked);
129 * Find the buffer associated with the given inode map
130 * We do basic validation checks on the buffer once it has been
131 * retrieved from disk.
147 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno,
148 (int)imap->im_len, buf_flags, &bp);
150 if (error != EAGAIN) {
152 "xfs_imap_to_bp: xfs_trans_read_buf()returned "
153 "an error %d on %s. Returning error.",
154 error, mp->m_fsname);
156 ASSERT(buf_flags & XFS_BUF_TRYLOCK);
162 * Validate the magic number and version of every inode in the buffer
163 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
166 ni = BBTOB(imap->im_len) >> mp->m_sb.sb_inodelog;
167 #else /* usual case */
171 for (i = 0; i < ni; i++) {
175 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
176 (i << mp->m_sb.sb_inodelog));
177 di_ok = be16_to_cpu(dip->di_core.di_magic) == XFS_DINODE_MAGIC &&
178 XFS_DINODE_GOOD_VERSION(dip->di_core.di_version);
179 if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
180 XFS_ERRTAG_ITOBP_INOTOBP,
181 XFS_RANDOM_ITOBP_INOTOBP))) {
182 if (imap_flags & XFS_IMAP_BULKSTAT) {
183 xfs_trans_brelse(tp, bp);
184 return XFS_ERROR(EINVAL);
186 XFS_CORRUPTION_ERROR("xfs_imap_to_bp",
187 XFS_ERRLEVEL_HIGH, mp, dip);
190 "Device %s - bad inode magic/vsn "
191 "daddr %lld #%d (magic=%x)",
192 XFS_BUFTARG_NAME(mp->m_ddev_targp),
193 (unsigned long long)imap->im_blkno, i,
194 be16_to_cpu(dip->di_core.di_magic));
196 xfs_trans_brelse(tp, bp);
197 return XFS_ERROR(EFSCORRUPTED);
201 xfs_inobp_check(mp, bp);
204 * Mark the buffer as an inode buffer now that it looks good
206 XFS_BUF_SET_VTYPE(bp, B_FS_INO);
213 * This routine is called to map an inode number within a file
214 * system to the buffer containing the on-disk version of the
215 * inode. It returns a pointer to the buffer containing the
216 * on-disk inode in the bpp parameter, and in the dip parameter
217 * it returns a pointer to the on-disk inode within that buffer.
219 * If a non-zero error is returned, then the contents of bpp and
220 * dipp are undefined.
222 * Use xfs_imap() to determine the size and location of the
223 * buffer to read from disk.
239 error = xfs_imap(mp, tp, ino, &imap, XFS_IMAP_LOOKUP);
243 error = xfs_imap_to_bp(mp, tp, &imap, &bp, XFS_BUF_LOCK, 0);
247 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
249 *offset = imap.im_boffset;
255 * This routine is called to map an inode to the buffer containing
256 * the on-disk version of the inode. It returns a pointer to the
257 * buffer containing the on-disk inode in the bpp parameter, and in
258 * the dip parameter it returns a pointer to the on-disk inode within
261 * If a non-zero error is returned, then the contents of bpp and
262 * dipp are undefined.
264 * If the inode is new and has not yet been initialized, use xfs_imap()
265 * to determine the size and location of the buffer to read from disk.
266 * If the inode has already been mapped to its buffer and read in once,
267 * then use the mapping information stored in the inode rather than
268 * calling xfs_imap(). This allows us to avoid the overhead of looking
269 * at the inode btree for small block file systems (see xfs_dilocate()).
270 * We can tell whether the inode has been mapped in before by comparing
271 * its disk block address to 0. Only uninitialized inodes will have
272 * 0 for the disk block address.
289 if (ip->i_blkno == (xfs_daddr_t)0) {
291 error = xfs_imap(mp, tp, ip->i_ino, &imap,
292 XFS_IMAP_LOOKUP | imap_flags);
297 * Fill in the fields in the inode that will be used to
298 * map the inode to its buffer from now on.
300 ip->i_blkno = imap.im_blkno;
301 ip->i_len = imap.im_len;
302 ip->i_boffset = imap.im_boffset;
305 * We've already mapped the inode once, so just use the
306 * mapping that we saved the first time.
308 imap.im_blkno = ip->i_blkno;
309 imap.im_len = ip->i_len;
310 imap.im_boffset = ip->i_boffset;
312 ASSERT(bno == 0 || bno == imap.im_blkno);
314 error = xfs_imap_to_bp(mp, tp, &imap, &bp, buf_flags, imap_flags);
319 ASSERT(buf_flags & XFS_BUF_TRYLOCK);
325 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
331 * Move inode type and inode format specific information from the
332 * on-disk inode to the in-core inode. For fifos, devs, and sockets
333 * this means set if_rdev to the proper value. For files, directories,
334 * and symlinks this means to bring in the in-line data or extent
335 * pointers. For a file in B-tree format, only the root is immediately
336 * brought in-core. The rest will be in-lined in if_extents when it
337 * is first referenced (see xfs_iread_extents()).
344 xfs_attr_shortform_t *atp;
348 ip->i_df.if_ext_max =
349 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
352 if (unlikely(be32_to_cpu(dip->di_core.di_nextents) +
353 be16_to_cpu(dip->di_core.di_anextents) >
354 be64_to_cpu(dip->di_core.di_nblocks))) {
355 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
356 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
357 (unsigned long long)ip->i_ino,
358 (int)(be32_to_cpu(dip->di_core.di_nextents) +
359 be16_to_cpu(dip->di_core.di_anextents)),
361 be64_to_cpu(dip->di_core.di_nblocks));
362 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
364 return XFS_ERROR(EFSCORRUPTED);
367 if (unlikely(dip->di_core.di_forkoff > ip->i_mount->m_sb.sb_inodesize)) {
368 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
369 "corrupt dinode %Lu, forkoff = 0x%x.",
370 (unsigned long long)ip->i_ino,
371 dip->di_core.di_forkoff);
372 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
374 return XFS_ERROR(EFSCORRUPTED);
377 switch (ip->i_d.di_mode & S_IFMT) {
382 if (unlikely(dip->di_core.di_format != XFS_DINODE_FMT_DEV)) {
383 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
385 return XFS_ERROR(EFSCORRUPTED);
389 ip->i_df.if_u2.if_rdev = be32_to_cpu(dip->di_u.di_dev);
395 switch (dip->di_core.di_format) {
396 case XFS_DINODE_FMT_LOCAL:
398 * no local regular files yet
400 if (unlikely((be16_to_cpu(dip->di_core.di_mode) & S_IFMT) == S_IFREG)) {
401 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
403 "(local format for regular file).",
404 (unsigned long long) ip->i_ino);
405 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
408 return XFS_ERROR(EFSCORRUPTED);
411 di_size = be64_to_cpu(dip->di_core.di_size);
412 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
413 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
415 "(bad size %Ld for local inode).",
416 (unsigned long long) ip->i_ino,
417 (long long) di_size);
418 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
421 return XFS_ERROR(EFSCORRUPTED);
425 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
427 case XFS_DINODE_FMT_EXTENTS:
428 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
430 case XFS_DINODE_FMT_BTREE:
431 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
434 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
436 return XFS_ERROR(EFSCORRUPTED);
441 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
442 return XFS_ERROR(EFSCORRUPTED);
447 if (!XFS_DFORK_Q(dip))
449 ASSERT(ip->i_afp == NULL);
450 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
451 ip->i_afp->if_ext_max =
452 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
453 switch (dip->di_core.di_aformat) {
454 case XFS_DINODE_FMT_LOCAL:
455 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
456 size = be16_to_cpu(atp->hdr.totsize);
457 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
459 case XFS_DINODE_FMT_EXTENTS:
460 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
462 case XFS_DINODE_FMT_BTREE:
463 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
466 error = XFS_ERROR(EFSCORRUPTED);
470 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
472 xfs_idestroy_fork(ip, XFS_DATA_FORK);
478 * The file is in-lined in the on-disk inode.
479 * If it fits into if_inline_data, then copy
480 * it there, otherwise allocate a buffer for it
481 * and copy the data there. Either way, set
482 * if_data to point at the data.
483 * If we allocate a buffer for the data, make
484 * sure that its size is a multiple of 4 and
485 * record the real size in i_real_bytes.
498 * If the size is unreasonable, then something
499 * is wrong and we just bail out rather than crash in
500 * kmem_alloc() or memcpy() below.
502 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
503 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
505 "(bad size %d for local fork, size = %d).",
506 (unsigned long long) ip->i_ino, size,
507 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
508 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
510 return XFS_ERROR(EFSCORRUPTED);
512 ifp = XFS_IFORK_PTR(ip, whichfork);
515 ifp->if_u1.if_data = NULL;
516 else if (size <= sizeof(ifp->if_u2.if_inline_data))
517 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
519 real_size = roundup(size, 4);
520 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
522 ifp->if_bytes = size;
523 ifp->if_real_bytes = real_size;
525 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
526 ifp->if_flags &= ~XFS_IFEXTENTS;
527 ifp->if_flags |= XFS_IFINLINE;
532 * The file consists of a set of extents all
533 * of which fit into the on-disk inode.
534 * If there are few enough extents to fit into
535 * the if_inline_ext, then copy them there.
536 * Otherwise allocate a buffer for them and copy
537 * them into it. Either way, set if_extents
538 * to point at the extents.
552 ifp = XFS_IFORK_PTR(ip, whichfork);
553 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
554 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
557 * If the number of extents is unreasonable, then something
558 * is wrong and we just bail out rather than crash in
559 * kmem_alloc() or memcpy() below.
561 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
562 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
563 "corrupt inode %Lu ((a)extents = %d).",
564 (unsigned long long) ip->i_ino, nex);
565 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
567 return XFS_ERROR(EFSCORRUPTED);
570 ifp->if_real_bytes = 0;
572 ifp->if_u1.if_extents = NULL;
573 else if (nex <= XFS_INLINE_EXTS)
574 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
576 xfs_iext_add(ifp, 0, nex);
578 ifp->if_bytes = size;
580 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
581 xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip));
582 for (i = 0; i < nex; i++, dp++) {
583 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
584 ep->l0 = get_unaligned_be64(&dp->l0);
585 ep->l1 = get_unaligned_be64(&dp->l1);
587 XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork);
588 if (whichfork != XFS_DATA_FORK ||
589 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
590 if (unlikely(xfs_check_nostate_extents(
592 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
595 return XFS_ERROR(EFSCORRUPTED);
598 ifp->if_flags |= XFS_IFEXTENTS;
603 * The file has too many extents to fit into
604 * the inode, so they are in B-tree format.
605 * Allocate a buffer for the root of the B-tree
606 * and copy the root into it. The i_extents
607 * field will remain NULL until all of the
608 * extents are read in (when they are needed).
616 xfs_bmdr_block_t *dfp;
622 ifp = XFS_IFORK_PTR(ip, whichfork);
623 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
624 size = XFS_BMAP_BROOT_SPACE(dfp);
625 nrecs = be16_to_cpu(dfp->bb_numrecs);
628 * blow out if -- fork has less extents than can fit in
629 * fork (fork shouldn't be a btree format), root btree
630 * block has more records than can fit into the fork,
631 * or the number of extents is greater than the number of
634 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
635 || XFS_BMDR_SPACE_CALC(nrecs) >
636 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
637 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
638 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
639 "corrupt inode %Lu (btree).",
640 (unsigned long long) ip->i_ino);
641 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
643 return XFS_ERROR(EFSCORRUPTED);
646 ifp->if_broot_bytes = size;
647 ifp->if_broot = kmem_alloc(size, KM_SLEEP);
648 ASSERT(ifp->if_broot != NULL);
650 * Copy and convert from the on-disk structure
651 * to the in-memory structure.
653 xfs_bmdr_to_bmbt(ip->i_mount, dfp,
654 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
655 ifp->if_broot, size);
656 ifp->if_flags &= ~XFS_IFEXTENTS;
657 ifp->if_flags |= XFS_IFBROOT;
663 xfs_dinode_from_disk(
665 xfs_dinode_core_t *from)
667 to->di_magic = be16_to_cpu(from->di_magic);
668 to->di_mode = be16_to_cpu(from->di_mode);
669 to->di_version = from ->di_version;
670 to->di_format = from->di_format;
671 to->di_onlink = be16_to_cpu(from->di_onlink);
672 to->di_uid = be32_to_cpu(from->di_uid);
673 to->di_gid = be32_to_cpu(from->di_gid);
674 to->di_nlink = be32_to_cpu(from->di_nlink);
675 to->di_projid = be16_to_cpu(from->di_projid);
676 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
677 to->di_flushiter = be16_to_cpu(from->di_flushiter);
678 to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec);
679 to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec);
680 to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec);
681 to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec);
682 to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec);
683 to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec);
684 to->di_size = be64_to_cpu(from->di_size);
685 to->di_nblocks = be64_to_cpu(from->di_nblocks);
686 to->di_extsize = be32_to_cpu(from->di_extsize);
687 to->di_nextents = be32_to_cpu(from->di_nextents);
688 to->di_anextents = be16_to_cpu(from->di_anextents);
689 to->di_forkoff = from->di_forkoff;
690 to->di_aformat = from->di_aformat;
691 to->di_dmevmask = be32_to_cpu(from->di_dmevmask);
692 to->di_dmstate = be16_to_cpu(from->di_dmstate);
693 to->di_flags = be16_to_cpu(from->di_flags);
694 to->di_gen = be32_to_cpu(from->di_gen);
699 xfs_dinode_core_t *to,
700 xfs_icdinode_t *from)
702 to->di_magic = cpu_to_be16(from->di_magic);
703 to->di_mode = cpu_to_be16(from->di_mode);
704 to->di_version = from ->di_version;
705 to->di_format = from->di_format;
706 to->di_onlink = cpu_to_be16(from->di_onlink);
707 to->di_uid = cpu_to_be32(from->di_uid);
708 to->di_gid = cpu_to_be32(from->di_gid);
709 to->di_nlink = cpu_to_be32(from->di_nlink);
710 to->di_projid = cpu_to_be16(from->di_projid);
711 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
712 to->di_flushiter = cpu_to_be16(from->di_flushiter);
713 to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec);
714 to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec);
715 to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec);
716 to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec);
717 to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec);
718 to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec);
719 to->di_size = cpu_to_be64(from->di_size);
720 to->di_nblocks = cpu_to_be64(from->di_nblocks);
721 to->di_extsize = cpu_to_be32(from->di_extsize);
722 to->di_nextents = cpu_to_be32(from->di_nextents);
723 to->di_anextents = cpu_to_be16(from->di_anextents);
724 to->di_forkoff = from->di_forkoff;
725 to->di_aformat = from->di_aformat;
726 to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
727 to->di_dmstate = cpu_to_be16(from->di_dmstate);
728 to->di_flags = cpu_to_be16(from->di_flags);
729 to->di_gen = cpu_to_be32(from->di_gen);
738 if (di_flags & XFS_DIFLAG_ANY) {
739 if (di_flags & XFS_DIFLAG_REALTIME)
740 flags |= XFS_XFLAG_REALTIME;
741 if (di_flags & XFS_DIFLAG_PREALLOC)
742 flags |= XFS_XFLAG_PREALLOC;
743 if (di_flags & XFS_DIFLAG_IMMUTABLE)
744 flags |= XFS_XFLAG_IMMUTABLE;
745 if (di_flags & XFS_DIFLAG_APPEND)
746 flags |= XFS_XFLAG_APPEND;
747 if (di_flags & XFS_DIFLAG_SYNC)
748 flags |= XFS_XFLAG_SYNC;
749 if (di_flags & XFS_DIFLAG_NOATIME)
750 flags |= XFS_XFLAG_NOATIME;
751 if (di_flags & XFS_DIFLAG_NODUMP)
752 flags |= XFS_XFLAG_NODUMP;
753 if (di_flags & XFS_DIFLAG_RTINHERIT)
754 flags |= XFS_XFLAG_RTINHERIT;
755 if (di_flags & XFS_DIFLAG_PROJINHERIT)
756 flags |= XFS_XFLAG_PROJINHERIT;
757 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
758 flags |= XFS_XFLAG_NOSYMLINKS;
759 if (di_flags & XFS_DIFLAG_EXTSIZE)
760 flags |= XFS_XFLAG_EXTSIZE;
761 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
762 flags |= XFS_XFLAG_EXTSZINHERIT;
763 if (di_flags & XFS_DIFLAG_NODEFRAG)
764 flags |= XFS_XFLAG_NODEFRAG;
765 if (di_flags & XFS_DIFLAG_FILESTREAM)
766 flags |= XFS_XFLAG_FILESTREAM;
776 xfs_icdinode_t *dic = &ip->i_d;
778 return _xfs_dic2xflags(dic->di_flags) |
779 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
786 xfs_dinode_core_t *dic = &dip->di_core;
788 return _xfs_dic2xflags(be16_to_cpu(dic->di_flags)) |
789 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
793 * Allocate and initialise an xfs_inode.
797 struct xfs_mount *mp,
800 struct xfs_inode *ip;
803 * if this didn't occur in transactions, we could use
804 * KM_MAYFAIL and return NULL here on ENOMEM. Set the
805 * code up to do this anyway.
807 ip = kmem_zone_alloc(xfs_inode_zone, KM_SLEEP);
811 ASSERT(atomic_read(&ip->i_iocount) == 0);
812 ASSERT(atomic_read(&ip->i_pincount) == 0);
813 ASSERT(!spin_is_locked(&ip->i_flags_lock));
814 ASSERT(completion_done(&ip->i_flush));
817 * initialise the VFS inode here to get failures
818 * out of the way early.
820 if (!inode_init_always(mp->m_super, VFS_I(ip))) {
821 kmem_zone_free(xfs_inode_zone, ip);
825 /* initialise the xfs inode */
832 memset(&ip->i_df, 0, sizeof(xfs_ifork_t));
834 ip->i_update_core = 0;
835 ip->i_update_size = 0;
836 ip->i_delayed_blks = 0;
837 memset(&ip->i_d, 0, sizeof(xfs_icdinode_t));
842 * Initialize inode's trace buffers.
844 #ifdef XFS_INODE_TRACE
845 ip->i_trace = ktrace_alloc(INODE_TRACE_SIZE, KM_NOFS);
847 #ifdef XFS_BMAP_TRACE
848 ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_NOFS);
850 #ifdef XFS_BTREE_TRACE
851 ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_NOFS);
854 ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_NOFS);
856 #ifdef XFS_ILOCK_TRACE
857 ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_NOFS);
859 #ifdef XFS_DIR2_TRACE
860 ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_NOFS);
867 * Given a mount structure and an inode number, return a pointer
868 * to a newly allocated in-core inode corresponding to the given
871 * Initialize the inode's attributes and extent pointers if it
872 * already has them (it will not if the inode has no links).
888 ip = xfs_inode_alloc(mp, ino);
893 * Get pointer's to the on-disk inode and the buffer containing it.
894 * If the inode number refers to a block outside the file system
895 * then xfs_itobp() will return NULL. In this case we should
896 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
897 * know that this is a new incore inode.
899 error = xfs_itobp(mp, tp, ip, &dip, &bp, bno, imap_flags, XFS_BUF_LOCK);
906 * If we got something that isn't an inode it means someone
907 * (nfs or dmi) has a stale handle.
909 if (be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC) {
911 xfs_trans_brelse(tp, bp);
913 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
914 "dip->di_core.di_magic (0x%x) != "
915 "XFS_DINODE_MAGIC (0x%x)",
916 be16_to_cpu(dip->di_core.di_magic),
919 return XFS_ERROR(EINVAL);
923 * If the on-disk inode is already linked to a directory
924 * entry, copy all of the inode into the in-core inode.
925 * xfs_iformat() handles copying in the inode format
926 * specific information.
927 * Otherwise, just get the truly permanent information.
929 if (dip->di_core.di_mode) {
930 xfs_dinode_from_disk(&ip->i_d, &dip->di_core);
931 error = xfs_iformat(ip, dip);
934 xfs_trans_brelse(tp, bp);
936 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
937 "xfs_iformat() returned error %d",
943 ip->i_d.di_magic = be16_to_cpu(dip->di_core.di_magic);
944 ip->i_d.di_version = dip->di_core.di_version;
945 ip->i_d.di_gen = be32_to_cpu(dip->di_core.di_gen);
946 ip->i_d.di_flushiter = be16_to_cpu(dip->di_core.di_flushiter);
948 * Make sure to pull in the mode here as well in
949 * case the inode is released without being used.
950 * This ensures that xfs_inactive() will see that
951 * the inode is already free and not try to mess
952 * with the uninitialized part of it.
956 * Initialize the per-fork minima and maxima for a new
957 * inode here. xfs_iformat will do it for old inodes.
959 ip->i_df.if_ext_max =
960 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
964 * The inode format changed when we moved the link count and
965 * made it 32 bits long. If this is an old format inode,
966 * convert it in memory to look like a new one. If it gets
967 * flushed to disk we will convert back before flushing or
968 * logging it. We zero out the new projid field and the old link
969 * count field. We'll handle clearing the pad field (the remains
970 * of the old uuid field) when we actually convert the inode to
971 * the new format. We don't change the version number so that we
972 * can distinguish this from a real new format inode.
974 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
975 ip->i_d.di_nlink = ip->i_d.di_onlink;
976 ip->i_d.di_onlink = 0;
977 ip->i_d.di_projid = 0;
980 ip->i_delayed_blks = 0;
981 ip->i_size = ip->i_d.di_size;
984 * Mark the buffer containing the inode as something to keep
985 * around for a while. This helps to keep recently accessed
986 * meta-data in-core longer.
988 XFS_BUF_SET_REF(bp, XFS_INO_REF);
991 * Use xfs_trans_brelse() to release the buffer containing the
992 * on-disk inode, because it was acquired with xfs_trans_read_buf()
993 * in xfs_itobp() above. If tp is NULL, this is just a normal
994 * brelse(). If we're within a transaction, then xfs_trans_brelse()
995 * will only release the buffer if it is not dirty within the
996 * transaction. It will be OK to release the buffer in this case,
997 * because inodes on disk are never destroyed and we will be
998 * locking the new in-core inode before putting it in the hash
999 * table where other processes can find it. Thus we don't have
1000 * to worry about the inode being changed just because we released
1003 xfs_trans_brelse(tp, bp);
1009 * Read in extents from a btree-format inode.
1010 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1020 xfs_extnum_t nextents;
1023 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
1024 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
1026 return XFS_ERROR(EFSCORRUPTED);
1028 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
1029 size = nextents * sizeof(xfs_bmbt_rec_t);
1030 ifp = XFS_IFORK_PTR(ip, whichfork);
1033 * We know that the size is valid (it's checked in iformat_btree)
1035 ifp->if_lastex = NULLEXTNUM;
1036 ifp->if_bytes = ifp->if_real_bytes = 0;
1037 ifp->if_flags |= XFS_IFEXTENTS;
1038 xfs_iext_add(ifp, 0, nextents);
1039 error = xfs_bmap_read_extents(tp, ip, whichfork);
1041 xfs_iext_destroy(ifp);
1042 ifp->if_flags &= ~XFS_IFEXTENTS;
1045 xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip));
1050 * Allocate an inode on disk and return a copy of its in-core version.
1051 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1052 * appropriately within the inode. The uid and gid for the inode are
1053 * set according to the contents of the given cred structure.
1055 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1056 * has a free inode available, call xfs_iget()
1057 * to obtain the in-core version of the allocated inode. Finally,
1058 * fill in the inode and log its initial contents. In this case,
1059 * ialloc_context would be set to NULL and call_again set to false.
1061 * If xfs_dialloc() does not have an available inode,
1062 * it will replenish its supply by doing an allocation. Since we can
1063 * only do one allocation within a transaction without deadlocks, we
1064 * must commit the current transaction before returning the inode itself.
1065 * In this case, therefore, we will set call_again to true and return.
1066 * The caller should then commit the current transaction, start a new
1067 * transaction, and call xfs_ialloc() again to actually get the inode.
1069 * To ensure that some other process does not grab the inode that
1070 * was allocated during the first call to xfs_ialloc(), this routine
1071 * also returns the [locked] bp pointing to the head of the freelist
1072 * as ialloc_context. The caller should hold this buffer across
1073 * the commit and pass it back into this routine on the second call.
1075 * If we are allocating quota inodes, we do not have a parent inode
1076 * to attach to or associate with (i.e. pip == NULL) because they
1077 * are not linked into the directory structure - they are attached
1078 * directly to the superblock - and so have no parent.
1090 xfs_buf_t **ialloc_context,
1091 boolean_t *call_again,
1099 int filestreams = 0;
1102 * Call the space management code to pick
1103 * the on-disk inode to be allocated.
1105 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
1106 ialloc_context, call_again, &ino);
1109 if (*call_again || ino == NULLFSINO) {
1113 ASSERT(*ialloc_context == NULL);
1116 * Get the in-core inode with the lock held exclusively.
1117 * This is because we're setting fields here we need
1118 * to prevent others from looking at until we're done.
1120 error = xfs_trans_iget(tp->t_mountp, tp, ino,
1121 XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
1126 ip->i_d.di_mode = (__uint16_t)mode;
1127 ip->i_d.di_onlink = 0;
1128 ip->i_d.di_nlink = nlink;
1129 ASSERT(ip->i_d.di_nlink == nlink);
1130 ip->i_d.di_uid = current_fsuid();
1131 ip->i_d.di_gid = current_fsgid();
1132 ip->i_d.di_projid = prid;
1133 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1136 * If the superblock version is up to where we support new format
1137 * inodes and this is currently an old format inode, then change
1138 * the inode version number now. This way we only do the conversion
1139 * here rather than here and in the flush/logging code.
1141 if (xfs_sb_version_hasnlink(&tp->t_mountp->m_sb) &&
1142 ip->i_d.di_version == XFS_DINODE_VERSION_1) {
1143 ip->i_d.di_version = XFS_DINODE_VERSION_2;
1145 * We've already zeroed the old link count, the projid field,
1146 * and the pad field.
1151 * Project ids won't be stored on disk if we are using a version 1 inode.
1153 if ((prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1))
1154 xfs_bump_ino_vers2(tp, ip);
1156 if (pip && XFS_INHERIT_GID(pip)) {
1157 ip->i_d.di_gid = pip->i_d.di_gid;
1158 if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1159 ip->i_d.di_mode |= S_ISGID;
1164 * If the group ID of the new file does not match the effective group
1165 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1166 * (and only if the irix_sgid_inherit compatibility variable is set).
1168 if ((irix_sgid_inherit) &&
1169 (ip->i_d.di_mode & S_ISGID) &&
1170 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1171 ip->i_d.di_mode &= ~S_ISGID;
1174 ip->i_d.di_size = 0;
1176 ip->i_d.di_nextents = 0;
1177 ASSERT(ip->i_d.di_nblocks == 0);
1180 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
1181 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
1182 ip->i_d.di_atime = ip->i_d.di_mtime;
1183 ip->i_d.di_ctime = ip->i_d.di_mtime;
1186 * di_gen will have been taken care of in xfs_iread.
1188 ip->i_d.di_extsize = 0;
1189 ip->i_d.di_dmevmask = 0;
1190 ip->i_d.di_dmstate = 0;
1191 ip->i_d.di_flags = 0;
1192 flags = XFS_ILOG_CORE;
1193 switch (mode & S_IFMT) {
1198 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1199 ip->i_df.if_u2.if_rdev = rdev;
1200 ip->i_df.if_flags = 0;
1201 flags |= XFS_ILOG_DEV;
1205 * we can't set up filestreams until after the VFS inode
1206 * is set up properly.
1208 if (pip && xfs_inode_is_filestream(pip))
1212 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
1215 if ((mode & S_IFMT) == S_IFDIR) {
1216 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1217 di_flags |= XFS_DIFLAG_RTINHERIT;
1218 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1219 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1220 ip->i_d.di_extsize = pip->i_d.di_extsize;
1222 } else if ((mode & S_IFMT) == S_IFREG) {
1223 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1224 di_flags |= XFS_DIFLAG_REALTIME;
1225 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1226 di_flags |= XFS_DIFLAG_EXTSIZE;
1227 ip->i_d.di_extsize = pip->i_d.di_extsize;
1230 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1231 xfs_inherit_noatime)
1232 di_flags |= XFS_DIFLAG_NOATIME;
1233 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1235 di_flags |= XFS_DIFLAG_NODUMP;
1236 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1238 di_flags |= XFS_DIFLAG_SYNC;
1239 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1240 xfs_inherit_nosymlinks)
1241 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1242 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1243 di_flags |= XFS_DIFLAG_PROJINHERIT;
1244 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1245 xfs_inherit_nodefrag)
1246 di_flags |= XFS_DIFLAG_NODEFRAG;
1247 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
1248 di_flags |= XFS_DIFLAG_FILESTREAM;
1249 ip->i_d.di_flags |= di_flags;
1253 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1254 ip->i_df.if_flags = XFS_IFEXTENTS;
1255 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1256 ip->i_df.if_u1.if_extents = NULL;
1262 * Attribute fork settings for new inode.
1264 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1265 ip->i_d.di_anextents = 0;
1268 * Log the new values stuffed into the inode.
1270 xfs_trans_log_inode(tp, ip, flags);
1272 /* now that we have an i_mode we can setup inode ops and unlock */
1273 xfs_setup_inode(ip);
1275 /* now we have set up the vfs inode we can associate the filestream */
1277 error = xfs_filestream_associate(pip, ip);
1281 xfs_iflags_set(ip, XFS_IFILESTREAM);
1289 * Check to make sure that there are no blocks allocated to the
1290 * file beyond the size of the file. We don't check this for
1291 * files with fixed size extents or real time extents, but we
1292 * at least do it for regular files.
1301 xfs_fileoff_t map_first;
1303 xfs_bmbt_irec_t imaps[2];
1305 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1308 if (XFS_IS_REALTIME_INODE(ip))
1311 if (ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE)
1315 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1317 * The filesystem could be shutting down, so bmapi may return
1320 if (xfs_bmapi(NULL, ip, map_first,
1322 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1324 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1327 ASSERT(nimaps == 1);
1328 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1333 * Calculate the last possible buffered byte in a file. This must
1334 * include data that was buffered beyond the EOF by the write code.
1335 * This also needs to deal with overflowing the xfs_fsize_t type
1336 * which can happen for sizes near the limit.
1338 * We also need to take into account any blocks beyond the EOF. It
1339 * may be the case that they were buffered by a write which failed.
1340 * In that case the pages will still be in memory, but the inode size
1341 * will never have been updated.
1348 xfs_fsize_t last_byte;
1349 xfs_fileoff_t last_block;
1350 xfs_fileoff_t size_last_block;
1353 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED));
1357 * Only check for blocks beyond the EOF if the extents have
1358 * been read in. This eliminates the need for the inode lock,
1359 * and it also saves us from looking when it really isn't
1362 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1363 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1371 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_size);
1372 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1374 last_byte = XFS_FSB_TO_B(mp, last_block);
1375 if (last_byte < 0) {
1376 return XFS_MAXIOFFSET(mp);
1378 last_byte += (1 << mp->m_writeio_log);
1379 if (last_byte < 0) {
1380 return XFS_MAXIOFFSET(mp);
1385 #if defined(XFS_RW_TRACE)
1391 xfs_fsize_t new_size,
1392 xfs_off_t toss_start,
1393 xfs_off_t toss_finish)
1395 if (ip->i_rwtrace == NULL) {
1399 ktrace_enter(ip->i_rwtrace,
1402 (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
1403 (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
1404 (void*)((long)flag),
1405 (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
1406 (void*)(unsigned long)(new_size & 0xffffffff),
1407 (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
1408 (void*)(unsigned long)(toss_start & 0xffffffff),
1409 (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
1410 (void*)(unsigned long)(toss_finish & 0xffffffff),
1411 (void*)(unsigned long)current_cpu(),
1412 (void*)(unsigned long)current_pid(),
1418 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1422 * Start the truncation of the file to new_size. The new size
1423 * must be smaller than the current size. This routine will
1424 * clear the buffer and page caches of file data in the removed
1425 * range, and xfs_itruncate_finish() will remove the underlying
1428 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1429 * must NOT have the inode lock held at all. This is because we're
1430 * calling into the buffer/page cache code and we can't hold the
1431 * inode lock when we do so.
1433 * We need to wait for any direct I/Os in flight to complete before we
1434 * proceed with the truncate. This is needed to prevent the extents
1435 * being read or written by the direct I/Os from being removed while the
1436 * I/O is in flight as there is no other method of synchronising
1437 * direct I/O with the truncate operation. Also, because we hold
1438 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1439 * started until the truncate completes and drops the lock. Essentially,
1440 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1441 * between direct I/Os and the truncate operation.
1443 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1444 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1445 * in the case that the caller is locking things out of order and
1446 * may not be able to call xfs_itruncate_finish() with the inode lock
1447 * held without dropping the I/O lock. If the caller must drop the
1448 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1449 * must be called again with all the same restrictions as the initial
1453 xfs_itruncate_start(
1456 xfs_fsize_t new_size)
1458 xfs_fsize_t last_byte;
1459 xfs_off_t toss_start;
1463 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1464 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1465 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1466 (flags == XFS_ITRUNC_MAYBE));
1470 /* wait for the completion of any pending DIOs */
1471 if (new_size == 0 || new_size < ip->i_size)
1475 * Call toss_pages or flushinval_pages to get rid of pages
1476 * overlapping the region being removed. We have to use
1477 * the less efficient flushinval_pages in the case that the
1478 * caller may not be able to finish the truncate without
1479 * dropping the inode's I/O lock. Make sure
1480 * to catch any pages brought in by buffers overlapping
1481 * the EOF by searching out beyond the isize by our
1482 * block size. We round new_size up to a block boundary
1483 * so that we don't toss things on the same block as
1484 * new_size but before it.
1486 * Before calling toss_page or flushinval_pages, make sure to
1487 * call remapf() over the same region if the file is mapped.
1488 * This frees up mapped file references to the pages in the
1489 * given range and for the flushinval_pages case it ensures
1490 * that we get the latest mapped changes flushed out.
1492 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1493 toss_start = XFS_FSB_TO_B(mp, toss_start);
1494 if (toss_start < 0) {
1496 * The place to start tossing is beyond our maximum
1497 * file size, so there is no way that the data extended
1502 last_byte = xfs_file_last_byte(ip);
1503 xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
1505 if (last_byte > toss_start) {
1506 if (flags & XFS_ITRUNC_DEFINITE) {
1507 xfs_tosspages(ip, toss_start,
1508 -1, FI_REMAPF_LOCKED);
1510 error = xfs_flushinval_pages(ip, toss_start,
1511 -1, FI_REMAPF_LOCKED);
1516 if (new_size == 0) {
1517 ASSERT(VN_CACHED(VFS_I(ip)) == 0);
1524 * Shrink the file to the given new_size. The new size must be smaller than
1525 * the current size. This will free up the underlying blocks in the removed
1526 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
1528 * The transaction passed to this routine must have made a permanent log
1529 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1530 * given transaction and start new ones, so make sure everything involved in
1531 * the transaction is tidy before calling here. Some transaction will be
1532 * returned to the caller to be committed. The incoming transaction must
1533 * already include the inode, and both inode locks must be held exclusively.
1534 * The inode must also be "held" within the transaction. On return the inode
1535 * will be "held" within the returned transaction. This routine does NOT
1536 * require any disk space to be reserved for it within the transaction.
1538 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1539 * indicates the fork which is to be truncated. For the attribute fork we only
1540 * support truncation to size 0.
1542 * We use the sync parameter to indicate whether or not the first transaction
1543 * we perform might have to be synchronous. For the attr fork, it needs to be
1544 * so if the unlink of the inode is not yet known to be permanent in the log.
1545 * This keeps us from freeing and reusing the blocks of the attribute fork
1546 * before the unlink of the inode becomes permanent.
1548 * For the data fork, we normally have to run synchronously if we're being
1549 * called out of the inactive path or we're being called out of the create path
1550 * where we're truncating an existing file. Either way, the truncate needs to
1551 * be sync so blocks don't reappear in the file with altered data in case of a
1552 * crash. wsync filesystems can run the first case async because anything that
1553 * shrinks the inode has to run sync so by the time we're called here from
1554 * inactive, the inode size is permanently set to 0.
1556 * Calls from the truncate path always need to be sync unless we're in a wsync
1557 * filesystem and the file has already been unlinked.
1559 * The caller is responsible for correctly setting the sync parameter. It gets
1560 * too hard for us to guess here which path we're being called out of just
1561 * based on inode state.
1563 * If we get an error, we must return with the inode locked and linked into the
1564 * current transaction. This keeps things simple for the higher level code,
1565 * because it always knows that the inode is locked and held in the transaction
1566 * that returns to it whether errors occur or not. We don't mark the inode
1567 * dirty on error so that transactions can be easily aborted if possible.
1570 xfs_itruncate_finish(
1573 xfs_fsize_t new_size,
1577 xfs_fsblock_t first_block;
1578 xfs_fileoff_t first_unmap_block;
1579 xfs_fileoff_t last_block;
1580 xfs_filblks_t unmap_len=0;
1585 xfs_bmap_free_t free_list;
1588 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
1589 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1590 ASSERT(*tp != NULL);
1591 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1592 ASSERT(ip->i_transp == *tp);
1593 ASSERT(ip->i_itemp != NULL);
1594 ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
1598 mp = (ntp)->t_mountp;
1599 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1602 * We only support truncating the entire attribute fork.
1604 if (fork == XFS_ATTR_FORK) {
1607 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1608 xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
1610 * The first thing we do is set the size to new_size permanently
1611 * on disk. This way we don't have to worry about anyone ever
1612 * being able to look at the data being freed even in the face
1613 * of a crash. What we're getting around here is the case where
1614 * we free a block, it is allocated to another file, it is written
1615 * to, and then we crash. If the new data gets written to the
1616 * file but the log buffers containing the free and reallocation
1617 * don't, then we'd end up with garbage in the blocks being freed.
1618 * As long as we make the new_size permanent before actually
1619 * freeing any blocks it doesn't matter if they get writtten to.
1621 * The callers must signal into us whether or not the size
1622 * setting here must be synchronous. There are a few cases
1623 * where it doesn't have to be synchronous. Those cases
1624 * occur if the file is unlinked and we know the unlink is
1625 * permanent or if the blocks being truncated are guaranteed
1626 * to be beyond the inode eof (regardless of the link count)
1627 * and the eof value is permanent. Both of these cases occur
1628 * only on wsync-mounted filesystems. In those cases, we're
1629 * guaranteed that no user will ever see the data in the blocks
1630 * that are being truncated so the truncate can run async.
1631 * In the free beyond eof case, the file may wind up with
1632 * more blocks allocated to it than it needs if we crash
1633 * and that won't get fixed until the next time the file
1634 * is re-opened and closed but that's ok as that shouldn't
1635 * be too many blocks.
1637 * However, we can't just make all wsync xactions run async
1638 * because there's one call out of the create path that needs
1639 * to run sync where it's truncating an existing file to size
1640 * 0 whose size is > 0.
1642 * It's probably possible to come up with a test in this
1643 * routine that would correctly distinguish all the above
1644 * cases from the values of the function parameters and the
1645 * inode state but for sanity's sake, I've decided to let the
1646 * layers above just tell us. It's simpler to correctly figure
1647 * out in the layer above exactly under what conditions we
1648 * can run async and I think it's easier for others read and
1649 * follow the logic in case something has to be changed.
1650 * cscope is your friend -- rcc.
1652 * The attribute fork is much simpler.
1654 * For the attribute fork we allow the caller to tell us whether
1655 * the unlink of the inode that led to this call is yet permanent
1656 * in the on disk log. If it is not and we will be freeing extents
1657 * in this inode then we make the first transaction synchronous
1658 * to make sure that the unlink is permanent by the time we free
1661 if (fork == XFS_DATA_FORK) {
1662 if (ip->i_d.di_nextents > 0) {
1664 * If we are not changing the file size then do
1665 * not update the on-disk file size - we may be
1666 * called from xfs_inactive_free_eofblocks(). If we
1667 * update the on-disk file size and then the system
1668 * crashes before the contents of the file are
1669 * flushed to disk then the files may be full of
1670 * holes (ie NULL files bug).
1672 if (ip->i_size != new_size) {
1673 ip->i_d.di_size = new_size;
1674 ip->i_size = new_size;
1675 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1679 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1680 if (ip->i_d.di_anextents > 0)
1681 xfs_trans_set_sync(ntp);
1683 ASSERT(fork == XFS_DATA_FORK ||
1684 (fork == XFS_ATTR_FORK &&
1685 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1686 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1689 * Since it is possible for space to become allocated beyond
1690 * the end of the file (in a crash where the space is allocated
1691 * but the inode size is not yet updated), simply remove any
1692 * blocks which show up between the new EOF and the maximum
1693 * possible file size. If the first block to be removed is
1694 * beyond the maximum file size (ie it is the same as last_block),
1695 * then there is nothing to do.
1697 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1698 ASSERT(first_unmap_block <= last_block);
1700 if (last_block == first_unmap_block) {
1703 unmap_len = last_block - first_unmap_block + 1;
1707 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1708 * will tell us whether it freed the entire range or
1709 * not. If this is a synchronous mount (wsync),
1710 * then we can tell bunmapi to keep all the
1711 * transactions asynchronous since the unlink
1712 * transaction that made this inode inactive has
1713 * already hit the disk. There's no danger of
1714 * the freed blocks being reused, there being a
1715 * crash, and the reused blocks suddenly reappearing
1716 * in this file with garbage in them once recovery
1719 XFS_BMAP_INIT(&free_list, &first_block);
1720 error = xfs_bunmapi(ntp, ip,
1721 first_unmap_block, unmap_len,
1722 XFS_BMAPI_AFLAG(fork) |
1723 (sync ? 0 : XFS_BMAPI_ASYNC),
1724 XFS_ITRUNC_MAX_EXTENTS,
1725 &first_block, &free_list,
1729 * If the bunmapi call encounters an error,
1730 * return to the caller where the transaction
1731 * can be properly aborted. We just need to
1732 * make sure we're not holding any resources
1733 * that we were not when we came in.
1735 xfs_bmap_cancel(&free_list);
1740 * Duplicate the transaction that has the permanent
1741 * reservation and commit the old transaction.
1743 error = xfs_bmap_finish(tp, &free_list, &committed);
1746 /* link the inode into the next xact in the chain */
1747 xfs_trans_ijoin(ntp, ip,
1748 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1749 xfs_trans_ihold(ntp, ip);
1754 * If the bmap finish call encounters an error, return
1755 * to the caller where the transaction can be properly
1756 * aborted. We just need to make sure we're not
1757 * holding any resources that we were not when we came
1760 * Aborting from this point might lose some blocks in
1761 * the file system, but oh well.
1763 xfs_bmap_cancel(&free_list);
1769 * Mark the inode dirty so it will be logged and
1770 * moved forward in the log as part of every commit.
1772 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1775 ntp = xfs_trans_dup(ntp);
1776 error = xfs_trans_commit(*tp, 0);
1779 /* link the inode into the next transaction in the chain */
1780 xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1781 xfs_trans_ihold(ntp, ip);
1784 error = xfs_trans_reserve(ntp, 0,
1785 XFS_ITRUNCATE_LOG_RES(mp), 0,
1786 XFS_TRANS_PERM_LOG_RES,
1787 XFS_ITRUNCATE_LOG_COUNT);
1792 * Only update the size in the case of the data fork, but
1793 * always re-log the inode so that our permanent transaction
1794 * can keep on rolling it forward in the log.
1796 if (fork == XFS_DATA_FORK) {
1797 xfs_isize_check(mp, ip, new_size);
1799 * If we are not changing the file size then do
1800 * not update the on-disk file size - we may be
1801 * called from xfs_inactive_free_eofblocks(). If we
1802 * update the on-disk file size and then the system
1803 * crashes before the contents of the file are
1804 * flushed to disk then the files may be full of
1805 * holes (ie NULL files bug).
1807 if (ip->i_size != new_size) {
1808 ip->i_d.di_size = new_size;
1809 ip->i_size = new_size;
1812 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1813 ASSERT((new_size != 0) ||
1814 (fork == XFS_ATTR_FORK) ||
1815 (ip->i_delayed_blks == 0));
1816 ASSERT((new_size != 0) ||
1817 (fork == XFS_ATTR_FORK) ||
1818 (ip->i_d.di_nextents == 0));
1819 xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
1824 * This is called when the inode's link count goes to 0.
1825 * We place the on-disk inode on a list in the AGI. It
1826 * will be pulled from this list when the inode is freed.
1838 xfs_agnumber_t agno;
1839 xfs_daddr_t agdaddr;
1846 ASSERT(ip->i_d.di_nlink == 0);
1847 ASSERT(ip->i_d.di_mode != 0);
1848 ASSERT(ip->i_transp == tp);
1852 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1853 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1856 * Get the agi buffer first. It ensures lock ordering
1859 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1860 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1865 * Validate the magic number of the agi block.
1867 agi = XFS_BUF_TO_AGI(agibp);
1869 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1870 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
1871 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK,
1872 XFS_RANDOM_IUNLINK))) {
1873 XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi);
1874 xfs_trans_brelse(tp, agibp);
1875 return XFS_ERROR(EFSCORRUPTED);
1878 * Get the index into the agi hash table for the
1879 * list this inode will go on.
1881 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1883 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1884 ASSERT(agi->agi_unlinked[bucket_index]);
1885 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1887 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
1889 * There is already another inode in the bucket we need
1890 * to add ourselves to. Add us at the front of the list.
1891 * Here we put the head pointer into our next pointer,
1892 * and then we fall through to point the head at us.
1894 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
1898 ASSERT(be32_to_cpu(dip->di_next_unlinked) == NULLAGINO);
1899 /* both on-disk, don't endian flip twice */
1900 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1901 offset = ip->i_boffset +
1902 offsetof(xfs_dinode_t, di_next_unlinked);
1903 xfs_trans_inode_buf(tp, ibp);
1904 xfs_trans_log_buf(tp, ibp, offset,
1905 (offset + sizeof(xfs_agino_t) - 1));
1906 xfs_inobp_check(mp, ibp);
1910 * Point the bucket head pointer at the inode being inserted.
1913 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1914 offset = offsetof(xfs_agi_t, agi_unlinked) +
1915 (sizeof(xfs_agino_t) * bucket_index);
1916 xfs_trans_log_buf(tp, agibp, offset,
1917 (offset + sizeof(xfs_agino_t) - 1));
1922 * Pull the on-disk inode from the AGI unlinked list.
1935 xfs_agnumber_t agno;
1936 xfs_daddr_t agdaddr;
1938 xfs_agino_t next_agino;
1939 xfs_buf_t *last_ibp;
1940 xfs_dinode_t *last_dip = NULL;
1942 int offset, last_offset = 0;
1947 * First pull the on-disk inode from the AGI unlinked list.
1951 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1952 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1955 * Get the agi buffer first. It ensures lock ordering
1958 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1959 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1962 "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
1963 error, mp->m_fsname);
1967 * Validate the magic number of the agi block.
1969 agi = XFS_BUF_TO_AGI(agibp);
1971 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1972 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
1973 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE,
1974 XFS_RANDOM_IUNLINK_REMOVE))) {
1975 XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW,
1977 xfs_trans_brelse(tp, agibp);
1979 "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
1981 return XFS_ERROR(EFSCORRUPTED);
1984 * Get the index into the agi hash table for the
1985 * list this inode will go on.
1987 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1989 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1990 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
1991 ASSERT(agi->agi_unlinked[bucket_index]);
1993 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
1995 * We're at the head of the list. Get the inode's
1996 * on-disk buffer to see if there is anyone after us
1997 * on the list. Only modify our next pointer if it
1998 * is not already NULLAGINO. This saves us the overhead
1999 * of dealing with the buffer when there is no need to
2002 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
2005 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2006 error, mp->m_fsname);
2009 next_agino = be32_to_cpu(dip->di_next_unlinked);
2010 ASSERT(next_agino != 0);
2011 if (next_agino != NULLAGINO) {
2012 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2013 offset = ip->i_boffset +
2014 offsetof(xfs_dinode_t, di_next_unlinked);
2015 xfs_trans_inode_buf(tp, ibp);
2016 xfs_trans_log_buf(tp, ibp, offset,
2017 (offset + sizeof(xfs_agino_t) - 1));
2018 xfs_inobp_check(mp, ibp);
2020 xfs_trans_brelse(tp, ibp);
2023 * Point the bucket head pointer at the next inode.
2025 ASSERT(next_agino != 0);
2026 ASSERT(next_agino != agino);
2027 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2028 offset = offsetof(xfs_agi_t, agi_unlinked) +
2029 (sizeof(xfs_agino_t) * bucket_index);
2030 xfs_trans_log_buf(tp, agibp, offset,
2031 (offset + sizeof(xfs_agino_t) - 1));
2034 * We need to search the list for the inode being freed.
2036 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2038 while (next_agino != agino) {
2040 * If the last inode wasn't the one pointing to
2041 * us, then release its buffer since we're not
2042 * going to do anything with it.
2044 if (last_ibp != NULL) {
2045 xfs_trans_brelse(tp, last_ibp);
2047 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2048 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
2049 &last_ibp, &last_offset);
2052 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
2053 error, mp->m_fsname);
2056 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2057 ASSERT(next_agino != NULLAGINO);
2058 ASSERT(next_agino != 0);
2061 * Now last_ibp points to the buffer previous to us on
2062 * the unlinked list. Pull us from the list.
2064 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
2067 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2068 error, mp->m_fsname);
2071 next_agino = be32_to_cpu(dip->di_next_unlinked);
2072 ASSERT(next_agino != 0);
2073 ASSERT(next_agino != agino);
2074 if (next_agino != NULLAGINO) {
2075 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2076 offset = ip->i_boffset +
2077 offsetof(xfs_dinode_t, di_next_unlinked);
2078 xfs_trans_inode_buf(tp, ibp);
2079 xfs_trans_log_buf(tp, ibp, offset,
2080 (offset + sizeof(xfs_agino_t) - 1));
2081 xfs_inobp_check(mp, ibp);
2083 xfs_trans_brelse(tp, ibp);
2086 * Point the previous inode on the list to the next inode.
2088 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2089 ASSERT(next_agino != 0);
2090 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2091 xfs_trans_inode_buf(tp, last_ibp);
2092 xfs_trans_log_buf(tp, last_ibp, offset,
2093 (offset + sizeof(xfs_agino_t) - 1));
2094 xfs_inobp_check(mp, last_ibp);
2101 xfs_inode_t *free_ip,
2105 xfs_mount_t *mp = free_ip->i_mount;
2106 int blks_per_cluster;
2109 int i, j, found, pre_flushed;
2112 xfs_inode_t *ip, **ip_found;
2113 xfs_inode_log_item_t *iip;
2114 xfs_log_item_t *lip;
2115 xfs_perag_t *pag = xfs_get_perag(mp, inum);
2117 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
2118 blks_per_cluster = 1;
2119 ninodes = mp->m_sb.sb_inopblock;
2120 nbufs = XFS_IALLOC_BLOCKS(mp);
2122 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
2123 mp->m_sb.sb_blocksize;
2124 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
2125 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
2128 ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
2130 for (j = 0; j < nbufs; j++, inum += ninodes) {
2131 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2132 XFS_INO_TO_AGBNO(mp, inum));
2136 * Look for each inode in memory and attempt to lock it,
2137 * we can be racing with flush and tail pushing here.
2138 * any inode we get the locks on, add to an array of
2139 * inode items to process later.
2141 * The get the buffer lock, we could beat a flush
2142 * or tail pushing thread to the lock here, in which
2143 * case they will go looking for the inode buffer
2144 * and fail, we need some other form of interlock
2148 for (i = 0; i < ninodes; i++) {
2149 read_lock(&pag->pag_ici_lock);
2150 ip = radix_tree_lookup(&pag->pag_ici_root,
2151 XFS_INO_TO_AGINO(mp, (inum + i)));
2153 /* Inode not in memory or we found it already,
2156 if (!ip || xfs_iflags_test(ip, XFS_ISTALE)) {
2157 read_unlock(&pag->pag_ici_lock);
2161 if (xfs_inode_clean(ip)) {
2162 read_unlock(&pag->pag_ici_lock);
2166 /* If we can get the locks then add it to the
2167 * list, otherwise by the time we get the bp lock
2168 * below it will already be attached to the
2172 /* This inode will already be locked - by us, lets
2176 if (ip == free_ip) {
2177 if (xfs_iflock_nowait(ip)) {
2178 xfs_iflags_set(ip, XFS_ISTALE);
2179 if (xfs_inode_clean(ip)) {
2182 ip_found[found++] = ip;
2185 read_unlock(&pag->pag_ici_lock);
2189 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2190 if (xfs_iflock_nowait(ip)) {
2191 xfs_iflags_set(ip, XFS_ISTALE);
2193 if (xfs_inode_clean(ip)) {
2195 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2197 ip_found[found++] = ip;
2200 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2203 read_unlock(&pag->pag_ici_lock);
2206 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2207 mp->m_bsize * blks_per_cluster,
2211 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
2213 if (lip->li_type == XFS_LI_INODE) {
2214 iip = (xfs_inode_log_item_t *)lip;
2215 ASSERT(iip->ili_logged == 1);
2216 lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
2217 xfs_trans_ail_copy_lsn(mp->m_ail,
2218 &iip->ili_flush_lsn,
2219 &iip->ili_item.li_lsn);
2220 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2223 lip = lip->li_bio_list;
2226 for (i = 0; i < found; i++) {
2231 ip->i_update_core = 0;
2233 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2237 iip->ili_last_fields = iip->ili_format.ilf_fields;
2238 iip->ili_format.ilf_fields = 0;
2239 iip->ili_logged = 1;
2240 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2241 &iip->ili_item.li_lsn);
2243 xfs_buf_attach_iodone(bp,
2244 (void(*)(xfs_buf_t*,xfs_log_item_t*))
2245 xfs_istale_done, (xfs_log_item_t *)iip);
2246 if (ip != free_ip) {
2247 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2251 if (found || pre_flushed)
2252 xfs_trans_stale_inode_buf(tp, bp);
2253 xfs_trans_binval(tp, bp);
2256 kmem_free(ip_found);
2257 xfs_put_perag(mp, pag);
2261 * This is called to return an inode to the inode free list.
2262 * The inode should already be truncated to 0 length and have
2263 * no pages associated with it. This routine also assumes that
2264 * the inode is already a part of the transaction.
2266 * The on-disk copy of the inode will have been added to the list
2267 * of unlinked inodes in the AGI. We need to remove the inode from
2268 * that list atomically with respect to freeing it here.
2274 xfs_bmap_free_t *flist)
2278 xfs_ino_t first_ino;
2282 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2283 ASSERT(ip->i_transp == tp);
2284 ASSERT(ip->i_d.di_nlink == 0);
2285 ASSERT(ip->i_d.di_nextents == 0);
2286 ASSERT(ip->i_d.di_anextents == 0);
2287 ASSERT((ip->i_d.di_size == 0 && ip->i_size == 0) ||
2288 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2289 ASSERT(ip->i_d.di_nblocks == 0);
2292 * Pull the on-disk inode from the AGI unlinked list.
2294 error = xfs_iunlink_remove(tp, ip);
2299 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2303 ip->i_d.di_mode = 0; /* mark incore inode as free */
2304 ip->i_d.di_flags = 0;
2305 ip->i_d.di_dmevmask = 0;
2306 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2307 ip->i_df.if_ext_max =
2308 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2309 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2310 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2312 * Bump the generation count so no one will be confused
2313 * by reincarnations of this inode.
2317 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2319 error = xfs_itobp(ip->i_mount, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
2324 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2325 * from picking up this inode when it is reclaimed (its incore state
2326 * initialzed but not flushed to disk yet). The in-core di_mode is
2327 * already cleared and a corresponding transaction logged.
2328 * The hack here just synchronizes the in-core to on-disk
2329 * di_mode value in advance before the actual inode sync to disk.
2330 * This is OK because the inode is already unlinked and would never
2331 * change its di_mode again for this inode generation.
2332 * This is a temporary hack that would require a proper fix
2335 dip->di_core.di_mode = 0;
2338 xfs_ifree_cluster(ip, tp, first_ino);
2345 * Reallocate the space for if_broot based on the number of records
2346 * being added or deleted as indicated in rec_diff. Move the records
2347 * and pointers in if_broot to fit the new size. When shrinking this
2348 * will eliminate holes between the records and pointers created by
2349 * the caller. When growing this will create holes to be filled in
2352 * The caller must not request to add more records than would fit in
2353 * the on-disk inode root. If the if_broot is currently NULL, then
2354 * if we adding records one will be allocated. The caller must also
2355 * not request that the number of records go below zero, although
2356 * it can go to zero.
2358 * ip -- the inode whose if_broot area is changing
2359 * ext_diff -- the change in the number of records, positive or negative,
2360 * requested for the if_broot array.
2368 struct xfs_mount *mp = ip->i_mount;
2371 struct xfs_btree_block *new_broot;
2378 * Handle the degenerate case quietly.
2380 if (rec_diff == 0) {
2384 ifp = XFS_IFORK_PTR(ip, whichfork);
2387 * If there wasn't any memory allocated before, just
2388 * allocate it now and get out.
2390 if (ifp->if_broot_bytes == 0) {
2391 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
2392 ifp->if_broot = kmem_alloc(new_size, KM_SLEEP);
2393 ifp->if_broot_bytes = (int)new_size;
2398 * If there is already an existing if_broot, then we need
2399 * to realloc() it and shift the pointers to their new
2400 * location. The records don't change location because
2401 * they are kept butted up against the btree block header.
2403 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
2404 new_max = cur_max + rec_diff;
2405 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2406 ifp->if_broot = kmem_realloc(ifp->if_broot, new_size,
2407 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2409 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2410 ifp->if_broot_bytes);
2411 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2413 ifp->if_broot_bytes = (int)new_size;
2414 ASSERT(ifp->if_broot_bytes <=
2415 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2416 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2421 * rec_diff is less than 0. In this case, we are shrinking the
2422 * if_broot buffer. It must already exist. If we go to zero
2423 * records, just get rid of the root and clear the status bit.
2425 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
2426 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
2427 new_max = cur_max + rec_diff;
2428 ASSERT(new_max >= 0);
2430 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2434 new_broot = kmem_alloc(new_size, KM_SLEEP);
2436 * First copy over the btree block header.
2438 memcpy(new_broot, ifp->if_broot, XFS_BTREE_LBLOCK_LEN);
2441 ifp->if_flags &= ~XFS_IFBROOT;
2445 * Only copy the records and pointers if there are any.
2449 * First copy the records.
2451 op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1);
2452 np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1);
2453 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2456 * Then copy the pointers.
2458 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2459 ifp->if_broot_bytes);
2460 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1,
2462 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2464 kmem_free(ifp->if_broot);
2465 ifp->if_broot = new_broot;
2466 ifp->if_broot_bytes = (int)new_size;
2467 ASSERT(ifp->if_broot_bytes <=
2468 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2474 * This is called when the amount of space needed for if_data
2475 * is increased or decreased. The change in size is indicated by
2476 * the number of bytes that need to be added or deleted in the
2477 * byte_diff parameter.
2479 * If the amount of space needed has decreased below the size of the
2480 * inline buffer, then switch to using the inline buffer. Otherwise,
2481 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2482 * to what is needed.
2484 * ip -- the inode whose if_data area is changing
2485 * byte_diff -- the change in the number of bytes, positive or negative,
2486 * requested for the if_data array.
2498 if (byte_diff == 0) {
2502 ifp = XFS_IFORK_PTR(ip, whichfork);
2503 new_size = (int)ifp->if_bytes + byte_diff;
2504 ASSERT(new_size >= 0);
2506 if (new_size == 0) {
2507 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2508 kmem_free(ifp->if_u1.if_data);
2510 ifp->if_u1.if_data = NULL;
2512 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2514 * If the valid extents/data can fit in if_inline_ext/data,
2515 * copy them from the malloc'd vector and free it.
2517 if (ifp->if_u1.if_data == NULL) {
2518 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2519 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2520 ASSERT(ifp->if_real_bytes != 0);
2521 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2523 kmem_free(ifp->if_u1.if_data);
2524 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2529 * Stuck with malloc/realloc.
2530 * For inline data, the underlying buffer must be
2531 * a multiple of 4 bytes in size so that it can be
2532 * logged and stay on word boundaries. We enforce
2535 real_size = roundup(new_size, 4);
2536 if (ifp->if_u1.if_data == NULL) {
2537 ASSERT(ifp->if_real_bytes == 0);
2538 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2539 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2541 * Only do the realloc if the underlying size
2542 * is really changing.
2544 if (ifp->if_real_bytes != real_size) {
2545 ifp->if_u1.if_data =
2546 kmem_realloc(ifp->if_u1.if_data,
2552 ASSERT(ifp->if_real_bytes == 0);
2553 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2554 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2558 ifp->if_real_bytes = real_size;
2559 ifp->if_bytes = new_size;
2560 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2567 * Map inode to disk block and offset.
2569 * mp -- the mount point structure for the current file system
2570 * tp -- the current transaction
2571 * ino -- the inode number of the inode to be located
2572 * imap -- this structure is filled in with the information necessary
2573 * to retrieve the given inode from disk
2574 * flags -- flags to pass to xfs_dilocate indicating whether or not
2575 * lookups in the inode btree were OK or not
2585 xfs_fsblock_t fsbno;
2590 fsbno = imap->im_blkno ?
2591 XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK;
2592 error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags);
2596 imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno);
2597 imap->im_len = XFS_FSB_TO_BB(mp, len);
2598 imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno);
2599 imap->im_ioffset = (ushort)off;
2600 imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog);
2603 * If the inode number maps to a block outside the bounds
2604 * of the file system then return NULL rather than calling
2605 * read_buf and panicing when we get an error from the
2608 if ((imap->im_blkno + imap->im_len) >
2609 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
2610 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
2611 "(imap->im_blkno (0x%llx) + imap->im_len (0x%llx)) > "
2612 " XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks) (0x%llx)",
2613 (unsigned long long) imap->im_blkno,
2614 (unsigned long long) imap->im_len,
2615 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
2628 ifp = XFS_IFORK_PTR(ip, whichfork);
2629 if (ifp->if_broot != NULL) {
2630 kmem_free(ifp->if_broot);
2631 ifp->if_broot = NULL;
2635 * If the format is local, then we can't have an extents
2636 * array so just look for an inline data array. If we're
2637 * not local then we may or may not have an extents list,
2638 * so check and free it up if we do.
2640 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2641 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2642 (ifp->if_u1.if_data != NULL)) {
2643 ASSERT(ifp->if_real_bytes != 0);
2644 kmem_free(ifp->if_u1.if_data);
2645 ifp->if_u1.if_data = NULL;
2646 ifp->if_real_bytes = 0;
2648 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2649 ((ifp->if_flags & XFS_IFEXTIREC) ||
2650 ((ifp->if_u1.if_extents != NULL) &&
2651 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
2652 ASSERT(ifp->if_real_bytes != 0);
2653 xfs_iext_destroy(ifp);
2655 ASSERT(ifp->if_u1.if_extents == NULL ||
2656 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2657 ASSERT(ifp->if_real_bytes == 0);
2658 if (whichfork == XFS_ATTR_FORK) {
2659 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2665 * This is called free all the memory associated with an inode.
2666 * It must free the inode itself and any buffers allocated for
2667 * if_extents/if_data and if_broot. It must also free the lock
2668 * associated with the inode.
2670 * Note: because we don't initialise everything on reallocation out
2671 * of the zone, we must ensure we nullify everything correctly before
2672 * freeing the structure.
2678 switch (ip->i_d.di_mode & S_IFMT) {
2682 xfs_idestroy_fork(ip, XFS_DATA_FORK);
2686 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
2688 #ifdef XFS_INODE_TRACE
2689 ktrace_free(ip->i_trace);
2691 #ifdef XFS_BMAP_TRACE
2692 ktrace_free(ip->i_xtrace);
2694 #ifdef XFS_BTREE_TRACE
2695 ktrace_free(ip->i_btrace);
2698 ktrace_free(ip->i_rwtrace);
2700 #ifdef XFS_ILOCK_TRACE
2701 ktrace_free(ip->i_lock_trace);
2703 #ifdef XFS_DIR2_TRACE
2704 ktrace_free(ip->i_dir_trace);
2708 * Only if we are shutting down the fs will we see an
2709 * inode still in the AIL. If it is there, we should remove
2710 * it to prevent a use-after-free from occurring.
2712 xfs_mount_t *mp = ip->i_mount;
2713 xfs_log_item_t *lip = &ip->i_itemp->ili_item;
2715 ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) ||
2716 XFS_FORCED_SHUTDOWN(ip->i_mount));
2717 if (lip->li_flags & XFS_LI_IN_AIL) {
2718 spin_lock(&mp->m_ail->xa_lock);
2719 if (lip->li_flags & XFS_LI_IN_AIL)
2720 xfs_trans_delete_ail(mp, lip);
2722 spin_unlock(&mp->m_ail->xa_lock);
2724 xfs_inode_item_destroy(ip);
2727 /* asserts to verify all state is correct here */
2728 ASSERT(atomic_read(&ip->i_iocount) == 0);
2729 ASSERT(atomic_read(&ip->i_pincount) == 0);
2730 ASSERT(!spin_is_locked(&ip->i_flags_lock));
2731 ASSERT(completion_done(&ip->i_flush));
2732 kmem_zone_free(xfs_inode_zone, ip);
2737 * Increment the pin count of the given buffer.
2738 * This value is protected by ipinlock spinlock in the mount structure.
2744 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2746 atomic_inc(&ip->i_pincount);
2750 * Decrement the pin count of the given inode, and wake up
2751 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2752 * inode must have been previously pinned with a call to xfs_ipin().
2758 ASSERT(atomic_read(&ip->i_pincount) > 0);
2760 if (atomic_dec_and_test(&ip->i_pincount))
2761 wake_up(&ip->i_ipin_wait);
2765 * This is called to unpin an inode. It can be directed to wait or to return
2766 * immediately without waiting for the inode to be unpinned. The caller must
2767 * have the inode locked in at least shared mode so that the buffer cannot be
2768 * subsequently pinned once someone is waiting for it to be unpinned.
2775 xfs_inode_log_item_t *iip = ip->i_itemp;
2777 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2778 if (atomic_read(&ip->i_pincount) == 0)
2781 /* Give the log a push to start the unpinning I/O */
2782 xfs_log_force(ip->i_mount, (iip && iip->ili_last_lsn) ?
2783 iip->ili_last_lsn : 0, XFS_LOG_FORCE);
2785 wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
2792 __xfs_iunpin_wait(ip, 1);
2799 __xfs_iunpin_wait(ip, 0);
2804 * xfs_iextents_copy()
2806 * This is called to copy the REAL extents (as opposed to the delayed
2807 * allocation extents) from the inode into the given buffer. It
2808 * returns the number of bytes copied into the buffer.
2810 * If there are no delayed allocation extents, then we can just
2811 * memcpy() the extents into the buffer. Otherwise, we need to
2812 * examine each extent in turn and skip those which are delayed.
2824 xfs_fsblock_t start_block;
2826 ifp = XFS_IFORK_PTR(ip, whichfork);
2827 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2828 ASSERT(ifp->if_bytes > 0);
2830 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2831 XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork);
2835 * There are some delayed allocation extents in the
2836 * inode, so copy the extents one at a time and skip
2837 * the delayed ones. There must be at least one
2838 * non-delayed extent.
2841 for (i = 0; i < nrecs; i++) {
2842 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
2843 start_block = xfs_bmbt_get_startblock(ep);
2844 if (ISNULLSTARTBLOCK(start_block)) {
2846 * It's a delayed allocation extent, so skip it.
2851 /* Translate to on disk format */
2852 put_unaligned(cpu_to_be64(ep->l0), &dp->l0);
2853 put_unaligned(cpu_to_be64(ep->l1), &dp->l1);
2857 ASSERT(copied != 0);
2858 xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip));
2860 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2864 * Each of the following cases stores data into the same region
2865 * of the on-disk inode, so only one of them can be valid at
2866 * any given time. While it is possible to have conflicting formats
2867 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2868 * in EXTENTS format, this can only happen when the fork has
2869 * changed formats after being modified but before being flushed.
2870 * In these cases, the format always takes precedence, because the
2871 * format indicates the current state of the fork.
2878 xfs_inode_log_item_t *iip,
2885 #ifdef XFS_TRANS_DEBUG
2888 static const short brootflag[2] =
2889 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2890 static const short dataflag[2] =
2891 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2892 static const short extflag[2] =
2893 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2897 ifp = XFS_IFORK_PTR(ip, whichfork);
2899 * This can happen if we gave up in iformat in an error path,
2900 * for the attribute fork.
2903 ASSERT(whichfork == XFS_ATTR_FORK);
2906 cp = XFS_DFORK_PTR(dip, whichfork);
2908 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2909 case XFS_DINODE_FMT_LOCAL:
2910 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2911 (ifp->if_bytes > 0)) {
2912 ASSERT(ifp->if_u1.if_data != NULL);
2913 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2914 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2918 case XFS_DINODE_FMT_EXTENTS:
2919 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2920 !(iip->ili_format.ilf_fields & extflag[whichfork]));
2921 ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
2922 (ifp->if_bytes == 0));
2923 ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
2924 (ifp->if_bytes > 0));
2925 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
2926 (ifp->if_bytes > 0)) {
2927 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2928 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2933 case XFS_DINODE_FMT_BTREE:
2934 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
2935 (ifp->if_broot_bytes > 0)) {
2936 ASSERT(ifp->if_broot != NULL);
2937 ASSERT(ifp->if_broot_bytes <=
2938 (XFS_IFORK_SIZE(ip, whichfork) +
2939 XFS_BROOT_SIZE_ADJ));
2940 xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes,
2941 (xfs_bmdr_block_t *)cp,
2942 XFS_DFORK_SIZE(dip, mp, whichfork));
2946 case XFS_DINODE_FMT_DEV:
2947 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
2948 ASSERT(whichfork == XFS_DATA_FORK);
2949 dip->di_u.di_dev = cpu_to_be32(ip->i_df.if_u2.if_rdev);
2953 case XFS_DINODE_FMT_UUID:
2954 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
2955 ASSERT(whichfork == XFS_DATA_FORK);
2956 memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid,
2972 xfs_mount_t *mp = ip->i_mount;
2973 xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino);
2974 unsigned long first_index, mask;
2975 unsigned long inodes_per_cluster;
2977 xfs_inode_t **ilist;
2984 ASSERT(pag->pagi_inodeok);
2985 ASSERT(pag->pag_ici_init);
2987 inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog;
2988 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
2989 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
2993 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
2994 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
2995 read_lock(&pag->pag_ici_lock);
2996 /* really need a gang lookup range call here */
2997 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
2998 first_index, inodes_per_cluster);
3002 for (i = 0; i < nr_found; i++) {
3006 /* if the inode lies outside this cluster, we're done. */
3007 if ((XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index)
3010 * Do an un-protected check to see if the inode is dirty and
3011 * is a candidate for flushing. These checks will be repeated
3012 * later after the appropriate locks are acquired.
3014 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
3018 * Try to get locks. If any are unavailable or it is pinned,
3019 * then this inode cannot be flushed and is skipped.
3022 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
3024 if (!xfs_iflock_nowait(iq)) {
3025 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3028 if (xfs_ipincount(iq)) {
3030 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3035 * arriving here means that this inode can be flushed. First
3036 * re-check that it's dirty before flushing.
3038 if (!xfs_inode_clean(iq)) {
3040 error = xfs_iflush_int(iq, bp);
3042 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3043 goto cluster_corrupt_out;
3049 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3053 XFS_STATS_INC(xs_icluster_flushcnt);
3054 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
3058 read_unlock(&pag->pag_ici_lock);
3063 cluster_corrupt_out:
3065 * Corruption detected in the clustering loop. Invalidate the
3066 * inode buffer and shut down the filesystem.
3068 read_unlock(&pag->pag_ici_lock);
3070 * Clean up the buffer. If it was B_DELWRI, just release it --
3071 * brelse can handle it with no problems. If not, shut down the
3072 * filesystem before releasing the buffer.
3074 bufwasdelwri = XFS_BUF_ISDELAYWRITE(bp);
3078 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3080 if (!bufwasdelwri) {
3082 * Just like incore_relse: if we have b_iodone functions,
3083 * mark the buffer as an error and call them. Otherwise
3084 * mark it as stale and brelse.
3086 if (XFS_BUF_IODONE_FUNC(bp)) {
3087 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
3091 XFS_BUF_ERROR(bp,EIO);
3100 * Unlocks the flush lock
3102 xfs_iflush_abort(iq);
3104 return XFS_ERROR(EFSCORRUPTED);
3108 * xfs_iflush() will write a modified inode's changes out to the
3109 * inode's on disk home. The caller must have the inode lock held
3110 * in at least shared mode and the inode flush completion must be
3111 * active as well. The inode lock will still be held upon return from
3112 * the call and the caller is free to unlock it.
3113 * The inode flush will be completed when the inode reaches the disk.
3114 * The flags indicate how the inode's buffer should be written out.
3121 xfs_inode_log_item_t *iip;
3126 int noblock = (flags == XFS_IFLUSH_ASYNC_NOBLOCK);
3127 enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
3129 XFS_STATS_INC(xs_iflush_count);
3131 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3132 ASSERT(!completion_done(&ip->i_flush));
3133 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3134 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3140 * If the inode isn't dirty, then just release the inode
3141 * flush lock and do nothing.
3143 if (xfs_inode_clean(ip)) {
3149 * We can't flush the inode until it is unpinned, so wait for it if we
3150 * are allowed to block. We know noone new can pin it, because we are
3151 * holding the inode lock shared and you need to hold it exclusively to
3154 * If we are not allowed to block, force the log out asynchronously so
3155 * that when we come back the inode will be unpinned. If other inodes
3156 * in the same cluster are dirty, they will probably write the inode
3157 * out for us if they occur after the log force completes.
3159 if (noblock && xfs_ipincount(ip)) {
3160 xfs_iunpin_nowait(ip);
3164 xfs_iunpin_wait(ip);
3167 * This may have been unpinned because the filesystem is shutting
3168 * down forcibly. If that's the case we must not write this inode
3169 * to disk, because the log record didn't make it to disk!
3171 if (XFS_FORCED_SHUTDOWN(mp)) {
3172 ip->i_update_core = 0;
3174 iip->ili_format.ilf_fields = 0;
3176 return XFS_ERROR(EIO);
3180 * Decide how buffer will be flushed out. This is done before
3181 * the call to xfs_iflush_int because this field is zeroed by it.
3183 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3185 * Flush out the inode buffer according to the directions
3186 * of the caller. In the cases where the caller has given
3187 * us a choice choose the non-delwri case. This is because
3188 * the inode is in the AIL and we need to get it out soon.
3191 case XFS_IFLUSH_SYNC:
3192 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3195 case XFS_IFLUSH_ASYNC_NOBLOCK:
3196 case XFS_IFLUSH_ASYNC:
3197 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3200 case XFS_IFLUSH_DELWRI:
3210 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3211 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3212 case XFS_IFLUSH_DELWRI:
3215 case XFS_IFLUSH_ASYNC_NOBLOCK:
3216 case XFS_IFLUSH_ASYNC:
3219 case XFS_IFLUSH_SYNC:
3230 * Get the buffer containing the on-disk inode.
3232 error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0, 0,
3233 noblock ? XFS_BUF_TRYLOCK : XFS_BUF_LOCK);
3240 * First flush out the inode that xfs_iflush was called with.
3242 error = xfs_iflush_int(ip, bp);
3247 * If the buffer is pinned then push on the log now so we won't
3248 * get stuck waiting in the write for too long.
3250 if (XFS_BUF_ISPINNED(bp))
3251 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
3255 * see if other inodes can be gathered into this write
3257 error = xfs_iflush_cluster(ip, bp);
3259 goto cluster_corrupt_out;
3261 if (flags & INT_DELWRI) {
3262 xfs_bdwrite(mp, bp);
3263 } else if (flags & INT_ASYNC) {
3264 error = xfs_bawrite(mp, bp);
3266 error = xfs_bwrite(mp, bp);
3272 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3273 cluster_corrupt_out:
3275 * Unlocks the flush lock
3277 xfs_iflush_abort(ip);
3278 return XFS_ERROR(EFSCORRUPTED);
3287 xfs_inode_log_item_t *iip;
3290 #ifdef XFS_TRANS_DEBUG
3294 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3295 ASSERT(!completion_done(&ip->i_flush));
3296 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3297 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3304 * If the inode isn't dirty, then just release the inode
3305 * flush lock and do nothing.
3307 if (xfs_inode_clean(ip)) {
3312 /* set *dip = inode's place in the buffer */
3313 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset);
3316 * Clear i_update_core before copying out the data.
3317 * This is for coordination with our timestamp updates
3318 * that don't hold the inode lock. They will always
3319 * update the timestamps BEFORE setting i_update_core,
3320 * so if we clear i_update_core after they set it we
3321 * are guaranteed to see their updates to the timestamps.
3322 * I believe that this depends on strongly ordered memory
3323 * semantics, but we have that. We use the SYNCHRONIZE
3324 * macro to make sure that the compiler does not reorder
3325 * the i_update_core access below the data copy below.
3327 ip->i_update_core = 0;
3331 * Make sure to get the latest atime from the Linux inode.
3333 xfs_synchronize_atime(ip);
3335 if (XFS_TEST_ERROR(be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC,
3336 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3337 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3338 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3339 ip->i_ino, be16_to_cpu(dip->di_core.di_magic), dip);
3342 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3343 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3344 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3345 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3346 ip->i_ino, ip, ip->i_d.di_magic);
3349 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3351 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3352 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3353 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3354 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3355 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3359 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
3361 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3362 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3363 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3364 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3365 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3366 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3371 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3372 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3373 XFS_RANDOM_IFLUSH_5)) {
3374 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3375 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3377 ip->i_d.di_nextents + ip->i_d.di_anextents,
3382 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3383 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3384 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3385 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3386 ip->i_ino, ip->i_d.di_forkoff, ip);
3390 * bump the flush iteration count, used to detect flushes which
3391 * postdate a log record during recovery.
3394 ip->i_d.di_flushiter++;
3397 * Copy the dirty parts of the inode into the on-disk
3398 * inode. We always copy out the core of the inode,
3399 * because if the inode is dirty at all the core must
3402 xfs_dinode_to_disk(&dip->di_core, &ip->i_d);
3404 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3405 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3406 ip->i_d.di_flushiter = 0;
3409 * If this is really an old format inode and the superblock version
3410 * has not been updated to support only new format inodes, then
3411 * convert back to the old inode format. If the superblock version
3412 * has been updated, then make the conversion permanent.
3414 ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 ||
3415 xfs_sb_version_hasnlink(&mp->m_sb));
3416 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
3417 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
3421 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3422 dip->di_core.di_onlink = cpu_to_be16(ip->i_d.di_nlink);
3425 * The superblock version has already been bumped,
3426 * so just make the conversion to the new inode
3429 ip->i_d.di_version = XFS_DINODE_VERSION_2;
3430 dip->di_core.di_version = XFS_DINODE_VERSION_2;
3431 ip->i_d.di_onlink = 0;
3432 dip->di_core.di_onlink = 0;
3433 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3434 memset(&(dip->di_core.di_pad[0]), 0,
3435 sizeof(dip->di_core.di_pad));
3436 ASSERT(ip->i_d.di_projid == 0);
3440 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
3441 if (XFS_IFORK_Q(ip))
3442 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3443 xfs_inobp_check(mp, bp);
3446 * We've recorded everything logged in the inode, so we'd
3447 * like to clear the ilf_fields bits so we don't log and
3448 * flush things unnecessarily. However, we can't stop
3449 * logging all this information until the data we've copied
3450 * into the disk buffer is written to disk. If we did we might
3451 * overwrite the copy of the inode in the log with all the
3452 * data after re-logging only part of it, and in the face of
3453 * a crash we wouldn't have all the data we need to recover.
3455 * What we do is move the bits to the ili_last_fields field.
3456 * When logging the inode, these bits are moved back to the
3457 * ilf_fields field. In the xfs_iflush_done() routine we
3458 * clear ili_last_fields, since we know that the information
3459 * those bits represent is permanently on disk. As long as
3460 * the flush completes before the inode is logged again, then
3461 * both ilf_fields and ili_last_fields will be cleared.
3463 * We can play with the ilf_fields bits here, because the inode
3464 * lock must be held exclusively in order to set bits there
3465 * and the flush lock protects the ili_last_fields bits.
3466 * Set ili_logged so the flush done
3467 * routine can tell whether or not to look in the AIL.
3468 * Also, store the current LSN of the inode so that we can tell
3469 * whether the item has moved in the AIL from xfs_iflush_done().
3470 * In order to read the lsn we need the AIL lock, because
3471 * it is a 64 bit value that cannot be read atomically.
3473 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3474 iip->ili_last_fields = iip->ili_format.ilf_fields;
3475 iip->ili_format.ilf_fields = 0;
3476 iip->ili_logged = 1;
3478 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3479 &iip->ili_item.li_lsn);
3482 * Attach the function xfs_iflush_done to the inode's
3483 * buffer. This will remove the inode from the AIL
3484 * and unlock the inode's flush lock when the inode is
3485 * completely written to disk.
3487 xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
3488 xfs_iflush_done, (xfs_log_item_t *)iip);
3490 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3491 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3494 * We're flushing an inode which is not in the AIL and has
3495 * not been logged but has i_update_core set. For this
3496 * case we can use a B_DELWRI flush and immediately drop
3497 * the inode flush lock because we can avoid the whole
3498 * AIL state thing. It's OK to drop the flush lock now,
3499 * because we've already locked the buffer and to do anything
3500 * you really need both.
3503 ASSERT(iip->ili_logged == 0);
3504 ASSERT(iip->ili_last_fields == 0);
3505 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3513 return XFS_ERROR(EFSCORRUPTED);
3518 #ifdef XFS_ILOCK_TRACE
3519 ktrace_t *xfs_ilock_trace_buf;
3522 xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
3524 ktrace_enter(ip->i_lock_trace,
3526 (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
3527 (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
3528 (void *)ra, /* caller of ilock */
3529 (void *)(unsigned long)current_cpu(),
3530 (void *)(unsigned long)current_pid(),
3531 NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL);
3536 * Return a pointer to the extent record at file index idx.
3538 xfs_bmbt_rec_host_t *
3540 xfs_ifork_t *ifp, /* inode fork pointer */
3541 xfs_extnum_t idx) /* index of target extent */
3544 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
3545 return ifp->if_u1.if_ext_irec->er_extbuf;
3546 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3547 xfs_ext_irec_t *erp; /* irec pointer */
3548 int erp_idx = 0; /* irec index */
3549 xfs_extnum_t page_idx = idx; /* ext index in target list */
3551 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3552 return &erp->er_extbuf[page_idx];
3553 } else if (ifp->if_bytes) {
3554 return &ifp->if_u1.if_extents[idx];
3561 * Insert new item(s) into the extent records for incore inode
3562 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3566 xfs_ifork_t *ifp, /* inode fork pointer */
3567 xfs_extnum_t idx, /* starting index of new items */
3568 xfs_extnum_t count, /* number of inserted items */
3569 xfs_bmbt_irec_t *new) /* items to insert */
3571 xfs_extnum_t i; /* extent record index */
3573 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3574 xfs_iext_add(ifp, idx, count);
3575 for (i = idx; i < idx + count; i++, new++)
3576 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new);
3580 * This is called when the amount of space required for incore file
3581 * extents needs to be increased. The ext_diff parameter stores the
3582 * number of new extents being added and the idx parameter contains
3583 * the extent index where the new extents will be added. If the new
3584 * extents are being appended, then we just need to (re)allocate and
3585 * initialize the space. Otherwise, if the new extents are being
3586 * inserted into the middle of the existing entries, a bit more work
3587 * is required to make room for the new extents to be inserted. The
3588 * caller is responsible for filling in the new extent entries upon
3593 xfs_ifork_t *ifp, /* inode fork pointer */
3594 xfs_extnum_t idx, /* index to begin adding exts */
3595 int ext_diff) /* number of extents to add */
3597 int byte_diff; /* new bytes being added */
3598 int new_size; /* size of extents after adding */
3599 xfs_extnum_t nextents; /* number of extents in file */
3601 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3602 ASSERT((idx >= 0) && (idx <= nextents));
3603 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
3604 new_size = ifp->if_bytes + byte_diff;
3606 * If the new number of extents (nextents + ext_diff)
3607 * fits inside the inode, then continue to use the inline
3610 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
3611 if (idx < nextents) {
3612 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
3613 &ifp->if_u2.if_inline_ext[idx],
3614 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3615 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
3617 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3618 ifp->if_real_bytes = 0;
3619 ifp->if_lastex = nextents + ext_diff;
3622 * Otherwise use a linear (direct) extent list.
3623 * If the extents are currently inside the inode,
3624 * xfs_iext_realloc_direct will switch us from
3625 * inline to direct extent allocation mode.
3627 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
3628 xfs_iext_realloc_direct(ifp, new_size);
3629 if (idx < nextents) {
3630 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
3631 &ifp->if_u1.if_extents[idx],
3632 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3633 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
3636 /* Indirection array */
3638 xfs_ext_irec_t *erp;
3642 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
3643 if (ifp->if_flags & XFS_IFEXTIREC) {
3644 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3646 xfs_iext_irec_init(ifp);
3647 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3648 erp = ifp->if_u1.if_ext_irec;
3650 /* Extents fit in target extent page */
3651 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3652 if (page_idx < erp->er_extcount) {
3653 memmove(&erp->er_extbuf[page_idx + ext_diff],
3654 &erp->er_extbuf[page_idx],
3655 (erp->er_extcount - page_idx) *
3656 sizeof(xfs_bmbt_rec_t));
3657 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3659 erp->er_extcount += ext_diff;
3660 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3662 /* Insert a new extent page */
3664 xfs_iext_add_indirect_multi(ifp,
3665 erp_idx, page_idx, ext_diff);
3668 * If extent(s) are being appended to the last page in
3669 * the indirection array and the new extent(s) don't fit
3670 * in the page, then erp is NULL and erp_idx is set to
3671 * the next index needed in the indirection array.
3674 int count = ext_diff;
3677 erp = xfs_iext_irec_new(ifp, erp_idx);
3678 erp->er_extcount = count;
3679 count -= MIN(count, (int)XFS_LINEAR_EXTS);
3686 ifp->if_bytes = new_size;
3690 * This is called when incore extents are being added to the indirection
3691 * array and the new extents do not fit in the target extent list. The
3692 * erp_idx parameter contains the irec index for the target extent list
3693 * in the indirection array, and the idx parameter contains the extent
3694 * index within the list. The number of extents being added is stored
3695 * in the count parameter.
3697 * |-------| |-------|
3698 * | | | | idx - number of extents before idx
3700 * | | | | count - number of extents being inserted at idx
3701 * |-------| |-------|
3702 * | count | | nex2 | nex2 - number of extents after idx + count
3703 * |-------| |-------|
3706 xfs_iext_add_indirect_multi(
3707 xfs_ifork_t *ifp, /* inode fork pointer */
3708 int erp_idx, /* target extent irec index */
3709 xfs_extnum_t idx, /* index within target list */
3710 int count) /* new extents being added */
3712 int byte_diff; /* new bytes being added */
3713 xfs_ext_irec_t *erp; /* pointer to irec entry */
3714 xfs_extnum_t ext_diff; /* number of extents to add */
3715 xfs_extnum_t ext_cnt; /* new extents still needed */
3716 xfs_extnum_t nex2; /* extents after idx + count */
3717 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3718 int nlists; /* number of irec's (lists) */
3720 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3721 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3722 nex2 = erp->er_extcount - idx;
3723 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3726 * Save second part of target extent list
3727 * (all extents past */
3729 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3730 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS);
3731 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3732 erp->er_extcount -= nex2;
3733 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3734 memset(&erp->er_extbuf[idx], 0, byte_diff);
3738 * Add the new extents to the end of the target
3739 * list, then allocate new irec record(s) and
3740 * extent buffer(s) as needed to store the rest
3741 * of the new extents.
3744 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3746 erp->er_extcount += ext_diff;
3747 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3748 ext_cnt -= ext_diff;
3752 erp = xfs_iext_irec_new(ifp, erp_idx);
3753 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3754 erp->er_extcount = ext_diff;
3755 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3756 ext_cnt -= ext_diff;
3759 /* Add nex2 extents back to indirection array */
3761 xfs_extnum_t ext_avail;
3764 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3765 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3768 * If nex2 extents fit in the current page, append
3769 * nex2_ep after the new extents.
3771 if (nex2 <= ext_avail) {
3772 i = erp->er_extcount;
3775 * Otherwise, check if space is available in the
3778 else if ((erp_idx < nlists - 1) &&
3779 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3780 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3783 /* Create a hole for nex2 extents */
3784 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3785 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3788 * Final choice, create a new extent page for
3793 erp = xfs_iext_irec_new(ifp, erp_idx);
3795 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
3797 erp->er_extcount += nex2;
3798 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3803 * This is called when the amount of space required for incore file
3804 * extents needs to be decreased. The ext_diff parameter stores the
3805 * number of extents to be removed and the idx parameter contains
3806 * the extent index where the extents will be removed from.
3808 * If the amount of space needed has decreased below the linear
3809 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3810 * extent array. Otherwise, use kmem_realloc() to adjust the
3811 * size to what is needed.
3815 xfs_ifork_t *ifp, /* inode fork pointer */
3816 xfs_extnum_t idx, /* index to begin removing exts */
3817 int ext_diff) /* number of extents to remove */
3819 xfs_extnum_t nextents; /* number of extents in file */
3820 int new_size; /* size of extents after removal */
3822 ASSERT(ext_diff > 0);
3823 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3824 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3826 if (new_size == 0) {
3827 xfs_iext_destroy(ifp);
3828 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3829 xfs_iext_remove_indirect(ifp, idx, ext_diff);
3830 } else if (ifp->if_real_bytes) {
3831 xfs_iext_remove_direct(ifp, idx, ext_diff);
3833 xfs_iext_remove_inline(ifp, idx, ext_diff);
3835 ifp->if_bytes = new_size;
3839 * This removes ext_diff extents from the inline buffer, beginning
3840 * at extent index idx.
3843 xfs_iext_remove_inline(
3844 xfs_ifork_t *ifp, /* inode fork pointer */
3845 xfs_extnum_t idx, /* index to begin removing exts */
3846 int ext_diff) /* number of extents to remove */
3848 int nextents; /* number of extents in file */
3850 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3851 ASSERT(idx < XFS_INLINE_EXTS);
3852 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3853 ASSERT(((nextents - ext_diff) > 0) &&
3854 (nextents - ext_diff) < XFS_INLINE_EXTS);
3856 if (idx + ext_diff < nextents) {
3857 memmove(&ifp->if_u2.if_inline_ext[idx],
3858 &ifp->if_u2.if_inline_ext[idx + ext_diff],
3859 (nextents - (idx + ext_diff)) *
3860 sizeof(xfs_bmbt_rec_t));
3861 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
3862 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3864 memset(&ifp->if_u2.if_inline_ext[idx], 0,
3865 ext_diff * sizeof(xfs_bmbt_rec_t));
3870 * This removes ext_diff extents from a linear (direct) extent list,
3871 * beginning at extent index idx. If the extents are being removed
3872 * from the end of the list (ie. truncate) then we just need to re-
3873 * allocate the list to remove the extra space. Otherwise, if the
3874 * extents are being removed from the middle of the existing extent
3875 * entries, then we first need to move the extent records beginning
3876 * at idx + ext_diff up in the list to overwrite the records being
3877 * removed, then remove the extra space via kmem_realloc.
3880 xfs_iext_remove_direct(
3881 xfs_ifork_t *ifp, /* inode fork pointer */
3882 xfs_extnum_t idx, /* index to begin removing exts */
3883 int ext_diff) /* number of extents to remove */
3885 xfs_extnum_t nextents; /* number of extents in file */
3886 int new_size; /* size of extents after removal */
3888 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3889 new_size = ifp->if_bytes -
3890 (ext_diff * sizeof(xfs_bmbt_rec_t));
3891 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3893 if (new_size == 0) {
3894 xfs_iext_destroy(ifp);
3897 /* Move extents up in the list (if needed) */
3898 if (idx + ext_diff < nextents) {
3899 memmove(&ifp->if_u1.if_extents[idx],
3900 &ifp->if_u1.if_extents[idx + ext_diff],
3901 (nextents - (idx + ext_diff)) *
3902 sizeof(xfs_bmbt_rec_t));
3904 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
3905 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3907 * Reallocate the direct extent list. If the extents
3908 * will fit inside the inode then xfs_iext_realloc_direct
3909 * will switch from direct to inline extent allocation
3912 xfs_iext_realloc_direct(ifp, new_size);
3913 ifp->if_bytes = new_size;
3917 * This is called when incore extents are being removed from the
3918 * indirection array and the extents being removed span multiple extent
3919 * buffers. The idx parameter contains the file extent index where we
3920 * want to begin removing extents, and the count parameter contains
3921 * how many extents need to be removed.
3923 * |-------| |-------|
3924 * | nex1 | | | nex1 - number of extents before idx
3925 * |-------| | count |
3926 * | | | | count - number of extents being removed at idx
3927 * | count | |-------|
3928 * | | | nex2 | nex2 - number of extents after idx + count
3929 * |-------| |-------|
3932 xfs_iext_remove_indirect(
3933 xfs_ifork_t *ifp, /* inode fork pointer */
3934 xfs_extnum_t idx, /* index to begin removing extents */
3935 int count) /* number of extents to remove */
3937 xfs_ext_irec_t *erp; /* indirection array pointer */
3938 int erp_idx = 0; /* indirection array index */
3939 xfs_extnum_t ext_cnt; /* extents left to remove */
3940 xfs_extnum_t ext_diff; /* extents to remove in current list */
3941 xfs_extnum_t nex1; /* number of extents before idx */
3942 xfs_extnum_t nex2; /* extents after idx + count */
3943 int nlists; /* entries in indirection array */
3944 int page_idx = idx; /* index in target extent list */
3946 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3947 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3948 ASSERT(erp != NULL);
3949 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3953 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
3954 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
3956 * Check for deletion of entire list;
3957 * xfs_iext_irec_remove() updates extent offsets.
3959 if (ext_diff == erp->er_extcount) {
3960 xfs_iext_irec_remove(ifp, erp_idx);
3961 ext_cnt -= ext_diff;
3964 ASSERT(erp_idx < ifp->if_real_bytes /
3966 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3973 /* Move extents up (if needed) */
3975 memmove(&erp->er_extbuf[nex1],
3976 &erp->er_extbuf[nex1 + ext_diff],
3977 nex2 * sizeof(xfs_bmbt_rec_t));
3979 /* Zero out rest of page */
3980 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
3981 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
3982 /* Update remaining counters */
3983 erp->er_extcount -= ext_diff;
3984 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
3985 ext_cnt -= ext_diff;
3990 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
3991 xfs_iext_irec_compact(ifp);
3995 * Create, destroy, or resize a linear (direct) block of extents.
3998 xfs_iext_realloc_direct(
3999 xfs_ifork_t *ifp, /* inode fork pointer */
4000 int new_size) /* new size of extents */
4002 int rnew_size; /* real new size of extents */
4004 rnew_size = new_size;
4006 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
4007 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
4008 (new_size != ifp->if_real_bytes)));
4010 /* Free extent records */
4011 if (new_size == 0) {
4012 xfs_iext_destroy(ifp);
4014 /* Resize direct extent list and zero any new bytes */
4015 else if (ifp->if_real_bytes) {
4016 /* Check if extents will fit inside the inode */
4017 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
4018 xfs_iext_direct_to_inline(ifp, new_size /
4019 (uint)sizeof(xfs_bmbt_rec_t));
4020 ifp->if_bytes = new_size;
4023 if (!is_power_of_2(new_size)){
4024 rnew_size = roundup_pow_of_two(new_size);
4026 if (rnew_size != ifp->if_real_bytes) {
4027 ifp->if_u1.if_extents =
4028 kmem_realloc(ifp->if_u1.if_extents,
4030 ifp->if_real_bytes, KM_NOFS);
4032 if (rnew_size > ifp->if_real_bytes) {
4033 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
4034 (uint)sizeof(xfs_bmbt_rec_t)], 0,
4035 rnew_size - ifp->if_real_bytes);
4039 * Switch from the inline extent buffer to a direct
4040 * extent list. Be sure to include the inline extent
4041 * bytes in new_size.
4044 new_size += ifp->if_bytes;
4045 if (!is_power_of_2(new_size)) {
4046 rnew_size = roundup_pow_of_two(new_size);
4048 xfs_iext_inline_to_direct(ifp, rnew_size);
4050 ifp->if_real_bytes = rnew_size;
4051 ifp->if_bytes = new_size;
4055 * Switch from linear (direct) extent records to inline buffer.
4058 xfs_iext_direct_to_inline(
4059 xfs_ifork_t *ifp, /* inode fork pointer */
4060 xfs_extnum_t nextents) /* number of extents in file */
4062 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4063 ASSERT(nextents <= XFS_INLINE_EXTS);
4065 * The inline buffer was zeroed when we switched
4066 * from inline to direct extent allocation mode,
4067 * so we don't need to clear it here.
4069 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
4070 nextents * sizeof(xfs_bmbt_rec_t));
4071 kmem_free(ifp->if_u1.if_extents);
4072 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
4073 ifp->if_real_bytes = 0;
4077 * Switch from inline buffer to linear (direct) extent records.
4078 * new_size should already be rounded up to the next power of 2
4079 * by the caller (when appropriate), so use new_size as it is.
4080 * However, since new_size may be rounded up, we can't update
4081 * if_bytes here. It is the caller's responsibility to update
4082 * if_bytes upon return.
4085 xfs_iext_inline_to_direct(
4086 xfs_ifork_t *ifp, /* inode fork pointer */
4087 int new_size) /* number of extents in file */
4089 ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS);
4090 memset(ifp->if_u1.if_extents, 0, new_size);
4091 if (ifp->if_bytes) {
4092 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
4094 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4095 sizeof(xfs_bmbt_rec_t));
4097 ifp->if_real_bytes = new_size;
4101 * Resize an extent indirection array to new_size bytes.
4104 xfs_iext_realloc_indirect(
4105 xfs_ifork_t *ifp, /* inode fork pointer */
4106 int new_size) /* new indirection array size */
4108 int nlists; /* number of irec's (ex lists) */
4109 int size; /* current indirection array size */
4111 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4112 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4113 size = nlists * sizeof(xfs_ext_irec_t);
4114 ASSERT(ifp->if_real_bytes);
4115 ASSERT((new_size >= 0) && (new_size != size));
4116 if (new_size == 0) {
4117 xfs_iext_destroy(ifp);
4119 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
4120 kmem_realloc(ifp->if_u1.if_ext_irec,
4121 new_size, size, KM_NOFS);
4126 * Switch from indirection array to linear (direct) extent allocations.
4129 xfs_iext_indirect_to_direct(
4130 xfs_ifork_t *ifp) /* inode fork pointer */
4132 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
4133 xfs_extnum_t nextents; /* number of extents in file */
4134 int size; /* size of file extents */
4136 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4137 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4138 ASSERT(nextents <= XFS_LINEAR_EXTS);
4139 size = nextents * sizeof(xfs_bmbt_rec_t);
4141 xfs_iext_irec_compact_pages(ifp);
4142 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
4144 ep = ifp->if_u1.if_ext_irec->er_extbuf;
4145 kmem_free(ifp->if_u1.if_ext_irec);
4146 ifp->if_flags &= ~XFS_IFEXTIREC;
4147 ifp->if_u1.if_extents = ep;
4148 ifp->if_bytes = size;
4149 if (nextents < XFS_LINEAR_EXTS) {
4150 xfs_iext_realloc_direct(ifp, size);
4155 * Free incore file extents.
4159 xfs_ifork_t *ifp) /* inode fork pointer */
4161 if (ifp->if_flags & XFS_IFEXTIREC) {
4165 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4166 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
4167 xfs_iext_irec_remove(ifp, erp_idx);
4169 ifp->if_flags &= ~XFS_IFEXTIREC;
4170 } else if (ifp->if_real_bytes) {
4171 kmem_free(ifp->if_u1.if_extents);
4172 } else if (ifp->if_bytes) {
4173 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4174 sizeof(xfs_bmbt_rec_t));
4176 ifp->if_u1.if_extents = NULL;
4177 ifp->if_real_bytes = 0;
4182 * Return a pointer to the extent record for file system block bno.
4184 xfs_bmbt_rec_host_t * /* pointer to found extent record */
4185 xfs_iext_bno_to_ext(
4186 xfs_ifork_t *ifp, /* inode fork pointer */
4187 xfs_fileoff_t bno, /* block number to search for */
4188 xfs_extnum_t *idxp) /* index of target extent */
4190 xfs_bmbt_rec_host_t *base; /* pointer to first extent */
4191 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
4192 xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */
4193 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4194 int high; /* upper boundary in search */
4195 xfs_extnum_t idx = 0; /* index of target extent */
4196 int low; /* lower boundary in search */
4197 xfs_extnum_t nextents; /* number of file extents */
4198 xfs_fileoff_t startoff = 0; /* start offset of extent */
4200 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4201 if (nextents == 0) {
4206 if (ifp->if_flags & XFS_IFEXTIREC) {
4207 /* Find target extent list */
4209 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
4210 base = erp->er_extbuf;
4211 high = erp->er_extcount - 1;
4213 base = ifp->if_u1.if_extents;
4214 high = nextents - 1;
4216 /* Binary search extent records */
4217 while (low <= high) {
4218 idx = (low + high) >> 1;
4220 startoff = xfs_bmbt_get_startoff(ep);
4221 blockcount = xfs_bmbt_get_blockcount(ep);
4222 if (bno < startoff) {
4224 } else if (bno >= startoff + blockcount) {
4227 /* Convert back to file-based extent index */
4228 if (ifp->if_flags & XFS_IFEXTIREC) {
4229 idx += erp->er_extoff;
4235 /* Convert back to file-based extent index */
4236 if (ifp->if_flags & XFS_IFEXTIREC) {
4237 idx += erp->er_extoff;
4239 if (bno >= startoff + blockcount) {
4240 if (++idx == nextents) {
4243 ep = xfs_iext_get_ext(ifp, idx);
4251 * Return a pointer to the indirection array entry containing the
4252 * extent record for filesystem block bno. Store the index of the
4253 * target irec in *erp_idxp.
4255 xfs_ext_irec_t * /* pointer to found extent record */
4256 xfs_iext_bno_to_irec(
4257 xfs_ifork_t *ifp, /* inode fork pointer */
4258 xfs_fileoff_t bno, /* block number to search for */
4259 int *erp_idxp) /* irec index of target ext list */
4261 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4262 xfs_ext_irec_t *erp_next; /* next indirection array entry */
4263 int erp_idx; /* indirection array index */
4264 int nlists; /* number of extent irec's (lists) */
4265 int high; /* binary search upper limit */
4266 int low; /* binary search lower limit */
4268 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4269 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4273 while (low <= high) {
4274 erp_idx = (low + high) >> 1;
4275 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4276 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
4277 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
4279 } else if (erp_next && bno >=
4280 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
4286 *erp_idxp = erp_idx;
4291 * Return a pointer to the indirection array entry containing the
4292 * extent record at file extent index *idxp. Store the index of the
4293 * target irec in *erp_idxp and store the page index of the target
4294 * extent record in *idxp.
4297 xfs_iext_idx_to_irec(
4298 xfs_ifork_t *ifp, /* inode fork pointer */
4299 xfs_extnum_t *idxp, /* extent index (file -> page) */
4300 int *erp_idxp, /* pointer to target irec */
4301 int realloc) /* new bytes were just added */
4303 xfs_ext_irec_t *prev; /* pointer to previous irec */
4304 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
4305 int erp_idx; /* indirection array index */
4306 int nlists; /* number of irec's (ex lists) */
4307 int high; /* binary search upper limit */
4308 int low; /* binary search lower limit */
4309 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
4311 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4312 ASSERT(page_idx >= 0 && page_idx <=
4313 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
4314 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4319 /* Binary search extent irec's */
4320 while (low <= high) {
4321 erp_idx = (low + high) >> 1;
4322 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4323 prev = erp_idx > 0 ? erp - 1 : NULL;
4324 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
4325 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
4327 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
4328 (page_idx == erp->er_extoff + erp->er_extcount &&
4331 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
4332 erp->er_extcount == XFS_LINEAR_EXTS) {
4336 erp = erp_idx < nlists ? erp + 1 : NULL;
4339 page_idx -= erp->er_extoff;
4344 *erp_idxp = erp_idx;
4349 * Allocate and initialize an indirection array once the space needed
4350 * for incore extents increases above XFS_IEXT_BUFSZ.
4354 xfs_ifork_t *ifp) /* inode fork pointer */
4356 xfs_ext_irec_t *erp; /* indirection array pointer */
4357 xfs_extnum_t nextents; /* number of extents in file */
4359 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4360 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4361 ASSERT(nextents <= XFS_LINEAR_EXTS);
4363 erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS);
4365 if (nextents == 0) {
4366 ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
4367 } else if (!ifp->if_real_bytes) {
4368 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
4369 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
4370 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
4372 erp->er_extbuf = ifp->if_u1.if_extents;
4373 erp->er_extcount = nextents;
4376 ifp->if_flags |= XFS_IFEXTIREC;
4377 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
4378 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
4379 ifp->if_u1.if_ext_irec = erp;
4385 * Allocate and initialize a new entry in the indirection array.
4389 xfs_ifork_t *ifp, /* inode fork pointer */
4390 int erp_idx) /* index for new irec */
4392 xfs_ext_irec_t *erp; /* indirection array pointer */
4393 int i; /* loop counter */
4394 int nlists; /* number of irec's (ex lists) */
4396 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4397 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4399 /* Resize indirection array */
4400 xfs_iext_realloc_indirect(ifp, ++nlists *
4401 sizeof(xfs_ext_irec_t));
4403 * Move records down in the array so the
4404 * new page can use erp_idx.
4406 erp = ifp->if_u1.if_ext_irec;
4407 for (i = nlists - 1; i > erp_idx; i--) {
4408 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
4410 ASSERT(i == erp_idx);
4412 /* Initialize new extent record */
4413 erp = ifp->if_u1.if_ext_irec;
4414 erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
4415 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4416 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
4417 erp[erp_idx].er_extcount = 0;
4418 erp[erp_idx].er_extoff = erp_idx > 0 ?
4419 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
4420 return (&erp[erp_idx]);
4424 * Remove a record from the indirection array.
4427 xfs_iext_irec_remove(
4428 xfs_ifork_t *ifp, /* inode fork pointer */
4429 int erp_idx) /* irec index to remove */
4431 xfs_ext_irec_t *erp; /* indirection array pointer */
4432 int i; /* loop counter */
4433 int nlists; /* number of irec's (ex lists) */
4435 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4436 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4437 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4438 if (erp->er_extbuf) {
4439 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
4441 kmem_free(erp->er_extbuf);
4443 /* Compact extent records */
4444 erp = ifp->if_u1.if_ext_irec;
4445 for (i = erp_idx; i < nlists - 1; i++) {
4446 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
4449 * Manually free the last extent record from the indirection
4450 * array. A call to xfs_iext_realloc_indirect() with a size
4451 * of zero would result in a call to xfs_iext_destroy() which
4452 * would in turn call this function again, creating a nasty
4456 xfs_iext_realloc_indirect(ifp,
4457 nlists * sizeof(xfs_ext_irec_t));
4459 kmem_free(ifp->if_u1.if_ext_irec);
4461 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4465 * This is called to clean up large amounts of unused memory allocated
4466 * by the indirection array. Before compacting anything though, verify
4467 * that the indirection array is still needed and switch back to the
4468 * linear extent list (or even the inline buffer) if possible. The
4469 * compaction policy is as follows:
4471 * Full Compaction: Extents fit into a single page (or inline buffer)
4472 * Partial Compaction: Extents occupy less than 50% of allocated space
4473 * No Compaction: Extents occupy at least 50% of allocated space
4476 xfs_iext_irec_compact(
4477 xfs_ifork_t *ifp) /* inode fork pointer */
4479 xfs_extnum_t nextents; /* number of extents in file */
4480 int nlists; /* number of irec's (ex lists) */
4482 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4483 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4484 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4486 if (nextents == 0) {
4487 xfs_iext_destroy(ifp);
4488 } else if (nextents <= XFS_INLINE_EXTS) {
4489 xfs_iext_indirect_to_direct(ifp);
4490 xfs_iext_direct_to_inline(ifp, nextents);
4491 } else if (nextents <= XFS_LINEAR_EXTS) {
4492 xfs_iext_indirect_to_direct(ifp);
4493 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
4494 xfs_iext_irec_compact_pages(ifp);
4499 * Combine extents from neighboring extent pages.
4502 xfs_iext_irec_compact_pages(
4503 xfs_ifork_t *ifp) /* inode fork pointer */
4505 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
4506 int erp_idx = 0; /* indirection array index */
4507 int nlists; /* number of irec's (ex lists) */
4509 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4510 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4511 while (erp_idx < nlists - 1) {
4512 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4514 if (erp_next->er_extcount <=
4515 (XFS_LINEAR_EXTS - erp->er_extcount)) {
4516 memcpy(&erp->er_extbuf[erp->er_extcount],
4517 erp_next->er_extbuf, erp_next->er_extcount *
4518 sizeof(xfs_bmbt_rec_t));
4519 erp->er_extcount += erp_next->er_extcount;
4521 * Free page before removing extent record
4522 * so er_extoffs don't get modified in
4523 * xfs_iext_irec_remove.
4525 kmem_free(erp_next->er_extbuf);
4526 erp_next->er_extbuf = NULL;
4527 xfs_iext_irec_remove(ifp, erp_idx + 1);
4528 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4536 * This is called to update the er_extoff field in the indirection
4537 * array when extents have been added or removed from one of the
4538 * extent lists. erp_idx contains the irec index to begin updating
4539 * at and ext_diff contains the number of extents that were added
4543 xfs_iext_irec_update_extoffs(
4544 xfs_ifork_t *ifp, /* inode fork pointer */
4545 int erp_idx, /* irec index to update */
4546 int ext_diff) /* number of new extents */
4548 int i; /* loop counter */
4549 int nlists; /* number of irec's (ex lists */
4551 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4552 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4553 for (i = erp_idx; i < nlists; i++) {
4554 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;