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_magic) == XFS_DINODE_MAGIC &&
178 XFS_DINODE_GOOD_VERSION(dip->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_magic));
196 xfs_trans_brelse(tp, bp);
197 return XFS_ERROR(EFSCORRUPTED);
201 xfs_inobp_check(mp, bp);
204 * Mark the buffer as an inode buffer now that it looks good
206 XFS_BUF_SET_VTYPE(bp, B_FS_INO);
213 * This routine is called to map an inode number within a file
214 * system to the buffer containing the on-disk version of the
215 * inode. It returns a pointer to the buffer containing the
216 * on-disk inode in the bpp parameter, and in the dip parameter
217 * it returns a pointer to the on-disk inode within that buffer.
219 * If a non-zero error is returned, then the contents of bpp and
220 * dipp are undefined.
222 * Use xfs_imap() to determine the size and location of the
223 * buffer to read from disk.
240 error = xfs_imap(mp, tp, ino, &imap, imap_flags | XFS_IMAP_LOOKUP);
244 error = xfs_imap_to_bp(mp, tp, &imap, &bp, XFS_BUF_LOCK, imap_flags);
248 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
250 *offset = imap.im_boffset;
256 * This routine is called to map an inode to the buffer containing
257 * the on-disk version of the inode. It returns a pointer to the
258 * buffer containing the on-disk inode in the bpp parameter, and in
259 * the dip parameter it returns a pointer to the on-disk inode within
262 * If a non-zero error is returned, then the contents of bpp and
263 * dipp are undefined.
265 * If the inode is new and has not yet been initialized, use xfs_imap()
266 * to determine the size and location of the buffer to read from disk.
267 * If the inode has already been mapped to its buffer and read in once,
268 * then use the mapping information stored in the inode rather than
269 * calling xfs_imap(). This allows us to avoid the overhead of looking
270 * at the inode btree for small block file systems (see xfs_dilocate()).
271 * We can tell whether the inode has been mapped in before by comparing
272 * its disk block address to 0. Only uninitialized inodes will have
273 * 0 for the disk block address.
290 if (ip->i_blkno == (xfs_daddr_t)0) {
292 error = xfs_imap(mp, tp, ip->i_ino, &imap,
293 XFS_IMAP_LOOKUP | imap_flags);
298 * Fill in the fields in the inode that will be used to
299 * map the inode to its buffer from now on.
301 ip->i_blkno = imap.im_blkno;
302 ip->i_len = imap.im_len;
303 ip->i_boffset = imap.im_boffset;
306 * We've already mapped the inode once, so just use the
307 * mapping that we saved the first time.
309 imap.im_blkno = ip->i_blkno;
310 imap.im_len = ip->i_len;
311 imap.im_boffset = ip->i_boffset;
313 ASSERT(bno == 0 || bno == imap.im_blkno);
315 error = xfs_imap_to_bp(mp, tp, &imap, &bp, buf_flags, imap_flags);
320 ASSERT(buf_flags & XFS_BUF_TRYLOCK);
326 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
332 * Move inode type and inode format specific information from the
333 * on-disk inode to the in-core inode. For fifos, devs, and sockets
334 * this means set if_rdev to the proper value. For files, directories,
335 * and symlinks this means to bring in the in-line data or extent
336 * pointers. For a file in B-tree format, only the root is immediately
337 * brought in-core. The rest will be in-lined in if_extents when it
338 * is first referenced (see xfs_iread_extents()).
345 xfs_attr_shortform_t *atp;
349 ip->i_df.if_ext_max =
350 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
353 if (unlikely(be32_to_cpu(dip->di_nextents) +
354 be16_to_cpu(dip->di_anextents) >
355 be64_to_cpu(dip->di_nblocks))) {
356 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
357 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
358 (unsigned long long)ip->i_ino,
359 (int)(be32_to_cpu(dip->di_nextents) +
360 be16_to_cpu(dip->di_anextents)),
362 be64_to_cpu(dip->di_nblocks));
363 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
365 return XFS_ERROR(EFSCORRUPTED);
368 if (unlikely(dip->di_forkoff > ip->i_mount->m_sb.sb_inodesize)) {
369 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
370 "corrupt dinode %Lu, forkoff = 0x%x.",
371 (unsigned long long)ip->i_ino,
373 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
375 return XFS_ERROR(EFSCORRUPTED);
378 switch (ip->i_d.di_mode & S_IFMT) {
383 if (unlikely(dip->di_format != XFS_DINODE_FMT_DEV)) {
384 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
386 return XFS_ERROR(EFSCORRUPTED);
390 ip->i_df.if_u2.if_rdev = xfs_dinode_get_rdev(dip);
396 switch (dip->di_format) {
397 case XFS_DINODE_FMT_LOCAL:
399 * no local regular files yet
401 if (unlikely((be16_to_cpu(dip->di_mode) & S_IFMT) == S_IFREG)) {
402 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
404 "(local format for regular file).",
405 (unsigned long long) ip->i_ino);
406 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
409 return XFS_ERROR(EFSCORRUPTED);
412 di_size = be64_to_cpu(dip->di_size);
413 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
414 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
416 "(bad size %Ld for local inode).",
417 (unsigned long long) ip->i_ino,
418 (long long) di_size);
419 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
422 return XFS_ERROR(EFSCORRUPTED);
426 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
428 case XFS_DINODE_FMT_EXTENTS:
429 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
431 case XFS_DINODE_FMT_BTREE:
432 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
435 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
437 return XFS_ERROR(EFSCORRUPTED);
442 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
443 return XFS_ERROR(EFSCORRUPTED);
448 if (!XFS_DFORK_Q(dip))
450 ASSERT(ip->i_afp == NULL);
451 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
452 ip->i_afp->if_ext_max =
453 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
454 switch (dip->di_aformat) {
455 case XFS_DINODE_FMT_LOCAL:
456 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
457 size = be16_to_cpu(atp->hdr.totsize);
458 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
460 case XFS_DINODE_FMT_EXTENTS:
461 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
463 case XFS_DINODE_FMT_BTREE:
464 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
467 error = XFS_ERROR(EFSCORRUPTED);
471 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
473 xfs_idestroy_fork(ip, XFS_DATA_FORK);
479 * The file is in-lined in the on-disk inode.
480 * If it fits into if_inline_data, then copy
481 * it there, otherwise allocate a buffer for it
482 * and copy the data there. Either way, set
483 * if_data to point at the data.
484 * If we allocate a buffer for the data, make
485 * sure that its size is a multiple of 4 and
486 * record the real size in i_real_bytes.
499 * If the size is unreasonable, then something
500 * is wrong and we just bail out rather than crash in
501 * kmem_alloc() or memcpy() below.
503 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
504 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
506 "(bad size %d for local fork, size = %d).",
507 (unsigned long long) ip->i_ino, size,
508 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
509 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
511 return XFS_ERROR(EFSCORRUPTED);
513 ifp = XFS_IFORK_PTR(ip, whichfork);
516 ifp->if_u1.if_data = NULL;
517 else if (size <= sizeof(ifp->if_u2.if_inline_data))
518 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
520 real_size = roundup(size, 4);
521 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
523 ifp->if_bytes = size;
524 ifp->if_real_bytes = real_size;
526 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
527 ifp->if_flags &= ~XFS_IFEXTENTS;
528 ifp->if_flags |= XFS_IFINLINE;
533 * The file consists of a set of extents all
534 * of which fit into the on-disk inode.
535 * If there are few enough extents to fit into
536 * the if_inline_ext, then copy them there.
537 * Otherwise allocate a buffer for them and copy
538 * them into it. Either way, set if_extents
539 * to point at the extents.
553 ifp = XFS_IFORK_PTR(ip, whichfork);
554 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
555 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
558 * If the number of extents is unreasonable, then something
559 * is wrong and we just bail out rather than crash in
560 * kmem_alloc() or memcpy() below.
562 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
563 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
564 "corrupt inode %Lu ((a)extents = %d).",
565 (unsigned long long) ip->i_ino, nex);
566 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
568 return XFS_ERROR(EFSCORRUPTED);
571 ifp->if_real_bytes = 0;
573 ifp->if_u1.if_extents = NULL;
574 else if (nex <= XFS_INLINE_EXTS)
575 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
577 xfs_iext_add(ifp, 0, nex);
579 ifp->if_bytes = size;
581 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
582 xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip));
583 for (i = 0; i < nex; i++, dp++) {
584 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
585 ep->l0 = get_unaligned_be64(&dp->l0);
586 ep->l1 = get_unaligned_be64(&dp->l1);
588 XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork);
589 if (whichfork != XFS_DATA_FORK ||
590 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
591 if (unlikely(xfs_check_nostate_extents(
593 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
596 return XFS_ERROR(EFSCORRUPTED);
599 ifp->if_flags |= XFS_IFEXTENTS;
604 * The file has too many extents to fit into
605 * the inode, so they are in B-tree format.
606 * Allocate a buffer for the root of the B-tree
607 * and copy the root into it. The i_extents
608 * field will remain NULL until all of the
609 * extents are read in (when they are needed).
617 xfs_bmdr_block_t *dfp;
623 ifp = XFS_IFORK_PTR(ip, whichfork);
624 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
625 size = XFS_BMAP_BROOT_SPACE(dfp);
626 nrecs = be16_to_cpu(dfp->bb_numrecs);
629 * blow out if -- fork has less extents than can fit in
630 * fork (fork shouldn't be a btree format), root btree
631 * block has more records than can fit into the fork,
632 * or the number of extents is greater than the number of
635 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
636 || XFS_BMDR_SPACE_CALC(nrecs) >
637 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
638 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
639 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
640 "corrupt inode %Lu (btree).",
641 (unsigned long long) ip->i_ino);
642 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
644 return XFS_ERROR(EFSCORRUPTED);
647 ifp->if_broot_bytes = size;
648 ifp->if_broot = kmem_alloc(size, KM_SLEEP);
649 ASSERT(ifp->if_broot != NULL);
651 * Copy and convert from the on-disk structure
652 * to the in-memory structure.
654 xfs_bmdr_to_bmbt(ip->i_mount, dfp,
655 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
656 ifp->if_broot, size);
657 ifp->if_flags &= ~XFS_IFEXTENTS;
658 ifp->if_flags |= XFS_IFBROOT;
664 xfs_dinode_from_disk(
668 to->di_magic = be16_to_cpu(from->di_magic);
669 to->di_mode = be16_to_cpu(from->di_mode);
670 to->di_version = from ->di_version;
671 to->di_format = from->di_format;
672 to->di_onlink = be16_to_cpu(from->di_onlink);
673 to->di_uid = be32_to_cpu(from->di_uid);
674 to->di_gid = be32_to_cpu(from->di_gid);
675 to->di_nlink = be32_to_cpu(from->di_nlink);
676 to->di_projid = be16_to_cpu(from->di_projid);
677 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
678 to->di_flushiter = be16_to_cpu(from->di_flushiter);
679 to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec);
680 to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec);
681 to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec);
682 to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec);
683 to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec);
684 to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec);
685 to->di_size = be64_to_cpu(from->di_size);
686 to->di_nblocks = be64_to_cpu(from->di_nblocks);
687 to->di_extsize = be32_to_cpu(from->di_extsize);
688 to->di_nextents = be32_to_cpu(from->di_nextents);
689 to->di_anextents = be16_to_cpu(from->di_anextents);
690 to->di_forkoff = from->di_forkoff;
691 to->di_aformat = from->di_aformat;
692 to->di_dmevmask = be32_to_cpu(from->di_dmevmask);
693 to->di_dmstate = be16_to_cpu(from->di_dmstate);
694 to->di_flags = be16_to_cpu(from->di_flags);
695 to->di_gen = be32_to_cpu(from->di_gen);
701 xfs_icdinode_t *from)
703 to->di_magic = cpu_to_be16(from->di_magic);
704 to->di_mode = cpu_to_be16(from->di_mode);
705 to->di_version = from ->di_version;
706 to->di_format = from->di_format;
707 to->di_onlink = cpu_to_be16(from->di_onlink);
708 to->di_uid = cpu_to_be32(from->di_uid);
709 to->di_gid = cpu_to_be32(from->di_gid);
710 to->di_nlink = cpu_to_be32(from->di_nlink);
711 to->di_projid = cpu_to_be16(from->di_projid);
712 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
713 to->di_flushiter = cpu_to_be16(from->di_flushiter);
714 to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec);
715 to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec);
716 to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec);
717 to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec);
718 to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec);
719 to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec);
720 to->di_size = cpu_to_be64(from->di_size);
721 to->di_nblocks = cpu_to_be64(from->di_nblocks);
722 to->di_extsize = cpu_to_be32(from->di_extsize);
723 to->di_nextents = cpu_to_be32(from->di_nextents);
724 to->di_anextents = cpu_to_be16(from->di_anextents);
725 to->di_forkoff = from->di_forkoff;
726 to->di_aformat = from->di_aformat;
727 to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
728 to->di_dmstate = cpu_to_be16(from->di_dmstate);
729 to->di_flags = cpu_to_be16(from->di_flags);
730 to->di_gen = cpu_to_be32(from->di_gen);
739 if (di_flags & XFS_DIFLAG_ANY) {
740 if (di_flags & XFS_DIFLAG_REALTIME)
741 flags |= XFS_XFLAG_REALTIME;
742 if (di_flags & XFS_DIFLAG_PREALLOC)
743 flags |= XFS_XFLAG_PREALLOC;
744 if (di_flags & XFS_DIFLAG_IMMUTABLE)
745 flags |= XFS_XFLAG_IMMUTABLE;
746 if (di_flags & XFS_DIFLAG_APPEND)
747 flags |= XFS_XFLAG_APPEND;
748 if (di_flags & XFS_DIFLAG_SYNC)
749 flags |= XFS_XFLAG_SYNC;
750 if (di_flags & XFS_DIFLAG_NOATIME)
751 flags |= XFS_XFLAG_NOATIME;
752 if (di_flags & XFS_DIFLAG_NODUMP)
753 flags |= XFS_XFLAG_NODUMP;
754 if (di_flags & XFS_DIFLAG_RTINHERIT)
755 flags |= XFS_XFLAG_RTINHERIT;
756 if (di_flags & XFS_DIFLAG_PROJINHERIT)
757 flags |= XFS_XFLAG_PROJINHERIT;
758 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
759 flags |= XFS_XFLAG_NOSYMLINKS;
760 if (di_flags & XFS_DIFLAG_EXTSIZE)
761 flags |= XFS_XFLAG_EXTSIZE;
762 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
763 flags |= XFS_XFLAG_EXTSZINHERIT;
764 if (di_flags & XFS_DIFLAG_NODEFRAG)
765 flags |= XFS_XFLAG_NODEFRAG;
766 if (di_flags & XFS_DIFLAG_FILESTREAM)
767 flags |= XFS_XFLAG_FILESTREAM;
777 xfs_icdinode_t *dic = &ip->i_d;
779 return _xfs_dic2xflags(dic->di_flags) |
780 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
787 return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) |
788 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
792 * Allocate and initialise an xfs_inode.
794 STATIC struct 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(completion_done(&ip->i_flush));
816 * initialise the VFS inode here to get failures
817 * out of the way early.
819 if (!inode_init_always(mp->m_super, VFS_I(ip))) {
820 kmem_zone_free(xfs_inode_zone, ip);
824 /* initialise the xfs inode */
831 memset(&ip->i_df, 0, sizeof(xfs_ifork_t));
833 ip->i_update_core = 0;
834 ip->i_update_size = 0;
835 ip->i_delayed_blks = 0;
836 memset(&ip->i_d, 0, sizeof(xfs_icdinode_t));
841 * Initialize inode's trace buffers.
843 #ifdef XFS_INODE_TRACE
844 ip->i_trace = ktrace_alloc(INODE_TRACE_SIZE, KM_NOFS);
846 #ifdef XFS_BMAP_TRACE
847 ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_NOFS);
849 #ifdef XFS_BTREE_TRACE
850 ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_NOFS);
853 ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_NOFS);
855 #ifdef XFS_ILOCK_TRACE
856 ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_NOFS);
858 #ifdef XFS_DIR2_TRACE
859 ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_NOFS);
866 * Given a mount structure and an inode number, return a pointer
867 * to a newly allocated in-core inode corresponding to the given
870 * Initialize the inode's attributes and extent pointers if it
871 * already has them (it will not if the inode has no links).
887 ip = xfs_inode_alloc(mp, ino);
892 * Get pointer's to the on-disk inode and the buffer containing it.
893 * If the inode number refers to a block outside the file system
894 * then xfs_itobp() will return NULL. In this case we should
895 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
896 * know that this is a new incore inode.
898 error = xfs_itobp(mp, tp, ip, &dip, &bp, bno, imap_flags, XFS_BUF_LOCK);
900 goto out_destroy_inode;
903 * If we got something that isn't an inode it means someone
904 * (nfs or dmi) has a stale handle.
906 if (be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC) {
908 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
909 "dip->di_magic (0x%x) != "
910 "XFS_DINODE_MAGIC (0x%x)",
911 be16_to_cpu(dip->di_magic),
914 error = XFS_ERROR(EINVAL);
919 * If the on-disk inode is already linked to a directory
920 * entry, copy all of the inode into the in-core inode.
921 * xfs_iformat() handles copying in the inode format
922 * specific information.
923 * Otherwise, just get the truly permanent information.
926 xfs_dinode_from_disk(&ip->i_d, dip);
927 error = xfs_iformat(ip, dip);
930 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
931 "xfs_iformat() returned error %d",
937 ip->i_d.di_magic = be16_to_cpu(dip->di_magic);
938 ip->i_d.di_version = dip->di_version;
939 ip->i_d.di_gen = be32_to_cpu(dip->di_gen);
940 ip->i_d.di_flushiter = be16_to_cpu(dip->di_flushiter);
942 * Make sure to pull in the mode here as well in
943 * case the inode is released without being used.
944 * This ensures that xfs_inactive() will see that
945 * the inode is already free and not try to mess
946 * with the uninitialized part of it.
950 * Initialize the per-fork minima and maxima for a new
951 * inode here. xfs_iformat will do it for old inodes.
953 ip->i_df.if_ext_max =
954 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
958 * The inode format changed when we moved the link count and
959 * made it 32 bits long. If this is an old format inode,
960 * convert it in memory to look like a new one. If it gets
961 * flushed to disk we will convert back before flushing or
962 * logging it. We zero out the new projid field and the old link
963 * count field. We'll handle clearing the pad field (the remains
964 * of the old uuid field) when we actually convert the inode to
965 * the new format. We don't change the version number so that we
966 * can distinguish this from a real new format inode.
968 if (ip->i_d.di_version == 1) {
969 ip->i_d.di_nlink = ip->i_d.di_onlink;
970 ip->i_d.di_onlink = 0;
971 ip->i_d.di_projid = 0;
974 ip->i_delayed_blks = 0;
975 ip->i_size = ip->i_d.di_size;
978 * Mark the buffer containing the inode as something to keep
979 * around for a while. This helps to keep recently accessed
980 * meta-data in-core longer.
982 XFS_BUF_SET_REF(bp, XFS_INO_REF);
985 * Use xfs_trans_brelse() to release the buffer containing the
986 * on-disk inode, because it was acquired with xfs_trans_read_buf()
987 * in xfs_itobp() above. If tp is NULL, this is just a normal
988 * brelse(). If we're within a transaction, then xfs_trans_brelse()
989 * will only release the buffer if it is not dirty within the
990 * transaction. It will be OK to release the buffer in this case,
991 * because inodes on disk are never destroyed and we will be
992 * locking the new in-core inode before putting it in the hash
993 * table where other processes can find it. Thus we don't have
994 * to worry about the inode being changed just because we released
997 xfs_trans_brelse(tp, bp);
1002 xfs_trans_brelse(tp, bp);
1004 xfs_destroy_inode(ip);
1009 * Read in extents from a btree-format inode.
1010 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1020 xfs_extnum_t nextents;
1023 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
1024 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
1026 return XFS_ERROR(EFSCORRUPTED);
1028 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
1029 size = nextents * sizeof(xfs_bmbt_rec_t);
1030 ifp = XFS_IFORK_PTR(ip, whichfork);
1033 * We know that the size is valid (it's checked in iformat_btree)
1035 ifp->if_lastex = NULLEXTNUM;
1036 ifp->if_bytes = ifp->if_real_bytes = 0;
1037 ifp->if_flags |= XFS_IFEXTENTS;
1038 xfs_iext_add(ifp, 0, nextents);
1039 error = xfs_bmap_read_extents(tp, ip, whichfork);
1041 xfs_iext_destroy(ifp);
1042 ifp->if_flags &= ~XFS_IFEXTENTS;
1045 xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip));
1050 * Allocate an inode on disk and return a copy of its in-core version.
1051 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1052 * appropriately within the inode. The uid and gid for the inode are
1053 * set according to the contents of the given cred structure.
1055 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1056 * has a free inode available, call xfs_iget()
1057 * to obtain the in-core version of the allocated inode. Finally,
1058 * fill in the inode and log its initial contents. In this case,
1059 * ialloc_context would be set to NULL and call_again set to false.
1061 * If xfs_dialloc() does not have an available inode,
1062 * it will replenish its supply by doing an allocation. Since we can
1063 * only do one allocation within a transaction without deadlocks, we
1064 * must commit the current transaction before returning the inode itself.
1065 * In this case, therefore, we will set call_again to true and return.
1066 * The caller should then commit the current transaction, start a new
1067 * transaction, and call xfs_ialloc() again to actually get the inode.
1069 * To ensure that some other process does not grab the inode that
1070 * was allocated during the first call to xfs_ialloc(), this routine
1071 * also returns the [locked] bp pointing to the head of the freelist
1072 * as ialloc_context. The caller should hold this buffer across
1073 * the commit and pass it back into this routine on the second call.
1075 * If we are allocating quota inodes, we do not have a parent inode
1076 * to attach to or associate with (i.e. pip == NULL) because they
1077 * are not linked into the directory structure - they are attached
1078 * directly to the superblock - and so have no parent.
1090 xfs_buf_t **ialloc_context,
1091 boolean_t *call_again,
1099 int filestreams = 0;
1102 * Call the space management code to pick
1103 * the on-disk inode to be allocated.
1105 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
1106 ialloc_context, call_again, &ino);
1109 if (*call_again || ino == NULLFSINO) {
1113 ASSERT(*ialloc_context == NULL);
1116 * Get the in-core inode with the lock held exclusively.
1117 * This is because we're setting fields here we need
1118 * to prevent others from looking at until we're done.
1120 error = xfs_trans_iget(tp->t_mountp, tp, ino,
1121 XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
1126 ip->i_d.di_mode = (__uint16_t)mode;
1127 ip->i_d.di_onlink = 0;
1128 ip->i_d.di_nlink = nlink;
1129 ASSERT(ip->i_d.di_nlink == nlink);
1130 ip->i_d.di_uid = current_fsuid();
1131 ip->i_d.di_gid = current_fsgid();
1132 ip->i_d.di_projid = prid;
1133 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1136 * If the superblock version is up to where we support new format
1137 * inodes and this is currently an old format inode, then change
1138 * the inode version number now. This way we only do the conversion
1139 * here rather than here and in the flush/logging code.
1141 if (xfs_sb_version_hasnlink(&tp->t_mountp->m_sb) &&
1142 ip->i_d.di_version == 1) {
1143 ip->i_d.di_version = 2;
1145 * We've already zeroed the old link count, the projid field,
1146 * and the pad field.
1151 * Project ids won't be stored on disk if we are using a version 1 inode.
1153 if ((prid != 0) && (ip->i_d.di_version == 1))
1154 xfs_bump_ino_vers2(tp, ip);
1156 if (pip && XFS_INHERIT_GID(pip)) {
1157 ip->i_d.di_gid = pip->i_d.di_gid;
1158 if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1159 ip->i_d.di_mode |= S_ISGID;
1164 * If the group ID of the new file does not match the effective group
1165 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1166 * (and only if the irix_sgid_inherit compatibility variable is set).
1168 if ((irix_sgid_inherit) &&
1169 (ip->i_d.di_mode & S_ISGID) &&
1170 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1171 ip->i_d.di_mode &= ~S_ISGID;
1174 ip->i_d.di_size = 0;
1176 ip->i_d.di_nextents = 0;
1177 ASSERT(ip->i_d.di_nblocks == 0);
1180 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
1181 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
1182 ip->i_d.di_atime = ip->i_d.di_mtime;
1183 ip->i_d.di_ctime = ip->i_d.di_mtime;
1186 * di_gen will have been taken care of in xfs_iread.
1188 ip->i_d.di_extsize = 0;
1189 ip->i_d.di_dmevmask = 0;
1190 ip->i_d.di_dmstate = 0;
1191 ip->i_d.di_flags = 0;
1192 flags = XFS_ILOG_CORE;
1193 switch (mode & S_IFMT) {
1198 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1199 ip->i_df.if_u2.if_rdev = rdev;
1200 ip->i_df.if_flags = 0;
1201 flags |= XFS_ILOG_DEV;
1205 * we can't set up filestreams until after the VFS inode
1206 * is set up properly.
1208 if (pip && xfs_inode_is_filestream(pip))
1212 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
1215 if ((mode & S_IFMT) == S_IFDIR) {
1216 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1217 di_flags |= XFS_DIFLAG_RTINHERIT;
1218 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1219 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1220 ip->i_d.di_extsize = pip->i_d.di_extsize;
1222 } else if ((mode & S_IFMT) == S_IFREG) {
1223 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1224 di_flags |= XFS_DIFLAG_REALTIME;
1225 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1226 di_flags |= XFS_DIFLAG_EXTSIZE;
1227 ip->i_d.di_extsize = pip->i_d.di_extsize;
1230 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1231 xfs_inherit_noatime)
1232 di_flags |= XFS_DIFLAG_NOATIME;
1233 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1235 di_flags |= XFS_DIFLAG_NODUMP;
1236 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1238 di_flags |= XFS_DIFLAG_SYNC;
1239 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1240 xfs_inherit_nosymlinks)
1241 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1242 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1243 di_flags |= XFS_DIFLAG_PROJINHERIT;
1244 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1245 xfs_inherit_nodefrag)
1246 di_flags |= XFS_DIFLAG_NODEFRAG;
1247 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
1248 di_flags |= XFS_DIFLAG_FILESTREAM;
1249 ip->i_d.di_flags |= di_flags;
1253 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1254 ip->i_df.if_flags = XFS_IFEXTENTS;
1255 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1256 ip->i_df.if_u1.if_extents = NULL;
1262 * Attribute fork settings for new inode.
1264 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1265 ip->i_d.di_anextents = 0;
1268 * Log the new values stuffed into the inode.
1270 xfs_trans_log_inode(tp, ip, flags);
1272 /* now that we have an i_mode we can setup inode ops and unlock */
1273 xfs_setup_inode(ip);
1275 /* now we have set up the vfs inode we can associate the filestream */
1277 error = xfs_filestream_associate(pip, ip);
1281 xfs_iflags_set(ip, XFS_IFILESTREAM);
1289 * Check to make sure that there are no blocks allocated to the
1290 * file beyond the size of the file. We don't check this for
1291 * files with fixed size extents or real time extents, but we
1292 * at least do it for regular files.
1301 xfs_fileoff_t map_first;
1303 xfs_bmbt_irec_t imaps[2];
1305 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1308 if (XFS_IS_REALTIME_INODE(ip))
1311 if (ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE)
1315 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1317 * The filesystem could be shutting down, so bmapi may return
1320 if (xfs_bmapi(NULL, ip, map_first,
1322 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1324 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1327 ASSERT(nimaps == 1);
1328 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1333 * Calculate the last possible buffered byte in a file. This must
1334 * include data that was buffered beyond the EOF by the write code.
1335 * This also needs to deal with overflowing the xfs_fsize_t type
1336 * which can happen for sizes near the limit.
1338 * We also need to take into account any blocks beyond the EOF. It
1339 * may be the case that they were buffered by a write which failed.
1340 * In that case the pages will still be in memory, but the inode size
1341 * will never have been updated.
1348 xfs_fsize_t last_byte;
1349 xfs_fileoff_t last_block;
1350 xfs_fileoff_t size_last_block;
1353 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED));
1357 * Only check for blocks beyond the EOF if the extents have
1358 * been read in. This eliminates the need for the inode lock,
1359 * and it also saves us from looking when it really isn't
1362 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1363 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1371 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_size);
1372 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1374 last_byte = XFS_FSB_TO_B(mp, last_block);
1375 if (last_byte < 0) {
1376 return XFS_MAXIOFFSET(mp);
1378 last_byte += (1 << mp->m_writeio_log);
1379 if (last_byte < 0) {
1380 return XFS_MAXIOFFSET(mp);
1385 #if defined(XFS_RW_TRACE)
1391 xfs_fsize_t new_size,
1392 xfs_off_t toss_start,
1393 xfs_off_t toss_finish)
1395 if (ip->i_rwtrace == NULL) {
1399 ktrace_enter(ip->i_rwtrace,
1402 (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
1403 (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
1404 (void*)((long)flag),
1405 (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
1406 (void*)(unsigned long)(new_size & 0xffffffff),
1407 (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
1408 (void*)(unsigned long)(toss_start & 0xffffffff),
1409 (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
1410 (void*)(unsigned long)(toss_finish & 0xffffffff),
1411 (void*)(unsigned long)current_cpu(),
1412 (void*)(unsigned long)current_pid(),
1418 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1422 * Start the truncation of the file to new_size. The new size
1423 * must be smaller than the current size. This routine will
1424 * clear the buffer and page caches of file data in the removed
1425 * range, and xfs_itruncate_finish() will remove the underlying
1428 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1429 * must NOT have the inode lock held at all. This is because we're
1430 * calling into the buffer/page cache code and we can't hold the
1431 * inode lock when we do so.
1433 * We need to wait for any direct I/Os in flight to complete before we
1434 * proceed with the truncate. This is needed to prevent the extents
1435 * being read or written by the direct I/Os from being removed while the
1436 * I/O is in flight as there is no other method of synchronising
1437 * direct I/O with the truncate operation. Also, because we hold
1438 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1439 * started until the truncate completes and drops the lock. Essentially,
1440 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1441 * between direct I/Os and the truncate operation.
1443 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1444 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1445 * in the case that the caller is locking things out of order and
1446 * may not be able to call xfs_itruncate_finish() with the inode lock
1447 * held without dropping the I/O lock. If the caller must drop the
1448 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1449 * must be called again with all the same restrictions as the initial
1453 xfs_itruncate_start(
1456 xfs_fsize_t new_size)
1458 xfs_fsize_t last_byte;
1459 xfs_off_t toss_start;
1463 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1464 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1465 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1466 (flags == XFS_ITRUNC_MAYBE));
1470 /* wait for the completion of any pending DIOs */
1471 if (new_size == 0 || new_size < ip->i_size)
1475 * Call toss_pages or flushinval_pages to get rid of pages
1476 * overlapping the region being removed. We have to use
1477 * the less efficient flushinval_pages in the case that the
1478 * caller may not be able to finish the truncate without
1479 * dropping the inode's I/O lock. Make sure
1480 * to catch any pages brought in by buffers overlapping
1481 * the EOF by searching out beyond the isize by our
1482 * block size. We round new_size up to a block boundary
1483 * so that we don't toss things on the same block as
1484 * new_size but before it.
1486 * Before calling toss_page or flushinval_pages, make sure to
1487 * call remapf() over the same region if the file is mapped.
1488 * This frees up mapped file references to the pages in the
1489 * given range and for the flushinval_pages case it ensures
1490 * that we get the latest mapped changes flushed out.
1492 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1493 toss_start = XFS_FSB_TO_B(mp, toss_start);
1494 if (toss_start < 0) {
1496 * The place to start tossing is beyond our maximum
1497 * file size, so there is no way that the data extended
1502 last_byte = xfs_file_last_byte(ip);
1503 xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
1505 if (last_byte > toss_start) {
1506 if (flags & XFS_ITRUNC_DEFINITE) {
1507 xfs_tosspages(ip, toss_start,
1508 -1, FI_REMAPF_LOCKED);
1510 error = xfs_flushinval_pages(ip, toss_start,
1511 -1, FI_REMAPF_LOCKED);
1516 if (new_size == 0) {
1517 ASSERT(VN_CACHED(VFS_I(ip)) == 0);
1524 * Shrink the file to the given new_size. The new size must be smaller than
1525 * the current size. This will free up the underlying blocks in the removed
1526 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
1528 * The transaction passed to this routine must have made a permanent log
1529 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1530 * given transaction and start new ones, so make sure everything involved in
1531 * the transaction is tidy before calling here. Some transaction will be
1532 * returned to the caller to be committed. The incoming transaction must
1533 * already include the inode, and both inode locks must be held exclusively.
1534 * The inode must also be "held" within the transaction. On return the inode
1535 * will be "held" within the returned transaction. This routine does NOT
1536 * require any disk space to be reserved for it within the transaction.
1538 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1539 * indicates the fork which is to be truncated. For the attribute fork we only
1540 * support truncation to size 0.
1542 * We use the sync parameter to indicate whether or not the first transaction
1543 * we perform might have to be synchronous. For the attr fork, it needs to be
1544 * so if the unlink of the inode is not yet known to be permanent in the log.
1545 * This keeps us from freeing and reusing the blocks of the attribute fork
1546 * before the unlink of the inode becomes permanent.
1548 * For the data fork, we normally have to run synchronously if we're being
1549 * called out of the inactive path or we're being called out of the create path
1550 * where we're truncating an existing file. Either way, the truncate needs to
1551 * be sync so blocks don't reappear in the file with altered data in case of a
1552 * crash. wsync filesystems can run the first case async because anything that
1553 * shrinks the inode has to run sync so by the time we're called here from
1554 * inactive, the inode size is permanently set to 0.
1556 * Calls from the truncate path always need to be sync unless we're in a wsync
1557 * filesystem and the file has already been unlinked.
1559 * The caller is responsible for correctly setting the sync parameter. It gets
1560 * too hard for us to guess here which path we're being called out of just
1561 * based on inode state.
1563 * If we get an error, we must return with the inode locked and linked into the
1564 * current transaction. This keeps things simple for the higher level code,
1565 * because it always knows that the inode is locked and held in the transaction
1566 * that returns to it whether errors occur or not. We don't mark the inode
1567 * dirty on error so that transactions can be easily aborted if possible.
1570 xfs_itruncate_finish(
1573 xfs_fsize_t new_size,
1577 xfs_fsblock_t first_block;
1578 xfs_fileoff_t first_unmap_block;
1579 xfs_fileoff_t last_block;
1580 xfs_filblks_t unmap_len=0;
1585 xfs_bmap_free_t free_list;
1588 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
1589 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1590 ASSERT(*tp != NULL);
1591 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1592 ASSERT(ip->i_transp == *tp);
1593 ASSERT(ip->i_itemp != NULL);
1594 ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
1598 mp = (ntp)->t_mountp;
1599 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1602 * We only support truncating the entire attribute fork.
1604 if (fork == XFS_ATTR_FORK) {
1607 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1608 xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
1610 * The first thing we do is set the size to new_size permanently
1611 * on disk. This way we don't have to worry about anyone ever
1612 * being able to look at the data being freed even in the face
1613 * of a crash. What we're getting around here is the case where
1614 * we free a block, it is allocated to another file, it is written
1615 * to, and then we crash. If the new data gets written to the
1616 * file but the log buffers containing the free and reallocation
1617 * don't, then we'd end up with garbage in the blocks being freed.
1618 * As long as we make the new_size permanent before actually
1619 * freeing any blocks it doesn't matter if they get writtten to.
1621 * The callers must signal into us whether or not the size
1622 * setting here must be synchronous. There are a few cases
1623 * where it doesn't have to be synchronous. Those cases
1624 * occur if the file is unlinked and we know the unlink is
1625 * permanent or if the blocks being truncated are guaranteed
1626 * to be beyond the inode eof (regardless of the link count)
1627 * and the eof value is permanent. Both of these cases occur
1628 * only on wsync-mounted filesystems. In those cases, we're
1629 * guaranteed that no user will ever see the data in the blocks
1630 * that are being truncated so the truncate can run async.
1631 * In the free beyond eof case, the file may wind up with
1632 * more blocks allocated to it than it needs if we crash
1633 * and that won't get fixed until the next time the file
1634 * is re-opened and closed but that's ok as that shouldn't
1635 * be too many blocks.
1637 * However, we can't just make all wsync xactions run async
1638 * because there's one call out of the create path that needs
1639 * to run sync where it's truncating an existing file to size
1640 * 0 whose size is > 0.
1642 * It's probably possible to come up with a test in this
1643 * routine that would correctly distinguish all the above
1644 * cases from the values of the function parameters and the
1645 * inode state but for sanity's sake, I've decided to let the
1646 * layers above just tell us. It's simpler to correctly figure
1647 * out in the layer above exactly under what conditions we
1648 * can run async and I think it's easier for others read and
1649 * follow the logic in case something has to be changed.
1650 * cscope is your friend -- rcc.
1652 * The attribute fork is much simpler.
1654 * For the attribute fork we allow the caller to tell us whether
1655 * the unlink of the inode that led to this call is yet permanent
1656 * in the on disk log. If it is not and we will be freeing extents
1657 * in this inode then we make the first transaction synchronous
1658 * to make sure that the unlink is permanent by the time we free
1661 if (fork == XFS_DATA_FORK) {
1662 if (ip->i_d.di_nextents > 0) {
1664 * If we are not changing the file size then do
1665 * not update the on-disk file size - we may be
1666 * called from xfs_inactive_free_eofblocks(). If we
1667 * update the on-disk file size and then the system
1668 * crashes before the contents of the file are
1669 * flushed to disk then the files may be full of
1670 * holes (ie NULL files bug).
1672 if (ip->i_size != new_size) {
1673 ip->i_d.di_size = new_size;
1674 ip->i_size = new_size;
1675 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1679 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1680 if (ip->i_d.di_anextents > 0)
1681 xfs_trans_set_sync(ntp);
1683 ASSERT(fork == XFS_DATA_FORK ||
1684 (fork == XFS_ATTR_FORK &&
1685 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1686 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1689 * Since it is possible for space to become allocated beyond
1690 * the end of the file (in a crash where the space is allocated
1691 * but the inode size is not yet updated), simply remove any
1692 * blocks which show up between the new EOF and the maximum
1693 * possible file size. If the first block to be removed is
1694 * beyond the maximum file size (ie it is the same as last_block),
1695 * then there is nothing to do.
1697 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1698 ASSERT(first_unmap_block <= last_block);
1700 if (last_block == first_unmap_block) {
1703 unmap_len = last_block - first_unmap_block + 1;
1707 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1708 * will tell us whether it freed the entire range or
1709 * not. If this is a synchronous mount (wsync),
1710 * then we can tell bunmapi to keep all the
1711 * transactions asynchronous since the unlink
1712 * transaction that made this inode inactive has
1713 * already hit the disk. There's no danger of
1714 * the freed blocks being reused, there being a
1715 * crash, and the reused blocks suddenly reappearing
1716 * in this file with garbage in them once recovery
1719 XFS_BMAP_INIT(&free_list, &first_block);
1720 error = xfs_bunmapi(ntp, ip,
1721 first_unmap_block, unmap_len,
1722 XFS_BMAPI_AFLAG(fork) |
1723 (sync ? 0 : XFS_BMAPI_ASYNC),
1724 XFS_ITRUNC_MAX_EXTENTS,
1725 &first_block, &free_list,
1729 * If the bunmapi call encounters an error,
1730 * return to the caller where the transaction
1731 * can be properly aborted. We just need to
1732 * make sure we're not holding any resources
1733 * that we were not when we came in.
1735 xfs_bmap_cancel(&free_list);
1740 * Duplicate the transaction that has the permanent
1741 * reservation and commit the old transaction.
1743 error = xfs_bmap_finish(tp, &free_list, &committed);
1746 /* link the inode into the next xact in the chain */
1747 xfs_trans_ijoin(ntp, ip,
1748 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1749 xfs_trans_ihold(ntp, ip);
1754 * If the bmap finish call encounters an error, return
1755 * to the caller where the transaction can be properly
1756 * aborted. We just need to make sure we're not
1757 * holding any resources that we were not when we came
1760 * Aborting from this point might lose some blocks in
1761 * the file system, but oh well.
1763 xfs_bmap_cancel(&free_list);
1769 * Mark the inode dirty so it will be logged and
1770 * moved forward in the log as part of every commit.
1772 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1775 ntp = xfs_trans_dup(ntp);
1776 error = xfs_trans_commit(*tp, 0);
1779 /* link the inode into the next transaction in the chain */
1780 xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1781 xfs_trans_ihold(ntp, ip);
1786 * transaction commit worked ok so we can drop the extra ticket
1787 * reference that we gained in xfs_trans_dup()
1789 xfs_log_ticket_put(ntp->t_ticket);
1790 error = xfs_trans_reserve(ntp, 0,
1791 XFS_ITRUNCATE_LOG_RES(mp), 0,
1792 XFS_TRANS_PERM_LOG_RES,
1793 XFS_ITRUNCATE_LOG_COUNT);
1798 * Only update the size in the case of the data fork, but
1799 * always re-log the inode so that our permanent transaction
1800 * can keep on rolling it forward in the log.
1802 if (fork == XFS_DATA_FORK) {
1803 xfs_isize_check(mp, ip, new_size);
1805 * If we are not changing the file size then do
1806 * not update the on-disk file size - we may be
1807 * called from xfs_inactive_free_eofblocks(). If we
1808 * update the on-disk file size and then the system
1809 * crashes before the contents of the file are
1810 * flushed to disk then the files may be full of
1811 * holes (ie NULL files bug).
1813 if (ip->i_size != new_size) {
1814 ip->i_d.di_size = new_size;
1815 ip->i_size = new_size;
1818 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1819 ASSERT((new_size != 0) ||
1820 (fork == XFS_ATTR_FORK) ||
1821 (ip->i_delayed_blks == 0));
1822 ASSERT((new_size != 0) ||
1823 (fork == XFS_ATTR_FORK) ||
1824 (ip->i_d.di_nextents == 0));
1825 xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
1830 * This is called when the inode's link count goes to 0.
1831 * We place the on-disk inode on a list in the AGI. It
1832 * will be pulled from this list when the inode is freed.
1849 ASSERT(ip->i_d.di_nlink == 0);
1850 ASSERT(ip->i_d.di_mode != 0);
1851 ASSERT(ip->i_transp == tp);
1856 * Get the agi buffer first. It ensures lock ordering
1859 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1862 agi = XFS_BUF_TO_AGI(agibp);
1865 * Get the index into the agi hash table for the
1866 * list this inode will go on.
1868 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1870 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1871 ASSERT(agi->agi_unlinked[bucket_index]);
1872 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1874 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
1876 * There is already another inode in the bucket we need
1877 * to add ourselves to. Add us at the front of the list.
1878 * Here we put the head pointer into our next pointer,
1879 * and then we fall through to point the head at us.
1881 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
1885 ASSERT(be32_to_cpu(dip->di_next_unlinked) == NULLAGINO);
1886 /* both on-disk, don't endian flip twice */
1887 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1888 offset = ip->i_boffset +
1889 offsetof(xfs_dinode_t, di_next_unlinked);
1890 xfs_trans_inode_buf(tp, ibp);
1891 xfs_trans_log_buf(tp, ibp, offset,
1892 (offset + sizeof(xfs_agino_t) - 1));
1893 xfs_inobp_check(mp, ibp);
1897 * Point the bucket head pointer at the inode being inserted.
1900 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1901 offset = offsetof(xfs_agi_t, agi_unlinked) +
1902 (sizeof(xfs_agino_t) * bucket_index);
1903 xfs_trans_log_buf(tp, agibp, offset,
1904 (offset + sizeof(xfs_agino_t) - 1));
1909 * Pull the on-disk inode from the AGI unlinked list.
1922 xfs_agnumber_t agno;
1924 xfs_agino_t next_agino;
1925 xfs_buf_t *last_ibp;
1926 xfs_dinode_t *last_dip = NULL;
1928 int offset, last_offset = 0;
1932 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1935 * Get the agi buffer first. It ensures lock ordering
1938 error = xfs_read_agi(mp, tp, agno, &agibp);
1942 agi = XFS_BUF_TO_AGI(agibp);
1945 * Get the index into the agi hash table for the
1946 * list this inode will go on.
1948 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1950 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1951 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
1952 ASSERT(agi->agi_unlinked[bucket_index]);
1954 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
1956 * We're at the head of the list. Get the inode's
1957 * on-disk buffer to see if there is anyone after us
1958 * on the list. Only modify our next pointer if it
1959 * is not already NULLAGINO. This saves us the overhead
1960 * of dealing with the buffer when there is no need to
1963 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
1966 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
1967 error, mp->m_fsname);
1970 next_agino = be32_to_cpu(dip->di_next_unlinked);
1971 ASSERT(next_agino != 0);
1972 if (next_agino != NULLAGINO) {
1973 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
1974 offset = ip->i_boffset +
1975 offsetof(xfs_dinode_t, di_next_unlinked);
1976 xfs_trans_inode_buf(tp, ibp);
1977 xfs_trans_log_buf(tp, ibp, offset,
1978 (offset + sizeof(xfs_agino_t) - 1));
1979 xfs_inobp_check(mp, ibp);
1981 xfs_trans_brelse(tp, ibp);
1984 * Point the bucket head pointer at the next inode.
1986 ASSERT(next_agino != 0);
1987 ASSERT(next_agino != agino);
1988 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
1989 offset = offsetof(xfs_agi_t, agi_unlinked) +
1990 (sizeof(xfs_agino_t) * bucket_index);
1991 xfs_trans_log_buf(tp, agibp, offset,
1992 (offset + sizeof(xfs_agino_t) - 1));
1995 * We need to search the list for the inode being freed.
1997 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
1999 while (next_agino != agino) {
2001 * If the last inode wasn't the one pointing to
2002 * us, then release its buffer since we're not
2003 * going to do anything with it.
2005 if (last_ibp != NULL) {
2006 xfs_trans_brelse(tp, last_ibp);
2008 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2009 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
2010 &last_ibp, &last_offset, 0);
2013 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
2014 error, mp->m_fsname);
2017 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2018 ASSERT(next_agino != NULLAGINO);
2019 ASSERT(next_agino != 0);
2022 * Now last_ibp points to the buffer previous to us on
2023 * the unlinked list. Pull us from the list.
2025 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
2028 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2029 error, mp->m_fsname);
2032 next_agino = be32_to_cpu(dip->di_next_unlinked);
2033 ASSERT(next_agino != 0);
2034 ASSERT(next_agino != agino);
2035 if (next_agino != NULLAGINO) {
2036 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2037 offset = ip->i_boffset +
2038 offsetof(xfs_dinode_t, di_next_unlinked);
2039 xfs_trans_inode_buf(tp, ibp);
2040 xfs_trans_log_buf(tp, ibp, offset,
2041 (offset + sizeof(xfs_agino_t) - 1));
2042 xfs_inobp_check(mp, ibp);
2044 xfs_trans_brelse(tp, ibp);
2047 * Point the previous inode on the list to the next inode.
2049 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2050 ASSERT(next_agino != 0);
2051 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2052 xfs_trans_inode_buf(tp, last_ibp);
2053 xfs_trans_log_buf(tp, last_ibp, offset,
2054 (offset + sizeof(xfs_agino_t) - 1));
2055 xfs_inobp_check(mp, last_ibp);
2062 xfs_inode_t *free_ip,
2066 xfs_mount_t *mp = free_ip->i_mount;
2067 int blks_per_cluster;
2070 int i, j, found, pre_flushed;
2073 xfs_inode_t *ip, **ip_found;
2074 xfs_inode_log_item_t *iip;
2075 xfs_log_item_t *lip;
2076 xfs_perag_t *pag = xfs_get_perag(mp, inum);
2078 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
2079 blks_per_cluster = 1;
2080 ninodes = mp->m_sb.sb_inopblock;
2081 nbufs = XFS_IALLOC_BLOCKS(mp);
2083 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
2084 mp->m_sb.sb_blocksize;
2085 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
2086 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
2089 ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
2091 for (j = 0; j < nbufs; j++, inum += ninodes) {
2092 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2093 XFS_INO_TO_AGBNO(mp, inum));
2097 * Look for each inode in memory and attempt to lock it,
2098 * we can be racing with flush and tail pushing here.
2099 * any inode we get the locks on, add to an array of
2100 * inode items to process later.
2102 * The get the buffer lock, we could beat a flush
2103 * or tail pushing thread to the lock here, in which
2104 * case they will go looking for the inode buffer
2105 * and fail, we need some other form of interlock
2109 for (i = 0; i < ninodes; i++) {
2110 read_lock(&pag->pag_ici_lock);
2111 ip = radix_tree_lookup(&pag->pag_ici_root,
2112 XFS_INO_TO_AGINO(mp, (inum + i)));
2114 /* Inode not in memory or we found it already,
2117 if (!ip || xfs_iflags_test(ip, XFS_ISTALE)) {
2118 read_unlock(&pag->pag_ici_lock);
2122 if (xfs_inode_clean(ip)) {
2123 read_unlock(&pag->pag_ici_lock);
2127 /* If we can get the locks then add it to the
2128 * list, otherwise by the time we get the bp lock
2129 * below it will already be attached to the
2133 /* This inode will already be locked - by us, lets
2137 if (ip == free_ip) {
2138 if (xfs_iflock_nowait(ip)) {
2139 xfs_iflags_set(ip, XFS_ISTALE);
2140 if (xfs_inode_clean(ip)) {
2143 ip_found[found++] = ip;
2146 read_unlock(&pag->pag_ici_lock);
2150 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2151 if (xfs_iflock_nowait(ip)) {
2152 xfs_iflags_set(ip, XFS_ISTALE);
2154 if (xfs_inode_clean(ip)) {
2156 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2158 ip_found[found++] = ip;
2161 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2164 read_unlock(&pag->pag_ici_lock);
2167 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2168 mp->m_bsize * blks_per_cluster,
2172 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
2174 if (lip->li_type == XFS_LI_INODE) {
2175 iip = (xfs_inode_log_item_t *)lip;
2176 ASSERT(iip->ili_logged == 1);
2177 lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
2178 xfs_trans_ail_copy_lsn(mp->m_ail,
2179 &iip->ili_flush_lsn,
2180 &iip->ili_item.li_lsn);
2181 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2184 lip = lip->li_bio_list;
2187 for (i = 0; i < found; i++) {
2192 ip->i_update_core = 0;
2194 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2198 iip->ili_last_fields = iip->ili_format.ilf_fields;
2199 iip->ili_format.ilf_fields = 0;
2200 iip->ili_logged = 1;
2201 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2202 &iip->ili_item.li_lsn);
2204 xfs_buf_attach_iodone(bp,
2205 (void(*)(xfs_buf_t*,xfs_log_item_t*))
2206 xfs_istale_done, (xfs_log_item_t *)iip);
2207 if (ip != free_ip) {
2208 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2212 if (found || pre_flushed)
2213 xfs_trans_stale_inode_buf(tp, bp);
2214 xfs_trans_binval(tp, bp);
2217 kmem_free(ip_found);
2218 xfs_put_perag(mp, pag);
2222 * This is called to return an inode to the inode free list.
2223 * The inode should already be truncated to 0 length and have
2224 * no pages associated with it. This routine also assumes that
2225 * the inode is already a part of the transaction.
2227 * The on-disk copy of the inode will have been added to the list
2228 * of unlinked inodes in the AGI. We need to remove the inode from
2229 * that list atomically with respect to freeing it here.
2235 xfs_bmap_free_t *flist)
2239 xfs_ino_t first_ino;
2243 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2244 ASSERT(ip->i_transp == tp);
2245 ASSERT(ip->i_d.di_nlink == 0);
2246 ASSERT(ip->i_d.di_nextents == 0);
2247 ASSERT(ip->i_d.di_anextents == 0);
2248 ASSERT((ip->i_d.di_size == 0 && ip->i_size == 0) ||
2249 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2250 ASSERT(ip->i_d.di_nblocks == 0);
2253 * Pull the on-disk inode from the AGI unlinked list.
2255 error = xfs_iunlink_remove(tp, ip);
2260 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2264 ip->i_d.di_mode = 0; /* mark incore inode as free */
2265 ip->i_d.di_flags = 0;
2266 ip->i_d.di_dmevmask = 0;
2267 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2268 ip->i_df.if_ext_max =
2269 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2270 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2271 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2273 * Bump the generation count so no one will be confused
2274 * by reincarnations of this inode.
2278 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2280 error = xfs_itobp(ip->i_mount, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
2285 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2286 * from picking up this inode when it is reclaimed (its incore state
2287 * initialzed but not flushed to disk yet). The in-core di_mode is
2288 * already cleared and a corresponding transaction logged.
2289 * The hack here just synchronizes the in-core to on-disk
2290 * di_mode value in advance before the actual inode sync to disk.
2291 * This is OK because the inode is already unlinked and would never
2292 * change its di_mode again for this inode generation.
2293 * This is a temporary hack that would require a proper fix
2299 xfs_ifree_cluster(ip, tp, first_ino);
2306 * Reallocate the space for if_broot based on the number of records
2307 * being added or deleted as indicated in rec_diff. Move the records
2308 * and pointers in if_broot to fit the new size. When shrinking this
2309 * will eliminate holes between the records and pointers created by
2310 * the caller. When growing this will create holes to be filled in
2313 * The caller must not request to add more records than would fit in
2314 * the on-disk inode root. If the if_broot is currently NULL, then
2315 * if we adding records one will be allocated. The caller must also
2316 * not request that the number of records go below zero, although
2317 * it can go to zero.
2319 * ip -- the inode whose if_broot area is changing
2320 * ext_diff -- the change in the number of records, positive or negative,
2321 * requested for the if_broot array.
2329 struct xfs_mount *mp = ip->i_mount;
2332 struct xfs_btree_block *new_broot;
2339 * Handle the degenerate case quietly.
2341 if (rec_diff == 0) {
2345 ifp = XFS_IFORK_PTR(ip, whichfork);
2348 * If there wasn't any memory allocated before, just
2349 * allocate it now and get out.
2351 if (ifp->if_broot_bytes == 0) {
2352 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
2353 ifp->if_broot = kmem_alloc(new_size, KM_SLEEP);
2354 ifp->if_broot_bytes = (int)new_size;
2359 * If there is already an existing if_broot, then we need
2360 * to realloc() it and shift the pointers to their new
2361 * location. The records don't change location because
2362 * they are kept butted up against the btree block header.
2364 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
2365 new_max = cur_max + rec_diff;
2366 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2367 ifp->if_broot = kmem_realloc(ifp->if_broot, new_size,
2368 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2370 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2371 ifp->if_broot_bytes);
2372 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2374 ifp->if_broot_bytes = (int)new_size;
2375 ASSERT(ifp->if_broot_bytes <=
2376 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2377 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2382 * rec_diff is less than 0. In this case, we are shrinking the
2383 * if_broot buffer. It must already exist. If we go to zero
2384 * records, just get rid of the root and clear the status bit.
2386 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
2387 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
2388 new_max = cur_max + rec_diff;
2389 ASSERT(new_max >= 0);
2391 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2395 new_broot = kmem_alloc(new_size, KM_SLEEP);
2397 * First copy over the btree block header.
2399 memcpy(new_broot, ifp->if_broot, XFS_BTREE_LBLOCK_LEN);
2402 ifp->if_flags &= ~XFS_IFBROOT;
2406 * Only copy the records and pointers if there are any.
2410 * First copy the records.
2412 op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1);
2413 np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1);
2414 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2417 * Then copy the pointers.
2419 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2420 ifp->if_broot_bytes);
2421 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1,
2423 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2425 kmem_free(ifp->if_broot);
2426 ifp->if_broot = new_broot;
2427 ifp->if_broot_bytes = (int)new_size;
2428 ASSERT(ifp->if_broot_bytes <=
2429 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2435 * This is called when the amount of space needed for if_data
2436 * is increased or decreased. The change in size is indicated by
2437 * the number of bytes that need to be added or deleted in the
2438 * byte_diff parameter.
2440 * If the amount of space needed has decreased below the size of the
2441 * inline buffer, then switch to using the inline buffer. Otherwise,
2442 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2443 * to what is needed.
2445 * ip -- the inode whose if_data area is changing
2446 * byte_diff -- the change in the number of bytes, positive or negative,
2447 * requested for the if_data array.
2459 if (byte_diff == 0) {
2463 ifp = XFS_IFORK_PTR(ip, whichfork);
2464 new_size = (int)ifp->if_bytes + byte_diff;
2465 ASSERT(new_size >= 0);
2467 if (new_size == 0) {
2468 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2469 kmem_free(ifp->if_u1.if_data);
2471 ifp->if_u1.if_data = NULL;
2473 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2475 * If the valid extents/data can fit in if_inline_ext/data,
2476 * copy them from the malloc'd vector and free it.
2478 if (ifp->if_u1.if_data == NULL) {
2479 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2480 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2481 ASSERT(ifp->if_real_bytes != 0);
2482 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2484 kmem_free(ifp->if_u1.if_data);
2485 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2490 * Stuck with malloc/realloc.
2491 * For inline data, the underlying buffer must be
2492 * a multiple of 4 bytes in size so that it can be
2493 * logged and stay on word boundaries. We enforce
2496 real_size = roundup(new_size, 4);
2497 if (ifp->if_u1.if_data == NULL) {
2498 ASSERT(ifp->if_real_bytes == 0);
2499 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2500 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2502 * Only do the realloc if the underlying size
2503 * is really changing.
2505 if (ifp->if_real_bytes != real_size) {
2506 ifp->if_u1.if_data =
2507 kmem_realloc(ifp->if_u1.if_data,
2513 ASSERT(ifp->if_real_bytes == 0);
2514 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2515 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2519 ifp->if_real_bytes = real_size;
2520 ifp->if_bytes = new_size;
2521 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2528 * Map inode to disk block and offset.
2530 * mp -- the mount point structure for the current file system
2531 * tp -- the current transaction
2532 * ino -- the inode number of the inode to be located
2533 * imap -- this structure is filled in with the information necessary
2534 * to retrieve the given inode from disk
2535 * flags -- flags to pass to xfs_dilocate indicating whether or not
2536 * lookups in the inode btree were OK or not
2546 xfs_fsblock_t fsbno;
2551 fsbno = imap->im_blkno ?
2552 XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK;
2553 error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags);
2557 imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno);
2558 imap->im_len = XFS_FSB_TO_BB(mp, len);
2559 imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno);
2560 imap->im_ioffset = (ushort)off;
2561 imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog);
2564 * If the inode number maps to a block outside the bounds
2565 * of the file system then return NULL rather than calling
2566 * read_buf and panicing when we get an error from the
2569 if ((imap->im_blkno + imap->im_len) >
2570 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
2571 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
2572 "(imap->im_blkno (0x%llx) + imap->im_len (0x%llx)) > "
2573 " XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks) (0x%llx)",
2574 (unsigned long long) imap->im_blkno,
2575 (unsigned long long) imap->im_len,
2576 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
2589 ifp = XFS_IFORK_PTR(ip, whichfork);
2590 if (ifp->if_broot != NULL) {
2591 kmem_free(ifp->if_broot);
2592 ifp->if_broot = NULL;
2596 * If the format is local, then we can't have an extents
2597 * array so just look for an inline data array. If we're
2598 * not local then we may or may not have an extents list,
2599 * so check and free it up if we do.
2601 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2602 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2603 (ifp->if_u1.if_data != NULL)) {
2604 ASSERT(ifp->if_real_bytes != 0);
2605 kmem_free(ifp->if_u1.if_data);
2606 ifp->if_u1.if_data = NULL;
2607 ifp->if_real_bytes = 0;
2609 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2610 ((ifp->if_flags & XFS_IFEXTIREC) ||
2611 ((ifp->if_u1.if_extents != NULL) &&
2612 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
2613 ASSERT(ifp->if_real_bytes != 0);
2614 xfs_iext_destroy(ifp);
2616 ASSERT(ifp->if_u1.if_extents == NULL ||
2617 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2618 ASSERT(ifp->if_real_bytes == 0);
2619 if (whichfork == XFS_ATTR_FORK) {
2620 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2626 * This is called free all the memory associated with an inode.
2627 * It must free the inode itself and any buffers allocated for
2628 * if_extents/if_data and if_broot. It must also free the lock
2629 * associated with the inode.
2631 * Note: because we don't initialise everything on reallocation out
2632 * of the zone, we must ensure we nullify everything correctly before
2633 * freeing the structure.
2639 switch (ip->i_d.di_mode & S_IFMT) {
2643 xfs_idestroy_fork(ip, XFS_DATA_FORK);
2647 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
2649 #ifdef XFS_INODE_TRACE
2650 ktrace_free(ip->i_trace);
2652 #ifdef XFS_BMAP_TRACE
2653 ktrace_free(ip->i_xtrace);
2655 #ifdef XFS_BTREE_TRACE
2656 ktrace_free(ip->i_btrace);
2659 ktrace_free(ip->i_rwtrace);
2661 #ifdef XFS_ILOCK_TRACE
2662 ktrace_free(ip->i_lock_trace);
2664 #ifdef XFS_DIR2_TRACE
2665 ktrace_free(ip->i_dir_trace);
2669 * Only if we are shutting down the fs will we see an
2670 * inode still in the AIL. If it is there, we should remove
2671 * it to prevent a use-after-free from occurring.
2673 xfs_log_item_t *lip = &ip->i_itemp->ili_item;
2674 struct xfs_ail *ailp = lip->li_ailp;
2676 ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) ||
2677 XFS_FORCED_SHUTDOWN(ip->i_mount));
2678 if (lip->li_flags & XFS_LI_IN_AIL) {
2679 spin_lock(&ailp->xa_lock);
2680 if (lip->li_flags & XFS_LI_IN_AIL)
2681 xfs_trans_ail_delete(ailp, lip);
2683 spin_unlock(&ailp->xa_lock);
2685 xfs_inode_item_destroy(ip);
2688 /* asserts to verify all state is correct here */
2689 ASSERT(atomic_read(&ip->i_iocount) == 0);
2690 ASSERT(atomic_read(&ip->i_pincount) == 0);
2691 ASSERT(!spin_is_locked(&ip->i_flags_lock));
2692 ASSERT(completion_done(&ip->i_flush));
2693 kmem_zone_free(xfs_inode_zone, ip);
2698 * Increment the pin count of the given buffer.
2699 * This value is protected by ipinlock spinlock in the mount structure.
2705 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2707 atomic_inc(&ip->i_pincount);
2711 * Decrement the pin count of the given inode, and wake up
2712 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2713 * inode must have been previously pinned with a call to xfs_ipin().
2719 ASSERT(atomic_read(&ip->i_pincount) > 0);
2721 if (atomic_dec_and_test(&ip->i_pincount))
2722 wake_up(&ip->i_ipin_wait);
2726 * This is called to unpin an inode. It can be directed to wait or to return
2727 * immediately without waiting for the inode to be unpinned. The caller must
2728 * have the inode locked in at least shared mode so that the buffer cannot be
2729 * subsequently pinned once someone is waiting for it to be unpinned.
2736 xfs_inode_log_item_t *iip = ip->i_itemp;
2738 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2739 if (atomic_read(&ip->i_pincount) == 0)
2742 /* Give the log a push to start the unpinning I/O */
2743 xfs_log_force(ip->i_mount, (iip && iip->ili_last_lsn) ?
2744 iip->ili_last_lsn : 0, XFS_LOG_FORCE);
2746 wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
2753 __xfs_iunpin_wait(ip, 1);
2760 __xfs_iunpin_wait(ip, 0);
2765 * xfs_iextents_copy()
2767 * This is called to copy the REAL extents (as opposed to the delayed
2768 * allocation extents) from the inode into the given buffer. It
2769 * returns the number of bytes copied into the buffer.
2771 * If there are no delayed allocation extents, then we can just
2772 * memcpy() the extents into the buffer. Otherwise, we need to
2773 * examine each extent in turn and skip those which are delayed.
2785 xfs_fsblock_t start_block;
2787 ifp = XFS_IFORK_PTR(ip, whichfork);
2788 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2789 ASSERT(ifp->if_bytes > 0);
2791 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2792 XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork);
2796 * There are some delayed allocation extents in the
2797 * inode, so copy the extents one at a time and skip
2798 * the delayed ones. There must be at least one
2799 * non-delayed extent.
2802 for (i = 0; i < nrecs; i++) {
2803 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
2804 start_block = xfs_bmbt_get_startblock(ep);
2805 if (ISNULLSTARTBLOCK(start_block)) {
2807 * It's a delayed allocation extent, so skip it.
2812 /* Translate to on disk format */
2813 put_unaligned(cpu_to_be64(ep->l0), &dp->l0);
2814 put_unaligned(cpu_to_be64(ep->l1), &dp->l1);
2818 ASSERT(copied != 0);
2819 xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip));
2821 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2825 * Each of the following cases stores data into the same region
2826 * of the on-disk inode, so only one of them can be valid at
2827 * any given time. While it is possible to have conflicting formats
2828 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2829 * in EXTENTS format, this can only happen when the fork has
2830 * changed formats after being modified but before being flushed.
2831 * In these cases, the format always takes precedence, because the
2832 * format indicates the current state of the fork.
2839 xfs_inode_log_item_t *iip,
2846 #ifdef XFS_TRANS_DEBUG
2849 static const short brootflag[2] =
2850 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2851 static const short dataflag[2] =
2852 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2853 static const short extflag[2] =
2854 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2858 ifp = XFS_IFORK_PTR(ip, whichfork);
2860 * This can happen if we gave up in iformat in an error path,
2861 * for the attribute fork.
2864 ASSERT(whichfork == XFS_ATTR_FORK);
2867 cp = XFS_DFORK_PTR(dip, whichfork);
2869 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2870 case XFS_DINODE_FMT_LOCAL:
2871 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2872 (ifp->if_bytes > 0)) {
2873 ASSERT(ifp->if_u1.if_data != NULL);
2874 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2875 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2879 case XFS_DINODE_FMT_EXTENTS:
2880 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2881 !(iip->ili_format.ilf_fields & extflag[whichfork]));
2882 ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
2883 (ifp->if_bytes == 0));
2884 ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
2885 (ifp->if_bytes > 0));
2886 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
2887 (ifp->if_bytes > 0)) {
2888 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2889 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2894 case XFS_DINODE_FMT_BTREE:
2895 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
2896 (ifp->if_broot_bytes > 0)) {
2897 ASSERT(ifp->if_broot != NULL);
2898 ASSERT(ifp->if_broot_bytes <=
2899 (XFS_IFORK_SIZE(ip, whichfork) +
2900 XFS_BROOT_SIZE_ADJ));
2901 xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes,
2902 (xfs_bmdr_block_t *)cp,
2903 XFS_DFORK_SIZE(dip, mp, whichfork));
2907 case XFS_DINODE_FMT_DEV:
2908 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
2909 ASSERT(whichfork == XFS_DATA_FORK);
2910 xfs_dinode_put_rdev(dip, ip->i_df.if_u2.if_rdev);
2914 case XFS_DINODE_FMT_UUID:
2915 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
2916 ASSERT(whichfork == XFS_DATA_FORK);
2917 memcpy(XFS_DFORK_DPTR(dip),
2918 &ip->i_df.if_u2.if_uuid,
2934 xfs_mount_t *mp = ip->i_mount;
2935 xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino);
2936 unsigned long first_index, mask;
2937 unsigned long inodes_per_cluster;
2939 xfs_inode_t **ilist;
2946 ASSERT(pag->pagi_inodeok);
2947 ASSERT(pag->pag_ici_init);
2949 inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog;
2950 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
2951 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
2955 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
2956 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
2957 read_lock(&pag->pag_ici_lock);
2958 /* really need a gang lookup range call here */
2959 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
2960 first_index, inodes_per_cluster);
2964 for (i = 0; i < nr_found; i++) {
2968 /* if the inode lies outside this cluster, we're done. */
2969 if ((XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index)
2972 * Do an un-protected check to see if the inode is dirty and
2973 * is a candidate for flushing. These checks will be repeated
2974 * later after the appropriate locks are acquired.
2976 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
2980 * Try to get locks. If any are unavailable or it is pinned,
2981 * then this inode cannot be flushed and is skipped.
2984 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
2986 if (!xfs_iflock_nowait(iq)) {
2987 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2990 if (xfs_ipincount(iq)) {
2992 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2997 * arriving here means that this inode can be flushed. First
2998 * re-check that it's dirty before flushing.
3000 if (!xfs_inode_clean(iq)) {
3002 error = xfs_iflush_int(iq, bp);
3004 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3005 goto cluster_corrupt_out;
3011 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3015 XFS_STATS_INC(xs_icluster_flushcnt);
3016 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
3020 read_unlock(&pag->pag_ici_lock);
3025 cluster_corrupt_out:
3027 * Corruption detected in the clustering loop. Invalidate the
3028 * inode buffer and shut down the filesystem.
3030 read_unlock(&pag->pag_ici_lock);
3032 * Clean up the buffer. If it was B_DELWRI, just release it --
3033 * brelse can handle it with no problems. If not, shut down the
3034 * filesystem before releasing the buffer.
3036 bufwasdelwri = XFS_BUF_ISDELAYWRITE(bp);
3040 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3042 if (!bufwasdelwri) {
3044 * Just like incore_relse: if we have b_iodone functions,
3045 * mark the buffer as an error and call them. Otherwise
3046 * mark it as stale and brelse.
3048 if (XFS_BUF_IODONE_FUNC(bp)) {
3049 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
3053 XFS_BUF_ERROR(bp,EIO);
3062 * Unlocks the flush lock
3064 xfs_iflush_abort(iq);
3066 return XFS_ERROR(EFSCORRUPTED);
3070 * xfs_iflush() will write a modified inode's changes out to the
3071 * inode's on disk home. The caller must have the inode lock held
3072 * in at least shared mode and the inode flush completion must be
3073 * active as well. The inode lock will still be held upon return from
3074 * the call and the caller is free to unlock it.
3075 * The inode flush will be completed when the inode reaches the disk.
3076 * The flags indicate how the inode's buffer should be written out.
3083 xfs_inode_log_item_t *iip;
3088 int noblock = (flags == XFS_IFLUSH_ASYNC_NOBLOCK);
3089 enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
3091 XFS_STATS_INC(xs_iflush_count);
3093 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3094 ASSERT(!completion_done(&ip->i_flush));
3095 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3096 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3102 * If the inode isn't dirty, then just release the inode
3103 * flush lock and do nothing.
3105 if (xfs_inode_clean(ip)) {
3111 * We can't flush the inode until it is unpinned, so wait for it if we
3112 * are allowed to block. We know noone new can pin it, because we are
3113 * holding the inode lock shared and you need to hold it exclusively to
3116 * If we are not allowed to block, force the log out asynchronously so
3117 * that when we come back the inode will be unpinned. If other inodes
3118 * in the same cluster are dirty, they will probably write the inode
3119 * out for us if they occur after the log force completes.
3121 if (noblock && xfs_ipincount(ip)) {
3122 xfs_iunpin_nowait(ip);
3126 xfs_iunpin_wait(ip);
3129 * This may have been unpinned because the filesystem is shutting
3130 * down forcibly. If that's the case we must not write this inode
3131 * to disk, because the log record didn't make it to disk!
3133 if (XFS_FORCED_SHUTDOWN(mp)) {
3134 ip->i_update_core = 0;
3136 iip->ili_format.ilf_fields = 0;
3138 return XFS_ERROR(EIO);
3142 * Decide how buffer will be flushed out. This is done before
3143 * the call to xfs_iflush_int because this field is zeroed by it.
3145 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3147 * Flush out the inode buffer according to the directions
3148 * of the caller. In the cases where the caller has given
3149 * us a choice choose the non-delwri case. This is because
3150 * the inode is in the AIL and we need to get it out soon.
3153 case XFS_IFLUSH_SYNC:
3154 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3157 case XFS_IFLUSH_ASYNC_NOBLOCK:
3158 case XFS_IFLUSH_ASYNC:
3159 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3162 case XFS_IFLUSH_DELWRI:
3172 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3173 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3174 case XFS_IFLUSH_DELWRI:
3177 case XFS_IFLUSH_ASYNC_NOBLOCK:
3178 case XFS_IFLUSH_ASYNC:
3181 case XFS_IFLUSH_SYNC:
3192 * Get the buffer containing the on-disk inode.
3194 error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0, 0,
3195 noblock ? XFS_BUF_TRYLOCK : XFS_BUF_LOCK);
3202 * First flush out the inode that xfs_iflush was called with.
3204 error = xfs_iflush_int(ip, bp);
3209 * If the buffer is pinned then push on the log now so we won't
3210 * get stuck waiting in the write for too long.
3212 if (XFS_BUF_ISPINNED(bp))
3213 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
3217 * see if other inodes can be gathered into this write
3219 error = xfs_iflush_cluster(ip, bp);
3221 goto cluster_corrupt_out;
3223 if (flags & INT_DELWRI) {
3224 xfs_bdwrite(mp, bp);
3225 } else if (flags & INT_ASYNC) {
3226 error = xfs_bawrite(mp, bp);
3228 error = xfs_bwrite(mp, bp);
3234 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3235 cluster_corrupt_out:
3237 * Unlocks the flush lock
3239 xfs_iflush_abort(ip);
3240 return XFS_ERROR(EFSCORRUPTED);
3249 xfs_inode_log_item_t *iip;
3252 #ifdef XFS_TRANS_DEBUG
3256 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3257 ASSERT(!completion_done(&ip->i_flush));
3258 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3259 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3266 * If the inode isn't dirty, then just release the inode
3267 * flush lock and do nothing.
3269 if (xfs_inode_clean(ip)) {
3274 /* set *dip = inode's place in the buffer */
3275 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset);
3278 * Clear i_update_core before copying out the data.
3279 * This is for coordination with our timestamp updates
3280 * that don't hold the inode lock. They will always
3281 * update the timestamps BEFORE setting i_update_core,
3282 * so if we clear i_update_core after they set it we
3283 * are guaranteed to see their updates to the timestamps.
3284 * I believe that this depends on strongly ordered memory
3285 * semantics, but we have that. We use the SYNCHRONIZE
3286 * macro to make sure that the compiler does not reorder
3287 * the i_update_core access below the data copy below.
3289 ip->i_update_core = 0;
3293 * Make sure to get the latest atime from the Linux inode.
3295 xfs_synchronize_atime(ip);
3297 if (XFS_TEST_ERROR(be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC,
3298 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3299 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3300 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3301 ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3304 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3305 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3306 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3307 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3308 ip->i_ino, ip, ip->i_d.di_magic);
3311 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3313 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3314 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3315 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3316 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3317 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3321 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
3323 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3324 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3325 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3326 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3327 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3328 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3333 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3334 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3335 XFS_RANDOM_IFLUSH_5)) {
3336 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3337 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3339 ip->i_d.di_nextents + ip->i_d.di_anextents,
3344 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3345 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3346 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3347 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3348 ip->i_ino, ip->i_d.di_forkoff, ip);
3352 * bump the flush iteration count, used to detect flushes which
3353 * postdate a log record during recovery.
3356 ip->i_d.di_flushiter++;
3359 * Copy the dirty parts of the inode into the on-disk
3360 * inode. We always copy out the core of the inode,
3361 * because if the inode is dirty at all the core must
3364 xfs_dinode_to_disk(dip, &ip->i_d);
3366 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3367 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3368 ip->i_d.di_flushiter = 0;
3371 * If this is really an old format inode and the superblock version
3372 * has not been updated to support only new format inodes, then
3373 * convert back to the old inode format. If the superblock version
3374 * has been updated, then make the conversion permanent.
3376 ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
3377 if (ip->i_d.di_version == 1) {
3378 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
3382 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3383 dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink);
3386 * The superblock version has already been bumped,
3387 * so just make the conversion to the new inode
3390 ip->i_d.di_version = 2;
3391 dip->di_version = 2;
3392 ip->i_d.di_onlink = 0;
3394 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3395 memset(&(dip->di_pad[0]), 0,
3396 sizeof(dip->di_pad));
3397 ASSERT(ip->i_d.di_projid == 0);
3401 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
3402 if (XFS_IFORK_Q(ip))
3403 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3404 xfs_inobp_check(mp, bp);
3407 * We've recorded everything logged in the inode, so we'd
3408 * like to clear the ilf_fields bits so we don't log and
3409 * flush things unnecessarily. However, we can't stop
3410 * logging all this information until the data we've copied
3411 * into the disk buffer is written to disk. If we did we might
3412 * overwrite the copy of the inode in the log with all the
3413 * data after re-logging only part of it, and in the face of
3414 * a crash we wouldn't have all the data we need to recover.
3416 * What we do is move the bits to the ili_last_fields field.
3417 * When logging the inode, these bits are moved back to the
3418 * ilf_fields field. In the xfs_iflush_done() routine we
3419 * clear ili_last_fields, since we know that the information
3420 * those bits represent is permanently on disk. As long as
3421 * the flush completes before the inode is logged again, then
3422 * both ilf_fields and ili_last_fields will be cleared.
3424 * We can play with the ilf_fields bits here, because the inode
3425 * lock must be held exclusively in order to set bits there
3426 * and the flush lock protects the ili_last_fields bits.
3427 * Set ili_logged so the flush done
3428 * routine can tell whether or not to look in the AIL.
3429 * Also, store the current LSN of the inode so that we can tell
3430 * whether the item has moved in the AIL from xfs_iflush_done().
3431 * In order to read the lsn we need the AIL lock, because
3432 * it is a 64 bit value that cannot be read atomically.
3434 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3435 iip->ili_last_fields = iip->ili_format.ilf_fields;
3436 iip->ili_format.ilf_fields = 0;
3437 iip->ili_logged = 1;
3439 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3440 &iip->ili_item.li_lsn);
3443 * Attach the function xfs_iflush_done to the inode's
3444 * buffer. This will remove the inode from the AIL
3445 * and unlock the inode's flush lock when the inode is
3446 * completely written to disk.
3448 xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
3449 xfs_iflush_done, (xfs_log_item_t *)iip);
3451 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3452 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3455 * We're flushing an inode which is not in the AIL and has
3456 * not been logged but has i_update_core set. For this
3457 * case we can use a B_DELWRI flush and immediately drop
3458 * the inode flush lock because we can avoid the whole
3459 * AIL state thing. It's OK to drop the flush lock now,
3460 * because we've already locked the buffer and to do anything
3461 * you really need both.
3464 ASSERT(iip->ili_logged == 0);
3465 ASSERT(iip->ili_last_fields == 0);
3466 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3474 return XFS_ERROR(EFSCORRUPTED);
3479 #ifdef XFS_ILOCK_TRACE
3480 ktrace_t *xfs_ilock_trace_buf;
3483 xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
3485 ktrace_enter(ip->i_lock_trace,
3487 (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
3488 (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
3489 (void *)ra, /* caller of ilock */
3490 (void *)(unsigned long)current_cpu(),
3491 (void *)(unsigned long)current_pid(),
3492 NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL);
3497 * Return a pointer to the extent record at file index idx.
3499 xfs_bmbt_rec_host_t *
3501 xfs_ifork_t *ifp, /* inode fork pointer */
3502 xfs_extnum_t idx) /* index of target extent */
3505 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
3506 return ifp->if_u1.if_ext_irec->er_extbuf;
3507 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3508 xfs_ext_irec_t *erp; /* irec pointer */
3509 int erp_idx = 0; /* irec index */
3510 xfs_extnum_t page_idx = idx; /* ext index in target list */
3512 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3513 return &erp->er_extbuf[page_idx];
3514 } else if (ifp->if_bytes) {
3515 return &ifp->if_u1.if_extents[idx];
3522 * Insert new item(s) into the extent records for incore inode
3523 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3527 xfs_ifork_t *ifp, /* inode fork pointer */
3528 xfs_extnum_t idx, /* starting index of new items */
3529 xfs_extnum_t count, /* number of inserted items */
3530 xfs_bmbt_irec_t *new) /* items to insert */
3532 xfs_extnum_t i; /* extent record index */
3534 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3535 xfs_iext_add(ifp, idx, count);
3536 for (i = idx; i < idx + count; i++, new++)
3537 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new);
3541 * This is called when the amount of space required for incore file
3542 * extents needs to be increased. The ext_diff parameter stores the
3543 * number of new extents being added and the idx parameter contains
3544 * the extent index where the new extents will be added. If the new
3545 * extents are being appended, then we just need to (re)allocate and
3546 * initialize the space. Otherwise, if the new extents are being
3547 * inserted into the middle of the existing entries, a bit more work
3548 * is required to make room for the new extents to be inserted. The
3549 * caller is responsible for filling in the new extent entries upon
3554 xfs_ifork_t *ifp, /* inode fork pointer */
3555 xfs_extnum_t idx, /* index to begin adding exts */
3556 int ext_diff) /* number of extents to add */
3558 int byte_diff; /* new bytes being added */
3559 int new_size; /* size of extents after adding */
3560 xfs_extnum_t nextents; /* number of extents in file */
3562 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3563 ASSERT((idx >= 0) && (idx <= nextents));
3564 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
3565 new_size = ifp->if_bytes + byte_diff;
3567 * If the new number of extents (nextents + ext_diff)
3568 * fits inside the inode, then continue to use the inline
3571 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
3572 if (idx < nextents) {
3573 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
3574 &ifp->if_u2.if_inline_ext[idx],
3575 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3576 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
3578 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3579 ifp->if_real_bytes = 0;
3580 ifp->if_lastex = nextents + ext_diff;
3583 * Otherwise use a linear (direct) extent list.
3584 * If the extents are currently inside the inode,
3585 * xfs_iext_realloc_direct will switch us from
3586 * inline to direct extent allocation mode.
3588 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
3589 xfs_iext_realloc_direct(ifp, new_size);
3590 if (idx < nextents) {
3591 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
3592 &ifp->if_u1.if_extents[idx],
3593 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3594 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
3597 /* Indirection array */
3599 xfs_ext_irec_t *erp;
3603 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
3604 if (ifp->if_flags & XFS_IFEXTIREC) {
3605 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3607 xfs_iext_irec_init(ifp);
3608 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3609 erp = ifp->if_u1.if_ext_irec;
3611 /* Extents fit in target extent page */
3612 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3613 if (page_idx < erp->er_extcount) {
3614 memmove(&erp->er_extbuf[page_idx + ext_diff],
3615 &erp->er_extbuf[page_idx],
3616 (erp->er_extcount - page_idx) *
3617 sizeof(xfs_bmbt_rec_t));
3618 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3620 erp->er_extcount += ext_diff;
3621 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3623 /* Insert a new extent page */
3625 xfs_iext_add_indirect_multi(ifp,
3626 erp_idx, page_idx, ext_diff);
3629 * If extent(s) are being appended to the last page in
3630 * the indirection array and the new extent(s) don't fit
3631 * in the page, then erp is NULL and erp_idx is set to
3632 * the next index needed in the indirection array.
3635 int count = ext_diff;
3638 erp = xfs_iext_irec_new(ifp, erp_idx);
3639 erp->er_extcount = count;
3640 count -= MIN(count, (int)XFS_LINEAR_EXTS);
3647 ifp->if_bytes = new_size;
3651 * This is called when incore extents are being added to the indirection
3652 * array and the new extents do not fit in the target extent list. The
3653 * erp_idx parameter contains the irec index for the target extent list
3654 * in the indirection array, and the idx parameter contains the extent
3655 * index within the list. The number of extents being added is stored
3656 * in the count parameter.
3658 * |-------| |-------|
3659 * | | | | idx - number of extents before idx
3661 * | | | | count - number of extents being inserted at idx
3662 * |-------| |-------|
3663 * | count | | nex2 | nex2 - number of extents after idx + count
3664 * |-------| |-------|
3667 xfs_iext_add_indirect_multi(
3668 xfs_ifork_t *ifp, /* inode fork pointer */
3669 int erp_idx, /* target extent irec index */
3670 xfs_extnum_t idx, /* index within target list */
3671 int count) /* new extents being added */
3673 int byte_diff; /* new bytes being added */
3674 xfs_ext_irec_t *erp; /* pointer to irec entry */
3675 xfs_extnum_t ext_diff; /* number of extents to add */
3676 xfs_extnum_t ext_cnt; /* new extents still needed */
3677 xfs_extnum_t nex2; /* extents after idx + count */
3678 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3679 int nlists; /* number of irec's (lists) */
3681 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3682 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3683 nex2 = erp->er_extcount - idx;
3684 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3687 * Save second part of target extent list
3688 * (all extents past */
3690 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3691 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS);
3692 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3693 erp->er_extcount -= nex2;
3694 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3695 memset(&erp->er_extbuf[idx], 0, byte_diff);
3699 * Add the new extents to the end of the target
3700 * list, then allocate new irec record(s) and
3701 * extent buffer(s) as needed to store the rest
3702 * of the new extents.
3705 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3707 erp->er_extcount += ext_diff;
3708 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3709 ext_cnt -= ext_diff;
3713 erp = xfs_iext_irec_new(ifp, erp_idx);
3714 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3715 erp->er_extcount = ext_diff;
3716 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3717 ext_cnt -= ext_diff;
3720 /* Add nex2 extents back to indirection array */
3722 xfs_extnum_t ext_avail;
3725 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3726 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3729 * If nex2 extents fit in the current page, append
3730 * nex2_ep after the new extents.
3732 if (nex2 <= ext_avail) {
3733 i = erp->er_extcount;
3736 * Otherwise, check if space is available in the
3739 else if ((erp_idx < nlists - 1) &&
3740 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3741 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3744 /* Create a hole for nex2 extents */
3745 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3746 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3749 * Final choice, create a new extent page for
3754 erp = xfs_iext_irec_new(ifp, erp_idx);
3756 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
3758 erp->er_extcount += nex2;
3759 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3764 * This is called when the amount of space required for incore file
3765 * extents needs to be decreased. The ext_diff parameter stores the
3766 * number of extents to be removed and the idx parameter contains
3767 * the extent index where the extents will be removed from.
3769 * If the amount of space needed has decreased below the linear
3770 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3771 * extent array. Otherwise, use kmem_realloc() to adjust the
3772 * size to what is needed.
3776 xfs_ifork_t *ifp, /* inode fork pointer */
3777 xfs_extnum_t idx, /* index to begin removing exts */
3778 int ext_diff) /* number of extents to remove */
3780 xfs_extnum_t nextents; /* number of extents in file */
3781 int new_size; /* size of extents after removal */
3783 ASSERT(ext_diff > 0);
3784 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3785 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3787 if (new_size == 0) {
3788 xfs_iext_destroy(ifp);
3789 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3790 xfs_iext_remove_indirect(ifp, idx, ext_diff);
3791 } else if (ifp->if_real_bytes) {
3792 xfs_iext_remove_direct(ifp, idx, ext_diff);
3794 xfs_iext_remove_inline(ifp, idx, ext_diff);
3796 ifp->if_bytes = new_size;
3800 * This removes ext_diff extents from the inline buffer, beginning
3801 * at extent index idx.
3804 xfs_iext_remove_inline(
3805 xfs_ifork_t *ifp, /* inode fork pointer */
3806 xfs_extnum_t idx, /* index to begin removing exts */
3807 int ext_diff) /* number of extents to remove */
3809 int nextents; /* number of extents in file */
3811 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3812 ASSERT(idx < XFS_INLINE_EXTS);
3813 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3814 ASSERT(((nextents - ext_diff) > 0) &&
3815 (nextents - ext_diff) < XFS_INLINE_EXTS);
3817 if (idx + ext_diff < nextents) {
3818 memmove(&ifp->if_u2.if_inline_ext[idx],
3819 &ifp->if_u2.if_inline_ext[idx + ext_diff],
3820 (nextents - (idx + ext_diff)) *
3821 sizeof(xfs_bmbt_rec_t));
3822 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
3823 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3825 memset(&ifp->if_u2.if_inline_ext[idx], 0,
3826 ext_diff * sizeof(xfs_bmbt_rec_t));
3831 * This removes ext_diff extents from a linear (direct) extent list,
3832 * beginning at extent index idx. If the extents are being removed
3833 * from the end of the list (ie. truncate) then we just need to re-
3834 * allocate the list to remove the extra space. Otherwise, if the
3835 * extents are being removed from the middle of the existing extent
3836 * entries, then we first need to move the extent records beginning
3837 * at idx + ext_diff up in the list to overwrite the records being
3838 * removed, then remove the extra space via kmem_realloc.
3841 xfs_iext_remove_direct(
3842 xfs_ifork_t *ifp, /* inode fork pointer */
3843 xfs_extnum_t idx, /* index to begin removing exts */
3844 int ext_diff) /* number of extents to remove */
3846 xfs_extnum_t nextents; /* number of extents in file */
3847 int new_size; /* size of extents after removal */
3849 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3850 new_size = ifp->if_bytes -
3851 (ext_diff * sizeof(xfs_bmbt_rec_t));
3852 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3854 if (new_size == 0) {
3855 xfs_iext_destroy(ifp);
3858 /* Move extents up in the list (if needed) */
3859 if (idx + ext_diff < nextents) {
3860 memmove(&ifp->if_u1.if_extents[idx],
3861 &ifp->if_u1.if_extents[idx + ext_diff],
3862 (nextents - (idx + ext_diff)) *
3863 sizeof(xfs_bmbt_rec_t));
3865 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
3866 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3868 * Reallocate the direct extent list. If the extents
3869 * will fit inside the inode then xfs_iext_realloc_direct
3870 * will switch from direct to inline extent allocation
3873 xfs_iext_realloc_direct(ifp, new_size);
3874 ifp->if_bytes = new_size;
3878 * This is called when incore extents are being removed from the
3879 * indirection array and the extents being removed span multiple extent
3880 * buffers. The idx parameter contains the file extent index where we
3881 * want to begin removing extents, and the count parameter contains
3882 * how many extents need to be removed.
3884 * |-------| |-------|
3885 * | nex1 | | | nex1 - number of extents before idx
3886 * |-------| | count |
3887 * | | | | count - number of extents being removed at idx
3888 * | count | |-------|
3889 * | | | nex2 | nex2 - number of extents after idx + count
3890 * |-------| |-------|
3893 xfs_iext_remove_indirect(
3894 xfs_ifork_t *ifp, /* inode fork pointer */
3895 xfs_extnum_t idx, /* index to begin removing extents */
3896 int count) /* number of extents to remove */
3898 xfs_ext_irec_t *erp; /* indirection array pointer */
3899 int erp_idx = 0; /* indirection array index */
3900 xfs_extnum_t ext_cnt; /* extents left to remove */
3901 xfs_extnum_t ext_diff; /* extents to remove in current list */
3902 xfs_extnum_t nex1; /* number of extents before idx */
3903 xfs_extnum_t nex2; /* extents after idx + count */
3904 int nlists; /* entries in indirection array */
3905 int page_idx = idx; /* index in target extent list */
3907 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3908 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3909 ASSERT(erp != NULL);
3910 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3914 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
3915 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
3917 * Check for deletion of entire list;
3918 * xfs_iext_irec_remove() updates extent offsets.
3920 if (ext_diff == erp->er_extcount) {
3921 xfs_iext_irec_remove(ifp, erp_idx);
3922 ext_cnt -= ext_diff;
3925 ASSERT(erp_idx < ifp->if_real_bytes /
3927 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3934 /* Move extents up (if needed) */
3936 memmove(&erp->er_extbuf[nex1],
3937 &erp->er_extbuf[nex1 + ext_diff],
3938 nex2 * sizeof(xfs_bmbt_rec_t));
3940 /* Zero out rest of page */
3941 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
3942 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
3943 /* Update remaining counters */
3944 erp->er_extcount -= ext_diff;
3945 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
3946 ext_cnt -= ext_diff;
3951 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
3952 xfs_iext_irec_compact(ifp);
3956 * Create, destroy, or resize a linear (direct) block of extents.
3959 xfs_iext_realloc_direct(
3960 xfs_ifork_t *ifp, /* inode fork pointer */
3961 int new_size) /* new size of extents */
3963 int rnew_size; /* real new size of extents */
3965 rnew_size = new_size;
3967 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
3968 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
3969 (new_size != ifp->if_real_bytes)));
3971 /* Free extent records */
3972 if (new_size == 0) {
3973 xfs_iext_destroy(ifp);
3975 /* Resize direct extent list and zero any new bytes */
3976 else if (ifp->if_real_bytes) {
3977 /* Check if extents will fit inside the inode */
3978 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
3979 xfs_iext_direct_to_inline(ifp, new_size /
3980 (uint)sizeof(xfs_bmbt_rec_t));
3981 ifp->if_bytes = new_size;
3984 if (!is_power_of_2(new_size)){
3985 rnew_size = roundup_pow_of_two(new_size);
3987 if (rnew_size != ifp->if_real_bytes) {
3988 ifp->if_u1.if_extents =
3989 kmem_realloc(ifp->if_u1.if_extents,
3991 ifp->if_real_bytes, KM_NOFS);
3993 if (rnew_size > ifp->if_real_bytes) {
3994 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
3995 (uint)sizeof(xfs_bmbt_rec_t)], 0,
3996 rnew_size - ifp->if_real_bytes);
4000 * Switch from the inline extent buffer to a direct
4001 * extent list. Be sure to include the inline extent
4002 * bytes in new_size.
4005 new_size += ifp->if_bytes;
4006 if (!is_power_of_2(new_size)) {
4007 rnew_size = roundup_pow_of_two(new_size);
4009 xfs_iext_inline_to_direct(ifp, rnew_size);
4011 ifp->if_real_bytes = rnew_size;
4012 ifp->if_bytes = new_size;
4016 * Switch from linear (direct) extent records to inline buffer.
4019 xfs_iext_direct_to_inline(
4020 xfs_ifork_t *ifp, /* inode fork pointer */
4021 xfs_extnum_t nextents) /* number of extents in file */
4023 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4024 ASSERT(nextents <= XFS_INLINE_EXTS);
4026 * The inline buffer was zeroed when we switched
4027 * from inline to direct extent allocation mode,
4028 * so we don't need to clear it here.
4030 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
4031 nextents * sizeof(xfs_bmbt_rec_t));
4032 kmem_free(ifp->if_u1.if_extents);
4033 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
4034 ifp->if_real_bytes = 0;
4038 * Switch from inline buffer to linear (direct) extent records.
4039 * new_size should already be rounded up to the next power of 2
4040 * by the caller (when appropriate), so use new_size as it is.
4041 * However, since new_size may be rounded up, we can't update
4042 * if_bytes here. It is the caller's responsibility to update
4043 * if_bytes upon return.
4046 xfs_iext_inline_to_direct(
4047 xfs_ifork_t *ifp, /* inode fork pointer */
4048 int new_size) /* number of extents in file */
4050 ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS);
4051 memset(ifp->if_u1.if_extents, 0, new_size);
4052 if (ifp->if_bytes) {
4053 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
4055 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4056 sizeof(xfs_bmbt_rec_t));
4058 ifp->if_real_bytes = new_size;
4062 * Resize an extent indirection array to new_size bytes.
4065 xfs_iext_realloc_indirect(
4066 xfs_ifork_t *ifp, /* inode fork pointer */
4067 int new_size) /* new indirection array size */
4069 int nlists; /* number of irec's (ex lists) */
4070 int size; /* current indirection array size */
4072 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4073 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4074 size = nlists * sizeof(xfs_ext_irec_t);
4075 ASSERT(ifp->if_real_bytes);
4076 ASSERT((new_size >= 0) && (new_size != size));
4077 if (new_size == 0) {
4078 xfs_iext_destroy(ifp);
4080 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
4081 kmem_realloc(ifp->if_u1.if_ext_irec,
4082 new_size, size, KM_NOFS);
4087 * Switch from indirection array to linear (direct) extent allocations.
4090 xfs_iext_indirect_to_direct(
4091 xfs_ifork_t *ifp) /* inode fork pointer */
4093 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
4094 xfs_extnum_t nextents; /* number of extents in file */
4095 int size; /* size of file extents */
4097 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4098 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4099 ASSERT(nextents <= XFS_LINEAR_EXTS);
4100 size = nextents * sizeof(xfs_bmbt_rec_t);
4102 xfs_iext_irec_compact_pages(ifp);
4103 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
4105 ep = ifp->if_u1.if_ext_irec->er_extbuf;
4106 kmem_free(ifp->if_u1.if_ext_irec);
4107 ifp->if_flags &= ~XFS_IFEXTIREC;
4108 ifp->if_u1.if_extents = ep;
4109 ifp->if_bytes = size;
4110 if (nextents < XFS_LINEAR_EXTS) {
4111 xfs_iext_realloc_direct(ifp, size);
4116 * Free incore file extents.
4120 xfs_ifork_t *ifp) /* inode fork pointer */
4122 if (ifp->if_flags & XFS_IFEXTIREC) {
4126 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4127 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
4128 xfs_iext_irec_remove(ifp, erp_idx);
4130 ifp->if_flags &= ~XFS_IFEXTIREC;
4131 } else if (ifp->if_real_bytes) {
4132 kmem_free(ifp->if_u1.if_extents);
4133 } else if (ifp->if_bytes) {
4134 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4135 sizeof(xfs_bmbt_rec_t));
4137 ifp->if_u1.if_extents = NULL;
4138 ifp->if_real_bytes = 0;
4143 * Return a pointer to the extent record for file system block bno.
4145 xfs_bmbt_rec_host_t * /* pointer to found extent record */
4146 xfs_iext_bno_to_ext(
4147 xfs_ifork_t *ifp, /* inode fork pointer */
4148 xfs_fileoff_t bno, /* block number to search for */
4149 xfs_extnum_t *idxp) /* index of target extent */
4151 xfs_bmbt_rec_host_t *base; /* pointer to first extent */
4152 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
4153 xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */
4154 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4155 int high; /* upper boundary in search */
4156 xfs_extnum_t idx = 0; /* index of target extent */
4157 int low; /* lower boundary in search */
4158 xfs_extnum_t nextents; /* number of file extents */
4159 xfs_fileoff_t startoff = 0; /* start offset of extent */
4161 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4162 if (nextents == 0) {
4167 if (ifp->if_flags & XFS_IFEXTIREC) {
4168 /* Find target extent list */
4170 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
4171 base = erp->er_extbuf;
4172 high = erp->er_extcount - 1;
4174 base = ifp->if_u1.if_extents;
4175 high = nextents - 1;
4177 /* Binary search extent records */
4178 while (low <= high) {
4179 idx = (low + high) >> 1;
4181 startoff = xfs_bmbt_get_startoff(ep);
4182 blockcount = xfs_bmbt_get_blockcount(ep);
4183 if (bno < startoff) {
4185 } else if (bno >= startoff + blockcount) {
4188 /* Convert back to file-based extent index */
4189 if (ifp->if_flags & XFS_IFEXTIREC) {
4190 idx += erp->er_extoff;
4196 /* Convert back to file-based extent index */
4197 if (ifp->if_flags & XFS_IFEXTIREC) {
4198 idx += erp->er_extoff;
4200 if (bno >= startoff + blockcount) {
4201 if (++idx == nextents) {
4204 ep = xfs_iext_get_ext(ifp, idx);
4212 * Return a pointer to the indirection array entry containing the
4213 * extent record for filesystem block bno. Store the index of the
4214 * target irec in *erp_idxp.
4216 xfs_ext_irec_t * /* pointer to found extent record */
4217 xfs_iext_bno_to_irec(
4218 xfs_ifork_t *ifp, /* inode fork pointer */
4219 xfs_fileoff_t bno, /* block number to search for */
4220 int *erp_idxp) /* irec index of target ext list */
4222 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4223 xfs_ext_irec_t *erp_next; /* next indirection array entry */
4224 int erp_idx; /* indirection array index */
4225 int nlists; /* number of extent irec's (lists) */
4226 int high; /* binary search upper limit */
4227 int low; /* binary search lower limit */
4229 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4230 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4234 while (low <= high) {
4235 erp_idx = (low + high) >> 1;
4236 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4237 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
4238 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
4240 } else if (erp_next && bno >=
4241 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
4247 *erp_idxp = erp_idx;
4252 * Return a pointer to the indirection array entry containing the
4253 * extent record at file extent index *idxp. Store the index of the
4254 * target irec in *erp_idxp and store the page index of the target
4255 * extent record in *idxp.
4258 xfs_iext_idx_to_irec(
4259 xfs_ifork_t *ifp, /* inode fork pointer */
4260 xfs_extnum_t *idxp, /* extent index (file -> page) */
4261 int *erp_idxp, /* pointer to target irec */
4262 int realloc) /* new bytes were just added */
4264 xfs_ext_irec_t *prev; /* pointer to previous irec */
4265 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
4266 int erp_idx; /* indirection array index */
4267 int nlists; /* number of irec's (ex lists) */
4268 int high; /* binary search upper limit */
4269 int low; /* binary search lower limit */
4270 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
4272 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4273 ASSERT(page_idx >= 0 && page_idx <=
4274 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
4275 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4280 /* Binary search extent irec's */
4281 while (low <= high) {
4282 erp_idx = (low + high) >> 1;
4283 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4284 prev = erp_idx > 0 ? erp - 1 : NULL;
4285 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
4286 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
4288 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
4289 (page_idx == erp->er_extoff + erp->er_extcount &&
4292 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
4293 erp->er_extcount == XFS_LINEAR_EXTS) {
4297 erp = erp_idx < nlists ? erp + 1 : NULL;
4300 page_idx -= erp->er_extoff;
4305 *erp_idxp = erp_idx;
4310 * Allocate and initialize an indirection array once the space needed
4311 * for incore extents increases above XFS_IEXT_BUFSZ.
4315 xfs_ifork_t *ifp) /* inode fork pointer */
4317 xfs_ext_irec_t *erp; /* indirection array pointer */
4318 xfs_extnum_t nextents; /* number of extents in file */
4320 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4321 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4322 ASSERT(nextents <= XFS_LINEAR_EXTS);
4324 erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS);
4326 if (nextents == 0) {
4327 ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
4328 } else if (!ifp->if_real_bytes) {
4329 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
4330 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
4331 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
4333 erp->er_extbuf = ifp->if_u1.if_extents;
4334 erp->er_extcount = nextents;
4337 ifp->if_flags |= XFS_IFEXTIREC;
4338 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
4339 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
4340 ifp->if_u1.if_ext_irec = erp;
4346 * Allocate and initialize a new entry in the indirection array.
4350 xfs_ifork_t *ifp, /* inode fork pointer */
4351 int erp_idx) /* index for new irec */
4353 xfs_ext_irec_t *erp; /* indirection array pointer */
4354 int i; /* loop counter */
4355 int nlists; /* number of irec's (ex lists) */
4357 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4358 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4360 /* Resize indirection array */
4361 xfs_iext_realloc_indirect(ifp, ++nlists *
4362 sizeof(xfs_ext_irec_t));
4364 * Move records down in the array so the
4365 * new page can use erp_idx.
4367 erp = ifp->if_u1.if_ext_irec;
4368 for (i = nlists - 1; i > erp_idx; i--) {
4369 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
4371 ASSERT(i == erp_idx);
4373 /* Initialize new extent record */
4374 erp = ifp->if_u1.if_ext_irec;
4375 erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
4376 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4377 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
4378 erp[erp_idx].er_extcount = 0;
4379 erp[erp_idx].er_extoff = erp_idx > 0 ?
4380 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
4381 return (&erp[erp_idx]);
4385 * Remove a record from the indirection array.
4388 xfs_iext_irec_remove(
4389 xfs_ifork_t *ifp, /* inode fork pointer */
4390 int erp_idx) /* irec index to remove */
4392 xfs_ext_irec_t *erp; /* indirection array pointer */
4393 int i; /* loop counter */
4394 int nlists; /* number of irec's (ex lists) */
4396 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4397 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4398 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4399 if (erp->er_extbuf) {
4400 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
4402 kmem_free(erp->er_extbuf);
4404 /* Compact extent records */
4405 erp = ifp->if_u1.if_ext_irec;
4406 for (i = erp_idx; i < nlists - 1; i++) {
4407 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
4410 * Manually free the last extent record from the indirection
4411 * array. A call to xfs_iext_realloc_indirect() with a size
4412 * of zero would result in a call to xfs_iext_destroy() which
4413 * would in turn call this function again, creating a nasty
4417 xfs_iext_realloc_indirect(ifp,
4418 nlists * sizeof(xfs_ext_irec_t));
4420 kmem_free(ifp->if_u1.if_ext_irec);
4422 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4426 * This is called to clean up large amounts of unused memory allocated
4427 * by the indirection array. Before compacting anything though, verify
4428 * that the indirection array is still needed and switch back to the
4429 * linear extent list (or even the inline buffer) if possible. The
4430 * compaction policy is as follows:
4432 * Full Compaction: Extents fit into a single page (or inline buffer)
4433 * Partial Compaction: Extents occupy less than 50% of allocated space
4434 * No Compaction: Extents occupy at least 50% of allocated space
4437 xfs_iext_irec_compact(
4438 xfs_ifork_t *ifp) /* inode fork pointer */
4440 xfs_extnum_t nextents; /* number of extents in file */
4441 int nlists; /* number of irec's (ex lists) */
4443 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4444 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4445 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4447 if (nextents == 0) {
4448 xfs_iext_destroy(ifp);
4449 } else if (nextents <= XFS_INLINE_EXTS) {
4450 xfs_iext_indirect_to_direct(ifp);
4451 xfs_iext_direct_to_inline(ifp, nextents);
4452 } else if (nextents <= XFS_LINEAR_EXTS) {
4453 xfs_iext_indirect_to_direct(ifp);
4454 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
4455 xfs_iext_irec_compact_pages(ifp);
4460 * Combine extents from neighboring extent pages.
4463 xfs_iext_irec_compact_pages(
4464 xfs_ifork_t *ifp) /* inode fork pointer */
4466 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
4467 int erp_idx = 0; /* indirection array index */
4468 int nlists; /* number of irec's (ex lists) */
4470 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4471 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4472 while (erp_idx < nlists - 1) {
4473 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4475 if (erp_next->er_extcount <=
4476 (XFS_LINEAR_EXTS - erp->er_extcount)) {
4477 memcpy(&erp->er_extbuf[erp->er_extcount],
4478 erp_next->er_extbuf, erp_next->er_extcount *
4479 sizeof(xfs_bmbt_rec_t));
4480 erp->er_extcount += erp_next->er_extcount;
4482 * Free page before removing extent record
4483 * so er_extoffs don't get modified in
4484 * xfs_iext_irec_remove.
4486 kmem_free(erp_next->er_extbuf);
4487 erp_next->er_extbuf = NULL;
4488 xfs_iext_irec_remove(ifp, erp_idx + 1);
4489 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4497 * This is called to update the er_extoff field in the indirection
4498 * array when extents have been added or removed from one of the
4499 * extent lists. erp_idx contains the irec index to begin updating
4500 * at and ext_diff contains the number of extents that were added
4504 xfs_iext_irec_update_extoffs(
4505 xfs_ifork_t *ifp, /* inode fork pointer */
4506 int erp_idx, /* irec index to update */
4507 int ext_diff) /* number of new extents */
4509 int i; /* loop counter */
4510 int nlists; /* number of irec's (ex lists */
4512 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4513 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4514 for (i = erp_idx; i < nlists; i++) {
4515 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;