2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 #include <linux/log2.h>
22 #include "xfs_types.h"
27 #include "xfs_trans.h"
28 #include "xfs_trans_priv.h"
32 #include "xfs_dmapi.h"
33 #include "xfs_mount.h"
34 #include "xfs_bmap_btree.h"
35 #include "xfs_alloc_btree.h"
36 #include "xfs_ialloc_btree.h"
37 #include "xfs_dir2_sf.h"
38 #include "xfs_attr_sf.h"
39 #include "xfs_dinode.h"
40 #include "xfs_inode.h"
41 #include "xfs_buf_item.h"
42 #include "xfs_inode_item.h"
43 #include "xfs_btree.h"
44 #include "xfs_btree_trace.h"
45 #include "xfs_alloc.h"
46 #include "xfs_ialloc.h"
49 #include "xfs_error.h"
50 #include "xfs_utils.h"
51 #include "xfs_dir2_trace.h"
52 #include "xfs_quota.h"
54 #include "xfs_filestream.h"
55 #include "xfs_vnodeops.h"
57 kmem_zone_t *xfs_ifork_zone;
58 kmem_zone_t *xfs_inode_zone;
61 * Used in xfs_itruncate(). This is the maximum number of extents
62 * freed from a file in a single transaction.
64 #define XFS_ITRUNC_MAX_EXTENTS 2
66 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
67 STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
68 STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
69 STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
73 * Make sure that the extents in the given memory buffer
83 xfs_bmbt_rec_host_t rec;
86 for (i = 0; i < nrecs; i++) {
87 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
88 rec.l0 = get_unaligned(&ep->l0);
89 rec.l1 = get_unaligned(&ep->l1);
90 xfs_bmbt_get_all(&rec, &irec);
91 if (fmt == XFS_EXTFMT_NOSTATE)
92 ASSERT(irec.br_state == XFS_EXT_NORM);
96 #define xfs_validate_extents(ifp, nrecs, fmt)
100 * Check that none of the inode's in the buffer have a next
101 * unlinked field of 0.
113 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
115 for (i = 0; i < j; i++) {
116 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
117 i * mp->m_sb.sb_inodesize);
118 if (!dip->di_next_unlinked) {
119 xfs_fs_cmn_err(CE_ALERT, mp,
120 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
122 ASSERT(dip->di_next_unlinked);
129 * Find the buffer associated with the given inode map
130 * We do basic validation checks on the buffer once it has been
131 * retrieved from disk.
147 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno,
148 (int)imap->im_len, buf_flags, &bp);
150 if (error != EAGAIN) {
152 "xfs_imap_to_bp: xfs_trans_read_buf()returned "
153 "an error %d on %s. Returning error.",
154 error, mp->m_fsname);
156 ASSERT(buf_flags & XFS_BUF_TRYLOCK);
162 * Validate the magic number and version of every inode in the buffer
163 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
166 ni = BBTOB(imap->im_len) >> mp->m_sb.sb_inodelog;
167 #else /* usual case */
171 for (i = 0; i < ni; i++) {
175 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
176 (i << mp->m_sb.sb_inodelog));
177 di_ok = be16_to_cpu(dip->di_core.di_magic) == XFS_DINODE_MAGIC &&
178 XFS_DINODE_GOOD_VERSION(dip->di_core.di_version);
179 if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
180 XFS_ERRTAG_ITOBP_INOTOBP,
181 XFS_RANDOM_ITOBP_INOTOBP))) {
182 if (imap_flags & XFS_IMAP_BULKSTAT) {
183 xfs_trans_brelse(tp, bp);
184 return XFS_ERROR(EINVAL);
186 XFS_CORRUPTION_ERROR("xfs_imap_to_bp",
187 XFS_ERRLEVEL_HIGH, mp, dip);
190 "Device %s - bad inode magic/vsn "
191 "daddr %lld #%d (magic=%x)",
192 XFS_BUFTARG_NAME(mp->m_ddev_targp),
193 (unsigned long long)imap->im_blkno, i,
194 be16_to_cpu(dip->di_core.di_magic));
196 xfs_trans_brelse(tp, bp);
197 return XFS_ERROR(EFSCORRUPTED);
201 xfs_inobp_check(mp, bp);
204 * Mark the buffer as an inode buffer now that it looks good
206 XFS_BUF_SET_VTYPE(bp, B_FS_INO);
213 * This routine is called to map an inode number within a file
214 * system to the buffer containing the on-disk version of the
215 * inode. It returns a pointer to the buffer containing the
216 * on-disk inode in the bpp parameter, and in the dip parameter
217 * it returns a pointer to the on-disk inode within that buffer.
219 * If a non-zero error is returned, then the contents of bpp and
220 * dipp are undefined.
222 * Use xfs_imap() to determine the size and location of the
223 * buffer to read from disk.
239 error = xfs_imap(mp, tp, ino, &imap, XFS_IMAP_LOOKUP);
243 error = xfs_imap_to_bp(mp, tp, &imap, &bp, XFS_BUF_LOCK, 0);
247 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
249 *offset = imap.im_boffset;
255 * This routine is called to map an inode to the buffer containing
256 * the on-disk version of the inode. It returns a pointer to the
257 * buffer containing the on-disk inode in the bpp parameter, and in
258 * the dip parameter it returns a pointer to the on-disk inode within
261 * If a non-zero error is returned, then the contents of bpp and
262 * dipp are undefined.
264 * If the inode is new and has not yet been initialized, use xfs_imap()
265 * to determine the size and location of the buffer to read from disk.
266 * If the inode has already been mapped to its buffer and read in once,
267 * then use the mapping information stored in the inode rather than
268 * calling xfs_imap(). This allows us to avoid the overhead of looking
269 * at the inode btree for small block file systems (see xfs_dilocate()).
270 * We can tell whether the inode has been mapped in before by comparing
271 * its disk block address to 0. Only uninitialized inodes will have
272 * 0 for the disk block address.
289 if (ip->i_blkno == (xfs_daddr_t)0) {
291 error = xfs_imap(mp, tp, ip->i_ino, &imap,
292 XFS_IMAP_LOOKUP | imap_flags);
297 * Fill in the fields in the inode that will be used to
298 * map the inode to its buffer from now on.
300 ip->i_blkno = imap.im_blkno;
301 ip->i_len = imap.im_len;
302 ip->i_boffset = imap.im_boffset;
305 * We've already mapped the inode once, so just use the
306 * mapping that we saved the first time.
308 imap.im_blkno = ip->i_blkno;
309 imap.im_len = ip->i_len;
310 imap.im_boffset = ip->i_boffset;
312 ASSERT(bno == 0 || bno == imap.im_blkno);
314 error = xfs_imap_to_bp(mp, tp, &imap, &bp, buf_flags, imap_flags);
319 ASSERT(buf_flags & XFS_BUF_TRYLOCK);
325 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
331 * Move inode type and inode format specific information from the
332 * on-disk inode to the in-core inode. For fifos, devs, and sockets
333 * this means set if_rdev to the proper value. For files, directories,
334 * and symlinks this means to bring in the in-line data or extent
335 * pointers. For a file in B-tree format, only the root is immediately
336 * brought in-core. The rest will be in-lined in if_extents when it
337 * is first referenced (see xfs_iread_extents()).
344 xfs_attr_shortform_t *atp;
348 ip->i_df.if_ext_max =
349 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
352 if (unlikely(be32_to_cpu(dip->di_core.di_nextents) +
353 be16_to_cpu(dip->di_core.di_anextents) >
354 be64_to_cpu(dip->di_core.di_nblocks))) {
355 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
356 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
357 (unsigned long long)ip->i_ino,
358 (int)(be32_to_cpu(dip->di_core.di_nextents) +
359 be16_to_cpu(dip->di_core.di_anextents)),
361 be64_to_cpu(dip->di_core.di_nblocks));
362 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
364 return XFS_ERROR(EFSCORRUPTED);
367 if (unlikely(dip->di_core.di_forkoff > ip->i_mount->m_sb.sb_inodesize)) {
368 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
369 "corrupt dinode %Lu, forkoff = 0x%x.",
370 (unsigned long long)ip->i_ino,
371 dip->di_core.di_forkoff);
372 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
374 return XFS_ERROR(EFSCORRUPTED);
377 switch (ip->i_d.di_mode & S_IFMT) {
382 if (unlikely(dip->di_core.di_format != XFS_DINODE_FMT_DEV)) {
383 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
385 return XFS_ERROR(EFSCORRUPTED);
389 ip->i_df.if_u2.if_rdev = be32_to_cpu(dip->di_u.di_dev);
395 switch (dip->di_core.di_format) {
396 case XFS_DINODE_FMT_LOCAL:
398 * no local regular files yet
400 if (unlikely((be16_to_cpu(dip->di_core.di_mode) & S_IFMT) == S_IFREG)) {
401 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
403 "(local format for regular file).",
404 (unsigned long long) ip->i_ino);
405 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
408 return XFS_ERROR(EFSCORRUPTED);
411 di_size = be64_to_cpu(dip->di_core.di_size);
412 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
413 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
415 "(bad size %Ld for local inode).",
416 (unsigned long long) ip->i_ino,
417 (long long) di_size);
418 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
421 return XFS_ERROR(EFSCORRUPTED);
425 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
427 case XFS_DINODE_FMT_EXTENTS:
428 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
430 case XFS_DINODE_FMT_BTREE:
431 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
434 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
436 return XFS_ERROR(EFSCORRUPTED);
441 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
442 return XFS_ERROR(EFSCORRUPTED);
447 if (!XFS_DFORK_Q(dip))
449 ASSERT(ip->i_afp == NULL);
450 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
451 ip->i_afp->if_ext_max =
452 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
453 switch (dip->di_core.di_aformat) {
454 case XFS_DINODE_FMT_LOCAL:
455 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
456 size = be16_to_cpu(atp->hdr.totsize);
457 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
459 case XFS_DINODE_FMT_EXTENTS:
460 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
462 case XFS_DINODE_FMT_BTREE:
463 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
466 error = XFS_ERROR(EFSCORRUPTED);
470 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
472 xfs_idestroy_fork(ip, XFS_DATA_FORK);
478 * The file is in-lined in the on-disk inode.
479 * If it fits into if_inline_data, then copy
480 * it there, otherwise allocate a buffer for it
481 * and copy the data there. Either way, set
482 * if_data to point at the data.
483 * If we allocate a buffer for the data, make
484 * sure that its size is a multiple of 4 and
485 * record the real size in i_real_bytes.
498 * If the size is unreasonable, then something
499 * is wrong and we just bail out rather than crash in
500 * kmem_alloc() or memcpy() below.
502 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
503 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
505 "(bad size %d for local fork, size = %d).",
506 (unsigned long long) ip->i_ino, size,
507 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
508 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
510 return XFS_ERROR(EFSCORRUPTED);
512 ifp = XFS_IFORK_PTR(ip, whichfork);
515 ifp->if_u1.if_data = NULL;
516 else if (size <= sizeof(ifp->if_u2.if_inline_data))
517 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
519 real_size = roundup(size, 4);
520 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
522 ifp->if_bytes = size;
523 ifp->if_real_bytes = real_size;
525 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
526 ifp->if_flags &= ~XFS_IFEXTENTS;
527 ifp->if_flags |= XFS_IFINLINE;
532 * The file consists of a set of extents all
533 * of which fit into the on-disk inode.
534 * If there are few enough extents to fit into
535 * the if_inline_ext, then copy them there.
536 * Otherwise allocate a buffer for them and copy
537 * them into it. Either way, set if_extents
538 * to point at the extents.
552 ifp = XFS_IFORK_PTR(ip, whichfork);
553 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
554 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
557 * If the number of extents is unreasonable, then something
558 * is wrong and we just bail out rather than crash in
559 * kmem_alloc() or memcpy() below.
561 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
562 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
563 "corrupt inode %Lu ((a)extents = %d).",
564 (unsigned long long) ip->i_ino, nex);
565 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
567 return XFS_ERROR(EFSCORRUPTED);
570 ifp->if_real_bytes = 0;
572 ifp->if_u1.if_extents = NULL;
573 else if (nex <= XFS_INLINE_EXTS)
574 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
576 xfs_iext_add(ifp, 0, nex);
578 ifp->if_bytes = size;
580 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
581 xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip));
582 for (i = 0; i < nex; i++, dp++) {
583 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
584 ep->l0 = get_unaligned_be64(&dp->l0);
585 ep->l1 = get_unaligned_be64(&dp->l1);
587 XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork);
588 if (whichfork != XFS_DATA_FORK ||
589 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
590 if (unlikely(xfs_check_nostate_extents(
592 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
595 return XFS_ERROR(EFSCORRUPTED);
598 ifp->if_flags |= XFS_IFEXTENTS;
603 * The file has too many extents to fit into
604 * the inode, so they are in B-tree format.
605 * Allocate a buffer for the root of the B-tree
606 * and copy the root into it. The i_extents
607 * field will remain NULL until all of the
608 * extents are read in (when they are needed).
616 xfs_bmdr_block_t *dfp;
622 ifp = XFS_IFORK_PTR(ip, whichfork);
623 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
624 size = XFS_BMAP_BROOT_SPACE(dfp);
625 nrecs = XFS_BMAP_BROOT_NUMRECS(dfp);
628 * blow out if -- fork has less extents than can fit in
629 * fork (fork shouldn't be a btree format), root btree
630 * block has more records than can fit into the fork,
631 * or the number of extents is greater than the number of
634 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
635 || XFS_BMDR_SPACE_CALC(nrecs) >
636 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
637 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
638 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
639 "corrupt inode %Lu (btree).",
640 (unsigned long long) ip->i_ino);
641 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
643 return XFS_ERROR(EFSCORRUPTED);
646 ifp->if_broot_bytes = size;
647 ifp->if_broot = kmem_alloc(size, KM_SLEEP);
648 ASSERT(ifp->if_broot != NULL);
650 * Copy and convert from the on-disk structure
651 * to the in-memory structure.
653 xfs_bmdr_to_bmbt(dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
654 ifp->if_broot, size);
655 ifp->if_flags &= ~XFS_IFEXTENTS;
656 ifp->if_flags |= XFS_IFBROOT;
662 xfs_dinode_from_disk(
664 xfs_dinode_core_t *from)
666 to->di_magic = be16_to_cpu(from->di_magic);
667 to->di_mode = be16_to_cpu(from->di_mode);
668 to->di_version = from ->di_version;
669 to->di_format = from->di_format;
670 to->di_onlink = be16_to_cpu(from->di_onlink);
671 to->di_uid = be32_to_cpu(from->di_uid);
672 to->di_gid = be32_to_cpu(from->di_gid);
673 to->di_nlink = be32_to_cpu(from->di_nlink);
674 to->di_projid = be16_to_cpu(from->di_projid);
675 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
676 to->di_flushiter = be16_to_cpu(from->di_flushiter);
677 to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec);
678 to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec);
679 to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec);
680 to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec);
681 to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec);
682 to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec);
683 to->di_size = be64_to_cpu(from->di_size);
684 to->di_nblocks = be64_to_cpu(from->di_nblocks);
685 to->di_extsize = be32_to_cpu(from->di_extsize);
686 to->di_nextents = be32_to_cpu(from->di_nextents);
687 to->di_anextents = be16_to_cpu(from->di_anextents);
688 to->di_forkoff = from->di_forkoff;
689 to->di_aformat = from->di_aformat;
690 to->di_dmevmask = be32_to_cpu(from->di_dmevmask);
691 to->di_dmstate = be16_to_cpu(from->di_dmstate);
692 to->di_flags = be16_to_cpu(from->di_flags);
693 to->di_gen = be32_to_cpu(from->di_gen);
698 xfs_dinode_core_t *to,
699 xfs_icdinode_t *from)
701 to->di_magic = cpu_to_be16(from->di_magic);
702 to->di_mode = cpu_to_be16(from->di_mode);
703 to->di_version = from ->di_version;
704 to->di_format = from->di_format;
705 to->di_onlink = cpu_to_be16(from->di_onlink);
706 to->di_uid = cpu_to_be32(from->di_uid);
707 to->di_gid = cpu_to_be32(from->di_gid);
708 to->di_nlink = cpu_to_be32(from->di_nlink);
709 to->di_projid = cpu_to_be16(from->di_projid);
710 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
711 to->di_flushiter = cpu_to_be16(from->di_flushiter);
712 to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec);
713 to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec);
714 to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec);
715 to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec);
716 to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec);
717 to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec);
718 to->di_size = cpu_to_be64(from->di_size);
719 to->di_nblocks = cpu_to_be64(from->di_nblocks);
720 to->di_extsize = cpu_to_be32(from->di_extsize);
721 to->di_nextents = cpu_to_be32(from->di_nextents);
722 to->di_anextents = cpu_to_be16(from->di_anextents);
723 to->di_forkoff = from->di_forkoff;
724 to->di_aformat = from->di_aformat;
725 to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
726 to->di_dmstate = cpu_to_be16(from->di_dmstate);
727 to->di_flags = cpu_to_be16(from->di_flags);
728 to->di_gen = cpu_to_be32(from->di_gen);
737 if (di_flags & XFS_DIFLAG_ANY) {
738 if (di_flags & XFS_DIFLAG_REALTIME)
739 flags |= XFS_XFLAG_REALTIME;
740 if (di_flags & XFS_DIFLAG_PREALLOC)
741 flags |= XFS_XFLAG_PREALLOC;
742 if (di_flags & XFS_DIFLAG_IMMUTABLE)
743 flags |= XFS_XFLAG_IMMUTABLE;
744 if (di_flags & XFS_DIFLAG_APPEND)
745 flags |= XFS_XFLAG_APPEND;
746 if (di_flags & XFS_DIFLAG_SYNC)
747 flags |= XFS_XFLAG_SYNC;
748 if (di_flags & XFS_DIFLAG_NOATIME)
749 flags |= XFS_XFLAG_NOATIME;
750 if (di_flags & XFS_DIFLAG_NODUMP)
751 flags |= XFS_XFLAG_NODUMP;
752 if (di_flags & XFS_DIFLAG_RTINHERIT)
753 flags |= XFS_XFLAG_RTINHERIT;
754 if (di_flags & XFS_DIFLAG_PROJINHERIT)
755 flags |= XFS_XFLAG_PROJINHERIT;
756 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
757 flags |= XFS_XFLAG_NOSYMLINKS;
758 if (di_flags & XFS_DIFLAG_EXTSIZE)
759 flags |= XFS_XFLAG_EXTSIZE;
760 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
761 flags |= XFS_XFLAG_EXTSZINHERIT;
762 if (di_flags & XFS_DIFLAG_NODEFRAG)
763 flags |= XFS_XFLAG_NODEFRAG;
764 if (di_flags & XFS_DIFLAG_FILESTREAM)
765 flags |= XFS_XFLAG_FILESTREAM;
775 xfs_icdinode_t *dic = &ip->i_d;
777 return _xfs_dic2xflags(dic->di_flags) |
778 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
785 xfs_dinode_core_t *dic = &dip->di_core;
787 return _xfs_dic2xflags(be16_to_cpu(dic->di_flags)) |
788 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
792 * Allocate and initialise an xfs_inode.
796 struct xfs_mount *mp,
799 struct xfs_inode *ip;
802 * if this didn't occur in transactions, we could use
803 * KM_MAYFAIL and return NULL here on ENOMEM. Set the
804 * code up to do this anyway.
806 ip = kmem_zone_alloc(xfs_inode_zone, KM_SLEEP);
810 ASSERT(atomic_read(&ip->i_iocount) == 0);
811 ASSERT(atomic_read(&ip->i_pincount) == 0);
812 ASSERT(!spin_is_locked(&ip->i_flags_lock));
813 ASSERT(list_empty(&ip->i_reclaim));
821 memset(&ip->i_df, 0, sizeof(xfs_ifork_t));
823 ip->i_update_core = 0;
824 ip->i_update_size = 0;
825 ip->i_delayed_blks = 0;
826 memset(&ip->i_d, 0, sizeof(xfs_icdinode_t));
831 * Initialize inode's trace buffers.
833 #ifdef XFS_INODE_TRACE
834 ip->i_trace = ktrace_alloc(INODE_TRACE_SIZE, KM_NOFS);
836 #ifdef XFS_BMAP_TRACE
837 ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_NOFS);
839 #ifdef XFS_BTREE_TRACE
840 ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_NOFS);
843 ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_NOFS);
845 #ifdef XFS_ILOCK_TRACE
846 ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_NOFS);
848 #ifdef XFS_DIR2_TRACE
849 ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_NOFS);
856 * Given a mount structure and an inode number, return a pointer
857 * to a newly allocated in-core inode corresponding to the given
860 * Initialize the inode's attributes and extent pointers if it
861 * already has them (it will not if the inode has no links).
877 ip = xfs_inode_alloc(mp, ino);
882 * Get pointer's to the on-disk inode and the buffer containing it.
883 * If the inode number refers to a block outside the file system
884 * then xfs_itobp() will return NULL. In this case we should
885 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
886 * know that this is a new incore inode.
888 error = xfs_itobp(mp, tp, ip, &dip, &bp, bno, imap_flags, XFS_BUF_LOCK);
895 * If we got something that isn't an inode it means someone
896 * (nfs or dmi) has a stale handle.
898 if (be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC) {
900 xfs_trans_brelse(tp, bp);
902 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
903 "dip->di_core.di_magic (0x%x) != "
904 "XFS_DINODE_MAGIC (0x%x)",
905 be16_to_cpu(dip->di_core.di_magic),
908 return XFS_ERROR(EINVAL);
912 * If the on-disk inode is already linked to a directory
913 * entry, copy all of the inode into the in-core inode.
914 * xfs_iformat() handles copying in the inode format
915 * specific information.
916 * Otherwise, just get the truly permanent information.
918 if (dip->di_core.di_mode) {
919 xfs_dinode_from_disk(&ip->i_d, &dip->di_core);
920 error = xfs_iformat(ip, dip);
923 xfs_trans_brelse(tp, bp);
925 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
926 "xfs_iformat() returned error %d",
932 ip->i_d.di_magic = be16_to_cpu(dip->di_core.di_magic);
933 ip->i_d.di_version = dip->di_core.di_version;
934 ip->i_d.di_gen = be32_to_cpu(dip->di_core.di_gen);
935 ip->i_d.di_flushiter = be16_to_cpu(dip->di_core.di_flushiter);
937 * Make sure to pull in the mode here as well in
938 * case the inode is released without being used.
939 * This ensures that xfs_inactive() will see that
940 * the inode is already free and not try to mess
941 * with the uninitialized part of it.
945 * Initialize the per-fork minima and maxima for a new
946 * inode here. xfs_iformat will do it for old inodes.
948 ip->i_df.if_ext_max =
949 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
953 * The inode format changed when we moved the link count and
954 * made it 32 bits long. If this is an old format inode,
955 * convert it in memory to look like a new one. If it gets
956 * flushed to disk we will convert back before flushing or
957 * logging it. We zero out the new projid field and the old link
958 * count field. We'll handle clearing the pad field (the remains
959 * of the old uuid field) when we actually convert the inode to
960 * the new format. We don't change the version number so that we
961 * can distinguish this from a real new format inode.
963 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
964 ip->i_d.di_nlink = ip->i_d.di_onlink;
965 ip->i_d.di_onlink = 0;
966 ip->i_d.di_projid = 0;
969 ip->i_delayed_blks = 0;
970 ip->i_size = ip->i_d.di_size;
973 * Mark the buffer containing the inode as something to keep
974 * around for a while. This helps to keep recently accessed
975 * meta-data in-core longer.
977 XFS_BUF_SET_REF(bp, XFS_INO_REF);
980 * Use xfs_trans_brelse() to release the buffer containing the
981 * on-disk inode, because it was acquired with xfs_trans_read_buf()
982 * in xfs_itobp() above. If tp is NULL, this is just a normal
983 * brelse(). If we're within a transaction, then xfs_trans_brelse()
984 * will only release the buffer if it is not dirty within the
985 * transaction. It will be OK to release the buffer in this case,
986 * because inodes on disk are never destroyed and we will be
987 * locking the new in-core inode before putting it in the hash
988 * table where other processes can find it. Thus we don't have
989 * to worry about the inode being changed just because we released
992 xfs_trans_brelse(tp, bp);
998 * Read in extents from a btree-format inode.
999 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1009 xfs_extnum_t nextents;
1012 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
1013 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
1015 return XFS_ERROR(EFSCORRUPTED);
1017 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
1018 size = nextents * sizeof(xfs_bmbt_rec_t);
1019 ifp = XFS_IFORK_PTR(ip, whichfork);
1022 * We know that the size is valid (it's checked in iformat_btree)
1024 ifp->if_lastex = NULLEXTNUM;
1025 ifp->if_bytes = ifp->if_real_bytes = 0;
1026 ifp->if_flags |= XFS_IFEXTENTS;
1027 xfs_iext_add(ifp, 0, nextents);
1028 error = xfs_bmap_read_extents(tp, ip, whichfork);
1030 xfs_iext_destroy(ifp);
1031 ifp->if_flags &= ~XFS_IFEXTENTS;
1034 xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip));
1039 * Allocate an inode on disk and return a copy of its in-core version.
1040 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1041 * appropriately within the inode. The uid and gid for the inode are
1042 * set according to the contents of the given cred structure.
1044 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1045 * has a free inode available, call xfs_iget()
1046 * to obtain the in-core version of the allocated inode. Finally,
1047 * fill in the inode and log its initial contents. In this case,
1048 * ialloc_context would be set to NULL and call_again set to false.
1050 * If xfs_dialloc() does not have an available inode,
1051 * it will replenish its supply by doing an allocation. Since we can
1052 * only do one allocation within a transaction without deadlocks, we
1053 * must commit the current transaction before returning the inode itself.
1054 * In this case, therefore, we will set call_again to true and return.
1055 * The caller should then commit the current transaction, start a new
1056 * transaction, and call xfs_ialloc() again to actually get the inode.
1058 * To ensure that some other process does not grab the inode that
1059 * was allocated during the first call to xfs_ialloc(), this routine
1060 * also returns the [locked] bp pointing to the head of the freelist
1061 * as ialloc_context. The caller should hold this buffer across
1062 * the commit and pass it back into this routine on the second call.
1064 * If we are allocating quota inodes, we do not have a parent inode
1065 * to attach to or associate with (i.e. pip == NULL) because they
1066 * are not linked into the directory structure - they are attached
1067 * directly to the superblock - and so have no parent.
1079 xfs_buf_t **ialloc_context,
1080 boolean_t *call_again,
1090 * Call the space management code to pick
1091 * the on-disk inode to be allocated.
1093 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
1094 ialloc_context, call_again, &ino);
1098 if (*call_again || ino == NULLFSINO) {
1102 ASSERT(*ialloc_context == NULL);
1105 * Get the in-core inode with the lock held exclusively.
1106 * This is because we're setting fields here we need
1107 * to prevent others from looking at until we're done.
1109 error = xfs_trans_iget(tp->t_mountp, tp, ino,
1110 XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
1116 ip->i_d.di_mode = (__uint16_t)mode;
1117 ip->i_d.di_onlink = 0;
1118 ip->i_d.di_nlink = nlink;
1119 ASSERT(ip->i_d.di_nlink == nlink);
1120 ip->i_d.di_uid = current_fsuid();
1121 ip->i_d.di_gid = current_fsgid();
1122 ip->i_d.di_projid = prid;
1123 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1126 * If the superblock version is up to where we support new format
1127 * inodes and this is currently an old format inode, then change
1128 * the inode version number now. This way we only do the conversion
1129 * here rather than here and in the flush/logging code.
1131 if (xfs_sb_version_hasnlink(&tp->t_mountp->m_sb) &&
1132 ip->i_d.di_version == XFS_DINODE_VERSION_1) {
1133 ip->i_d.di_version = XFS_DINODE_VERSION_2;
1135 * We've already zeroed the old link count, the projid field,
1136 * and the pad field.
1141 * Project ids won't be stored on disk if we are using a version 1 inode.
1143 if ((prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1))
1144 xfs_bump_ino_vers2(tp, ip);
1146 if (pip && XFS_INHERIT_GID(pip)) {
1147 ip->i_d.di_gid = pip->i_d.di_gid;
1148 if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1149 ip->i_d.di_mode |= S_ISGID;
1154 * If the group ID of the new file does not match the effective group
1155 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1156 * (and only if the irix_sgid_inherit compatibility variable is set).
1158 if ((irix_sgid_inherit) &&
1159 (ip->i_d.di_mode & S_ISGID) &&
1160 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1161 ip->i_d.di_mode &= ~S_ISGID;
1164 ip->i_d.di_size = 0;
1166 ip->i_d.di_nextents = 0;
1167 ASSERT(ip->i_d.di_nblocks == 0);
1170 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
1171 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
1172 ip->i_d.di_atime = ip->i_d.di_mtime;
1173 ip->i_d.di_ctime = ip->i_d.di_mtime;
1176 * di_gen will have been taken care of in xfs_iread.
1178 ip->i_d.di_extsize = 0;
1179 ip->i_d.di_dmevmask = 0;
1180 ip->i_d.di_dmstate = 0;
1181 ip->i_d.di_flags = 0;
1182 flags = XFS_ILOG_CORE;
1183 switch (mode & S_IFMT) {
1188 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1189 ip->i_df.if_u2.if_rdev = rdev;
1190 ip->i_df.if_flags = 0;
1191 flags |= XFS_ILOG_DEV;
1194 if (pip && xfs_inode_is_filestream(pip)) {
1195 error = xfs_filestream_associate(pip, ip);
1199 xfs_iflags_set(ip, XFS_IFILESTREAM);
1203 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
1206 if ((mode & S_IFMT) == S_IFDIR) {
1207 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1208 di_flags |= XFS_DIFLAG_RTINHERIT;
1209 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1210 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1211 ip->i_d.di_extsize = pip->i_d.di_extsize;
1213 } else if ((mode & S_IFMT) == S_IFREG) {
1214 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1215 di_flags |= XFS_DIFLAG_REALTIME;
1216 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1217 di_flags |= XFS_DIFLAG_EXTSIZE;
1218 ip->i_d.di_extsize = pip->i_d.di_extsize;
1221 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1222 xfs_inherit_noatime)
1223 di_flags |= XFS_DIFLAG_NOATIME;
1224 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1226 di_flags |= XFS_DIFLAG_NODUMP;
1227 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1229 di_flags |= XFS_DIFLAG_SYNC;
1230 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1231 xfs_inherit_nosymlinks)
1232 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1233 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1234 di_flags |= XFS_DIFLAG_PROJINHERIT;
1235 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1236 xfs_inherit_nodefrag)
1237 di_flags |= XFS_DIFLAG_NODEFRAG;
1238 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
1239 di_flags |= XFS_DIFLAG_FILESTREAM;
1240 ip->i_d.di_flags |= di_flags;
1244 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1245 ip->i_df.if_flags = XFS_IFEXTENTS;
1246 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1247 ip->i_df.if_u1.if_extents = NULL;
1253 * Attribute fork settings for new inode.
1255 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1256 ip->i_d.di_anextents = 0;
1259 * Log the new values stuffed into the inode.
1261 xfs_trans_log_inode(tp, ip, flags);
1263 /* now that we have an i_mode we can setup inode ops and unlock */
1264 xfs_setup_inode(ip);
1271 * Check to make sure that there are no blocks allocated to the
1272 * file beyond the size of the file. We don't check this for
1273 * files with fixed size extents or real time extents, but we
1274 * at least do it for regular files.
1283 xfs_fileoff_t map_first;
1285 xfs_bmbt_irec_t imaps[2];
1287 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1290 if (XFS_IS_REALTIME_INODE(ip))
1293 if (ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE)
1297 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1299 * The filesystem could be shutting down, so bmapi may return
1302 if (xfs_bmapi(NULL, ip, map_first,
1304 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1306 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1309 ASSERT(nimaps == 1);
1310 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1315 * Calculate the last possible buffered byte in a file. This must
1316 * include data that was buffered beyond the EOF by the write code.
1317 * This also needs to deal with overflowing the xfs_fsize_t type
1318 * which can happen for sizes near the limit.
1320 * We also need to take into account any blocks beyond the EOF. It
1321 * may be the case that they were buffered by a write which failed.
1322 * In that case the pages will still be in memory, but the inode size
1323 * will never have been updated.
1330 xfs_fsize_t last_byte;
1331 xfs_fileoff_t last_block;
1332 xfs_fileoff_t size_last_block;
1335 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED));
1339 * Only check for blocks beyond the EOF if the extents have
1340 * been read in. This eliminates the need for the inode lock,
1341 * and it also saves us from looking when it really isn't
1344 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1345 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1353 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_size);
1354 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1356 last_byte = XFS_FSB_TO_B(mp, last_block);
1357 if (last_byte < 0) {
1358 return XFS_MAXIOFFSET(mp);
1360 last_byte += (1 << mp->m_writeio_log);
1361 if (last_byte < 0) {
1362 return XFS_MAXIOFFSET(mp);
1367 #if defined(XFS_RW_TRACE)
1373 xfs_fsize_t new_size,
1374 xfs_off_t toss_start,
1375 xfs_off_t toss_finish)
1377 if (ip->i_rwtrace == NULL) {
1381 ktrace_enter(ip->i_rwtrace,
1384 (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
1385 (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
1386 (void*)((long)flag),
1387 (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
1388 (void*)(unsigned long)(new_size & 0xffffffff),
1389 (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
1390 (void*)(unsigned long)(toss_start & 0xffffffff),
1391 (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
1392 (void*)(unsigned long)(toss_finish & 0xffffffff),
1393 (void*)(unsigned long)current_cpu(),
1394 (void*)(unsigned long)current_pid(),
1400 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1404 * Start the truncation of the file to new_size. The new size
1405 * must be smaller than the current size. This routine will
1406 * clear the buffer and page caches of file data in the removed
1407 * range, and xfs_itruncate_finish() will remove the underlying
1410 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1411 * must NOT have the inode lock held at all. This is because we're
1412 * calling into the buffer/page cache code and we can't hold the
1413 * inode lock when we do so.
1415 * We need to wait for any direct I/Os in flight to complete before we
1416 * proceed with the truncate. This is needed to prevent the extents
1417 * being read or written by the direct I/Os from being removed while the
1418 * I/O is in flight as there is no other method of synchronising
1419 * direct I/O with the truncate operation. Also, because we hold
1420 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1421 * started until the truncate completes and drops the lock. Essentially,
1422 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1423 * between direct I/Os and the truncate operation.
1425 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1426 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1427 * in the case that the caller is locking things out of order and
1428 * may not be able to call xfs_itruncate_finish() with the inode lock
1429 * held without dropping the I/O lock. If the caller must drop the
1430 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1431 * must be called again with all the same restrictions as the initial
1435 xfs_itruncate_start(
1438 xfs_fsize_t new_size)
1440 xfs_fsize_t last_byte;
1441 xfs_off_t toss_start;
1445 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1446 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1447 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1448 (flags == XFS_ITRUNC_MAYBE));
1452 /* wait for the completion of any pending DIOs */
1453 if (new_size == 0 || new_size < ip->i_size)
1457 * Call toss_pages or flushinval_pages to get rid of pages
1458 * overlapping the region being removed. We have to use
1459 * the less efficient flushinval_pages in the case that the
1460 * caller may not be able to finish the truncate without
1461 * dropping the inode's I/O lock. Make sure
1462 * to catch any pages brought in by buffers overlapping
1463 * the EOF by searching out beyond the isize by our
1464 * block size. We round new_size up to a block boundary
1465 * so that we don't toss things on the same block as
1466 * new_size but before it.
1468 * Before calling toss_page or flushinval_pages, make sure to
1469 * call remapf() over the same region if the file is mapped.
1470 * This frees up mapped file references to the pages in the
1471 * given range and for the flushinval_pages case it ensures
1472 * that we get the latest mapped changes flushed out.
1474 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1475 toss_start = XFS_FSB_TO_B(mp, toss_start);
1476 if (toss_start < 0) {
1478 * The place to start tossing is beyond our maximum
1479 * file size, so there is no way that the data extended
1484 last_byte = xfs_file_last_byte(ip);
1485 xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
1487 if (last_byte > toss_start) {
1488 if (flags & XFS_ITRUNC_DEFINITE) {
1489 xfs_tosspages(ip, toss_start,
1490 -1, FI_REMAPF_LOCKED);
1492 error = xfs_flushinval_pages(ip, toss_start,
1493 -1, FI_REMAPF_LOCKED);
1498 if (new_size == 0) {
1499 ASSERT(VN_CACHED(VFS_I(ip)) == 0);
1506 * Shrink the file to the given new_size. The new size must be smaller than
1507 * the current size. This will free up the underlying blocks in the removed
1508 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
1510 * The transaction passed to this routine must have made a permanent log
1511 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1512 * given transaction and start new ones, so make sure everything involved in
1513 * the transaction is tidy before calling here. Some transaction will be
1514 * returned to the caller to be committed. The incoming transaction must
1515 * already include the inode, and both inode locks must be held exclusively.
1516 * The inode must also be "held" within the transaction. On return the inode
1517 * will be "held" within the returned transaction. This routine does NOT
1518 * require any disk space to be reserved for it within the transaction.
1520 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1521 * indicates the fork which is to be truncated. For the attribute fork we only
1522 * support truncation to size 0.
1524 * We use the sync parameter to indicate whether or not the first transaction
1525 * we perform might have to be synchronous. For the attr fork, it needs to be
1526 * so if the unlink of the inode is not yet known to be permanent in the log.
1527 * This keeps us from freeing and reusing the blocks of the attribute fork
1528 * before the unlink of the inode becomes permanent.
1530 * For the data fork, we normally have to run synchronously if we're being
1531 * called out of the inactive path or we're being called out of the create path
1532 * where we're truncating an existing file. Either way, the truncate needs to
1533 * be sync so blocks don't reappear in the file with altered data in case of a
1534 * crash. wsync filesystems can run the first case async because anything that
1535 * shrinks the inode has to run sync so by the time we're called here from
1536 * inactive, the inode size is permanently set to 0.
1538 * Calls from the truncate path always need to be sync unless we're in a wsync
1539 * filesystem and the file has already been unlinked.
1541 * The caller is responsible for correctly setting the sync parameter. It gets
1542 * too hard for us to guess here which path we're being called out of just
1543 * based on inode state.
1545 * If we get an error, we must return with the inode locked and linked into the
1546 * current transaction. This keeps things simple for the higher level code,
1547 * because it always knows that the inode is locked and held in the transaction
1548 * that returns to it whether errors occur or not. We don't mark the inode
1549 * dirty on error so that transactions can be easily aborted if possible.
1552 xfs_itruncate_finish(
1555 xfs_fsize_t new_size,
1559 xfs_fsblock_t first_block;
1560 xfs_fileoff_t first_unmap_block;
1561 xfs_fileoff_t last_block;
1562 xfs_filblks_t unmap_len=0;
1567 xfs_bmap_free_t free_list;
1570 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
1571 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1572 ASSERT(*tp != NULL);
1573 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1574 ASSERT(ip->i_transp == *tp);
1575 ASSERT(ip->i_itemp != NULL);
1576 ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
1580 mp = (ntp)->t_mountp;
1581 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1584 * We only support truncating the entire attribute fork.
1586 if (fork == XFS_ATTR_FORK) {
1589 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1590 xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
1592 * The first thing we do is set the size to new_size permanently
1593 * on disk. This way we don't have to worry about anyone ever
1594 * being able to look at the data being freed even in the face
1595 * of a crash. What we're getting around here is the case where
1596 * we free a block, it is allocated to another file, it is written
1597 * to, and then we crash. If the new data gets written to the
1598 * file but the log buffers containing the free and reallocation
1599 * don't, then we'd end up with garbage in the blocks being freed.
1600 * As long as we make the new_size permanent before actually
1601 * freeing any blocks it doesn't matter if they get writtten to.
1603 * The callers must signal into us whether or not the size
1604 * setting here must be synchronous. There are a few cases
1605 * where it doesn't have to be synchronous. Those cases
1606 * occur if the file is unlinked and we know the unlink is
1607 * permanent or if the blocks being truncated are guaranteed
1608 * to be beyond the inode eof (regardless of the link count)
1609 * and the eof value is permanent. Both of these cases occur
1610 * only on wsync-mounted filesystems. In those cases, we're
1611 * guaranteed that no user will ever see the data in the blocks
1612 * that are being truncated so the truncate can run async.
1613 * In the free beyond eof case, the file may wind up with
1614 * more blocks allocated to it than it needs if we crash
1615 * and that won't get fixed until the next time the file
1616 * is re-opened and closed but that's ok as that shouldn't
1617 * be too many blocks.
1619 * However, we can't just make all wsync xactions run async
1620 * because there's one call out of the create path that needs
1621 * to run sync where it's truncating an existing file to size
1622 * 0 whose size is > 0.
1624 * It's probably possible to come up with a test in this
1625 * routine that would correctly distinguish all the above
1626 * cases from the values of the function parameters and the
1627 * inode state but for sanity's sake, I've decided to let the
1628 * layers above just tell us. It's simpler to correctly figure
1629 * out in the layer above exactly under what conditions we
1630 * can run async and I think it's easier for others read and
1631 * follow the logic in case something has to be changed.
1632 * cscope is your friend -- rcc.
1634 * The attribute fork is much simpler.
1636 * For the attribute fork we allow the caller to tell us whether
1637 * the unlink of the inode that led to this call is yet permanent
1638 * in the on disk log. If it is not and we will be freeing extents
1639 * in this inode then we make the first transaction synchronous
1640 * to make sure that the unlink is permanent by the time we free
1643 if (fork == XFS_DATA_FORK) {
1644 if (ip->i_d.di_nextents > 0) {
1646 * If we are not changing the file size then do
1647 * not update the on-disk file size - we may be
1648 * called from xfs_inactive_free_eofblocks(). If we
1649 * update the on-disk file size and then the system
1650 * crashes before the contents of the file are
1651 * flushed to disk then the files may be full of
1652 * holes (ie NULL files bug).
1654 if (ip->i_size != new_size) {
1655 ip->i_d.di_size = new_size;
1656 ip->i_size = new_size;
1657 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1661 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1662 if (ip->i_d.di_anextents > 0)
1663 xfs_trans_set_sync(ntp);
1665 ASSERT(fork == XFS_DATA_FORK ||
1666 (fork == XFS_ATTR_FORK &&
1667 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1668 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1671 * Since it is possible for space to become allocated beyond
1672 * the end of the file (in a crash where the space is allocated
1673 * but the inode size is not yet updated), simply remove any
1674 * blocks which show up between the new EOF and the maximum
1675 * possible file size. If the first block to be removed is
1676 * beyond the maximum file size (ie it is the same as last_block),
1677 * then there is nothing to do.
1679 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1680 ASSERT(first_unmap_block <= last_block);
1682 if (last_block == first_unmap_block) {
1685 unmap_len = last_block - first_unmap_block + 1;
1689 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1690 * will tell us whether it freed the entire range or
1691 * not. If this is a synchronous mount (wsync),
1692 * then we can tell bunmapi to keep all the
1693 * transactions asynchronous since the unlink
1694 * transaction that made this inode inactive has
1695 * already hit the disk. There's no danger of
1696 * the freed blocks being reused, there being a
1697 * crash, and the reused blocks suddenly reappearing
1698 * in this file with garbage in them once recovery
1701 XFS_BMAP_INIT(&free_list, &first_block);
1702 error = xfs_bunmapi(ntp, ip,
1703 first_unmap_block, unmap_len,
1704 XFS_BMAPI_AFLAG(fork) |
1705 (sync ? 0 : XFS_BMAPI_ASYNC),
1706 XFS_ITRUNC_MAX_EXTENTS,
1707 &first_block, &free_list,
1711 * If the bunmapi call encounters an error,
1712 * return to the caller where the transaction
1713 * can be properly aborted. We just need to
1714 * make sure we're not holding any resources
1715 * that we were not when we came in.
1717 xfs_bmap_cancel(&free_list);
1722 * Duplicate the transaction that has the permanent
1723 * reservation and commit the old transaction.
1725 error = xfs_bmap_finish(tp, &free_list, &committed);
1728 /* link the inode into the next xact in the chain */
1729 xfs_trans_ijoin(ntp, ip,
1730 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1731 xfs_trans_ihold(ntp, ip);
1736 * If the bmap finish call encounters an error, return
1737 * to the caller where the transaction can be properly
1738 * aborted. We just need to make sure we're not
1739 * holding any resources that we were not when we came
1742 * Aborting from this point might lose some blocks in
1743 * the file system, but oh well.
1745 xfs_bmap_cancel(&free_list);
1751 * Mark the inode dirty so it will be logged and
1752 * moved forward in the log as part of every commit.
1754 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1757 ntp = xfs_trans_dup(ntp);
1758 error = xfs_trans_commit(*tp, 0);
1761 /* link the inode into the next transaction in the chain */
1762 xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1763 xfs_trans_ihold(ntp, ip);
1766 error = xfs_trans_reserve(ntp, 0,
1767 XFS_ITRUNCATE_LOG_RES(mp), 0,
1768 XFS_TRANS_PERM_LOG_RES,
1769 XFS_ITRUNCATE_LOG_COUNT);
1774 * Only update the size in the case of the data fork, but
1775 * always re-log the inode so that our permanent transaction
1776 * can keep on rolling it forward in the log.
1778 if (fork == XFS_DATA_FORK) {
1779 xfs_isize_check(mp, ip, new_size);
1781 * If we are not changing the file size then do
1782 * not update the on-disk file size - we may be
1783 * called from xfs_inactive_free_eofblocks(). If we
1784 * update the on-disk file size and then the system
1785 * crashes before the contents of the file are
1786 * flushed to disk then the files may be full of
1787 * holes (ie NULL files bug).
1789 if (ip->i_size != new_size) {
1790 ip->i_d.di_size = new_size;
1791 ip->i_size = new_size;
1794 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1795 ASSERT((new_size != 0) ||
1796 (fork == XFS_ATTR_FORK) ||
1797 (ip->i_delayed_blks == 0));
1798 ASSERT((new_size != 0) ||
1799 (fork == XFS_ATTR_FORK) ||
1800 (ip->i_d.di_nextents == 0));
1801 xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
1806 * This is called when the inode's link count goes to 0.
1807 * We place the on-disk inode on a list in the AGI. It
1808 * will be pulled from this list when the inode is freed.
1820 xfs_agnumber_t agno;
1821 xfs_daddr_t agdaddr;
1828 ASSERT(ip->i_d.di_nlink == 0);
1829 ASSERT(ip->i_d.di_mode != 0);
1830 ASSERT(ip->i_transp == tp);
1834 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1835 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1838 * Get the agi buffer first. It ensures lock ordering
1841 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1842 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1847 * Validate the magic number of the agi block.
1849 agi = XFS_BUF_TO_AGI(agibp);
1851 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1852 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
1853 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK,
1854 XFS_RANDOM_IUNLINK))) {
1855 XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi);
1856 xfs_trans_brelse(tp, agibp);
1857 return XFS_ERROR(EFSCORRUPTED);
1860 * Get the index into the agi hash table for the
1861 * list this inode will go on.
1863 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1865 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1866 ASSERT(agi->agi_unlinked[bucket_index]);
1867 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1869 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
1871 * There is already another inode in the bucket we need
1872 * to add ourselves to. Add us at the front of the list.
1873 * Here we put the head pointer into our next pointer,
1874 * and then we fall through to point the head at us.
1876 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
1880 ASSERT(be32_to_cpu(dip->di_next_unlinked) == NULLAGINO);
1881 /* both on-disk, don't endian flip twice */
1882 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1883 offset = ip->i_boffset +
1884 offsetof(xfs_dinode_t, di_next_unlinked);
1885 xfs_trans_inode_buf(tp, ibp);
1886 xfs_trans_log_buf(tp, ibp, offset,
1887 (offset + sizeof(xfs_agino_t) - 1));
1888 xfs_inobp_check(mp, ibp);
1892 * Point the bucket head pointer at the inode being inserted.
1895 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1896 offset = offsetof(xfs_agi_t, agi_unlinked) +
1897 (sizeof(xfs_agino_t) * bucket_index);
1898 xfs_trans_log_buf(tp, agibp, offset,
1899 (offset + sizeof(xfs_agino_t) - 1));
1904 * Pull the on-disk inode from the AGI unlinked list.
1917 xfs_agnumber_t agno;
1918 xfs_daddr_t agdaddr;
1920 xfs_agino_t next_agino;
1921 xfs_buf_t *last_ibp;
1922 xfs_dinode_t *last_dip = NULL;
1924 int offset, last_offset = 0;
1929 * First pull the on-disk inode from the AGI unlinked list.
1933 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1934 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1937 * Get the agi buffer first. It ensures lock ordering
1940 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1941 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1944 "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
1945 error, mp->m_fsname);
1949 * Validate the magic number of the agi block.
1951 agi = XFS_BUF_TO_AGI(agibp);
1953 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1954 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
1955 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE,
1956 XFS_RANDOM_IUNLINK_REMOVE))) {
1957 XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW,
1959 xfs_trans_brelse(tp, agibp);
1961 "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
1963 return XFS_ERROR(EFSCORRUPTED);
1966 * Get the index into the agi hash table for the
1967 * list this inode will go on.
1969 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1971 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1972 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
1973 ASSERT(agi->agi_unlinked[bucket_index]);
1975 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
1977 * We're at the head of the list. Get the inode's
1978 * on-disk buffer to see if there is anyone after us
1979 * on the list. Only modify our next pointer if it
1980 * is not already NULLAGINO. This saves us the overhead
1981 * of dealing with the buffer when there is no need to
1984 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
1987 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
1988 error, mp->m_fsname);
1991 next_agino = be32_to_cpu(dip->di_next_unlinked);
1992 ASSERT(next_agino != 0);
1993 if (next_agino != NULLAGINO) {
1994 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
1995 offset = ip->i_boffset +
1996 offsetof(xfs_dinode_t, di_next_unlinked);
1997 xfs_trans_inode_buf(tp, ibp);
1998 xfs_trans_log_buf(tp, ibp, offset,
1999 (offset + sizeof(xfs_agino_t) - 1));
2000 xfs_inobp_check(mp, ibp);
2002 xfs_trans_brelse(tp, ibp);
2005 * Point the bucket head pointer at the next inode.
2007 ASSERT(next_agino != 0);
2008 ASSERT(next_agino != agino);
2009 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2010 offset = offsetof(xfs_agi_t, agi_unlinked) +
2011 (sizeof(xfs_agino_t) * bucket_index);
2012 xfs_trans_log_buf(tp, agibp, offset,
2013 (offset + sizeof(xfs_agino_t) - 1));
2016 * We need to search the list for the inode being freed.
2018 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2020 while (next_agino != agino) {
2022 * If the last inode wasn't the one pointing to
2023 * us, then release its buffer since we're not
2024 * going to do anything with it.
2026 if (last_ibp != NULL) {
2027 xfs_trans_brelse(tp, last_ibp);
2029 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2030 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
2031 &last_ibp, &last_offset);
2034 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
2035 error, mp->m_fsname);
2038 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2039 ASSERT(next_agino != NULLAGINO);
2040 ASSERT(next_agino != 0);
2043 * Now last_ibp points to the buffer previous to us on
2044 * the unlinked list. Pull us from the list.
2046 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
2049 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2050 error, mp->m_fsname);
2053 next_agino = be32_to_cpu(dip->di_next_unlinked);
2054 ASSERT(next_agino != 0);
2055 ASSERT(next_agino != agino);
2056 if (next_agino != NULLAGINO) {
2057 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2058 offset = ip->i_boffset +
2059 offsetof(xfs_dinode_t, di_next_unlinked);
2060 xfs_trans_inode_buf(tp, ibp);
2061 xfs_trans_log_buf(tp, ibp, offset,
2062 (offset + sizeof(xfs_agino_t) - 1));
2063 xfs_inobp_check(mp, ibp);
2065 xfs_trans_brelse(tp, ibp);
2068 * Point the previous inode on the list to the next inode.
2070 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2071 ASSERT(next_agino != 0);
2072 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2073 xfs_trans_inode_buf(tp, last_ibp);
2074 xfs_trans_log_buf(tp, last_ibp, offset,
2075 (offset + sizeof(xfs_agino_t) - 1));
2076 xfs_inobp_check(mp, last_ibp);
2083 xfs_inode_t *free_ip,
2087 xfs_mount_t *mp = free_ip->i_mount;
2088 int blks_per_cluster;
2091 int i, j, found, pre_flushed;
2094 xfs_inode_t *ip, **ip_found;
2095 xfs_inode_log_item_t *iip;
2096 xfs_log_item_t *lip;
2097 xfs_perag_t *pag = xfs_get_perag(mp, inum);
2099 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
2100 blks_per_cluster = 1;
2101 ninodes = mp->m_sb.sb_inopblock;
2102 nbufs = XFS_IALLOC_BLOCKS(mp);
2104 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
2105 mp->m_sb.sb_blocksize;
2106 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
2107 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
2110 ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
2112 for (j = 0; j < nbufs; j++, inum += ninodes) {
2113 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2114 XFS_INO_TO_AGBNO(mp, inum));
2118 * Look for each inode in memory and attempt to lock it,
2119 * we can be racing with flush and tail pushing here.
2120 * any inode we get the locks on, add to an array of
2121 * inode items to process later.
2123 * The get the buffer lock, we could beat a flush
2124 * or tail pushing thread to the lock here, in which
2125 * case they will go looking for the inode buffer
2126 * and fail, we need some other form of interlock
2130 for (i = 0; i < ninodes; i++) {
2131 read_lock(&pag->pag_ici_lock);
2132 ip = radix_tree_lookup(&pag->pag_ici_root,
2133 XFS_INO_TO_AGINO(mp, (inum + i)));
2135 /* Inode not in memory or we found it already,
2138 if (!ip || xfs_iflags_test(ip, XFS_ISTALE)) {
2139 read_unlock(&pag->pag_ici_lock);
2143 if (xfs_inode_clean(ip)) {
2144 read_unlock(&pag->pag_ici_lock);
2148 /* If we can get the locks then add it to the
2149 * list, otherwise by the time we get the bp lock
2150 * below it will already be attached to the
2154 /* This inode will already be locked - by us, lets
2158 if (ip == free_ip) {
2159 if (xfs_iflock_nowait(ip)) {
2160 xfs_iflags_set(ip, XFS_ISTALE);
2161 if (xfs_inode_clean(ip)) {
2164 ip_found[found++] = ip;
2167 read_unlock(&pag->pag_ici_lock);
2171 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2172 if (xfs_iflock_nowait(ip)) {
2173 xfs_iflags_set(ip, XFS_ISTALE);
2175 if (xfs_inode_clean(ip)) {
2177 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2179 ip_found[found++] = ip;
2182 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2185 read_unlock(&pag->pag_ici_lock);
2188 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2189 mp->m_bsize * blks_per_cluster,
2193 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
2195 if (lip->li_type == XFS_LI_INODE) {
2196 iip = (xfs_inode_log_item_t *)lip;
2197 ASSERT(iip->ili_logged == 1);
2198 lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
2199 spin_lock(&mp->m_ail_lock);
2200 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2201 spin_unlock(&mp->m_ail_lock);
2202 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2205 lip = lip->li_bio_list;
2208 for (i = 0; i < found; i++) {
2213 ip->i_update_core = 0;
2215 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2219 iip->ili_last_fields = iip->ili_format.ilf_fields;
2220 iip->ili_format.ilf_fields = 0;
2221 iip->ili_logged = 1;
2222 spin_lock(&mp->m_ail_lock);
2223 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2224 spin_unlock(&mp->m_ail_lock);
2226 xfs_buf_attach_iodone(bp,
2227 (void(*)(xfs_buf_t*,xfs_log_item_t*))
2228 xfs_istale_done, (xfs_log_item_t *)iip);
2229 if (ip != free_ip) {
2230 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2234 if (found || pre_flushed)
2235 xfs_trans_stale_inode_buf(tp, bp);
2236 xfs_trans_binval(tp, bp);
2239 kmem_free(ip_found);
2240 xfs_put_perag(mp, pag);
2244 * This is called to return an inode to the inode free list.
2245 * The inode should already be truncated to 0 length and have
2246 * no pages associated with it. This routine also assumes that
2247 * the inode is already a part of the transaction.
2249 * The on-disk copy of the inode will have been added to the list
2250 * of unlinked inodes in the AGI. We need to remove the inode from
2251 * that list atomically with respect to freeing it here.
2257 xfs_bmap_free_t *flist)
2261 xfs_ino_t first_ino;
2265 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2266 ASSERT(ip->i_transp == tp);
2267 ASSERT(ip->i_d.di_nlink == 0);
2268 ASSERT(ip->i_d.di_nextents == 0);
2269 ASSERT(ip->i_d.di_anextents == 0);
2270 ASSERT((ip->i_d.di_size == 0 && ip->i_size == 0) ||
2271 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2272 ASSERT(ip->i_d.di_nblocks == 0);
2275 * Pull the on-disk inode from the AGI unlinked list.
2277 error = xfs_iunlink_remove(tp, ip);
2282 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2286 ip->i_d.di_mode = 0; /* mark incore inode as free */
2287 ip->i_d.di_flags = 0;
2288 ip->i_d.di_dmevmask = 0;
2289 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2290 ip->i_df.if_ext_max =
2291 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2292 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2293 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2295 * Bump the generation count so no one will be confused
2296 * by reincarnations of this inode.
2300 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2302 error = xfs_itobp(ip->i_mount, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
2307 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2308 * from picking up this inode when it is reclaimed (its incore state
2309 * initialzed but not flushed to disk yet). The in-core di_mode is
2310 * already cleared and a corresponding transaction logged.
2311 * The hack here just synchronizes the in-core to on-disk
2312 * di_mode value in advance before the actual inode sync to disk.
2313 * This is OK because the inode is already unlinked and would never
2314 * change its di_mode again for this inode generation.
2315 * This is a temporary hack that would require a proper fix
2318 dip->di_core.di_mode = 0;
2321 xfs_ifree_cluster(ip, tp, first_ino);
2328 * Reallocate the space for if_broot based on the number of records
2329 * being added or deleted as indicated in rec_diff. Move the records
2330 * and pointers in if_broot to fit the new size. When shrinking this
2331 * will eliminate holes between the records and pointers created by
2332 * the caller. When growing this will create holes to be filled in
2335 * The caller must not request to add more records than would fit in
2336 * the on-disk inode root. If the if_broot is currently NULL, then
2337 * if we adding records one will be allocated. The caller must also
2338 * not request that the number of records go below zero, although
2339 * it can go to zero.
2341 * ip -- the inode whose if_broot area is changing
2342 * ext_diff -- the change in the number of records, positive or negative,
2343 * requested for the if_broot array.
2353 xfs_bmbt_block_t *new_broot;
2360 * Handle the degenerate case quietly.
2362 if (rec_diff == 0) {
2366 ifp = XFS_IFORK_PTR(ip, whichfork);
2369 * If there wasn't any memory allocated before, just
2370 * allocate it now and get out.
2372 if (ifp->if_broot_bytes == 0) {
2373 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
2374 ifp->if_broot = (xfs_bmbt_block_t*)kmem_alloc(new_size,
2376 ifp->if_broot_bytes = (int)new_size;
2381 * If there is already an existing if_broot, then we need
2382 * to realloc() it and shift the pointers to their new
2383 * location. The records don't change location because
2384 * they are kept butted up against the btree block header.
2386 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2387 new_max = cur_max + rec_diff;
2388 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2389 ifp->if_broot = (xfs_bmbt_block_t *)
2390 kmem_realloc(ifp->if_broot,
2392 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2394 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2395 ifp->if_broot_bytes);
2396 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2398 ifp->if_broot_bytes = (int)new_size;
2399 ASSERT(ifp->if_broot_bytes <=
2400 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2401 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2406 * rec_diff is less than 0. In this case, we are shrinking the
2407 * if_broot buffer. It must already exist. If we go to zero
2408 * records, just get rid of the root and clear the status bit.
2410 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
2411 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2412 new_max = cur_max + rec_diff;
2413 ASSERT(new_max >= 0);
2415 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2419 new_broot = (xfs_bmbt_block_t *)kmem_alloc(new_size, KM_SLEEP);
2421 * First copy over the btree block header.
2423 memcpy(new_broot, ifp->if_broot, sizeof(xfs_bmbt_block_t));
2426 ifp->if_flags &= ~XFS_IFBROOT;
2430 * Only copy the records and pointers if there are any.
2434 * First copy the records.
2436 op = (char *)XFS_BMAP_BROOT_REC_ADDR(ifp->if_broot, 1,
2437 ifp->if_broot_bytes);
2438 np = (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot, 1,
2440 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2443 * Then copy the pointers.
2445 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2446 ifp->if_broot_bytes);
2447 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot, 1,
2449 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2451 kmem_free(ifp->if_broot);
2452 ifp->if_broot = new_broot;
2453 ifp->if_broot_bytes = (int)new_size;
2454 ASSERT(ifp->if_broot_bytes <=
2455 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2461 * This is called when the amount of space needed for if_data
2462 * is increased or decreased. The change in size is indicated by
2463 * the number of bytes that need to be added or deleted in the
2464 * byte_diff parameter.
2466 * If the amount of space needed has decreased below the size of the
2467 * inline buffer, then switch to using the inline buffer. Otherwise,
2468 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2469 * to what is needed.
2471 * ip -- the inode whose if_data area is changing
2472 * byte_diff -- the change in the number of bytes, positive or negative,
2473 * requested for the if_data array.
2485 if (byte_diff == 0) {
2489 ifp = XFS_IFORK_PTR(ip, whichfork);
2490 new_size = (int)ifp->if_bytes + byte_diff;
2491 ASSERT(new_size >= 0);
2493 if (new_size == 0) {
2494 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2495 kmem_free(ifp->if_u1.if_data);
2497 ifp->if_u1.if_data = NULL;
2499 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2501 * If the valid extents/data can fit in if_inline_ext/data,
2502 * copy them from the malloc'd vector and free it.
2504 if (ifp->if_u1.if_data == NULL) {
2505 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2506 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2507 ASSERT(ifp->if_real_bytes != 0);
2508 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2510 kmem_free(ifp->if_u1.if_data);
2511 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2516 * Stuck with malloc/realloc.
2517 * For inline data, the underlying buffer must be
2518 * a multiple of 4 bytes in size so that it can be
2519 * logged and stay on word boundaries. We enforce
2522 real_size = roundup(new_size, 4);
2523 if (ifp->if_u1.if_data == NULL) {
2524 ASSERT(ifp->if_real_bytes == 0);
2525 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2526 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2528 * Only do the realloc if the underlying size
2529 * is really changing.
2531 if (ifp->if_real_bytes != real_size) {
2532 ifp->if_u1.if_data =
2533 kmem_realloc(ifp->if_u1.if_data,
2539 ASSERT(ifp->if_real_bytes == 0);
2540 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2541 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2545 ifp->if_real_bytes = real_size;
2546 ifp->if_bytes = new_size;
2547 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2554 * Map inode to disk block and offset.
2556 * mp -- the mount point structure for the current file system
2557 * tp -- the current transaction
2558 * ino -- the inode number of the inode to be located
2559 * imap -- this structure is filled in with the information necessary
2560 * to retrieve the given inode from disk
2561 * flags -- flags to pass to xfs_dilocate indicating whether or not
2562 * lookups in the inode btree were OK or not
2572 xfs_fsblock_t fsbno;
2577 fsbno = imap->im_blkno ?
2578 XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK;
2579 error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags);
2583 imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno);
2584 imap->im_len = XFS_FSB_TO_BB(mp, len);
2585 imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno);
2586 imap->im_ioffset = (ushort)off;
2587 imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog);
2590 * If the inode number maps to a block outside the bounds
2591 * of the file system then return NULL rather than calling
2592 * read_buf and panicing when we get an error from the
2595 if ((imap->im_blkno + imap->im_len) >
2596 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
2597 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
2598 "(imap->im_blkno (0x%llx) + imap->im_len (0x%llx)) > "
2599 " XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks) (0x%llx)",
2600 (unsigned long long) imap->im_blkno,
2601 (unsigned long long) imap->im_len,
2602 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
2615 ifp = XFS_IFORK_PTR(ip, whichfork);
2616 if (ifp->if_broot != NULL) {
2617 kmem_free(ifp->if_broot);
2618 ifp->if_broot = NULL;
2622 * If the format is local, then we can't have an extents
2623 * array so just look for an inline data array. If we're
2624 * not local then we may or may not have an extents list,
2625 * so check and free it up if we do.
2627 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2628 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2629 (ifp->if_u1.if_data != NULL)) {
2630 ASSERT(ifp->if_real_bytes != 0);
2631 kmem_free(ifp->if_u1.if_data);
2632 ifp->if_u1.if_data = NULL;
2633 ifp->if_real_bytes = 0;
2635 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2636 ((ifp->if_flags & XFS_IFEXTIREC) ||
2637 ((ifp->if_u1.if_extents != NULL) &&
2638 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
2639 ASSERT(ifp->if_real_bytes != 0);
2640 xfs_iext_destroy(ifp);
2642 ASSERT(ifp->if_u1.if_extents == NULL ||
2643 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2644 ASSERT(ifp->if_real_bytes == 0);
2645 if (whichfork == XFS_ATTR_FORK) {
2646 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2652 * This is called free all the memory associated with an inode.
2653 * It must free the inode itself and any buffers allocated for
2654 * if_extents/if_data and if_broot. It must also free the lock
2655 * associated with the inode.
2661 switch (ip->i_d.di_mode & S_IFMT) {
2665 xfs_idestroy_fork(ip, XFS_DATA_FORK);
2669 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
2671 #ifdef XFS_INODE_TRACE
2672 ktrace_free(ip->i_trace);
2674 #ifdef XFS_BMAP_TRACE
2675 ktrace_free(ip->i_xtrace);
2677 #ifdef XFS_BTREE_TRACE
2678 ktrace_free(ip->i_btrace);
2681 ktrace_free(ip->i_rwtrace);
2683 #ifdef XFS_ILOCK_TRACE
2684 ktrace_free(ip->i_lock_trace);
2686 #ifdef XFS_DIR2_TRACE
2687 ktrace_free(ip->i_dir_trace);
2691 * Only if we are shutting down the fs will we see an
2692 * inode still in the AIL. If it is there, we should remove
2693 * it to prevent a use-after-free from occurring.
2695 xfs_mount_t *mp = ip->i_mount;
2696 xfs_log_item_t *lip = &ip->i_itemp->ili_item;
2698 ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) ||
2699 XFS_FORCED_SHUTDOWN(ip->i_mount));
2700 if (lip->li_flags & XFS_LI_IN_AIL) {
2701 spin_lock(&mp->m_ail_lock);
2702 if (lip->li_flags & XFS_LI_IN_AIL)
2703 xfs_trans_delete_ail(mp, lip);
2705 spin_unlock(&mp->m_ail_lock);
2707 xfs_inode_item_destroy(ip);
2710 /* asserts to verify all state is correct here */
2711 ASSERT(atomic_read(&ip->i_iocount) == 0);
2712 ASSERT(atomic_read(&ip->i_pincount) == 0);
2713 ASSERT(!spin_is_locked(&ip->i_flags_lock));
2714 ASSERT(list_empty(&ip->i_reclaim));
2715 kmem_zone_free(xfs_inode_zone, ip);
2720 * Increment the pin count of the given buffer.
2721 * This value is protected by ipinlock spinlock in the mount structure.
2727 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2729 atomic_inc(&ip->i_pincount);
2733 * Decrement the pin count of the given inode, and wake up
2734 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2735 * inode must have been previously pinned with a call to xfs_ipin().
2741 ASSERT(atomic_read(&ip->i_pincount) > 0);
2743 if (atomic_dec_and_test(&ip->i_pincount))
2744 wake_up(&ip->i_ipin_wait);
2748 * This is called to unpin an inode. It can be directed to wait or to return
2749 * immediately without waiting for the inode to be unpinned. The caller must
2750 * have the inode locked in at least shared mode so that the buffer cannot be
2751 * subsequently pinned once someone is waiting for it to be unpinned.
2758 xfs_inode_log_item_t *iip = ip->i_itemp;
2760 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2761 if (atomic_read(&ip->i_pincount) == 0)
2764 /* Give the log a push to start the unpinning I/O */
2765 xfs_log_force(ip->i_mount, (iip && iip->ili_last_lsn) ?
2766 iip->ili_last_lsn : 0, XFS_LOG_FORCE);
2768 wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
2775 __xfs_iunpin_wait(ip, 1);
2782 __xfs_iunpin_wait(ip, 0);
2787 * xfs_iextents_copy()
2789 * This is called to copy the REAL extents (as opposed to the delayed
2790 * allocation extents) from the inode into the given buffer. It
2791 * returns the number of bytes copied into the buffer.
2793 * If there are no delayed allocation extents, then we can just
2794 * memcpy() the extents into the buffer. Otherwise, we need to
2795 * examine each extent in turn and skip those which are delayed.
2807 xfs_fsblock_t start_block;
2809 ifp = XFS_IFORK_PTR(ip, whichfork);
2810 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2811 ASSERT(ifp->if_bytes > 0);
2813 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2814 XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork);
2818 * There are some delayed allocation extents in the
2819 * inode, so copy the extents one at a time and skip
2820 * the delayed ones. There must be at least one
2821 * non-delayed extent.
2824 for (i = 0; i < nrecs; i++) {
2825 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
2826 start_block = xfs_bmbt_get_startblock(ep);
2827 if (ISNULLSTARTBLOCK(start_block)) {
2829 * It's a delayed allocation extent, so skip it.
2834 /* Translate to on disk format */
2835 put_unaligned(cpu_to_be64(ep->l0), &dp->l0);
2836 put_unaligned(cpu_to_be64(ep->l1), &dp->l1);
2840 ASSERT(copied != 0);
2841 xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip));
2843 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2847 * Each of the following cases stores data into the same region
2848 * of the on-disk inode, so only one of them can be valid at
2849 * any given time. While it is possible to have conflicting formats
2850 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2851 * in EXTENTS format, this can only happen when the fork has
2852 * changed formats after being modified but before being flushed.
2853 * In these cases, the format always takes precedence, because the
2854 * format indicates the current state of the fork.
2861 xfs_inode_log_item_t *iip,
2868 #ifdef XFS_TRANS_DEBUG
2871 static const short brootflag[2] =
2872 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2873 static const short dataflag[2] =
2874 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2875 static const short extflag[2] =
2876 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2880 ifp = XFS_IFORK_PTR(ip, whichfork);
2882 * This can happen if we gave up in iformat in an error path,
2883 * for the attribute fork.
2886 ASSERT(whichfork == XFS_ATTR_FORK);
2889 cp = XFS_DFORK_PTR(dip, whichfork);
2891 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2892 case XFS_DINODE_FMT_LOCAL:
2893 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2894 (ifp->if_bytes > 0)) {
2895 ASSERT(ifp->if_u1.if_data != NULL);
2896 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2897 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2901 case XFS_DINODE_FMT_EXTENTS:
2902 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2903 !(iip->ili_format.ilf_fields & extflag[whichfork]));
2904 ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
2905 (ifp->if_bytes == 0));
2906 ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
2907 (ifp->if_bytes > 0));
2908 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
2909 (ifp->if_bytes > 0)) {
2910 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2911 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2916 case XFS_DINODE_FMT_BTREE:
2917 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
2918 (ifp->if_broot_bytes > 0)) {
2919 ASSERT(ifp->if_broot != NULL);
2920 ASSERT(ifp->if_broot_bytes <=
2921 (XFS_IFORK_SIZE(ip, whichfork) +
2922 XFS_BROOT_SIZE_ADJ));
2923 xfs_bmbt_to_bmdr(ifp->if_broot, ifp->if_broot_bytes,
2924 (xfs_bmdr_block_t *)cp,
2925 XFS_DFORK_SIZE(dip, mp, whichfork));
2929 case XFS_DINODE_FMT_DEV:
2930 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
2931 ASSERT(whichfork == XFS_DATA_FORK);
2932 dip->di_u.di_dev = cpu_to_be32(ip->i_df.if_u2.if_rdev);
2936 case XFS_DINODE_FMT_UUID:
2937 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
2938 ASSERT(whichfork == XFS_DATA_FORK);
2939 memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid,
2955 xfs_mount_t *mp = ip->i_mount;
2956 xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino);
2957 unsigned long first_index, mask;
2958 unsigned long inodes_per_cluster;
2960 xfs_inode_t **ilist;
2967 ASSERT(pag->pagi_inodeok);
2968 ASSERT(pag->pag_ici_init);
2970 inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog;
2971 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
2972 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
2976 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
2977 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
2978 read_lock(&pag->pag_ici_lock);
2979 /* really need a gang lookup range call here */
2980 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
2981 first_index, inodes_per_cluster);
2985 for (i = 0; i < nr_found; i++) {
2989 /* if the inode lies outside this cluster, we're done. */
2990 if ((XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index)
2993 * Do an un-protected check to see if the inode is dirty and
2994 * is a candidate for flushing. These checks will be repeated
2995 * later after the appropriate locks are acquired.
2997 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
3001 * Try to get locks. If any are unavailable or it is pinned,
3002 * then this inode cannot be flushed and is skipped.
3005 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
3007 if (!xfs_iflock_nowait(iq)) {
3008 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3011 if (xfs_ipincount(iq)) {
3013 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3018 * arriving here means that this inode can be flushed. First
3019 * re-check that it's dirty before flushing.
3021 if (!xfs_inode_clean(iq)) {
3023 error = xfs_iflush_int(iq, bp);
3025 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3026 goto cluster_corrupt_out;
3032 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3036 XFS_STATS_INC(xs_icluster_flushcnt);
3037 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
3041 read_unlock(&pag->pag_ici_lock);
3046 cluster_corrupt_out:
3048 * Corruption detected in the clustering loop. Invalidate the
3049 * inode buffer and shut down the filesystem.
3051 read_unlock(&pag->pag_ici_lock);
3053 * Clean up the buffer. If it was B_DELWRI, just release it --
3054 * brelse can handle it with no problems. If not, shut down the
3055 * filesystem before releasing the buffer.
3057 bufwasdelwri = XFS_BUF_ISDELAYWRITE(bp);
3061 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3063 if (!bufwasdelwri) {
3065 * Just like incore_relse: if we have b_iodone functions,
3066 * mark the buffer as an error and call them. Otherwise
3067 * mark it as stale and brelse.
3069 if (XFS_BUF_IODONE_FUNC(bp)) {
3070 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
3074 XFS_BUF_ERROR(bp,EIO);
3083 * Unlocks the flush lock
3085 xfs_iflush_abort(iq);
3087 return XFS_ERROR(EFSCORRUPTED);
3091 * xfs_iflush() will write a modified inode's changes out to the
3092 * inode's on disk home. The caller must have the inode lock held
3093 * in at least shared mode and the inode flush completion must be
3094 * active as well. The inode lock will still be held upon return from
3095 * the call and the caller is free to unlock it.
3096 * The inode flush will be completed when the inode reaches the disk.
3097 * The flags indicate how the inode's buffer should be written out.
3104 xfs_inode_log_item_t *iip;
3109 int noblock = (flags == XFS_IFLUSH_ASYNC_NOBLOCK);
3110 enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
3112 XFS_STATS_INC(xs_iflush_count);
3114 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3115 ASSERT(!completion_done(&ip->i_flush));
3116 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3117 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3123 * If the inode isn't dirty, then just release the inode
3124 * flush lock and do nothing.
3126 if (xfs_inode_clean(ip)) {
3132 * We can't flush the inode until it is unpinned, so wait for it if we
3133 * are allowed to block. We know noone new can pin it, because we are
3134 * holding the inode lock shared and you need to hold it exclusively to
3137 * If we are not allowed to block, force the log out asynchronously so
3138 * that when we come back the inode will be unpinned. If other inodes
3139 * in the same cluster are dirty, they will probably write the inode
3140 * out for us if they occur after the log force completes.
3142 if (noblock && xfs_ipincount(ip)) {
3143 xfs_iunpin_nowait(ip);
3147 xfs_iunpin_wait(ip);
3150 * This may have been unpinned because the filesystem is shutting
3151 * down forcibly. If that's the case we must not write this inode
3152 * to disk, because the log record didn't make it to disk!
3154 if (XFS_FORCED_SHUTDOWN(mp)) {
3155 ip->i_update_core = 0;
3157 iip->ili_format.ilf_fields = 0;
3159 return XFS_ERROR(EIO);
3163 * Decide how buffer will be flushed out. This is done before
3164 * the call to xfs_iflush_int because this field is zeroed by it.
3166 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3168 * Flush out the inode buffer according to the directions
3169 * of the caller. In the cases where the caller has given
3170 * us a choice choose the non-delwri case. This is because
3171 * the inode is in the AIL and we need to get it out soon.
3174 case XFS_IFLUSH_SYNC:
3175 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3178 case XFS_IFLUSH_ASYNC_NOBLOCK:
3179 case XFS_IFLUSH_ASYNC:
3180 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3183 case XFS_IFLUSH_DELWRI:
3193 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3194 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3195 case XFS_IFLUSH_DELWRI:
3198 case XFS_IFLUSH_ASYNC_NOBLOCK:
3199 case XFS_IFLUSH_ASYNC:
3202 case XFS_IFLUSH_SYNC:
3213 * Get the buffer containing the on-disk inode.
3215 error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0, 0,
3216 noblock ? XFS_BUF_TRYLOCK : XFS_BUF_LOCK);
3223 * First flush out the inode that xfs_iflush was called with.
3225 error = xfs_iflush_int(ip, bp);
3230 * If the buffer is pinned then push on the log now so we won't
3231 * get stuck waiting in the write for too long.
3233 if (XFS_BUF_ISPINNED(bp))
3234 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
3238 * see if other inodes can be gathered into this write
3240 error = xfs_iflush_cluster(ip, bp);
3242 goto cluster_corrupt_out;
3244 if (flags & INT_DELWRI) {
3245 xfs_bdwrite(mp, bp);
3246 } else if (flags & INT_ASYNC) {
3247 error = xfs_bawrite(mp, bp);
3249 error = xfs_bwrite(mp, bp);
3255 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3256 cluster_corrupt_out:
3258 * Unlocks the flush lock
3260 xfs_iflush_abort(ip);
3261 return XFS_ERROR(EFSCORRUPTED);
3270 xfs_inode_log_item_t *iip;
3273 #ifdef XFS_TRANS_DEBUG
3277 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3278 ASSERT(!completion_done(&ip->i_flush));
3279 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3280 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3287 * If the inode isn't dirty, then just release the inode
3288 * flush lock and do nothing.
3290 if (xfs_inode_clean(ip)) {
3295 /* set *dip = inode's place in the buffer */
3296 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset);
3299 * Clear i_update_core before copying out the data.
3300 * This is for coordination with our timestamp updates
3301 * that don't hold the inode lock. They will always
3302 * update the timestamps BEFORE setting i_update_core,
3303 * so if we clear i_update_core after they set it we
3304 * are guaranteed to see their updates to the timestamps.
3305 * I believe that this depends on strongly ordered memory
3306 * semantics, but we have that. We use the SYNCHRONIZE
3307 * macro to make sure that the compiler does not reorder
3308 * the i_update_core access below the data copy below.
3310 ip->i_update_core = 0;
3314 * Make sure to get the latest atime from the Linux inode.
3316 xfs_synchronize_atime(ip);
3318 if (XFS_TEST_ERROR(be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC,
3319 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3320 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3321 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3322 ip->i_ino, be16_to_cpu(dip->di_core.di_magic), dip);
3325 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3326 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3327 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3328 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3329 ip->i_ino, ip, ip->i_d.di_magic);
3332 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3334 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3335 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3336 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3337 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3338 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3342 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
3344 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3345 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3346 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3347 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3348 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3349 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3354 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3355 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3356 XFS_RANDOM_IFLUSH_5)) {
3357 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3358 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3360 ip->i_d.di_nextents + ip->i_d.di_anextents,
3365 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3366 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3367 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3368 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3369 ip->i_ino, ip->i_d.di_forkoff, ip);
3373 * bump the flush iteration count, used to detect flushes which
3374 * postdate a log record during recovery.
3377 ip->i_d.di_flushiter++;
3380 * Copy the dirty parts of the inode into the on-disk
3381 * inode. We always copy out the core of the inode,
3382 * because if the inode is dirty at all the core must
3385 xfs_dinode_to_disk(&dip->di_core, &ip->i_d);
3387 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3388 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3389 ip->i_d.di_flushiter = 0;
3392 * If this is really an old format inode and the superblock version
3393 * has not been updated to support only new format inodes, then
3394 * convert back to the old inode format. If the superblock version
3395 * has been updated, then make the conversion permanent.
3397 ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 ||
3398 xfs_sb_version_hasnlink(&mp->m_sb));
3399 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
3400 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
3404 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3405 dip->di_core.di_onlink = cpu_to_be16(ip->i_d.di_nlink);
3408 * The superblock version has already been bumped,
3409 * so just make the conversion to the new inode
3412 ip->i_d.di_version = XFS_DINODE_VERSION_2;
3413 dip->di_core.di_version = XFS_DINODE_VERSION_2;
3414 ip->i_d.di_onlink = 0;
3415 dip->di_core.di_onlink = 0;
3416 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3417 memset(&(dip->di_core.di_pad[0]), 0,
3418 sizeof(dip->di_core.di_pad));
3419 ASSERT(ip->i_d.di_projid == 0);
3423 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
3424 if (XFS_IFORK_Q(ip))
3425 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3426 xfs_inobp_check(mp, bp);
3429 * We've recorded everything logged in the inode, so we'd
3430 * like to clear the ilf_fields bits so we don't log and
3431 * flush things unnecessarily. However, we can't stop
3432 * logging all this information until the data we've copied
3433 * into the disk buffer is written to disk. If we did we might
3434 * overwrite the copy of the inode in the log with all the
3435 * data after re-logging only part of it, and in the face of
3436 * a crash we wouldn't have all the data we need to recover.
3438 * What we do is move the bits to the ili_last_fields field.
3439 * When logging the inode, these bits are moved back to the
3440 * ilf_fields field. In the xfs_iflush_done() routine we
3441 * clear ili_last_fields, since we know that the information
3442 * those bits represent is permanently on disk. As long as
3443 * the flush completes before the inode is logged again, then
3444 * both ilf_fields and ili_last_fields will be cleared.
3446 * We can play with the ilf_fields bits here, because the inode
3447 * lock must be held exclusively in order to set bits there
3448 * and the flush lock protects the ili_last_fields bits.
3449 * Set ili_logged so the flush done
3450 * routine can tell whether or not to look in the AIL.
3451 * Also, store the current LSN of the inode so that we can tell
3452 * whether the item has moved in the AIL from xfs_iflush_done().
3453 * In order to read the lsn we need the AIL lock, because
3454 * it is a 64 bit value that cannot be read atomically.
3456 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3457 iip->ili_last_fields = iip->ili_format.ilf_fields;
3458 iip->ili_format.ilf_fields = 0;
3459 iip->ili_logged = 1;
3461 ASSERT(sizeof(xfs_lsn_t) == 8); /* don't lock if it shrinks */
3462 spin_lock(&mp->m_ail_lock);
3463 iip->ili_flush_lsn = iip->ili_item.li_lsn;
3464 spin_unlock(&mp->m_ail_lock);
3467 * Attach the function xfs_iflush_done to the inode's
3468 * buffer. This will remove the inode from the AIL
3469 * and unlock the inode's flush lock when the inode is
3470 * completely written to disk.
3472 xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
3473 xfs_iflush_done, (xfs_log_item_t *)iip);
3475 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3476 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3479 * We're flushing an inode which is not in the AIL and has
3480 * not been logged but has i_update_core set. For this
3481 * case we can use a B_DELWRI flush and immediately drop
3482 * the inode flush lock because we can avoid the whole
3483 * AIL state thing. It's OK to drop the flush lock now,
3484 * because we've already locked the buffer and to do anything
3485 * you really need both.
3488 ASSERT(iip->ili_logged == 0);
3489 ASSERT(iip->ili_last_fields == 0);
3490 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3498 return XFS_ERROR(EFSCORRUPTED);
3503 #ifdef XFS_ILOCK_TRACE
3504 ktrace_t *xfs_ilock_trace_buf;
3507 xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
3509 ktrace_enter(ip->i_lock_trace,
3511 (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
3512 (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
3513 (void *)ra, /* caller of ilock */
3514 (void *)(unsigned long)current_cpu(),
3515 (void *)(unsigned long)current_pid(),
3516 NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL);
3521 * Return a pointer to the extent record at file index idx.
3523 xfs_bmbt_rec_host_t *
3525 xfs_ifork_t *ifp, /* inode fork pointer */
3526 xfs_extnum_t idx) /* index of target extent */
3529 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
3530 return ifp->if_u1.if_ext_irec->er_extbuf;
3531 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3532 xfs_ext_irec_t *erp; /* irec pointer */
3533 int erp_idx = 0; /* irec index */
3534 xfs_extnum_t page_idx = idx; /* ext index in target list */
3536 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3537 return &erp->er_extbuf[page_idx];
3538 } else if (ifp->if_bytes) {
3539 return &ifp->if_u1.if_extents[idx];
3546 * Insert new item(s) into the extent records for incore inode
3547 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3551 xfs_ifork_t *ifp, /* inode fork pointer */
3552 xfs_extnum_t idx, /* starting index of new items */
3553 xfs_extnum_t count, /* number of inserted items */
3554 xfs_bmbt_irec_t *new) /* items to insert */
3556 xfs_extnum_t i; /* extent record index */
3558 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3559 xfs_iext_add(ifp, idx, count);
3560 for (i = idx; i < idx + count; i++, new++)
3561 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new);
3565 * This is called when the amount of space required for incore file
3566 * extents needs to be increased. The ext_diff parameter stores the
3567 * number of new extents being added and the idx parameter contains
3568 * the extent index where the new extents will be added. If the new
3569 * extents are being appended, then we just need to (re)allocate and
3570 * initialize the space. Otherwise, if the new extents are being
3571 * inserted into the middle of the existing entries, a bit more work
3572 * is required to make room for the new extents to be inserted. The
3573 * caller is responsible for filling in the new extent entries upon
3578 xfs_ifork_t *ifp, /* inode fork pointer */
3579 xfs_extnum_t idx, /* index to begin adding exts */
3580 int ext_diff) /* number of extents to add */
3582 int byte_diff; /* new bytes being added */
3583 int new_size; /* size of extents after adding */
3584 xfs_extnum_t nextents; /* number of extents in file */
3586 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3587 ASSERT((idx >= 0) && (idx <= nextents));
3588 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
3589 new_size = ifp->if_bytes + byte_diff;
3591 * If the new number of extents (nextents + ext_diff)
3592 * fits inside the inode, then continue to use the inline
3595 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
3596 if (idx < nextents) {
3597 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
3598 &ifp->if_u2.if_inline_ext[idx],
3599 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3600 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
3602 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3603 ifp->if_real_bytes = 0;
3604 ifp->if_lastex = nextents + ext_diff;
3607 * Otherwise use a linear (direct) extent list.
3608 * If the extents are currently inside the inode,
3609 * xfs_iext_realloc_direct will switch us from
3610 * inline to direct extent allocation mode.
3612 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
3613 xfs_iext_realloc_direct(ifp, new_size);
3614 if (idx < nextents) {
3615 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
3616 &ifp->if_u1.if_extents[idx],
3617 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3618 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
3621 /* Indirection array */
3623 xfs_ext_irec_t *erp;
3627 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
3628 if (ifp->if_flags & XFS_IFEXTIREC) {
3629 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3631 xfs_iext_irec_init(ifp);
3632 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3633 erp = ifp->if_u1.if_ext_irec;
3635 /* Extents fit in target extent page */
3636 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3637 if (page_idx < erp->er_extcount) {
3638 memmove(&erp->er_extbuf[page_idx + ext_diff],
3639 &erp->er_extbuf[page_idx],
3640 (erp->er_extcount - page_idx) *
3641 sizeof(xfs_bmbt_rec_t));
3642 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3644 erp->er_extcount += ext_diff;
3645 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3647 /* Insert a new extent page */
3649 xfs_iext_add_indirect_multi(ifp,
3650 erp_idx, page_idx, ext_diff);
3653 * If extent(s) are being appended to the last page in
3654 * the indirection array and the new extent(s) don't fit
3655 * in the page, then erp is NULL and erp_idx is set to
3656 * the next index needed in the indirection array.
3659 int count = ext_diff;
3662 erp = xfs_iext_irec_new(ifp, erp_idx);
3663 erp->er_extcount = count;
3664 count -= MIN(count, (int)XFS_LINEAR_EXTS);
3671 ifp->if_bytes = new_size;
3675 * This is called when incore extents are being added to the indirection
3676 * array and the new extents do not fit in the target extent list. The
3677 * erp_idx parameter contains the irec index for the target extent list
3678 * in the indirection array, and the idx parameter contains the extent
3679 * index within the list. The number of extents being added is stored
3680 * in the count parameter.
3682 * |-------| |-------|
3683 * | | | | idx - number of extents before idx
3685 * | | | | count - number of extents being inserted at idx
3686 * |-------| |-------|
3687 * | count | | nex2 | nex2 - number of extents after idx + count
3688 * |-------| |-------|
3691 xfs_iext_add_indirect_multi(
3692 xfs_ifork_t *ifp, /* inode fork pointer */
3693 int erp_idx, /* target extent irec index */
3694 xfs_extnum_t idx, /* index within target list */
3695 int count) /* new extents being added */
3697 int byte_diff; /* new bytes being added */
3698 xfs_ext_irec_t *erp; /* pointer to irec entry */
3699 xfs_extnum_t ext_diff; /* number of extents to add */
3700 xfs_extnum_t ext_cnt; /* new extents still needed */
3701 xfs_extnum_t nex2; /* extents after idx + count */
3702 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3703 int nlists; /* number of irec's (lists) */
3705 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3706 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3707 nex2 = erp->er_extcount - idx;
3708 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3711 * Save second part of target extent list
3712 * (all extents past */
3714 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3715 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS);
3716 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3717 erp->er_extcount -= nex2;
3718 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3719 memset(&erp->er_extbuf[idx], 0, byte_diff);
3723 * Add the new extents to the end of the target
3724 * list, then allocate new irec record(s) and
3725 * extent buffer(s) as needed to store the rest
3726 * of the new extents.
3729 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3731 erp->er_extcount += ext_diff;
3732 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3733 ext_cnt -= ext_diff;
3737 erp = xfs_iext_irec_new(ifp, erp_idx);
3738 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3739 erp->er_extcount = ext_diff;
3740 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3741 ext_cnt -= ext_diff;
3744 /* Add nex2 extents back to indirection array */
3746 xfs_extnum_t ext_avail;
3749 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3750 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3753 * If nex2 extents fit in the current page, append
3754 * nex2_ep after the new extents.
3756 if (nex2 <= ext_avail) {
3757 i = erp->er_extcount;
3760 * Otherwise, check if space is available in the
3763 else if ((erp_idx < nlists - 1) &&
3764 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3765 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3768 /* Create a hole for nex2 extents */
3769 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3770 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3773 * Final choice, create a new extent page for
3778 erp = xfs_iext_irec_new(ifp, erp_idx);
3780 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
3782 erp->er_extcount += nex2;
3783 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3788 * This is called when the amount of space required for incore file
3789 * extents needs to be decreased. The ext_diff parameter stores the
3790 * number of extents to be removed and the idx parameter contains
3791 * the extent index where the extents will be removed from.
3793 * If the amount of space needed has decreased below the linear
3794 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3795 * extent array. Otherwise, use kmem_realloc() to adjust the
3796 * size to what is needed.
3800 xfs_ifork_t *ifp, /* inode fork pointer */
3801 xfs_extnum_t idx, /* index to begin removing exts */
3802 int ext_diff) /* number of extents to remove */
3804 xfs_extnum_t nextents; /* number of extents in file */
3805 int new_size; /* size of extents after removal */
3807 ASSERT(ext_diff > 0);
3808 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3809 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3811 if (new_size == 0) {
3812 xfs_iext_destroy(ifp);
3813 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3814 xfs_iext_remove_indirect(ifp, idx, ext_diff);
3815 } else if (ifp->if_real_bytes) {
3816 xfs_iext_remove_direct(ifp, idx, ext_diff);
3818 xfs_iext_remove_inline(ifp, idx, ext_diff);
3820 ifp->if_bytes = new_size;
3824 * This removes ext_diff extents from the inline buffer, beginning
3825 * at extent index idx.
3828 xfs_iext_remove_inline(
3829 xfs_ifork_t *ifp, /* inode fork pointer */
3830 xfs_extnum_t idx, /* index to begin removing exts */
3831 int ext_diff) /* number of extents to remove */
3833 int nextents; /* number of extents in file */
3835 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3836 ASSERT(idx < XFS_INLINE_EXTS);
3837 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3838 ASSERT(((nextents - ext_diff) > 0) &&
3839 (nextents - ext_diff) < XFS_INLINE_EXTS);
3841 if (idx + ext_diff < nextents) {
3842 memmove(&ifp->if_u2.if_inline_ext[idx],
3843 &ifp->if_u2.if_inline_ext[idx + ext_diff],
3844 (nextents - (idx + ext_diff)) *
3845 sizeof(xfs_bmbt_rec_t));
3846 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
3847 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3849 memset(&ifp->if_u2.if_inline_ext[idx], 0,
3850 ext_diff * sizeof(xfs_bmbt_rec_t));
3855 * This removes ext_diff extents from a linear (direct) extent list,
3856 * beginning at extent index idx. If the extents are being removed
3857 * from the end of the list (ie. truncate) then we just need to re-
3858 * allocate the list to remove the extra space. Otherwise, if the
3859 * extents are being removed from the middle of the existing extent
3860 * entries, then we first need to move the extent records beginning
3861 * at idx + ext_diff up in the list to overwrite the records being
3862 * removed, then remove the extra space via kmem_realloc.
3865 xfs_iext_remove_direct(
3866 xfs_ifork_t *ifp, /* inode fork pointer */
3867 xfs_extnum_t idx, /* index to begin removing exts */
3868 int ext_diff) /* number of extents to remove */
3870 xfs_extnum_t nextents; /* number of extents in file */
3871 int new_size; /* size of extents after removal */
3873 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3874 new_size = ifp->if_bytes -
3875 (ext_diff * sizeof(xfs_bmbt_rec_t));
3876 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3878 if (new_size == 0) {
3879 xfs_iext_destroy(ifp);
3882 /* Move extents up in the list (if needed) */
3883 if (idx + ext_diff < nextents) {
3884 memmove(&ifp->if_u1.if_extents[idx],
3885 &ifp->if_u1.if_extents[idx + ext_diff],
3886 (nextents - (idx + ext_diff)) *
3887 sizeof(xfs_bmbt_rec_t));
3889 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
3890 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3892 * Reallocate the direct extent list. If the extents
3893 * will fit inside the inode then xfs_iext_realloc_direct
3894 * will switch from direct to inline extent allocation
3897 xfs_iext_realloc_direct(ifp, new_size);
3898 ifp->if_bytes = new_size;
3902 * This is called when incore extents are being removed from the
3903 * indirection array and the extents being removed span multiple extent
3904 * buffers. The idx parameter contains the file extent index where we
3905 * want to begin removing extents, and the count parameter contains
3906 * how many extents need to be removed.
3908 * |-------| |-------|
3909 * | nex1 | | | nex1 - number of extents before idx
3910 * |-------| | count |
3911 * | | | | count - number of extents being removed at idx
3912 * | count | |-------|
3913 * | | | nex2 | nex2 - number of extents after idx + count
3914 * |-------| |-------|
3917 xfs_iext_remove_indirect(
3918 xfs_ifork_t *ifp, /* inode fork pointer */
3919 xfs_extnum_t idx, /* index to begin removing extents */
3920 int count) /* number of extents to remove */
3922 xfs_ext_irec_t *erp; /* indirection array pointer */
3923 int erp_idx = 0; /* indirection array index */
3924 xfs_extnum_t ext_cnt; /* extents left to remove */
3925 xfs_extnum_t ext_diff; /* extents to remove in current list */
3926 xfs_extnum_t nex1; /* number of extents before idx */
3927 xfs_extnum_t nex2; /* extents after idx + count */
3928 int nlists; /* entries in indirection array */
3929 int page_idx = idx; /* index in target extent list */
3931 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3932 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3933 ASSERT(erp != NULL);
3934 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3938 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
3939 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
3941 * Check for deletion of entire list;
3942 * xfs_iext_irec_remove() updates extent offsets.
3944 if (ext_diff == erp->er_extcount) {
3945 xfs_iext_irec_remove(ifp, erp_idx);
3946 ext_cnt -= ext_diff;
3949 ASSERT(erp_idx < ifp->if_real_bytes /
3951 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3958 /* Move extents up (if needed) */
3960 memmove(&erp->er_extbuf[nex1],
3961 &erp->er_extbuf[nex1 + ext_diff],
3962 nex2 * sizeof(xfs_bmbt_rec_t));
3964 /* Zero out rest of page */
3965 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
3966 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
3967 /* Update remaining counters */
3968 erp->er_extcount -= ext_diff;
3969 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
3970 ext_cnt -= ext_diff;
3975 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
3976 xfs_iext_irec_compact(ifp);
3980 * Create, destroy, or resize a linear (direct) block of extents.
3983 xfs_iext_realloc_direct(
3984 xfs_ifork_t *ifp, /* inode fork pointer */
3985 int new_size) /* new size of extents */
3987 int rnew_size; /* real new size of extents */
3989 rnew_size = new_size;
3991 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
3992 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
3993 (new_size != ifp->if_real_bytes)));
3995 /* Free extent records */
3996 if (new_size == 0) {
3997 xfs_iext_destroy(ifp);
3999 /* Resize direct extent list and zero any new bytes */
4000 else if (ifp->if_real_bytes) {
4001 /* Check if extents will fit inside the inode */
4002 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
4003 xfs_iext_direct_to_inline(ifp, new_size /
4004 (uint)sizeof(xfs_bmbt_rec_t));
4005 ifp->if_bytes = new_size;
4008 if (!is_power_of_2(new_size)){
4009 rnew_size = roundup_pow_of_two(new_size);
4011 if (rnew_size != ifp->if_real_bytes) {
4012 ifp->if_u1.if_extents =
4013 kmem_realloc(ifp->if_u1.if_extents,
4015 ifp->if_real_bytes, KM_NOFS);
4017 if (rnew_size > ifp->if_real_bytes) {
4018 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
4019 (uint)sizeof(xfs_bmbt_rec_t)], 0,
4020 rnew_size - ifp->if_real_bytes);
4024 * Switch from the inline extent buffer to a direct
4025 * extent list. Be sure to include the inline extent
4026 * bytes in new_size.
4029 new_size += ifp->if_bytes;
4030 if (!is_power_of_2(new_size)) {
4031 rnew_size = roundup_pow_of_two(new_size);
4033 xfs_iext_inline_to_direct(ifp, rnew_size);
4035 ifp->if_real_bytes = rnew_size;
4036 ifp->if_bytes = new_size;
4040 * Switch from linear (direct) extent records to inline buffer.
4043 xfs_iext_direct_to_inline(
4044 xfs_ifork_t *ifp, /* inode fork pointer */
4045 xfs_extnum_t nextents) /* number of extents in file */
4047 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4048 ASSERT(nextents <= XFS_INLINE_EXTS);
4050 * The inline buffer was zeroed when we switched
4051 * from inline to direct extent allocation mode,
4052 * so we don't need to clear it here.
4054 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
4055 nextents * sizeof(xfs_bmbt_rec_t));
4056 kmem_free(ifp->if_u1.if_extents);
4057 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
4058 ifp->if_real_bytes = 0;
4062 * Switch from inline buffer to linear (direct) extent records.
4063 * new_size should already be rounded up to the next power of 2
4064 * by the caller (when appropriate), so use new_size as it is.
4065 * However, since new_size may be rounded up, we can't update
4066 * if_bytes here. It is the caller's responsibility to update
4067 * if_bytes upon return.
4070 xfs_iext_inline_to_direct(
4071 xfs_ifork_t *ifp, /* inode fork pointer */
4072 int new_size) /* number of extents in file */
4074 ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS);
4075 memset(ifp->if_u1.if_extents, 0, new_size);
4076 if (ifp->if_bytes) {
4077 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
4079 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4080 sizeof(xfs_bmbt_rec_t));
4082 ifp->if_real_bytes = new_size;
4086 * Resize an extent indirection array to new_size bytes.
4089 xfs_iext_realloc_indirect(
4090 xfs_ifork_t *ifp, /* inode fork pointer */
4091 int new_size) /* new indirection array size */
4093 int nlists; /* number of irec's (ex lists) */
4094 int size; /* current indirection array size */
4096 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4097 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4098 size = nlists * sizeof(xfs_ext_irec_t);
4099 ASSERT(ifp->if_real_bytes);
4100 ASSERT((new_size >= 0) && (new_size != size));
4101 if (new_size == 0) {
4102 xfs_iext_destroy(ifp);
4104 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
4105 kmem_realloc(ifp->if_u1.if_ext_irec,
4106 new_size, size, KM_NOFS);
4111 * Switch from indirection array to linear (direct) extent allocations.
4114 xfs_iext_indirect_to_direct(
4115 xfs_ifork_t *ifp) /* inode fork pointer */
4117 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
4118 xfs_extnum_t nextents; /* number of extents in file */
4119 int size; /* size of file extents */
4121 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4122 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4123 ASSERT(nextents <= XFS_LINEAR_EXTS);
4124 size = nextents * sizeof(xfs_bmbt_rec_t);
4126 xfs_iext_irec_compact_pages(ifp);
4127 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
4129 ep = ifp->if_u1.if_ext_irec->er_extbuf;
4130 kmem_free(ifp->if_u1.if_ext_irec);
4131 ifp->if_flags &= ~XFS_IFEXTIREC;
4132 ifp->if_u1.if_extents = ep;
4133 ifp->if_bytes = size;
4134 if (nextents < XFS_LINEAR_EXTS) {
4135 xfs_iext_realloc_direct(ifp, size);
4140 * Free incore file extents.
4144 xfs_ifork_t *ifp) /* inode fork pointer */
4146 if (ifp->if_flags & XFS_IFEXTIREC) {
4150 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4151 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
4152 xfs_iext_irec_remove(ifp, erp_idx);
4154 ifp->if_flags &= ~XFS_IFEXTIREC;
4155 } else if (ifp->if_real_bytes) {
4156 kmem_free(ifp->if_u1.if_extents);
4157 } else if (ifp->if_bytes) {
4158 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4159 sizeof(xfs_bmbt_rec_t));
4161 ifp->if_u1.if_extents = NULL;
4162 ifp->if_real_bytes = 0;
4167 * Return a pointer to the extent record for file system block bno.
4169 xfs_bmbt_rec_host_t * /* pointer to found extent record */
4170 xfs_iext_bno_to_ext(
4171 xfs_ifork_t *ifp, /* inode fork pointer */
4172 xfs_fileoff_t bno, /* block number to search for */
4173 xfs_extnum_t *idxp) /* index of target extent */
4175 xfs_bmbt_rec_host_t *base; /* pointer to first extent */
4176 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
4177 xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */
4178 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4179 int high; /* upper boundary in search */
4180 xfs_extnum_t idx = 0; /* index of target extent */
4181 int low; /* lower boundary in search */
4182 xfs_extnum_t nextents; /* number of file extents */
4183 xfs_fileoff_t startoff = 0; /* start offset of extent */
4185 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4186 if (nextents == 0) {
4191 if (ifp->if_flags & XFS_IFEXTIREC) {
4192 /* Find target extent list */
4194 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
4195 base = erp->er_extbuf;
4196 high = erp->er_extcount - 1;
4198 base = ifp->if_u1.if_extents;
4199 high = nextents - 1;
4201 /* Binary search extent records */
4202 while (low <= high) {
4203 idx = (low + high) >> 1;
4205 startoff = xfs_bmbt_get_startoff(ep);
4206 blockcount = xfs_bmbt_get_blockcount(ep);
4207 if (bno < startoff) {
4209 } else if (bno >= startoff + blockcount) {
4212 /* Convert back to file-based extent index */
4213 if (ifp->if_flags & XFS_IFEXTIREC) {
4214 idx += erp->er_extoff;
4220 /* Convert back to file-based extent index */
4221 if (ifp->if_flags & XFS_IFEXTIREC) {
4222 idx += erp->er_extoff;
4224 if (bno >= startoff + blockcount) {
4225 if (++idx == nextents) {
4228 ep = xfs_iext_get_ext(ifp, idx);
4236 * Return a pointer to the indirection array entry containing the
4237 * extent record for filesystem block bno. Store the index of the
4238 * target irec in *erp_idxp.
4240 xfs_ext_irec_t * /* pointer to found extent record */
4241 xfs_iext_bno_to_irec(
4242 xfs_ifork_t *ifp, /* inode fork pointer */
4243 xfs_fileoff_t bno, /* block number to search for */
4244 int *erp_idxp) /* irec index of target ext list */
4246 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4247 xfs_ext_irec_t *erp_next; /* next indirection array entry */
4248 int erp_idx; /* indirection array index */
4249 int nlists; /* number of extent irec's (lists) */
4250 int high; /* binary search upper limit */
4251 int low; /* binary search lower limit */
4253 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4254 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4258 while (low <= high) {
4259 erp_idx = (low + high) >> 1;
4260 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4261 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
4262 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
4264 } else if (erp_next && bno >=
4265 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
4271 *erp_idxp = erp_idx;
4276 * Return a pointer to the indirection array entry containing the
4277 * extent record at file extent index *idxp. Store the index of the
4278 * target irec in *erp_idxp and store the page index of the target
4279 * extent record in *idxp.
4282 xfs_iext_idx_to_irec(
4283 xfs_ifork_t *ifp, /* inode fork pointer */
4284 xfs_extnum_t *idxp, /* extent index (file -> page) */
4285 int *erp_idxp, /* pointer to target irec */
4286 int realloc) /* new bytes were just added */
4288 xfs_ext_irec_t *prev; /* pointer to previous irec */
4289 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
4290 int erp_idx; /* indirection array index */
4291 int nlists; /* number of irec's (ex lists) */
4292 int high; /* binary search upper limit */
4293 int low; /* binary search lower limit */
4294 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
4296 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4297 ASSERT(page_idx >= 0 && page_idx <=
4298 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
4299 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4304 /* Binary search extent irec's */
4305 while (low <= high) {
4306 erp_idx = (low + high) >> 1;
4307 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4308 prev = erp_idx > 0 ? erp - 1 : NULL;
4309 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
4310 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
4312 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
4313 (page_idx == erp->er_extoff + erp->er_extcount &&
4316 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
4317 erp->er_extcount == XFS_LINEAR_EXTS) {
4321 erp = erp_idx < nlists ? erp + 1 : NULL;
4324 page_idx -= erp->er_extoff;
4329 *erp_idxp = erp_idx;
4334 * Allocate and initialize an indirection array once the space needed
4335 * for incore extents increases above XFS_IEXT_BUFSZ.
4339 xfs_ifork_t *ifp) /* inode fork pointer */
4341 xfs_ext_irec_t *erp; /* indirection array pointer */
4342 xfs_extnum_t nextents; /* number of extents in file */
4344 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4345 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4346 ASSERT(nextents <= XFS_LINEAR_EXTS);
4348 erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS);
4350 if (nextents == 0) {
4351 ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
4352 } else if (!ifp->if_real_bytes) {
4353 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
4354 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
4355 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
4357 erp->er_extbuf = ifp->if_u1.if_extents;
4358 erp->er_extcount = nextents;
4361 ifp->if_flags |= XFS_IFEXTIREC;
4362 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
4363 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
4364 ifp->if_u1.if_ext_irec = erp;
4370 * Allocate and initialize a new entry in the indirection array.
4374 xfs_ifork_t *ifp, /* inode fork pointer */
4375 int erp_idx) /* index for new irec */
4377 xfs_ext_irec_t *erp; /* indirection array pointer */
4378 int i; /* loop counter */
4379 int nlists; /* number of irec's (ex lists) */
4381 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4382 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4384 /* Resize indirection array */
4385 xfs_iext_realloc_indirect(ifp, ++nlists *
4386 sizeof(xfs_ext_irec_t));
4388 * Move records down in the array so the
4389 * new page can use erp_idx.
4391 erp = ifp->if_u1.if_ext_irec;
4392 for (i = nlists - 1; i > erp_idx; i--) {
4393 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
4395 ASSERT(i == erp_idx);
4397 /* Initialize new extent record */
4398 erp = ifp->if_u1.if_ext_irec;
4399 erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
4400 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4401 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
4402 erp[erp_idx].er_extcount = 0;
4403 erp[erp_idx].er_extoff = erp_idx > 0 ?
4404 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
4405 return (&erp[erp_idx]);
4409 * Remove a record from the indirection array.
4412 xfs_iext_irec_remove(
4413 xfs_ifork_t *ifp, /* inode fork pointer */
4414 int erp_idx) /* irec index to remove */
4416 xfs_ext_irec_t *erp; /* indirection array pointer */
4417 int i; /* loop counter */
4418 int nlists; /* number of irec's (ex lists) */
4420 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4421 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4422 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4423 if (erp->er_extbuf) {
4424 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
4426 kmem_free(erp->er_extbuf);
4428 /* Compact extent records */
4429 erp = ifp->if_u1.if_ext_irec;
4430 for (i = erp_idx; i < nlists - 1; i++) {
4431 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
4434 * Manually free the last extent record from the indirection
4435 * array. A call to xfs_iext_realloc_indirect() with a size
4436 * of zero would result in a call to xfs_iext_destroy() which
4437 * would in turn call this function again, creating a nasty
4441 xfs_iext_realloc_indirect(ifp,
4442 nlists * sizeof(xfs_ext_irec_t));
4444 kmem_free(ifp->if_u1.if_ext_irec);
4446 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4450 * This is called to clean up large amounts of unused memory allocated
4451 * by the indirection array. Before compacting anything though, verify
4452 * that the indirection array is still needed and switch back to the
4453 * linear extent list (or even the inline buffer) if possible. The
4454 * compaction policy is as follows:
4456 * Full Compaction: Extents fit into a single page (or inline buffer)
4457 * Partial Compaction: Extents occupy less than 50% of allocated space
4458 * No Compaction: Extents occupy at least 50% of allocated space
4461 xfs_iext_irec_compact(
4462 xfs_ifork_t *ifp) /* inode fork pointer */
4464 xfs_extnum_t nextents; /* number of extents in file */
4465 int nlists; /* number of irec's (ex lists) */
4467 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4468 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4469 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4471 if (nextents == 0) {
4472 xfs_iext_destroy(ifp);
4473 } else if (nextents <= XFS_INLINE_EXTS) {
4474 xfs_iext_indirect_to_direct(ifp);
4475 xfs_iext_direct_to_inline(ifp, nextents);
4476 } else if (nextents <= XFS_LINEAR_EXTS) {
4477 xfs_iext_indirect_to_direct(ifp);
4478 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
4479 xfs_iext_irec_compact_pages(ifp);
4484 * Combine extents from neighboring extent pages.
4487 xfs_iext_irec_compact_pages(
4488 xfs_ifork_t *ifp) /* inode fork pointer */
4490 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
4491 int erp_idx = 0; /* indirection array index */
4492 int nlists; /* number of irec's (ex lists) */
4494 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4495 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4496 while (erp_idx < nlists - 1) {
4497 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4499 if (erp_next->er_extcount <=
4500 (XFS_LINEAR_EXTS - erp->er_extcount)) {
4501 memcpy(&erp->er_extbuf[erp->er_extcount],
4502 erp_next->er_extbuf, erp_next->er_extcount *
4503 sizeof(xfs_bmbt_rec_t));
4504 erp->er_extcount += erp_next->er_extcount;
4506 * Free page before removing extent record
4507 * so er_extoffs don't get modified in
4508 * xfs_iext_irec_remove.
4510 kmem_free(erp_next->er_extbuf);
4511 erp_next->er_extbuf = NULL;
4512 xfs_iext_irec_remove(ifp, erp_idx + 1);
4513 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4521 * This is called to update the er_extoff field in the indirection
4522 * array when extents have been added or removed from one of the
4523 * extent lists. erp_idx contains the irec index to begin updating
4524 * at and ext_diff contains the number of extents that were added
4528 xfs_iext_irec_update_extoffs(
4529 xfs_ifork_t *ifp, /* inode fork pointer */
4530 int erp_idx, /* irec index to update */
4531 int ext_diff) /* number of new extents */
4533 int i; /* loop counter */
4534 int nlists; /* number of irec's (ex lists */
4536 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4537 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4538 for (i = erp_idx; i < nlists; i++) {
4539 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;