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.
240 error = xfs_imap(mp, tp, ino, &imap, imap_flags | XFS_IMAP_LOOKUP);
244 error = xfs_imap_to_bp(mp, tp, &imap, &bp, XFS_BUF_LOCK, imap_flags);
248 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
250 *offset = imap.im_boffset;
256 * This routine is called to map an inode to the buffer containing
257 * the on-disk version of the inode. It returns a pointer to the
258 * buffer containing the on-disk inode in the bpp parameter, and in
259 * the dip parameter it returns a pointer to the on-disk inode within
262 * If a non-zero error is returned, then the contents of bpp and
263 * dipp are undefined.
265 * If the inode is new and has not yet been initialized, use xfs_imap()
266 * to determine the size and location of the buffer to read from disk.
267 * If the inode has already been mapped to its buffer and read in once,
268 * then use the mapping information stored in the inode rather than
269 * calling xfs_imap(). This allows us to avoid the overhead of looking
270 * at the inode btree for small block file systems (see xfs_dilocate()).
271 * We can tell whether the inode has been mapped in before by comparing
272 * its disk block address to 0. Only uninitialized inodes will have
273 * 0 for the disk block address.
290 if (ip->i_blkno == (xfs_daddr_t)0) {
292 error = xfs_imap(mp, tp, ip->i_ino, &imap,
293 XFS_IMAP_LOOKUP | imap_flags);
298 * Fill in the fields in the inode that will be used to
299 * map the inode to its buffer from now on.
301 ip->i_blkno = imap.im_blkno;
302 ip->i_len = imap.im_len;
303 ip->i_boffset = imap.im_boffset;
306 * We've already mapped the inode once, so just use the
307 * mapping that we saved the first time.
309 imap.im_blkno = ip->i_blkno;
310 imap.im_len = ip->i_len;
311 imap.im_boffset = ip->i_boffset;
313 ASSERT(bno == 0 || bno == imap.im_blkno);
315 error = xfs_imap_to_bp(mp, tp, &imap, &bp, buf_flags, imap_flags);
320 ASSERT(buf_flags & XFS_BUF_TRYLOCK);
326 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
332 * Move inode type and inode format specific information from the
333 * on-disk inode to the in-core inode. For fifos, devs, and sockets
334 * this means set if_rdev to the proper value. For files, directories,
335 * and symlinks this means to bring in the in-line data or extent
336 * pointers. For a file in B-tree format, only the root is immediately
337 * brought in-core. The rest will be in-lined in if_extents when it
338 * is first referenced (see xfs_iread_extents()).
345 xfs_attr_shortform_t *atp;
349 ip->i_df.if_ext_max =
350 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
353 if (unlikely(be32_to_cpu(dip->di_core.di_nextents) +
354 be16_to_cpu(dip->di_core.di_anextents) >
355 be64_to_cpu(dip->di_core.di_nblocks))) {
356 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
357 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
358 (unsigned long long)ip->i_ino,
359 (int)(be32_to_cpu(dip->di_core.di_nextents) +
360 be16_to_cpu(dip->di_core.di_anextents)),
362 be64_to_cpu(dip->di_core.di_nblocks));
363 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
365 return XFS_ERROR(EFSCORRUPTED);
368 if (unlikely(dip->di_core.di_forkoff > ip->i_mount->m_sb.sb_inodesize)) {
369 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
370 "corrupt dinode %Lu, forkoff = 0x%x.",
371 (unsigned long long)ip->i_ino,
372 dip->di_core.di_forkoff);
373 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
375 return XFS_ERROR(EFSCORRUPTED);
378 switch (ip->i_d.di_mode & S_IFMT) {
383 if (unlikely(dip->di_core.di_format != XFS_DINODE_FMT_DEV)) {
384 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
386 return XFS_ERROR(EFSCORRUPTED);
390 ip->i_df.if_u2.if_rdev = be32_to_cpu(dip->di_u.di_dev);
396 switch (dip->di_core.di_format) {
397 case XFS_DINODE_FMT_LOCAL:
399 * no local regular files yet
401 if (unlikely((be16_to_cpu(dip->di_core.di_mode) & S_IFMT) == S_IFREG)) {
402 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
404 "(local format for regular file).",
405 (unsigned long long) ip->i_ino);
406 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
409 return XFS_ERROR(EFSCORRUPTED);
412 di_size = be64_to_cpu(dip->di_core.di_size);
413 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
414 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
416 "(bad size %Ld for local inode).",
417 (unsigned long long) ip->i_ino,
418 (long long) di_size);
419 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
422 return XFS_ERROR(EFSCORRUPTED);
426 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
428 case XFS_DINODE_FMT_EXTENTS:
429 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
431 case XFS_DINODE_FMT_BTREE:
432 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
435 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
437 return XFS_ERROR(EFSCORRUPTED);
442 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
443 return XFS_ERROR(EFSCORRUPTED);
448 if (!XFS_DFORK_Q(dip))
450 ASSERT(ip->i_afp == NULL);
451 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
452 ip->i_afp->if_ext_max =
453 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
454 switch (dip->di_core.di_aformat) {
455 case XFS_DINODE_FMT_LOCAL:
456 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
457 size = be16_to_cpu(atp->hdr.totsize);
458 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
460 case XFS_DINODE_FMT_EXTENTS:
461 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
463 case XFS_DINODE_FMT_BTREE:
464 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
467 error = XFS_ERROR(EFSCORRUPTED);
471 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
473 xfs_idestroy_fork(ip, XFS_DATA_FORK);
479 * The file is in-lined in the on-disk inode.
480 * If it fits into if_inline_data, then copy
481 * it there, otherwise allocate a buffer for it
482 * and copy the data there. Either way, set
483 * if_data to point at the data.
484 * If we allocate a buffer for the data, make
485 * sure that its size is a multiple of 4 and
486 * record the real size in i_real_bytes.
499 * If the size is unreasonable, then something
500 * is wrong and we just bail out rather than crash in
501 * kmem_alloc() or memcpy() below.
503 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
504 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
506 "(bad size %d for local fork, size = %d).",
507 (unsigned long long) ip->i_ino, size,
508 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
509 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
511 return XFS_ERROR(EFSCORRUPTED);
513 ifp = XFS_IFORK_PTR(ip, whichfork);
516 ifp->if_u1.if_data = NULL;
517 else if (size <= sizeof(ifp->if_u2.if_inline_data))
518 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
520 real_size = roundup(size, 4);
521 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
523 ifp->if_bytes = size;
524 ifp->if_real_bytes = real_size;
526 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
527 ifp->if_flags &= ~XFS_IFEXTENTS;
528 ifp->if_flags |= XFS_IFINLINE;
533 * The file consists of a set of extents all
534 * of which fit into the on-disk inode.
535 * If there are few enough extents to fit into
536 * the if_inline_ext, then copy them there.
537 * Otherwise allocate a buffer for them and copy
538 * them into it. Either way, set if_extents
539 * to point at the extents.
553 ifp = XFS_IFORK_PTR(ip, whichfork);
554 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
555 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
558 * If the number of extents is unreasonable, then something
559 * is wrong and we just bail out rather than crash in
560 * kmem_alloc() or memcpy() below.
562 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
563 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
564 "corrupt inode %Lu ((a)extents = %d).",
565 (unsigned long long) ip->i_ino, nex);
566 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
568 return XFS_ERROR(EFSCORRUPTED);
571 ifp->if_real_bytes = 0;
573 ifp->if_u1.if_extents = NULL;
574 else if (nex <= XFS_INLINE_EXTS)
575 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
577 xfs_iext_add(ifp, 0, nex);
579 ifp->if_bytes = size;
581 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
582 xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip));
583 for (i = 0; i < nex; i++, dp++) {
584 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
585 ep->l0 = get_unaligned_be64(&dp->l0);
586 ep->l1 = get_unaligned_be64(&dp->l1);
588 XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork);
589 if (whichfork != XFS_DATA_FORK ||
590 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
591 if (unlikely(xfs_check_nostate_extents(
593 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
596 return XFS_ERROR(EFSCORRUPTED);
599 ifp->if_flags |= XFS_IFEXTENTS;
604 * The file has too many extents to fit into
605 * the inode, so they are in B-tree format.
606 * Allocate a buffer for the root of the B-tree
607 * and copy the root into it. The i_extents
608 * field will remain NULL until all of the
609 * extents are read in (when they are needed).
617 xfs_bmdr_block_t *dfp;
623 ifp = XFS_IFORK_PTR(ip, whichfork);
624 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
625 size = XFS_BMAP_BROOT_SPACE(dfp);
626 nrecs = be16_to_cpu(dfp->bb_numrecs);
629 * blow out if -- fork has less extents than can fit in
630 * fork (fork shouldn't be a btree format), root btree
631 * block has more records than can fit into the fork,
632 * or the number of extents is greater than the number of
635 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
636 || XFS_BMDR_SPACE_CALC(nrecs) >
637 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
638 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
639 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
640 "corrupt inode %Lu (btree).",
641 (unsigned long long) ip->i_ino);
642 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
644 return XFS_ERROR(EFSCORRUPTED);
647 ifp->if_broot_bytes = size;
648 ifp->if_broot = kmem_alloc(size, KM_SLEEP);
649 ASSERT(ifp->if_broot != NULL);
651 * Copy and convert from the on-disk structure
652 * to the in-memory structure.
654 xfs_bmdr_to_bmbt(ip->i_mount, dfp,
655 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
656 ifp->if_broot, size);
657 ifp->if_flags &= ~XFS_IFEXTENTS;
658 ifp->if_flags |= XFS_IFBROOT;
664 xfs_dinode_from_disk(
666 xfs_dinode_core_t *from)
668 to->di_magic = be16_to_cpu(from->di_magic);
669 to->di_mode = be16_to_cpu(from->di_mode);
670 to->di_version = from ->di_version;
671 to->di_format = from->di_format;
672 to->di_onlink = be16_to_cpu(from->di_onlink);
673 to->di_uid = be32_to_cpu(from->di_uid);
674 to->di_gid = be32_to_cpu(from->di_gid);
675 to->di_nlink = be32_to_cpu(from->di_nlink);
676 to->di_projid = be16_to_cpu(from->di_projid);
677 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
678 to->di_flushiter = be16_to_cpu(from->di_flushiter);
679 to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec);
680 to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec);
681 to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec);
682 to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec);
683 to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec);
684 to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec);
685 to->di_size = be64_to_cpu(from->di_size);
686 to->di_nblocks = be64_to_cpu(from->di_nblocks);
687 to->di_extsize = be32_to_cpu(from->di_extsize);
688 to->di_nextents = be32_to_cpu(from->di_nextents);
689 to->di_anextents = be16_to_cpu(from->di_anextents);
690 to->di_forkoff = from->di_forkoff;
691 to->di_aformat = from->di_aformat;
692 to->di_dmevmask = be32_to_cpu(from->di_dmevmask);
693 to->di_dmstate = be16_to_cpu(from->di_dmstate);
694 to->di_flags = be16_to_cpu(from->di_flags);
695 to->di_gen = be32_to_cpu(from->di_gen);
700 xfs_dinode_core_t *to,
701 xfs_icdinode_t *from)
703 to->di_magic = cpu_to_be16(from->di_magic);
704 to->di_mode = cpu_to_be16(from->di_mode);
705 to->di_version = from ->di_version;
706 to->di_format = from->di_format;
707 to->di_onlink = cpu_to_be16(from->di_onlink);
708 to->di_uid = cpu_to_be32(from->di_uid);
709 to->di_gid = cpu_to_be32(from->di_gid);
710 to->di_nlink = cpu_to_be32(from->di_nlink);
711 to->di_projid = cpu_to_be16(from->di_projid);
712 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
713 to->di_flushiter = cpu_to_be16(from->di_flushiter);
714 to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec);
715 to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec);
716 to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec);
717 to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec);
718 to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec);
719 to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec);
720 to->di_size = cpu_to_be64(from->di_size);
721 to->di_nblocks = cpu_to_be64(from->di_nblocks);
722 to->di_extsize = cpu_to_be32(from->di_extsize);
723 to->di_nextents = cpu_to_be32(from->di_nextents);
724 to->di_anextents = cpu_to_be16(from->di_anextents);
725 to->di_forkoff = from->di_forkoff;
726 to->di_aformat = from->di_aformat;
727 to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
728 to->di_dmstate = cpu_to_be16(from->di_dmstate);
729 to->di_flags = cpu_to_be16(from->di_flags);
730 to->di_gen = cpu_to_be32(from->di_gen);
739 if (di_flags & XFS_DIFLAG_ANY) {
740 if (di_flags & XFS_DIFLAG_REALTIME)
741 flags |= XFS_XFLAG_REALTIME;
742 if (di_flags & XFS_DIFLAG_PREALLOC)
743 flags |= XFS_XFLAG_PREALLOC;
744 if (di_flags & XFS_DIFLAG_IMMUTABLE)
745 flags |= XFS_XFLAG_IMMUTABLE;
746 if (di_flags & XFS_DIFLAG_APPEND)
747 flags |= XFS_XFLAG_APPEND;
748 if (di_flags & XFS_DIFLAG_SYNC)
749 flags |= XFS_XFLAG_SYNC;
750 if (di_flags & XFS_DIFLAG_NOATIME)
751 flags |= XFS_XFLAG_NOATIME;
752 if (di_flags & XFS_DIFLAG_NODUMP)
753 flags |= XFS_XFLAG_NODUMP;
754 if (di_flags & XFS_DIFLAG_RTINHERIT)
755 flags |= XFS_XFLAG_RTINHERIT;
756 if (di_flags & XFS_DIFLAG_PROJINHERIT)
757 flags |= XFS_XFLAG_PROJINHERIT;
758 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
759 flags |= XFS_XFLAG_NOSYMLINKS;
760 if (di_flags & XFS_DIFLAG_EXTSIZE)
761 flags |= XFS_XFLAG_EXTSIZE;
762 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
763 flags |= XFS_XFLAG_EXTSZINHERIT;
764 if (di_flags & XFS_DIFLAG_NODEFRAG)
765 flags |= XFS_XFLAG_NODEFRAG;
766 if (di_flags & XFS_DIFLAG_FILESTREAM)
767 flags |= XFS_XFLAG_FILESTREAM;
777 xfs_icdinode_t *dic = &ip->i_d;
779 return _xfs_dic2xflags(dic->di_flags) |
780 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
787 xfs_dinode_core_t *dic = &dip->di_core;
789 return _xfs_dic2xflags(be16_to_cpu(dic->di_flags)) |
790 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
794 * Allocate and initialise an xfs_inode.
796 STATIC struct xfs_inode *
798 struct xfs_mount *mp,
801 struct xfs_inode *ip;
804 * if this didn't occur in transactions, we could use
805 * KM_MAYFAIL and return NULL here on ENOMEM. Set the
806 * code up to do this anyway.
808 ip = kmem_zone_alloc(xfs_inode_zone, KM_SLEEP);
812 ASSERT(atomic_read(&ip->i_iocount) == 0);
813 ASSERT(atomic_read(&ip->i_pincount) == 0);
814 ASSERT(!spin_is_locked(&ip->i_flags_lock));
815 ASSERT(completion_done(&ip->i_flush));
818 * initialise the VFS inode here to get failures
819 * out of the way early.
821 if (!inode_init_always(mp->m_super, VFS_I(ip))) {
822 kmem_zone_free(xfs_inode_zone, ip);
826 /* initialise the xfs inode */
833 memset(&ip->i_df, 0, sizeof(xfs_ifork_t));
835 ip->i_update_core = 0;
836 ip->i_update_size = 0;
837 ip->i_delayed_blks = 0;
838 memset(&ip->i_d, 0, sizeof(xfs_icdinode_t));
843 * Initialize inode's trace buffers.
845 #ifdef XFS_INODE_TRACE
846 ip->i_trace = ktrace_alloc(INODE_TRACE_SIZE, KM_NOFS);
848 #ifdef XFS_BMAP_TRACE
849 ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_NOFS);
851 #ifdef XFS_BTREE_TRACE
852 ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_NOFS);
855 ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_NOFS);
857 #ifdef XFS_ILOCK_TRACE
858 ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_NOFS);
860 #ifdef XFS_DIR2_TRACE
861 ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_NOFS);
868 * Given a mount structure and an inode number, return a pointer
869 * to a newly allocated in-core inode corresponding to the given
872 * Initialize the inode's attributes and extent pointers if it
873 * already has them (it will not if the inode has no links).
889 ip = xfs_inode_alloc(mp, ino);
894 * Get pointer's to the on-disk inode and the buffer containing it.
895 * If the inode number refers to a block outside the file system
896 * then xfs_itobp() will return NULL. In this case we should
897 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
898 * know that this is a new incore inode.
900 error = xfs_itobp(mp, tp, ip, &dip, &bp, bno, imap_flags, XFS_BUF_LOCK);
902 goto out_destroy_inode;
905 * If we got something that isn't an inode it means someone
906 * (nfs or dmi) has a stale handle.
908 if (be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC) {
910 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
911 "dip->di_core.di_magic (0x%x) != "
912 "XFS_DINODE_MAGIC (0x%x)",
913 be16_to_cpu(dip->di_core.di_magic),
916 error = XFS_ERROR(EINVAL);
921 * If the on-disk inode is already linked to a directory
922 * entry, copy all of the inode into the in-core inode.
923 * xfs_iformat() handles copying in the inode format
924 * specific information.
925 * Otherwise, just get the truly permanent information.
927 if (dip->di_core.di_mode) {
928 xfs_dinode_from_disk(&ip->i_d, &dip->di_core);
929 error = xfs_iformat(ip, dip);
932 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
933 "xfs_iformat() returned error %d",
939 ip->i_d.di_magic = be16_to_cpu(dip->di_core.di_magic);
940 ip->i_d.di_version = dip->di_core.di_version;
941 ip->i_d.di_gen = be32_to_cpu(dip->di_core.di_gen);
942 ip->i_d.di_flushiter = be16_to_cpu(dip->di_core.di_flushiter);
944 * Make sure to pull in the mode here as well in
945 * case the inode is released without being used.
946 * This ensures that xfs_inactive() will see that
947 * the inode is already free and not try to mess
948 * with the uninitialized part of it.
952 * Initialize the per-fork minima and maxima for a new
953 * inode here. xfs_iformat will do it for old inodes.
955 ip->i_df.if_ext_max =
956 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
960 * The inode format changed when we moved the link count and
961 * made it 32 bits long. If this is an old format inode,
962 * convert it in memory to look like a new one. If it gets
963 * flushed to disk we will convert back before flushing or
964 * logging it. We zero out the new projid field and the old link
965 * count field. We'll handle clearing the pad field (the remains
966 * of the old uuid field) when we actually convert the inode to
967 * the new format. We don't change the version number so that we
968 * can distinguish this from a real new format inode.
970 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
971 ip->i_d.di_nlink = ip->i_d.di_onlink;
972 ip->i_d.di_onlink = 0;
973 ip->i_d.di_projid = 0;
976 ip->i_delayed_blks = 0;
977 ip->i_size = ip->i_d.di_size;
980 * Mark the buffer containing the inode as something to keep
981 * around for a while. This helps to keep recently accessed
982 * meta-data in-core longer.
984 XFS_BUF_SET_REF(bp, XFS_INO_REF);
987 * Use xfs_trans_brelse() to release the buffer containing the
988 * on-disk inode, because it was acquired with xfs_trans_read_buf()
989 * in xfs_itobp() above. If tp is NULL, this is just a normal
990 * brelse(). If we're within a transaction, then xfs_trans_brelse()
991 * will only release the buffer if it is not dirty within the
992 * transaction. It will be OK to release the buffer in this case,
993 * because inodes on disk are never destroyed and we will be
994 * locking the new in-core inode before putting it in the hash
995 * table where other processes can find it. Thus we don't have
996 * to worry about the inode being changed just because we released
999 xfs_trans_brelse(tp, bp);
1004 xfs_trans_brelse(tp, bp);
1006 xfs_destroy_inode(ip);
1011 * Read in extents from a btree-format inode.
1012 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1022 xfs_extnum_t nextents;
1025 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
1026 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
1028 return XFS_ERROR(EFSCORRUPTED);
1030 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
1031 size = nextents * sizeof(xfs_bmbt_rec_t);
1032 ifp = XFS_IFORK_PTR(ip, whichfork);
1035 * We know that the size is valid (it's checked in iformat_btree)
1037 ifp->if_lastex = NULLEXTNUM;
1038 ifp->if_bytes = ifp->if_real_bytes = 0;
1039 ifp->if_flags |= XFS_IFEXTENTS;
1040 xfs_iext_add(ifp, 0, nextents);
1041 error = xfs_bmap_read_extents(tp, ip, whichfork);
1043 xfs_iext_destroy(ifp);
1044 ifp->if_flags &= ~XFS_IFEXTENTS;
1047 xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip));
1052 * Allocate an inode on disk and return a copy of its in-core version.
1053 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1054 * appropriately within the inode. The uid and gid for the inode are
1055 * set according to the contents of the given cred structure.
1057 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1058 * has a free inode available, call xfs_iget()
1059 * to obtain the in-core version of the allocated inode. Finally,
1060 * fill in the inode and log its initial contents. In this case,
1061 * ialloc_context would be set to NULL and call_again set to false.
1063 * If xfs_dialloc() does not have an available inode,
1064 * it will replenish its supply by doing an allocation. Since we can
1065 * only do one allocation within a transaction without deadlocks, we
1066 * must commit the current transaction before returning the inode itself.
1067 * In this case, therefore, we will set call_again to true and return.
1068 * The caller should then commit the current transaction, start a new
1069 * transaction, and call xfs_ialloc() again to actually get the inode.
1071 * To ensure that some other process does not grab the inode that
1072 * was allocated during the first call to xfs_ialloc(), this routine
1073 * also returns the [locked] bp pointing to the head of the freelist
1074 * as ialloc_context. The caller should hold this buffer across
1075 * the commit and pass it back into this routine on the second call.
1077 * If we are allocating quota inodes, we do not have a parent inode
1078 * to attach to or associate with (i.e. pip == NULL) because they
1079 * are not linked into the directory structure - they are attached
1080 * directly to the superblock - and so have no parent.
1092 xfs_buf_t **ialloc_context,
1093 boolean_t *call_again,
1101 int filestreams = 0;
1104 * Call the space management code to pick
1105 * the on-disk inode to be allocated.
1107 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
1108 ialloc_context, call_again, &ino);
1111 if (*call_again || ino == NULLFSINO) {
1115 ASSERT(*ialloc_context == NULL);
1118 * Get the in-core inode with the lock held exclusively.
1119 * This is because we're setting fields here we need
1120 * to prevent others from looking at until we're done.
1122 error = xfs_trans_iget(tp->t_mountp, tp, ino,
1123 XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
1128 ip->i_d.di_mode = (__uint16_t)mode;
1129 ip->i_d.di_onlink = 0;
1130 ip->i_d.di_nlink = nlink;
1131 ASSERT(ip->i_d.di_nlink == nlink);
1132 ip->i_d.di_uid = current_fsuid();
1133 ip->i_d.di_gid = current_fsgid();
1134 ip->i_d.di_projid = prid;
1135 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1138 * If the superblock version is up to where we support new format
1139 * inodes and this is currently an old format inode, then change
1140 * the inode version number now. This way we only do the conversion
1141 * here rather than here and in the flush/logging code.
1143 if (xfs_sb_version_hasnlink(&tp->t_mountp->m_sb) &&
1144 ip->i_d.di_version == XFS_DINODE_VERSION_1) {
1145 ip->i_d.di_version = XFS_DINODE_VERSION_2;
1147 * We've already zeroed the old link count, the projid field,
1148 * and the pad field.
1153 * Project ids won't be stored on disk if we are using a version 1 inode.
1155 if ((prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1))
1156 xfs_bump_ino_vers2(tp, ip);
1158 if (pip && XFS_INHERIT_GID(pip)) {
1159 ip->i_d.di_gid = pip->i_d.di_gid;
1160 if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1161 ip->i_d.di_mode |= S_ISGID;
1166 * If the group ID of the new file does not match the effective group
1167 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1168 * (and only if the irix_sgid_inherit compatibility variable is set).
1170 if ((irix_sgid_inherit) &&
1171 (ip->i_d.di_mode & S_ISGID) &&
1172 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1173 ip->i_d.di_mode &= ~S_ISGID;
1176 ip->i_d.di_size = 0;
1178 ip->i_d.di_nextents = 0;
1179 ASSERT(ip->i_d.di_nblocks == 0);
1182 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
1183 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
1184 ip->i_d.di_atime = ip->i_d.di_mtime;
1185 ip->i_d.di_ctime = ip->i_d.di_mtime;
1188 * di_gen will have been taken care of in xfs_iread.
1190 ip->i_d.di_extsize = 0;
1191 ip->i_d.di_dmevmask = 0;
1192 ip->i_d.di_dmstate = 0;
1193 ip->i_d.di_flags = 0;
1194 flags = XFS_ILOG_CORE;
1195 switch (mode & S_IFMT) {
1200 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1201 ip->i_df.if_u2.if_rdev = rdev;
1202 ip->i_df.if_flags = 0;
1203 flags |= XFS_ILOG_DEV;
1207 * we can't set up filestreams until after the VFS inode
1208 * is set up properly.
1210 if (pip && xfs_inode_is_filestream(pip))
1214 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
1217 if ((mode & S_IFMT) == S_IFDIR) {
1218 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1219 di_flags |= XFS_DIFLAG_RTINHERIT;
1220 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1221 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1222 ip->i_d.di_extsize = pip->i_d.di_extsize;
1224 } else if ((mode & S_IFMT) == S_IFREG) {
1225 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1226 di_flags |= XFS_DIFLAG_REALTIME;
1227 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1228 di_flags |= XFS_DIFLAG_EXTSIZE;
1229 ip->i_d.di_extsize = pip->i_d.di_extsize;
1232 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1233 xfs_inherit_noatime)
1234 di_flags |= XFS_DIFLAG_NOATIME;
1235 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1237 di_flags |= XFS_DIFLAG_NODUMP;
1238 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1240 di_flags |= XFS_DIFLAG_SYNC;
1241 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1242 xfs_inherit_nosymlinks)
1243 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1244 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1245 di_flags |= XFS_DIFLAG_PROJINHERIT;
1246 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1247 xfs_inherit_nodefrag)
1248 di_flags |= XFS_DIFLAG_NODEFRAG;
1249 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
1250 di_flags |= XFS_DIFLAG_FILESTREAM;
1251 ip->i_d.di_flags |= di_flags;
1255 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1256 ip->i_df.if_flags = XFS_IFEXTENTS;
1257 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1258 ip->i_df.if_u1.if_extents = NULL;
1264 * Attribute fork settings for new inode.
1266 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1267 ip->i_d.di_anextents = 0;
1270 * Log the new values stuffed into the inode.
1272 xfs_trans_log_inode(tp, ip, flags);
1274 /* now that we have an i_mode we can setup inode ops and unlock */
1275 xfs_setup_inode(ip);
1277 /* now we have set up the vfs inode we can associate the filestream */
1279 error = xfs_filestream_associate(pip, ip);
1283 xfs_iflags_set(ip, XFS_IFILESTREAM);
1291 * Check to make sure that there are no blocks allocated to the
1292 * file beyond the size of the file. We don't check this for
1293 * files with fixed size extents or real time extents, but we
1294 * at least do it for regular files.
1303 xfs_fileoff_t map_first;
1305 xfs_bmbt_irec_t imaps[2];
1307 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1310 if (XFS_IS_REALTIME_INODE(ip))
1313 if (ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE)
1317 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1319 * The filesystem could be shutting down, so bmapi may return
1322 if (xfs_bmapi(NULL, ip, map_first,
1324 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1326 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1329 ASSERT(nimaps == 1);
1330 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1335 * Calculate the last possible buffered byte in a file. This must
1336 * include data that was buffered beyond the EOF by the write code.
1337 * This also needs to deal with overflowing the xfs_fsize_t type
1338 * which can happen for sizes near the limit.
1340 * We also need to take into account any blocks beyond the EOF. It
1341 * may be the case that they were buffered by a write which failed.
1342 * In that case the pages will still be in memory, but the inode size
1343 * will never have been updated.
1350 xfs_fsize_t last_byte;
1351 xfs_fileoff_t last_block;
1352 xfs_fileoff_t size_last_block;
1355 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED));
1359 * Only check for blocks beyond the EOF if the extents have
1360 * been read in. This eliminates the need for the inode lock,
1361 * and it also saves us from looking when it really isn't
1364 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1365 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1373 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_size);
1374 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1376 last_byte = XFS_FSB_TO_B(mp, last_block);
1377 if (last_byte < 0) {
1378 return XFS_MAXIOFFSET(mp);
1380 last_byte += (1 << mp->m_writeio_log);
1381 if (last_byte < 0) {
1382 return XFS_MAXIOFFSET(mp);
1387 #if defined(XFS_RW_TRACE)
1393 xfs_fsize_t new_size,
1394 xfs_off_t toss_start,
1395 xfs_off_t toss_finish)
1397 if (ip->i_rwtrace == NULL) {
1401 ktrace_enter(ip->i_rwtrace,
1404 (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
1405 (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
1406 (void*)((long)flag),
1407 (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
1408 (void*)(unsigned long)(new_size & 0xffffffff),
1409 (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
1410 (void*)(unsigned long)(toss_start & 0xffffffff),
1411 (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
1412 (void*)(unsigned long)(toss_finish & 0xffffffff),
1413 (void*)(unsigned long)current_cpu(),
1414 (void*)(unsigned long)current_pid(),
1420 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1424 * Start the truncation of the file to new_size. The new size
1425 * must be smaller than the current size. This routine will
1426 * clear the buffer and page caches of file data in the removed
1427 * range, and xfs_itruncate_finish() will remove the underlying
1430 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1431 * must NOT have the inode lock held at all. This is because we're
1432 * calling into the buffer/page cache code and we can't hold the
1433 * inode lock when we do so.
1435 * We need to wait for any direct I/Os in flight to complete before we
1436 * proceed with the truncate. This is needed to prevent the extents
1437 * being read or written by the direct I/Os from being removed while the
1438 * I/O is in flight as there is no other method of synchronising
1439 * direct I/O with the truncate operation. Also, because we hold
1440 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1441 * started until the truncate completes and drops the lock. Essentially,
1442 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1443 * between direct I/Os and the truncate operation.
1445 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1446 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1447 * in the case that the caller is locking things out of order and
1448 * may not be able to call xfs_itruncate_finish() with the inode lock
1449 * held without dropping the I/O lock. If the caller must drop the
1450 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1451 * must be called again with all the same restrictions as the initial
1455 xfs_itruncate_start(
1458 xfs_fsize_t new_size)
1460 xfs_fsize_t last_byte;
1461 xfs_off_t toss_start;
1465 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1466 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1467 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1468 (flags == XFS_ITRUNC_MAYBE));
1472 /* wait for the completion of any pending DIOs */
1473 if (new_size == 0 || new_size < ip->i_size)
1477 * Call toss_pages or flushinval_pages to get rid of pages
1478 * overlapping the region being removed. We have to use
1479 * the less efficient flushinval_pages in the case that the
1480 * caller may not be able to finish the truncate without
1481 * dropping the inode's I/O lock. Make sure
1482 * to catch any pages brought in by buffers overlapping
1483 * the EOF by searching out beyond the isize by our
1484 * block size. We round new_size up to a block boundary
1485 * so that we don't toss things on the same block as
1486 * new_size but before it.
1488 * Before calling toss_page or flushinval_pages, make sure to
1489 * call remapf() over the same region if the file is mapped.
1490 * This frees up mapped file references to the pages in the
1491 * given range and for the flushinval_pages case it ensures
1492 * that we get the latest mapped changes flushed out.
1494 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1495 toss_start = XFS_FSB_TO_B(mp, toss_start);
1496 if (toss_start < 0) {
1498 * The place to start tossing is beyond our maximum
1499 * file size, so there is no way that the data extended
1504 last_byte = xfs_file_last_byte(ip);
1505 xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
1507 if (last_byte > toss_start) {
1508 if (flags & XFS_ITRUNC_DEFINITE) {
1509 xfs_tosspages(ip, toss_start,
1510 -1, FI_REMAPF_LOCKED);
1512 error = xfs_flushinval_pages(ip, toss_start,
1513 -1, FI_REMAPF_LOCKED);
1518 if (new_size == 0) {
1519 ASSERT(VN_CACHED(VFS_I(ip)) == 0);
1526 * Shrink the file to the given new_size. The new size must be smaller than
1527 * the current size. This will free up the underlying blocks in the removed
1528 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
1530 * The transaction passed to this routine must have made a permanent log
1531 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1532 * given transaction and start new ones, so make sure everything involved in
1533 * the transaction is tidy before calling here. Some transaction will be
1534 * returned to the caller to be committed. The incoming transaction must
1535 * already include the inode, and both inode locks must be held exclusively.
1536 * The inode must also be "held" within the transaction. On return the inode
1537 * will be "held" within the returned transaction. This routine does NOT
1538 * require any disk space to be reserved for it within the transaction.
1540 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1541 * indicates the fork which is to be truncated. For the attribute fork we only
1542 * support truncation to size 0.
1544 * We use the sync parameter to indicate whether or not the first transaction
1545 * we perform might have to be synchronous. For the attr fork, it needs to be
1546 * so if the unlink of the inode is not yet known to be permanent in the log.
1547 * This keeps us from freeing and reusing the blocks of the attribute fork
1548 * before the unlink of the inode becomes permanent.
1550 * For the data fork, we normally have to run synchronously if we're being
1551 * called out of the inactive path or we're being called out of the create path
1552 * where we're truncating an existing file. Either way, the truncate needs to
1553 * be sync so blocks don't reappear in the file with altered data in case of a
1554 * crash. wsync filesystems can run the first case async because anything that
1555 * shrinks the inode has to run sync so by the time we're called here from
1556 * inactive, the inode size is permanently set to 0.
1558 * Calls from the truncate path always need to be sync unless we're in a wsync
1559 * filesystem and the file has already been unlinked.
1561 * The caller is responsible for correctly setting the sync parameter. It gets
1562 * too hard for us to guess here which path we're being called out of just
1563 * based on inode state.
1565 * If we get an error, we must return with the inode locked and linked into the
1566 * current transaction. This keeps things simple for the higher level code,
1567 * because it always knows that the inode is locked and held in the transaction
1568 * that returns to it whether errors occur or not. We don't mark the inode
1569 * dirty on error so that transactions can be easily aborted if possible.
1572 xfs_itruncate_finish(
1575 xfs_fsize_t new_size,
1579 xfs_fsblock_t first_block;
1580 xfs_fileoff_t first_unmap_block;
1581 xfs_fileoff_t last_block;
1582 xfs_filblks_t unmap_len=0;
1587 xfs_bmap_free_t free_list;
1590 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
1591 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1592 ASSERT(*tp != NULL);
1593 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1594 ASSERT(ip->i_transp == *tp);
1595 ASSERT(ip->i_itemp != NULL);
1596 ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
1600 mp = (ntp)->t_mountp;
1601 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1604 * We only support truncating the entire attribute fork.
1606 if (fork == XFS_ATTR_FORK) {
1609 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1610 xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
1612 * The first thing we do is set the size to new_size permanently
1613 * on disk. This way we don't have to worry about anyone ever
1614 * being able to look at the data being freed even in the face
1615 * of a crash. What we're getting around here is the case where
1616 * we free a block, it is allocated to another file, it is written
1617 * to, and then we crash. If the new data gets written to the
1618 * file but the log buffers containing the free and reallocation
1619 * don't, then we'd end up with garbage in the blocks being freed.
1620 * As long as we make the new_size permanent before actually
1621 * freeing any blocks it doesn't matter if they get writtten to.
1623 * The callers must signal into us whether or not the size
1624 * setting here must be synchronous. There are a few cases
1625 * where it doesn't have to be synchronous. Those cases
1626 * occur if the file is unlinked and we know the unlink is
1627 * permanent or if the blocks being truncated are guaranteed
1628 * to be beyond the inode eof (regardless of the link count)
1629 * and the eof value is permanent. Both of these cases occur
1630 * only on wsync-mounted filesystems. In those cases, we're
1631 * guaranteed that no user will ever see the data in the blocks
1632 * that are being truncated so the truncate can run async.
1633 * In the free beyond eof case, the file may wind up with
1634 * more blocks allocated to it than it needs if we crash
1635 * and that won't get fixed until the next time the file
1636 * is re-opened and closed but that's ok as that shouldn't
1637 * be too many blocks.
1639 * However, we can't just make all wsync xactions run async
1640 * because there's one call out of the create path that needs
1641 * to run sync where it's truncating an existing file to size
1642 * 0 whose size is > 0.
1644 * It's probably possible to come up with a test in this
1645 * routine that would correctly distinguish all the above
1646 * cases from the values of the function parameters and the
1647 * inode state but for sanity's sake, I've decided to let the
1648 * layers above just tell us. It's simpler to correctly figure
1649 * out in the layer above exactly under what conditions we
1650 * can run async and I think it's easier for others read and
1651 * follow the logic in case something has to be changed.
1652 * cscope is your friend -- rcc.
1654 * The attribute fork is much simpler.
1656 * For the attribute fork we allow the caller to tell us whether
1657 * the unlink of the inode that led to this call is yet permanent
1658 * in the on disk log. If it is not and we will be freeing extents
1659 * in this inode then we make the first transaction synchronous
1660 * to make sure that the unlink is permanent by the time we free
1663 if (fork == XFS_DATA_FORK) {
1664 if (ip->i_d.di_nextents > 0) {
1666 * If we are not changing the file size then do
1667 * not update the on-disk file size - we may be
1668 * called from xfs_inactive_free_eofblocks(). If we
1669 * update the on-disk file size and then the system
1670 * crashes before the contents of the file are
1671 * flushed to disk then the files may be full of
1672 * holes (ie NULL files bug).
1674 if (ip->i_size != new_size) {
1675 ip->i_d.di_size = new_size;
1676 ip->i_size = new_size;
1677 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1681 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1682 if (ip->i_d.di_anextents > 0)
1683 xfs_trans_set_sync(ntp);
1685 ASSERT(fork == XFS_DATA_FORK ||
1686 (fork == XFS_ATTR_FORK &&
1687 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1688 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1691 * Since it is possible for space to become allocated beyond
1692 * the end of the file (in a crash where the space is allocated
1693 * but the inode size is not yet updated), simply remove any
1694 * blocks which show up between the new EOF and the maximum
1695 * possible file size. If the first block to be removed is
1696 * beyond the maximum file size (ie it is the same as last_block),
1697 * then there is nothing to do.
1699 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1700 ASSERT(first_unmap_block <= last_block);
1702 if (last_block == first_unmap_block) {
1705 unmap_len = last_block - first_unmap_block + 1;
1709 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1710 * will tell us whether it freed the entire range or
1711 * not. If this is a synchronous mount (wsync),
1712 * then we can tell bunmapi to keep all the
1713 * transactions asynchronous since the unlink
1714 * transaction that made this inode inactive has
1715 * already hit the disk. There's no danger of
1716 * the freed blocks being reused, there being a
1717 * crash, and the reused blocks suddenly reappearing
1718 * in this file with garbage in them once recovery
1721 XFS_BMAP_INIT(&free_list, &first_block);
1722 error = xfs_bunmapi(ntp, ip,
1723 first_unmap_block, unmap_len,
1724 XFS_BMAPI_AFLAG(fork) |
1725 (sync ? 0 : XFS_BMAPI_ASYNC),
1726 XFS_ITRUNC_MAX_EXTENTS,
1727 &first_block, &free_list,
1731 * If the bunmapi call encounters an error,
1732 * return to the caller where the transaction
1733 * can be properly aborted. We just need to
1734 * make sure we're not holding any resources
1735 * that we were not when we came in.
1737 xfs_bmap_cancel(&free_list);
1742 * Duplicate the transaction that has the permanent
1743 * reservation and commit the old transaction.
1745 error = xfs_bmap_finish(tp, &free_list, &committed);
1748 /* link the inode into the next xact in the chain */
1749 xfs_trans_ijoin(ntp, ip,
1750 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1751 xfs_trans_ihold(ntp, ip);
1756 * If the bmap finish call encounters an error, return
1757 * to the caller where the transaction can be properly
1758 * aborted. We just need to make sure we're not
1759 * holding any resources that we were not when we came
1762 * Aborting from this point might lose some blocks in
1763 * the file system, but oh well.
1765 xfs_bmap_cancel(&free_list);
1771 * Mark the inode dirty so it will be logged and
1772 * moved forward in the log as part of every commit.
1774 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1777 ntp = xfs_trans_dup(ntp);
1778 error = xfs_trans_commit(*tp, 0);
1781 /* link the inode into the next transaction in the chain */
1782 xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1783 xfs_trans_ihold(ntp, ip);
1788 * transaction commit worked ok so we can drop the extra ticket
1789 * reference that we gained in xfs_trans_dup()
1791 xfs_log_ticket_put(ntp->t_ticket);
1792 error = xfs_trans_reserve(ntp, 0,
1793 XFS_ITRUNCATE_LOG_RES(mp), 0,
1794 XFS_TRANS_PERM_LOG_RES,
1795 XFS_ITRUNCATE_LOG_COUNT);
1800 * Only update the size in the case of the data fork, but
1801 * always re-log the inode so that our permanent transaction
1802 * can keep on rolling it forward in the log.
1804 if (fork == XFS_DATA_FORK) {
1805 xfs_isize_check(mp, ip, new_size);
1807 * If we are not changing the file size then do
1808 * not update the on-disk file size - we may be
1809 * called from xfs_inactive_free_eofblocks(). If we
1810 * update the on-disk file size and then the system
1811 * crashes before the contents of the file are
1812 * flushed to disk then the files may be full of
1813 * holes (ie NULL files bug).
1815 if (ip->i_size != new_size) {
1816 ip->i_d.di_size = new_size;
1817 ip->i_size = new_size;
1820 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1821 ASSERT((new_size != 0) ||
1822 (fork == XFS_ATTR_FORK) ||
1823 (ip->i_delayed_blks == 0));
1824 ASSERT((new_size != 0) ||
1825 (fork == XFS_ATTR_FORK) ||
1826 (ip->i_d.di_nextents == 0));
1827 xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
1832 * This is called when the inode's link count goes to 0.
1833 * We place the on-disk inode on a list in the AGI. It
1834 * will be pulled from this list when the inode is freed.
1846 xfs_agnumber_t agno;
1847 xfs_daddr_t agdaddr;
1854 ASSERT(ip->i_d.di_nlink == 0);
1855 ASSERT(ip->i_d.di_mode != 0);
1856 ASSERT(ip->i_transp == tp);
1860 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1861 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1864 * Get the agi buffer first. It ensures lock ordering
1867 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1868 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1873 * Validate the magic number of the agi block.
1875 agi = XFS_BUF_TO_AGI(agibp);
1877 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1878 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
1879 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK,
1880 XFS_RANDOM_IUNLINK))) {
1881 XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi);
1882 xfs_trans_brelse(tp, agibp);
1883 return XFS_ERROR(EFSCORRUPTED);
1886 * Get the index into the agi hash table for the
1887 * list this inode will go on.
1889 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1891 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1892 ASSERT(agi->agi_unlinked[bucket_index]);
1893 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1895 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
1897 * There is already another inode in the bucket we need
1898 * to add ourselves to. Add us at the front of the list.
1899 * Here we put the head pointer into our next pointer,
1900 * and then we fall through to point the head at us.
1902 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
1906 ASSERT(be32_to_cpu(dip->di_next_unlinked) == NULLAGINO);
1907 /* both on-disk, don't endian flip twice */
1908 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1909 offset = ip->i_boffset +
1910 offsetof(xfs_dinode_t, di_next_unlinked);
1911 xfs_trans_inode_buf(tp, ibp);
1912 xfs_trans_log_buf(tp, ibp, offset,
1913 (offset + sizeof(xfs_agino_t) - 1));
1914 xfs_inobp_check(mp, ibp);
1918 * Point the bucket head pointer at the inode being inserted.
1921 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1922 offset = offsetof(xfs_agi_t, agi_unlinked) +
1923 (sizeof(xfs_agino_t) * bucket_index);
1924 xfs_trans_log_buf(tp, agibp, offset,
1925 (offset + sizeof(xfs_agino_t) - 1));
1930 * Pull the on-disk inode from the AGI unlinked list.
1943 xfs_agnumber_t agno;
1944 xfs_daddr_t agdaddr;
1946 xfs_agino_t next_agino;
1947 xfs_buf_t *last_ibp;
1948 xfs_dinode_t *last_dip = NULL;
1950 int offset, last_offset = 0;
1955 * First pull the on-disk inode from the AGI unlinked list.
1959 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1960 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1963 * Get the agi buffer first. It ensures lock ordering
1966 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1967 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1970 "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
1971 error, mp->m_fsname);
1975 * Validate the magic number of the agi block.
1977 agi = XFS_BUF_TO_AGI(agibp);
1979 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1980 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
1981 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE,
1982 XFS_RANDOM_IUNLINK_REMOVE))) {
1983 XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW,
1985 xfs_trans_brelse(tp, agibp);
1987 "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
1989 return XFS_ERROR(EFSCORRUPTED);
1992 * Get the index into the agi hash table for the
1993 * list this inode will go on.
1995 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1997 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1998 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
1999 ASSERT(agi->agi_unlinked[bucket_index]);
2001 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2003 * We're at the head of the list. Get the inode's
2004 * on-disk buffer to see if there is anyone after us
2005 * on the list. Only modify our next pointer if it
2006 * is not already NULLAGINO. This saves us the overhead
2007 * of dealing with the buffer when there is no need to
2010 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
2013 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2014 error, mp->m_fsname);
2017 next_agino = be32_to_cpu(dip->di_next_unlinked);
2018 ASSERT(next_agino != 0);
2019 if (next_agino != NULLAGINO) {
2020 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2021 offset = ip->i_boffset +
2022 offsetof(xfs_dinode_t, di_next_unlinked);
2023 xfs_trans_inode_buf(tp, ibp);
2024 xfs_trans_log_buf(tp, ibp, offset,
2025 (offset + sizeof(xfs_agino_t) - 1));
2026 xfs_inobp_check(mp, ibp);
2028 xfs_trans_brelse(tp, ibp);
2031 * Point the bucket head pointer at the next inode.
2033 ASSERT(next_agino != 0);
2034 ASSERT(next_agino != agino);
2035 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2036 offset = offsetof(xfs_agi_t, agi_unlinked) +
2037 (sizeof(xfs_agino_t) * bucket_index);
2038 xfs_trans_log_buf(tp, agibp, offset,
2039 (offset + sizeof(xfs_agino_t) - 1));
2042 * We need to search the list for the inode being freed.
2044 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2046 while (next_agino != agino) {
2048 * If the last inode wasn't the one pointing to
2049 * us, then release its buffer since we're not
2050 * going to do anything with it.
2052 if (last_ibp != NULL) {
2053 xfs_trans_brelse(tp, last_ibp);
2055 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2056 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
2057 &last_ibp, &last_offset, 0);
2060 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
2061 error, mp->m_fsname);
2064 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2065 ASSERT(next_agino != NULLAGINO);
2066 ASSERT(next_agino != 0);
2069 * Now last_ibp points to the buffer previous to us on
2070 * the unlinked list. Pull us from the list.
2072 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
2075 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2076 error, mp->m_fsname);
2079 next_agino = be32_to_cpu(dip->di_next_unlinked);
2080 ASSERT(next_agino != 0);
2081 ASSERT(next_agino != agino);
2082 if (next_agino != NULLAGINO) {
2083 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2084 offset = ip->i_boffset +
2085 offsetof(xfs_dinode_t, di_next_unlinked);
2086 xfs_trans_inode_buf(tp, ibp);
2087 xfs_trans_log_buf(tp, ibp, offset,
2088 (offset + sizeof(xfs_agino_t) - 1));
2089 xfs_inobp_check(mp, ibp);
2091 xfs_trans_brelse(tp, ibp);
2094 * Point the previous inode on the list to the next inode.
2096 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2097 ASSERT(next_agino != 0);
2098 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2099 xfs_trans_inode_buf(tp, last_ibp);
2100 xfs_trans_log_buf(tp, last_ibp, offset,
2101 (offset + sizeof(xfs_agino_t) - 1));
2102 xfs_inobp_check(mp, last_ibp);
2109 xfs_inode_t *free_ip,
2113 xfs_mount_t *mp = free_ip->i_mount;
2114 int blks_per_cluster;
2117 int i, j, found, pre_flushed;
2120 xfs_inode_t *ip, **ip_found;
2121 xfs_inode_log_item_t *iip;
2122 xfs_log_item_t *lip;
2123 xfs_perag_t *pag = xfs_get_perag(mp, inum);
2125 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
2126 blks_per_cluster = 1;
2127 ninodes = mp->m_sb.sb_inopblock;
2128 nbufs = XFS_IALLOC_BLOCKS(mp);
2130 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
2131 mp->m_sb.sb_blocksize;
2132 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
2133 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
2136 ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
2138 for (j = 0; j < nbufs; j++, inum += ninodes) {
2139 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2140 XFS_INO_TO_AGBNO(mp, inum));
2144 * Look for each inode in memory and attempt to lock it,
2145 * we can be racing with flush and tail pushing here.
2146 * any inode we get the locks on, add to an array of
2147 * inode items to process later.
2149 * The get the buffer lock, we could beat a flush
2150 * or tail pushing thread to the lock here, in which
2151 * case they will go looking for the inode buffer
2152 * and fail, we need some other form of interlock
2156 for (i = 0; i < ninodes; i++) {
2157 read_lock(&pag->pag_ici_lock);
2158 ip = radix_tree_lookup(&pag->pag_ici_root,
2159 XFS_INO_TO_AGINO(mp, (inum + i)));
2161 /* Inode not in memory or we found it already,
2164 if (!ip || xfs_iflags_test(ip, XFS_ISTALE)) {
2165 read_unlock(&pag->pag_ici_lock);
2169 if (xfs_inode_clean(ip)) {
2170 read_unlock(&pag->pag_ici_lock);
2174 /* If we can get the locks then add it to the
2175 * list, otherwise by the time we get the bp lock
2176 * below it will already be attached to the
2180 /* This inode will already be locked - by us, lets
2184 if (ip == free_ip) {
2185 if (xfs_iflock_nowait(ip)) {
2186 xfs_iflags_set(ip, XFS_ISTALE);
2187 if (xfs_inode_clean(ip)) {
2190 ip_found[found++] = ip;
2193 read_unlock(&pag->pag_ici_lock);
2197 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2198 if (xfs_iflock_nowait(ip)) {
2199 xfs_iflags_set(ip, XFS_ISTALE);
2201 if (xfs_inode_clean(ip)) {
2203 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2205 ip_found[found++] = ip;
2208 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2211 read_unlock(&pag->pag_ici_lock);
2214 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2215 mp->m_bsize * blks_per_cluster,
2219 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
2221 if (lip->li_type == XFS_LI_INODE) {
2222 iip = (xfs_inode_log_item_t *)lip;
2223 ASSERT(iip->ili_logged == 1);
2224 lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
2225 xfs_trans_ail_copy_lsn(mp->m_ail,
2226 &iip->ili_flush_lsn,
2227 &iip->ili_item.li_lsn);
2228 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2231 lip = lip->li_bio_list;
2234 for (i = 0; i < found; i++) {
2239 ip->i_update_core = 0;
2241 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2245 iip->ili_last_fields = iip->ili_format.ilf_fields;
2246 iip->ili_format.ilf_fields = 0;
2247 iip->ili_logged = 1;
2248 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2249 &iip->ili_item.li_lsn);
2251 xfs_buf_attach_iodone(bp,
2252 (void(*)(xfs_buf_t*,xfs_log_item_t*))
2253 xfs_istale_done, (xfs_log_item_t *)iip);
2254 if (ip != free_ip) {
2255 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2259 if (found || pre_flushed)
2260 xfs_trans_stale_inode_buf(tp, bp);
2261 xfs_trans_binval(tp, bp);
2264 kmem_free(ip_found);
2265 xfs_put_perag(mp, pag);
2269 * This is called to return an inode to the inode free list.
2270 * The inode should already be truncated to 0 length and have
2271 * no pages associated with it. This routine also assumes that
2272 * the inode is already a part of the transaction.
2274 * The on-disk copy of the inode will have been added to the list
2275 * of unlinked inodes in the AGI. We need to remove the inode from
2276 * that list atomically with respect to freeing it here.
2282 xfs_bmap_free_t *flist)
2286 xfs_ino_t first_ino;
2290 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2291 ASSERT(ip->i_transp == tp);
2292 ASSERT(ip->i_d.di_nlink == 0);
2293 ASSERT(ip->i_d.di_nextents == 0);
2294 ASSERT(ip->i_d.di_anextents == 0);
2295 ASSERT((ip->i_d.di_size == 0 && ip->i_size == 0) ||
2296 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2297 ASSERT(ip->i_d.di_nblocks == 0);
2300 * Pull the on-disk inode from the AGI unlinked list.
2302 error = xfs_iunlink_remove(tp, ip);
2307 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2311 ip->i_d.di_mode = 0; /* mark incore inode as free */
2312 ip->i_d.di_flags = 0;
2313 ip->i_d.di_dmevmask = 0;
2314 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2315 ip->i_df.if_ext_max =
2316 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2317 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2318 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2320 * Bump the generation count so no one will be confused
2321 * by reincarnations of this inode.
2325 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2327 error = xfs_itobp(ip->i_mount, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
2332 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2333 * from picking up this inode when it is reclaimed (its incore state
2334 * initialzed but not flushed to disk yet). The in-core di_mode is
2335 * already cleared and a corresponding transaction logged.
2336 * The hack here just synchronizes the in-core to on-disk
2337 * di_mode value in advance before the actual inode sync to disk.
2338 * This is OK because the inode is already unlinked and would never
2339 * change its di_mode again for this inode generation.
2340 * This is a temporary hack that would require a proper fix
2343 dip->di_core.di_mode = 0;
2346 xfs_ifree_cluster(ip, tp, first_ino);
2353 * Reallocate the space for if_broot based on the number of records
2354 * being added or deleted as indicated in rec_diff. Move the records
2355 * and pointers in if_broot to fit the new size. When shrinking this
2356 * will eliminate holes between the records and pointers created by
2357 * the caller. When growing this will create holes to be filled in
2360 * The caller must not request to add more records than would fit in
2361 * the on-disk inode root. If the if_broot is currently NULL, then
2362 * if we adding records one will be allocated. The caller must also
2363 * not request that the number of records go below zero, although
2364 * it can go to zero.
2366 * ip -- the inode whose if_broot area is changing
2367 * ext_diff -- the change in the number of records, positive or negative,
2368 * requested for the if_broot array.
2376 struct xfs_mount *mp = ip->i_mount;
2379 struct xfs_btree_block *new_broot;
2386 * Handle the degenerate case quietly.
2388 if (rec_diff == 0) {
2392 ifp = XFS_IFORK_PTR(ip, whichfork);
2395 * If there wasn't any memory allocated before, just
2396 * allocate it now and get out.
2398 if (ifp->if_broot_bytes == 0) {
2399 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
2400 ifp->if_broot = kmem_alloc(new_size, KM_SLEEP);
2401 ifp->if_broot_bytes = (int)new_size;
2406 * If there is already an existing if_broot, then we need
2407 * to realloc() it and shift the pointers to their new
2408 * location. The records don't change location because
2409 * they are kept butted up against the btree block header.
2411 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
2412 new_max = cur_max + rec_diff;
2413 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2414 ifp->if_broot = kmem_realloc(ifp->if_broot, new_size,
2415 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2417 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2418 ifp->if_broot_bytes);
2419 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2421 ifp->if_broot_bytes = (int)new_size;
2422 ASSERT(ifp->if_broot_bytes <=
2423 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2424 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2429 * rec_diff is less than 0. In this case, we are shrinking the
2430 * if_broot buffer. It must already exist. If we go to zero
2431 * records, just get rid of the root and clear the status bit.
2433 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
2434 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
2435 new_max = cur_max + rec_diff;
2436 ASSERT(new_max >= 0);
2438 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2442 new_broot = kmem_alloc(new_size, KM_SLEEP);
2444 * First copy over the btree block header.
2446 memcpy(new_broot, ifp->if_broot, XFS_BTREE_LBLOCK_LEN);
2449 ifp->if_flags &= ~XFS_IFBROOT;
2453 * Only copy the records and pointers if there are any.
2457 * First copy the records.
2459 op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1);
2460 np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1);
2461 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2464 * Then copy the pointers.
2466 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2467 ifp->if_broot_bytes);
2468 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1,
2470 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2472 kmem_free(ifp->if_broot);
2473 ifp->if_broot = new_broot;
2474 ifp->if_broot_bytes = (int)new_size;
2475 ASSERT(ifp->if_broot_bytes <=
2476 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2482 * This is called when the amount of space needed for if_data
2483 * is increased or decreased. The change in size is indicated by
2484 * the number of bytes that need to be added or deleted in the
2485 * byte_diff parameter.
2487 * If the amount of space needed has decreased below the size of the
2488 * inline buffer, then switch to using the inline buffer. Otherwise,
2489 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2490 * to what is needed.
2492 * ip -- the inode whose if_data area is changing
2493 * byte_diff -- the change in the number of bytes, positive or negative,
2494 * requested for the if_data array.
2506 if (byte_diff == 0) {
2510 ifp = XFS_IFORK_PTR(ip, whichfork);
2511 new_size = (int)ifp->if_bytes + byte_diff;
2512 ASSERT(new_size >= 0);
2514 if (new_size == 0) {
2515 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2516 kmem_free(ifp->if_u1.if_data);
2518 ifp->if_u1.if_data = NULL;
2520 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2522 * If the valid extents/data can fit in if_inline_ext/data,
2523 * copy them from the malloc'd vector and free it.
2525 if (ifp->if_u1.if_data == NULL) {
2526 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2527 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2528 ASSERT(ifp->if_real_bytes != 0);
2529 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2531 kmem_free(ifp->if_u1.if_data);
2532 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2537 * Stuck with malloc/realloc.
2538 * For inline data, the underlying buffer must be
2539 * a multiple of 4 bytes in size so that it can be
2540 * logged and stay on word boundaries. We enforce
2543 real_size = roundup(new_size, 4);
2544 if (ifp->if_u1.if_data == NULL) {
2545 ASSERT(ifp->if_real_bytes == 0);
2546 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2547 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2549 * Only do the realloc if the underlying size
2550 * is really changing.
2552 if (ifp->if_real_bytes != real_size) {
2553 ifp->if_u1.if_data =
2554 kmem_realloc(ifp->if_u1.if_data,
2560 ASSERT(ifp->if_real_bytes == 0);
2561 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2562 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2566 ifp->if_real_bytes = real_size;
2567 ifp->if_bytes = new_size;
2568 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2575 * Map inode to disk block and offset.
2577 * mp -- the mount point structure for the current file system
2578 * tp -- the current transaction
2579 * ino -- the inode number of the inode to be located
2580 * imap -- this structure is filled in with the information necessary
2581 * to retrieve the given inode from disk
2582 * flags -- flags to pass to xfs_dilocate indicating whether or not
2583 * lookups in the inode btree were OK or not
2593 xfs_fsblock_t fsbno;
2598 fsbno = imap->im_blkno ?
2599 XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK;
2600 error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags);
2604 imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno);
2605 imap->im_len = XFS_FSB_TO_BB(mp, len);
2606 imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno);
2607 imap->im_ioffset = (ushort)off;
2608 imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog);
2611 * If the inode number maps to a block outside the bounds
2612 * of the file system then return NULL rather than calling
2613 * read_buf and panicing when we get an error from the
2616 if ((imap->im_blkno + imap->im_len) >
2617 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
2618 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
2619 "(imap->im_blkno (0x%llx) + imap->im_len (0x%llx)) > "
2620 " XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks) (0x%llx)",
2621 (unsigned long long) imap->im_blkno,
2622 (unsigned long long) imap->im_len,
2623 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
2636 ifp = XFS_IFORK_PTR(ip, whichfork);
2637 if (ifp->if_broot != NULL) {
2638 kmem_free(ifp->if_broot);
2639 ifp->if_broot = NULL;
2643 * If the format is local, then we can't have an extents
2644 * array so just look for an inline data array. If we're
2645 * not local then we may or may not have an extents list,
2646 * so check and free it up if we do.
2648 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2649 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2650 (ifp->if_u1.if_data != NULL)) {
2651 ASSERT(ifp->if_real_bytes != 0);
2652 kmem_free(ifp->if_u1.if_data);
2653 ifp->if_u1.if_data = NULL;
2654 ifp->if_real_bytes = 0;
2656 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2657 ((ifp->if_flags & XFS_IFEXTIREC) ||
2658 ((ifp->if_u1.if_extents != NULL) &&
2659 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
2660 ASSERT(ifp->if_real_bytes != 0);
2661 xfs_iext_destroy(ifp);
2663 ASSERT(ifp->if_u1.if_extents == NULL ||
2664 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2665 ASSERT(ifp->if_real_bytes == 0);
2666 if (whichfork == XFS_ATTR_FORK) {
2667 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2673 * This is called free all the memory associated with an inode.
2674 * It must free the inode itself and any buffers allocated for
2675 * if_extents/if_data and if_broot. It must also free the lock
2676 * associated with the inode.
2678 * Note: because we don't initialise everything on reallocation out
2679 * of the zone, we must ensure we nullify everything correctly before
2680 * freeing the structure.
2686 switch (ip->i_d.di_mode & S_IFMT) {
2690 xfs_idestroy_fork(ip, XFS_DATA_FORK);
2694 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
2696 #ifdef XFS_INODE_TRACE
2697 ktrace_free(ip->i_trace);
2699 #ifdef XFS_BMAP_TRACE
2700 ktrace_free(ip->i_xtrace);
2702 #ifdef XFS_BTREE_TRACE
2703 ktrace_free(ip->i_btrace);
2706 ktrace_free(ip->i_rwtrace);
2708 #ifdef XFS_ILOCK_TRACE
2709 ktrace_free(ip->i_lock_trace);
2711 #ifdef XFS_DIR2_TRACE
2712 ktrace_free(ip->i_dir_trace);
2716 * Only if we are shutting down the fs will we see an
2717 * inode still in the AIL. If it is there, we should remove
2718 * it to prevent a use-after-free from occurring.
2720 xfs_log_item_t *lip = &ip->i_itemp->ili_item;
2721 struct xfs_ail *ailp = lip->li_ailp;
2723 ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) ||
2724 XFS_FORCED_SHUTDOWN(ip->i_mount));
2725 if (lip->li_flags & XFS_LI_IN_AIL) {
2726 spin_lock(&ailp->xa_lock);
2727 if (lip->li_flags & XFS_LI_IN_AIL)
2728 xfs_trans_ail_delete(ailp, lip);
2730 spin_unlock(&ailp->xa_lock);
2732 xfs_inode_item_destroy(ip);
2735 /* asserts to verify all state is correct here */
2736 ASSERT(atomic_read(&ip->i_iocount) == 0);
2737 ASSERT(atomic_read(&ip->i_pincount) == 0);
2738 ASSERT(!spin_is_locked(&ip->i_flags_lock));
2739 ASSERT(completion_done(&ip->i_flush));
2740 kmem_zone_free(xfs_inode_zone, ip);
2745 * Increment the pin count of the given buffer.
2746 * This value is protected by ipinlock spinlock in the mount structure.
2752 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2754 atomic_inc(&ip->i_pincount);
2758 * Decrement the pin count of the given inode, and wake up
2759 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2760 * inode must have been previously pinned with a call to xfs_ipin().
2766 ASSERT(atomic_read(&ip->i_pincount) > 0);
2768 if (atomic_dec_and_test(&ip->i_pincount))
2769 wake_up(&ip->i_ipin_wait);
2773 * This is called to unpin an inode. It can be directed to wait or to return
2774 * immediately without waiting for the inode to be unpinned. The caller must
2775 * have the inode locked in at least shared mode so that the buffer cannot be
2776 * subsequently pinned once someone is waiting for it to be unpinned.
2783 xfs_inode_log_item_t *iip = ip->i_itemp;
2785 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2786 if (atomic_read(&ip->i_pincount) == 0)
2789 /* Give the log a push to start the unpinning I/O */
2790 xfs_log_force(ip->i_mount, (iip && iip->ili_last_lsn) ?
2791 iip->ili_last_lsn : 0, XFS_LOG_FORCE);
2793 wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
2800 __xfs_iunpin_wait(ip, 1);
2807 __xfs_iunpin_wait(ip, 0);
2812 * xfs_iextents_copy()
2814 * This is called to copy the REAL extents (as opposed to the delayed
2815 * allocation extents) from the inode into the given buffer. It
2816 * returns the number of bytes copied into the buffer.
2818 * If there are no delayed allocation extents, then we can just
2819 * memcpy() the extents into the buffer. Otherwise, we need to
2820 * examine each extent in turn and skip those which are delayed.
2832 xfs_fsblock_t start_block;
2834 ifp = XFS_IFORK_PTR(ip, whichfork);
2835 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2836 ASSERT(ifp->if_bytes > 0);
2838 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2839 XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork);
2843 * There are some delayed allocation extents in the
2844 * inode, so copy the extents one at a time and skip
2845 * the delayed ones. There must be at least one
2846 * non-delayed extent.
2849 for (i = 0; i < nrecs; i++) {
2850 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
2851 start_block = xfs_bmbt_get_startblock(ep);
2852 if (ISNULLSTARTBLOCK(start_block)) {
2854 * It's a delayed allocation extent, so skip it.
2859 /* Translate to on disk format */
2860 put_unaligned(cpu_to_be64(ep->l0), &dp->l0);
2861 put_unaligned(cpu_to_be64(ep->l1), &dp->l1);
2865 ASSERT(copied != 0);
2866 xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip));
2868 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2872 * Each of the following cases stores data into the same region
2873 * of the on-disk inode, so only one of them can be valid at
2874 * any given time. While it is possible to have conflicting formats
2875 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2876 * in EXTENTS format, this can only happen when the fork has
2877 * changed formats after being modified but before being flushed.
2878 * In these cases, the format always takes precedence, because the
2879 * format indicates the current state of the fork.
2886 xfs_inode_log_item_t *iip,
2893 #ifdef XFS_TRANS_DEBUG
2896 static const short brootflag[2] =
2897 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2898 static const short dataflag[2] =
2899 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2900 static const short extflag[2] =
2901 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2905 ifp = XFS_IFORK_PTR(ip, whichfork);
2907 * This can happen if we gave up in iformat in an error path,
2908 * for the attribute fork.
2911 ASSERT(whichfork == XFS_ATTR_FORK);
2914 cp = XFS_DFORK_PTR(dip, whichfork);
2916 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2917 case XFS_DINODE_FMT_LOCAL:
2918 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2919 (ifp->if_bytes > 0)) {
2920 ASSERT(ifp->if_u1.if_data != NULL);
2921 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2922 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2926 case XFS_DINODE_FMT_EXTENTS:
2927 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2928 !(iip->ili_format.ilf_fields & extflag[whichfork]));
2929 ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
2930 (ifp->if_bytes == 0));
2931 ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
2932 (ifp->if_bytes > 0));
2933 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
2934 (ifp->if_bytes > 0)) {
2935 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2936 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2941 case XFS_DINODE_FMT_BTREE:
2942 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
2943 (ifp->if_broot_bytes > 0)) {
2944 ASSERT(ifp->if_broot != NULL);
2945 ASSERT(ifp->if_broot_bytes <=
2946 (XFS_IFORK_SIZE(ip, whichfork) +
2947 XFS_BROOT_SIZE_ADJ));
2948 xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes,
2949 (xfs_bmdr_block_t *)cp,
2950 XFS_DFORK_SIZE(dip, mp, whichfork));
2954 case XFS_DINODE_FMT_DEV:
2955 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
2956 ASSERT(whichfork == XFS_DATA_FORK);
2957 dip->di_u.di_dev = cpu_to_be32(ip->i_df.if_u2.if_rdev);
2961 case XFS_DINODE_FMT_UUID:
2962 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
2963 ASSERT(whichfork == XFS_DATA_FORK);
2964 memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid,
2980 xfs_mount_t *mp = ip->i_mount;
2981 xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino);
2982 unsigned long first_index, mask;
2983 unsigned long inodes_per_cluster;
2985 xfs_inode_t **ilist;
2992 ASSERT(pag->pagi_inodeok);
2993 ASSERT(pag->pag_ici_init);
2995 inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog;
2996 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
2997 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
3001 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
3002 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
3003 read_lock(&pag->pag_ici_lock);
3004 /* really need a gang lookup range call here */
3005 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
3006 first_index, inodes_per_cluster);
3010 for (i = 0; i < nr_found; i++) {
3014 /* if the inode lies outside this cluster, we're done. */
3015 if ((XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index)
3018 * Do an un-protected check to see if the inode is dirty and
3019 * is a candidate for flushing. These checks will be repeated
3020 * later after the appropriate locks are acquired.
3022 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
3026 * Try to get locks. If any are unavailable or it is pinned,
3027 * then this inode cannot be flushed and is skipped.
3030 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
3032 if (!xfs_iflock_nowait(iq)) {
3033 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3036 if (xfs_ipincount(iq)) {
3038 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3043 * arriving here means that this inode can be flushed. First
3044 * re-check that it's dirty before flushing.
3046 if (!xfs_inode_clean(iq)) {
3048 error = xfs_iflush_int(iq, bp);
3050 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3051 goto cluster_corrupt_out;
3057 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3061 XFS_STATS_INC(xs_icluster_flushcnt);
3062 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
3066 read_unlock(&pag->pag_ici_lock);
3071 cluster_corrupt_out:
3073 * Corruption detected in the clustering loop. Invalidate the
3074 * inode buffer and shut down the filesystem.
3076 read_unlock(&pag->pag_ici_lock);
3078 * Clean up the buffer. If it was B_DELWRI, just release it --
3079 * brelse can handle it with no problems. If not, shut down the
3080 * filesystem before releasing the buffer.
3082 bufwasdelwri = XFS_BUF_ISDELAYWRITE(bp);
3086 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3088 if (!bufwasdelwri) {
3090 * Just like incore_relse: if we have b_iodone functions,
3091 * mark the buffer as an error and call them. Otherwise
3092 * mark it as stale and brelse.
3094 if (XFS_BUF_IODONE_FUNC(bp)) {
3095 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
3099 XFS_BUF_ERROR(bp,EIO);
3108 * Unlocks the flush lock
3110 xfs_iflush_abort(iq);
3112 return XFS_ERROR(EFSCORRUPTED);
3116 * xfs_iflush() will write a modified inode's changes out to the
3117 * inode's on disk home. The caller must have the inode lock held
3118 * in at least shared mode and the inode flush completion must be
3119 * active as well. The inode lock will still be held upon return from
3120 * the call and the caller is free to unlock it.
3121 * The inode flush will be completed when the inode reaches the disk.
3122 * The flags indicate how the inode's buffer should be written out.
3129 xfs_inode_log_item_t *iip;
3134 int noblock = (flags == XFS_IFLUSH_ASYNC_NOBLOCK);
3135 enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
3137 XFS_STATS_INC(xs_iflush_count);
3139 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3140 ASSERT(!completion_done(&ip->i_flush));
3141 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3142 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3148 * If the inode isn't dirty, then just release the inode
3149 * flush lock and do nothing.
3151 if (xfs_inode_clean(ip)) {
3157 * We can't flush the inode until it is unpinned, so wait for it if we
3158 * are allowed to block. We know noone new can pin it, because we are
3159 * holding the inode lock shared and you need to hold it exclusively to
3162 * If we are not allowed to block, force the log out asynchronously so
3163 * that when we come back the inode will be unpinned. If other inodes
3164 * in the same cluster are dirty, they will probably write the inode
3165 * out for us if they occur after the log force completes.
3167 if (noblock && xfs_ipincount(ip)) {
3168 xfs_iunpin_nowait(ip);
3172 xfs_iunpin_wait(ip);
3175 * This may have been unpinned because the filesystem is shutting
3176 * down forcibly. If that's the case we must not write this inode
3177 * to disk, because the log record didn't make it to disk!
3179 if (XFS_FORCED_SHUTDOWN(mp)) {
3180 ip->i_update_core = 0;
3182 iip->ili_format.ilf_fields = 0;
3184 return XFS_ERROR(EIO);
3188 * Decide how buffer will be flushed out. This is done before
3189 * the call to xfs_iflush_int because this field is zeroed by it.
3191 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3193 * Flush out the inode buffer according to the directions
3194 * of the caller. In the cases where the caller has given
3195 * us a choice choose the non-delwri case. This is because
3196 * the inode is in the AIL and we need to get it out soon.
3199 case XFS_IFLUSH_SYNC:
3200 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3203 case XFS_IFLUSH_ASYNC_NOBLOCK:
3204 case XFS_IFLUSH_ASYNC:
3205 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3208 case XFS_IFLUSH_DELWRI:
3218 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3219 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3220 case XFS_IFLUSH_DELWRI:
3223 case XFS_IFLUSH_ASYNC_NOBLOCK:
3224 case XFS_IFLUSH_ASYNC:
3227 case XFS_IFLUSH_SYNC:
3238 * Get the buffer containing the on-disk inode.
3240 error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0, 0,
3241 noblock ? XFS_BUF_TRYLOCK : XFS_BUF_LOCK);
3248 * First flush out the inode that xfs_iflush was called with.
3250 error = xfs_iflush_int(ip, bp);
3255 * If the buffer is pinned then push on the log now so we won't
3256 * get stuck waiting in the write for too long.
3258 if (XFS_BUF_ISPINNED(bp))
3259 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
3263 * see if other inodes can be gathered into this write
3265 error = xfs_iflush_cluster(ip, bp);
3267 goto cluster_corrupt_out;
3269 if (flags & INT_DELWRI) {
3270 xfs_bdwrite(mp, bp);
3271 } else if (flags & INT_ASYNC) {
3272 error = xfs_bawrite(mp, bp);
3274 error = xfs_bwrite(mp, bp);
3280 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3281 cluster_corrupt_out:
3283 * Unlocks the flush lock
3285 xfs_iflush_abort(ip);
3286 return XFS_ERROR(EFSCORRUPTED);
3295 xfs_inode_log_item_t *iip;
3298 #ifdef XFS_TRANS_DEBUG
3302 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3303 ASSERT(!completion_done(&ip->i_flush));
3304 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3305 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3312 * If the inode isn't dirty, then just release the inode
3313 * flush lock and do nothing.
3315 if (xfs_inode_clean(ip)) {
3320 /* set *dip = inode's place in the buffer */
3321 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset);
3324 * Clear i_update_core before copying out the data.
3325 * This is for coordination with our timestamp updates
3326 * that don't hold the inode lock. They will always
3327 * update the timestamps BEFORE setting i_update_core,
3328 * so if we clear i_update_core after they set it we
3329 * are guaranteed to see their updates to the timestamps.
3330 * I believe that this depends on strongly ordered memory
3331 * semantics, but we have that. We use the SYNCHRONIZE
3332 * macro to make sure that the compiler does not reorder
3333 * the i_update_core access below the data copy below.
3335 ip->i_update_core = 0;
3339 * Make sure to get the latest atime from the Linux inode.
3341 xfs_synchronize_atime(ip);
3343 if (XFS_TEST_ERROR(be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC,
3344 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3345 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3346 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3347 ip->i_ino, be16_to_cpu(dip->di_core.di_magic), dip);
3350 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3351 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3352 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3353 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3354 ip->i_ino, ip, ip->i_d.di_magic);
3357 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3359 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3360 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3361 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3362 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3363 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3367 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
3369 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3370 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3371 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3372 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3373 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3374 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3379 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3380 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3381 XFS_RANDOM_IFLUSH_5)) {
3382 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3383 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3385 ip->i_d.di_nextents + ip->i_d.di_anextents,
3390 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3391 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3392 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3393 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3394 ip->i_ino, ip->i_d.di_forkoff, ip);
3398 * bump the flush iteration count, used to detect flushes which
3399 * postdate a log record during recovery.
3402 ip->i_d.di_flushiter++;
3405 * Copy the dirty parts of the inode into the on-disk
3406 * inode. We always copy out the core of the inode,
3407 * because if the inode is dirty at all the core must
3410 xfs_dinode_to_disk(&dip->di_core, &ip->i_d);
3412 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3413 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3414 ip->i_d.di_flushiter = 0;
3417 * If this is really an old format inode and the superblock version
3418 * has not been updated to support only new format inodes, then
3419 * convert back to the old inode format. If the superblock version
3420 * has been updated, then make the conversion permanent.
3422 ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 ||
3423 xfs_sb_version_hasnlink(&mp->m_sb));
3424 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
3425 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
3429 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3430 dip->di_core.di_onlink = cpu_to_be16(ip->i_d.di_nlink);
3433 * The superblock version has already been bumped,
3434 * so just make the conversion to the new inode
3437 ip->i_d.di_version = XFS_DINODE_VERSION_2;
3438 dip->di_core.di_version = XFS_DINODE_VERSION_2;
3439 ip->i_d.di_onlink = 0;
3440 dip->di_core.di_onlink = 0;
3441 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3442 memset(&(dip->di_core.di_pad[0]), 0,
3443 sizeof(dip->di_core.di_pad));
3444 ASSERT(ip->i_d.di_projid == 0);
3448 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
3449 if (XFS_IFORK_Q(ip))
3450 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3451 xfs_inobp_check(mp, bp);
3454 * We've recorded everything logged in the inode, so we'd
3455 * like to clear the ilf_fields bits so we don't log and
3456 * flush things unnecessarily. However, we can't stop
3457 * logging all this information until the data we've copied
3458 * into the disk buffer is written to disk. If we did we might
3459 * overwrite the copy of the inode in the log with all the
3460 * data after re-logging only part of it, and in the face of
3461 * a crash we wouldn't have all the data we need to recover.
3463 * What we do is move the bits to the ili_last_fields field.
3464 * When logging the inode, these bits are moved back to the
3465 * ilf_fields field. In the xfs_iflush_done() routine we
3466 * clear ili_last_fields, since we know that the information
3467 * those bits represent is permanently on disk. As long as
3468 * the flush completes before the inode is logged again, then
3469 * both ilf_fields and ili_last_fields will be cleared.
3471 * We can play with the ilf_fields bits here, because the inode
3472 * lock must be held exclusively in order to set bits there
3473 * and the flush lock protects the ili_last_fields bits.
3474 * Set ili_logged so the flush done
3475 * routine can tell whether or not to look in the AIL.
3476 * Also, store the current LSN of the inode so that we can tell
3477 * whether the item has moved in the AIL from xfs_iflush_done().
3478 * In order to read the lsn we need the AIL lock, because
3479 * it is a 64 bit value that cannot be read atomically.
3481 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3482 iip->ili_last_fields = iip->ili_format.ilf_fields;
3483 iip->ili_format.ilf_fields = 0;
3484 iip->ili_logged = 1;
3486 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3487 &iip->ili_item.li_lsn);
3490 * Attach the function xfs_iflush_done to the inode's
3491 * buffer. This will remove the inode from the AIL
3492 * and unlock the inode's flush lock when the inode is
3493 * completely written to disk.
3495 xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
3496 xfs_iflush_done, (xfs_log_item_t *)iip);
3498 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3499 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3502 * We're flushing an inode which is not in the AIL and has
3503 * not been logged but has i_update_core set. For this
3504 * case we can use a B_DELWRI flush and immediately drop
3505 * the inode flush lock because we can avoid the whole
3506 * AIL state thing. It's OK to drop the flush lock now,
3507 * because we've already locked the buffer and to do anything
3508 * you really need both.
3511 ASSERT(iip->ili_logged == 0);
3512 ASSERT(iip->ili_last_fields == 0);
3513 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3521 return XFS_ERROR(EFSCORRUPTED);
3526 #ifdef XFS_ILOCK_TRACE
3527 ktrace_t *xfs_ilock_trace_buf;
3530 xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
3532 ktrace_enter(ip->i_lock_trace,
3534 (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
3535 (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
3536 (void *)ra, /* caller of ilock */
3537 (void *)(unsigned long)current_cpu(),
3538 (void *)(unsigned long)current_pid(),
3539 NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL);
3544 * Return a pointer to the extent record at file index idx.
3546 xfs_bmbt_rec_host_t *
3548 xfs_ifork_t *ifp, /* inode fork pointer */
3549 xfs_extnum_t idx) /* index of target extent */
3552 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
3553 return ifp->if_u1.if_ext_irec->er_extbuf;
3554 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3555 xfs_ext_irec_t *erp; /* irec pointer */
3556 int erp_idx = 0; /* irec index */
3557 xfs_extnum_t page_idx = idx; /* ext index in target list */
3559 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3560 return &erp->er_extbuf[page_idx];
3561 } else if (ifp->if_bytes) {
3562 return &ifp->if_u1.if_extents[idx];
3569 * Insert new item(s) into the extent records for incore inode
3570 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3574 xfs_ifork_t *ifp, /* inode fork pointer */
3575 xfs_extnum_t idx, /* starting index of new items */
3576 xfs_extnum_t count, /* number of inserted items */
3577 xfs_bmbt_irec_t *new) /* items to insert */
3579 xfs_extnum_t i; /* extent record index */
3581 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3582 xfs_iext_add(ifp, idx, count);
3583 for (i = idx; i < idx + count; i++, new++)
3584 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new);
3588 * This is called when the amount of space required for incore file
3589 * extents needs to be increased. The ext_diff parameter stores the
3590 * number of new extents being added and the idx parameter contains
3591 * the extent index where the new extents will be added. If the new
3592 * extents are being appended, then we just need to (re)allocate and
3593 * initialize the space. Otherwise, if the new extents are being
3594 * inserted into the middle of the existing entries, a bit more work
3595 * is required to make room for the new extents to be inserted. The
3596 * caller is responsible for filling in the new extent entries upon
3601 xfs_ifork_t *ifp, /* inode fork pointer */
3602 xfs_extnum_t idx, /* index to begin adding exts */
3603 int ext_diff) /* number of extents to add */
3605 int byte_diff; /* new bytes being added */
3606 int new_size; /* size of extents after adding */
3607 xfs_extnum_t nextents; /* number of extents in file */
3609 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3610 ASSERT((idx >= 0) && (idx <= nextents));
3611 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
3612 new_size = ifp->if_bytes + byte_diff;
3614 * If the new number of extents (nextents + ext_diff)
3615 * fits inside the inode, then continue to use the inline
3618 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
3619 if (idx < nextents) {
3620 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
3621 &ifp->if_u2.if_inline_ext[idx],
3622 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3623 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
3625 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3626 ifp->if_real_bytes = 0;
3627 ifp->if_lastex = nextents + ext_diff;
3630 * Otherwise use a linear (direct) extent list.
3631 * If the extents are currently inside the inode,
3632 * xfs_iext_realloc_direct will switch us from
3633 * inline to direct extent allocation mode.
3635 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
3636 xfs_iext_realloc_direct(ifp, new_size);
3637 if (idx < nextents) {
3638 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
3639 &ifp->if_u1.if_extents[idx],
3640 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3641 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
3644 /* Indirection array */
3646 xfs_ext_irec_t *erp;
3650 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
3651 if (ifp->if_flags & XFS_IFEXTIREC) {
3652 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3654 xfs_iext_irec_init(ifp);
3655 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3656 erp = ifp->if_u1.if_ext_irec;
3658 /* Extents fit in target extent page */
3659 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3660 if (page_idx < erp->er_extcount) {
3661 memmove(&erp->er_extbuf[page_idx + ext_diff],
3662 &erp->er_extbuf[page_idx],
3663 (erp->er_extcount - page_idx) *
3664 sizeof(xfs_bmbt_rec_t));
3665 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3667 erp->er_extcount += ext_diff;
3668 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3670 /* Insert a new extent page */
3672 xfs_iext_add_indirect_multi(ifp,
3673 erp_idx, page_idx, ext_diff);
3676 * If extent(s) are being appended to the last page in
3677 * the indirection array and the new extent(s) don't fit
3678 * in the page, then erp is NULL and erp_idx is set to
3679 * the next index needed in the indirection array.
3682 int count = ext_diff;
3685 erp = xfs_iext_irec_new(ifp, erp_idx);
3686 erp->er_extcount = count;
3687 count -= MIN(count, (int)XFS_LINEAR_EXTS);
3694 ifp->if_bytes = new_size;
3698 * This is called when incore extents are being added to the indirection
3699 * array and the new extents do not fit in the target extent list. The
3700 * erp_idx parameter contains the irec index for the target extent list
3701 * in the indirection array, and the idx parameter contains the extent
3702 * index within the list. The number of extents being added is stored
3703 * in the count parameter.
3705 * |-------| |-------|
3706 * | | | | idx - number of extents before idx
3708 * | | | | count - number of extents being inserted at idx
3709 * |-------| |-------|
3710 * | count | | nex2 | nex2 - number of extents after idx + count
3711 * |-------| |-------|
3714 xfs_iext_add_indirect_multi(
3715 xfs_ifork_t *ifp, /* inode fork pointer */
3716 int erp_idx, /* target extent irec index */
3717 xfs_extnum_t idx, /* index within target list */
3718 int count) /* new extents being added */
3720 int byte_diff; /* new bytes being added */
3721 xfs_ext_irec_t *erp; /* pointer to irec entry */
3722 xfs_extnum_t ext_diff; /* number of extents to add */
3723 xfs_extnum_t ext_cnt; /* new extents still needed */
3724 xfs_extnum_t nex2; /* extents after idx + count */
3725 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3726 int nlists; /* number of irec's (lists) */
3728 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3729 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3730 nex2 = erp->er_extcount - idx;
3731 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3734 * Save second part of target extent list
3735 * (all extents past */
3737 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3738 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS);
3739 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3740 erp->er_extcount -= nex2;
3741 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3742 memset(&erp->er_extbuf[idx], 0, byte_diff);
3746 * Add the new extents to the end of the target
3747 * list, then allocate new irec record(s) and
3748 * extent buffer(s) as needed to store the rest
3749 * of the new extents.
3752 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3754 erp->er_extcount += ext_diff;
3755 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3756 ext_cnt -= ext_diff;
3760 erp = xfs_iext_irec_new(ifp, erp_idx);
3761 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3762 erp->er_extcount = ext_diff;
3763 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3764 ext_cnt -= ext_diff;
3767 /* Add nex2 extents back to indirection array */
3769 xfs_extnum_t ext_avail;
3772 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3773 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3776 * If nex2 extents fit in the current page, append
3777 * nex2_ep after the new extents.
3779 if (nex2 <= ext_avail) {
3780 i = erp->er_extcount;
3783 * Otherwise, check if space is available in the
3786 else if ((erp_idx < nlists - 1) &&
3787 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3788 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3791 /* Create a hole for nex2 extents */
3792 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3793 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3796 * Final choice, create a new extent page for
3801 erp = xfs_iext_irec_new(ifp, erp_idx);
3803 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
3805 erp->er_extcount += nex2;
3806 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3811 * This is called when the amount of space required for incore file
3812 * extents needs to be decreased. The ext_diff parameter stores the
3813 * number of extents to be removed and the idx parameter contains
3814 * the extent index where the extents will be removed from.
3816 * If the amount of space needed has decreased below the linear
3817 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3818 * extent array. Otherwise, use kmem_realloc() to adjust the
3819 * size to what is needed.
3823 xfs_ifork_t *ifp, /* inode fork pointer */
3824 xfs_extnum_t idx, /* index to begin removing exts */
3825 int ext_diff) /* number of extents to remove */
3827 xfs_extnum_t nextents; /* number of extents in file */
3828 int new_size; /* size of extents after removal */
3830 ASSERT(ext_diff > 0);
3831 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3832 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3834 if (new_size == 0) {
3835 xfs_iext_destroy(ifp);
3836 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3837 xfs_iext_remove_indirect(ifp, idx, ext_diff);
3838 } else if (ifp->if_real_bytes) {
3839 xfs_iext_remove_direct(ifp, idx, ext_diff);
3841 xfs_iext_remove_inline(ifp, idx, ext_diff);
3843 ifp->if_bytes = new_size;
3847 * This removes ext_diff extents from the inline buffer, beginning
3848 * at extent index idx.
3851 xfs_iext_remove_inline(
3852 xfs_ifork_t *ifp, /* inode fork pointer */
3853 xfs_extnum_t idx, /* index to begin removing exts */
3854 int ext_diff) /* number of extents to remove */
3856 int nextents; /* number of extents in file */
3858 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3859 ASSERT(idx < XFS_INLINE_EXTS);
3860 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3861 ASSERT(((nextents - ext_diff) > 0) &&
3862 (nextents - ext_diff) < XFS_INLINE_EXTS);
3864 if (idx + ext_diff < nextents) {
3865 memmove(&ifp->if_u2.if_inline_ext[idx],
3866 &ifp->if_u2.if_inline_ext[idx + ext_diff],
3867 (nextents - (idx + ext_diff)) *
3868 sizeof(xfs_bmbt_rec_t));
3869 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
3870 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3872 memset(&ifp->if_u2.if_inline_ext[idx], 0,
3873 ext_diff * sizeof(xfs_bmbt_rec_t));
3878 * This removes ext_diff extents from a linear (direct) extent list,
3879 * beginning at extent index idx. If the extents are being removed
3880 * from the end of the list (ie. truncate) then we just need to re-
3881 * allocate the list to remove the extra space. Otherwise, if the
3882 * extents are being removed from the middle of the existing extent
3883 * entries, then we first need to move the extent records beginning
3884 * at idx + ext_diff up in the list to overwrite the records being
3885 * removed, then remove the extra space via kmem_realloc.
3888 xfs_iext_remove_direct(
3889 xfs_ifork_t *ifp, /* inode fork pointer */
3890 xfs_extnum_t idx, /* index to begin removing exts */
3891 int ext_diff) /* number of extents to remove */
3893 xfs_extnum_t nextents; /* number of extents in file */
3894 int new_size; /* size of extents after removal */
3896 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3897 new_size = ifp->if_bytes -
3898 (ext_diff * sizeof(xfs_bmbt_rec_t));
3899 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3901 if (new_size == 0) {
3902 xfs_iext_destroy(ifp);
3905 /* Move extents up in the list (if needed) */
3906 if (idx + ext_diff < nextents) {
3907 memmove(&ifp->if_u1.if_extents[idx],
3908 &ifp->if_u1.if_extents[idx + ext_diff],
3909 (nextents - (idx + ext_diff)) *
3910 sizeof(xfs_bmbt_rec_t));
3912 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
3913 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3915 * Reallocate the direct extent list. If the extents
3916 * will fit inside the inode then xfs_iext_realloc_direct
3917 * will switch from direct to inline extent allocation
3920 xfs_iext_realloc_direct(ifp, new_size);
3921 ifp->if_bytes = new_size;
3925 * This is called when incore extents are being removed from the
3926 * indirection array and the extents being removed span multiple extent
3927 * buffers. The idx parameter contains the file extent index where we
3928 * want to begin removing extents, and the count parameter contains
3929 * how many extents need to be removed.
3931 * |-------| |-------|
3932 * | nex1 | | | nex1 - number of extents before idx
3933 * |-------| | count |
3934 * | | | | count - number of extents being removed at idx
3935 * | count | |-------|
3936 * | | | nex2 | nex2 - number of extents after idx + count
3937 * |-------| |-------|
3940 xfs_iext_remove_indirect(
3941 xfs_ifork_t *ifp, /* inode fork pointer */
3942 xfs_extnum_t idx, /* index to begin removing extents */
3943 int count) /* number of extents to remove */
3945 xfs_ext_irec_t *erp; /* indirection array pointer */
3946 int erp_idx = 0; /* indirection array index */
3947 xfs_extnum_t ext_cnt; /* extents left to remove */
3948 xfs_extnum_t ext_diff; /* extents to remove in current list */
3949 xfs_extnum_t nex1; /* number of extents before idx */
3950 xfs_extnum_t nex2; /* extents after idx + count */
3951 int nlists; /* entries in indirection array */
3952 int page_idx = idx; /* index in target extent list */
3954 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3955 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3956 ASSERT(erp != NULL);
3957 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3961 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
3962 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
3964 * Check for deletion of entire list;
3965 * xfs_iext_irec_remove() updates extent offsets.
3967 if (ext_diff == erp->er_extcount) {
3968 xfs_iext_irec_remove(ifp, erp_idx);
3969 ext_cnt -= ext_diff;
3972 ASSERT(erp_idx < ifp->if_real_bytes /
3974 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3981 /* Move extents up (if needed) */
3983 memmove(&erp->er_extbuf[nex1],
3984 &erp->er_extbuf[nex1 + ext_diff],
3985 nex2 * sizeof(xfs_bmbt_rec_t));
3987 /* Zero out rest of page */
3988 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
3989 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
3990 /* Update remaining counters */
3991 erp->er_extcount -= ext_diff;
3992 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
3993 ext_cnt -= ext_diff;
3998 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
3999 xfs_iext_irec_compact(ifp);
4003 * Create, destroy, or resize a linear (direct) block of extents.
4006 xfs_iext_realloc_direct(
4007 xfs_ifork_t *ifp, /* inode fork pointer */
4008 int new_size) /* new size of extents */
4010 int rnew_size; /* real new size of extents */
4012 rnew_size = new_size;
4014 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
4015 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
4016 (new_size != ifp->if_real_bytes)));
4018 /* Free extent records */
4019 if (new_size == 0) {
4020 xfs_iext_destroy(ifp);
4022 /* Resize direct extent list and zero any new bytes */
4023 else if (ifp->if_real_bytes) {
4024 /* Check if extents will fit inside the inode */
4025 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
4026 xfs_iext_direct_to_inline(ifp, new_size /
4027 (uint)sizeof(xfs_bmbt_rec_t));
4028 ifp->if_bytes = new_size;
4031 if (!is_power_of_2(new_size)){
4032 rnew_size = roundup_pow_of_two(new_size);
4034 if (rnew_size != ifp->if_real_bytes) {
4035 ifp->if_u1.if_extents =
4036 kmem_realloc(ifp->if_u1.if_extents,
4038 ifp->if_real_bytes, KM_NOFS);
4040 if (rnew_size > ifp->if_real_bytes) {
4041 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
4042 (uint)sizeof(xfs_bmbt_rec_t)], 0,
4043 rnew_size - ifp->if_real_bytes);
4047 * Switch from the inline extent buffer to a direct
4048 * extent list. Be sure to include the inline extent
4049 * bytes in new_size.
4052 new_size += ifp->if_bytes;
4053 if (!is_power_of_2(new_size)) {
4054 rnew_size = roundup_pow_of_two(new_size);
4056 xfs_iext_inline_to_direct(ifp, rnew_size);
4058 ifp->if_real_bytes = rnew_size;
4059 ifp->if_bytes = new_size;
4063 * Switch from linear (direct) extent records to inline buffer.
4066 xfs_iext_direct_to_inline(
4067 xfs_ifork_t *ifp, /* inode fork pointer */
4068 xfs_extnum_t nextents) /* number of extents in file */
4070 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4071 ASSERT(nextents <= XFS_INLINE_EXTS);
4073 * The inline buffer was zeroed when we switched
4074 * from inline to direct extent allocation mode,
4075 * so we don't need to clear it here.
4077 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
4078 nextents * sizeof(xfs_bmbt_rec_t));
4079 kmem_free(ifp->if_u1.if_extents);
4080 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
4081 ifp->if_real_bytes = 0;
4085 * Switch from inline buffer to linear (direct) extent records.
4086 * new_size should already be rounded up to the next power of 2
4087 * by the caller (when appropriate), so use new_size as it is.
4088 * However, since new_size may be rounded up, we can't update
4089 * if_bytes here. It is the caller's responsibility to update
4090 * if_bytes upon return.
4093 xfs_iext_inline_to_direct(
4094 xfs_ifork_t *ifp, /* inode fork pointer */
4095 int new_size) /* number of extents in file */
4097 ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS);
4098 memset(ifp->if_u1.if_extents, 0, new_size);
4099 if (ifp->if_bytes) {
4100 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
4102 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4103 sizeof(xfs_bmbt_rec_t));
4105 ifp->if_real_bytes = new_size;
4109 * Resize an extent indirection array to new_size bytes.
4112 xfs_iext_realloc_indirect(
4113 xfs_ifork_t *ifp, /* inode fork pointer */
4114 int new_size) /* new indirection array size */
4116 int nlists; /* number of irec's (ex lists) */
4117 int size; /* current indirection array size */
4119 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4120 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4121 size = nlists * sizeof(xfs_ext_irec_t);
4122 ASSERT(ifp->if_real_bytes);
4123 ASSERT((new_size >= 0) && (new_size != size));
4124 if (new_size == 0) {
4125 xfs_iext_destroy(ifp);
4127 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
4128 kmem_realloc(ifp->if_u1.if_ext_irec,
4129 new_size, size, KM_NOFS);
4134 * Switch from indirection array to linear (direct) extent allocations.
4137 xfs_iext_indirect_to_direct(
4138 xfs_ifork_t *ifp) /* inode fork pointer */
4140 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
4141 xfs_extnum_t nextents; /* number of extents in file */
4142 int size; /* size of file extents */
4144 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4145 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4146 ASSERT(nextents <= XFS_LINEAR_EXTS);
4147 size = nextents * sizeof(xfs_bmbt_rec_t);
4149 xfs_iext_irec_compact_pages(ifp);
4150 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
4152 ep = ifp->if_u1.if_ext_irec->er_extbuf;
4153 kmem_free(ifp->if_u1.if_ext_irec);
4154 ifp->if_flags &= ~XFS_IFEXTIREC;
4155 ifp->if_u1.if_extents = ep;
4156 ifp->if_bytes = size;
4157 if (nextents < XFS_LINEAR_EXTS) {
4158 xfs_iext_realloc_direct(ifp, size);
4163 * Free incore file extents.
4167 xfs_ifork_t *ifp) /* inode fork pointer */
4169 if (ifp->if_flags & XFS_IFEXTIREC) {
4173 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4174 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
4175 xfs_iext_irec_remove(ifp, erp_idx);
4177 ifp->if_flags &= ~XFS_IFEXTIREC;
4178 } else if (ifp->if_real_bytes) {
4179 kmem_free(ifp->if_u1.if_extents);
4180 } else if (ifp->if_bytes) {
4181 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4182 sizeof(xfs_bmbt_rec_t));
4184 ifp->if_u1.if_extents = NULL;
4185 ifp->if_real_bytes = 0;
4190 * Return a pointer to the extent record for file system block bno.
4192 xfs_bmbt_rec_host_t * /* pointer to found extent record */
4193 xfs_iext_bno_to_ext(
4194 xfs_ifork_t *ifp, /* inode fork pointer */
4195 xfs_fileoff_t bno, /* block number to search for */
4196 xfs_extnum_t *idxp) /* index of target extent */
4198 xfs_bmbt_rec_host_t *base; /* pointer to first extent */
4199 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
4200 xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */
4201 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4202 int high; /* upper boundary in search */
4203 xfs_extnum_t idx = 0; /* index of target extent */
4204 int low; /* lower boundary in search */
4205 xfs_extnum_t nextents; /* number of file extents */
4206 xfs_fileoff_t startoff = 0; /* start offset of extent */
4208 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4209 if (nextents == 0) {
4214 if (ifp->if_flags & XFS_IFEXTIREC) {
4215 /* Find target extent list */
4217 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
4218 base = erp->er_extbuf;
4219 high = erp->er_extcount - 1;
4221 base = ifp->if_u1.if_extents;
4222 high = nextents - 1;
4224 /* Binary search extent records */
4225 while (low <= high) {
4226 idx = (low + high) >> 1;
4228 startoff = xfs_bmbt_get_startoff(ep);
4229 blockcount = xfs_bmbt_get_blockcount(ep);
4230 if (bno < startoff) {
4232 } else if (bno >= startoff + blockcount) {
4235 /* Convert back to file-based extent index */
4236 if (ifp->if_flags & XFS_IFEXTIREC) {
4237 idx += erp->er_extoff;
4243 /* Convert back to file-based extent index */
4244 if (ifp->if_flags & XFS_IFEXTIREC) {
4245 idx += erp->er_extoff;
4247 if (bno >= startoff + blockcount) {
4248 if (++idx == nextents) {
4251 ep = xfs_iext_get_ext(ifp, idx);
4259 * Return a pointer to the indirection array entry containing the
4260 * extent record for filesystem block bno. Store the index of the
4261 * target irec in *erp_idxp.
4263 xfs_ext_irec_t * /* pointer to found extent record */
4264 xfs_iext_bno_to_irec(
4265 xfs_ifork_t *ifp, /* inode fork pointer */
4266 xfs_fileoff_t bno, /* block number to search for */
4267 int *erp_idxp) /* irec index of target ext list */
4269 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4270 xfs_ext_irec_t *erp_next; /* next indirection array entry */
4271 int erp_idx; /* indirection array index */
4272 int nlists; /* number of extent irec's (lists) */
4273 int high; /* binary search upper limit */
4274 int low; /* binary search lower limit */
4276 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4277 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4281 while (low <= high) {
4282 erp_idx = (low + high) >> 1;
4283 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4284 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
4285 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
4287 } else if (erp_next && bno >=
4288 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
4294 *erp_idxp = erp_idx;
4299 * Return a pointer to the indirection array entry containing the
4300 * extent record at file extent index *idxp. Store the index of the
4301 * target irec in *erp_idxp and store the page index of the target
4302 * extent record in *idxp.
4305 xfs_iext_idx_to_irec(
4306 xfs_ifork_t *ifp, /* inode fork pointer */
4307 xfs_extnum_t *idxp, /* extent index (file -> page) */
4308 int *erp_idxp, /* pointer to target irec */
4309 int realloc) /* new bytes were just added */
4311 xfs_ext_irec_t *prev; /* pointer to previous irec */
4312 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
4313 int erp_idx; /* indirection array index */
4314 int nlists; /* number of irec's (ex lists) */
4315 int high; /* binary search upper limit */
4316 int low; /* binary search lower limit */
4317 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
4319 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4320 ASSERT(page_idx >= 0 && page_idx <=
4321 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
4322 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4327 /* Binary search extent irec's */
4328 while (low <= high) {
4329 erp_idx = (low + high) >> 1;
4330 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4331 prev = erp_idx > 0 ? erp - 1 : NULL;
4332 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
4333 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
4335 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
4336 (page_idx == erp->er_extoff + erp->er_extcount &&
4339 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
4340 erp->er_extcount == XFS_LINEAR_EXTS) {
4344 erp = erp_idx < nlists ? erp + 1 : NULL;
4347 page_idx -= erp->er_extoff;
4352 *erp_idxp = erp_idx;
4357 * Allocate and initialize an indirection array once the space needed
4358 * for incore extents increases above XFS_IEXT_BUFSZ.
4362 xfs_ifork_t *ifp) /* inode fork pointer */
4364 xfs_ext_irec_t *erp; /* indirection array pointer */
4365 xfs_extnum_t nextents; /* number of extents in file */
4367 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4368 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4369 ASSERT(nextents <= XFS_LINEAR_EXTS);
4371 erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS);
4373 if (nextents == 0) {
4374 ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
4375 } else if (!ifp->if_real_bytes) {
4376 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
4377 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
4378 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
4380 erp->er_extbuf = ifp->if_u1.if_extents;
4381 erp->er_extcount = nextents;
4384 ifp->if_flags |= XFS_IFEXTIREC;
4385 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
4386 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
4387 ifp->if_u1.if_ext_irec = erp;
4393 * Allocate and initialize a new entry in the indirection array.
4397 xfs_ifork_t *ifp, /* inode fork pointer */
4398 int erp_idx) /* index for new irec */
4400 xfs_ext_irec_t *erp; /* indirection array pointer */
4401 int i; /* loop counter */
4402 int nlists; /* number of irec's (ex lists) */
4404 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4405 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4407 /* Resize indirection array */
4408 xfs_iext_realloc_indirect(ifp, ++nlists *
4409 sizeof(xfs_ext_irec_t));
4411 * Move records down in the array so the
4412 * new page can use erp_idx.
4414 erp = ifp->if_u1.if_ext_irec;
4415 for (i = nlists - 1; i > erp_idx; i--) {
4416 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
4418 ASSERT(i == erp_idx);
4420 /* Initialize new extent record */
4421 erp = ifp->if_u1.if_ext_irec;
4422 erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
4423 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4424 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
4425 erp[erp_idx].er_extcount = 0;
4426 erp[erp_idx].er_extoff = erp_idx > 0 ?
4427 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
4428 return (&erp[erp_idx]);
4432 * Remove a record from the indirection array.
4435 xfs_iext_irec_remove(
4436 xfs_ifork_t *ifp, /* inode fork pointer */
4437 int erp_idx) /* irec index to remove */
4439 xfs_ext_irec_t *erp; /* indirection array pointer */
4440 int i; /* loop counter */
4441 int nlists; /* number of irec's (ex lists) */
4443 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4444 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4445 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4446 if (erp->er_extbuf) {
4447 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
4449 kmem_free(erp->er_extbuf);
4451 /* Compact extent records */
4452 erp = ifp->if_u1.if_ext_irec;
4453 for (i = erp_idx; i < nlists - 1; i++) {
4454 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
4457 * Manually free the last extent record from the indirection
4458 * array. A call to xfs_iext_realloc_indirect() with a size
4459 * of zero would result in a call to xfs_iext_destroy() which
4460 * would in turn call this function again, creating a nasty
4464 xfs_iext_realloc_indirect(ifp,
4465 nlists * sizeof(xfs_ext_irec_t));
4467 kmem_free(ifp->if_u1.if_ext_irec);
4469 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4473 * This is called to clean up large amounts of unused memory allocated
4474 * by the indirection array. Before compacting anything though, verify
4475 * that the indirection array is still needed and switch back to the
4476 * linear extent list (or even the inline buffer) if possible. The
4477 * compaction policy is as follows:
4479 * Full Compaction: Extents fit into a single page (or inline buffer)
4480 * Partial Compaction: Extents occupy less than 50% of allocated space
4481 * No Compaction: Extents occupy at least 50% of allocated space
4484 xfs_iext_irec_compact(
4485 xfs_ifork_t *ifp) /* inode fork pointer */
4487 xfs_extnum_t nextents; /* number of extents in file */
4488 int nlists; /* number of irec's (ex lists) */
4490 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4491 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4492 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4494 if (nextents == 0) {
4495 xfs_iext_destroy(ifp);
4496 } else if (nextents <= XFS_INLINE_EXTS) {
4497 xfs_iext_indirect_to_direct(ifp);
4498 xfs_iext_direct_to_inline(ifp, nextents);
4499 } else if (nextents <= XFS_LINEAR_EXTS) {
4500 xfs_iext_indirect_to_direct(ifp);
4501 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
4502 xfs_iext_irec_compact_pages(ifp);
4507 * Combine extents from neighboring extent pages.
4510 xfs_iext_irec_compact_pages(
4511 xfs_ifork_t *ifp) /* inode fork pointer */
4513 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
4514 int erp_idx = 0; /* indirection array index */
4515 int nlists; /* number of irec's (ex lists) */
4517 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4518 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4519 while (erp_idx < nlists - 1) {
4520 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4522 if (erp_next->er_extcount <=
4523 (XFS_LINEAR_EXTS - erp->er_extcount)) {
4524 memcpy(&erp->er_extbuf[erp->er_extcount],
4525 erp_next->er_extbuf, erp_next->er_extcount *
4526 sizeof(xfs_bmbt_rec_t));
4527 erp->er_extcount += erp_next->er_extcount;
4529 * Free page before removing extent record
4530 * so er_extoffs don't get modified in
4531 * xfs_iext_irec_remove.
4533 kmem_free(erp_next->er_extbuf);
4534 erp_next->er_extbuf = NULL;
4535 xfs_iext_irec_remove(ifp, erp_idx + 1);
4536 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4544 * This is called to update the er_extoff field in the indirection
4545 * array when extents have been added or removed from one of the
4546 * extent lists. erp_idx contains the irec index to begin updating
4547 * at and ext_diff contains the number of extents that were added
4551 xfs_iext_irec_update_extoffs(
4552 xfs_ifork_t *ifp, /* inode fork pointer */
4553 int erp_idx, /* irec index to update */
4554 int ext_diff) /* number of new extents */
4556 int i; /* loop counter */
4557 int nlists; /* number of irec's (ex lists */
4559 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4560 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4561 for (i = erp_idx; i < nlists; i++) {
4562 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;