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
20 #include "xfs_types.h"
25 #include "xfs_trans.h"
26 #include "xfs_trans_priv.h"
30 #include "xfs_dmapi.h"
31 #include "xfs_mount.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_ialloc_btree.h"
35 #include "xfs_dir2_sf.h"
36 #include "xfs_attr_sf.h"
37 #include "xfs_dinode.h"
38 #include "xfs_inode.h"
39 #include "xfs_buf_item.h"
40 #include "xfs_inode_item.h"
41 #include "xfs_btree.h"
42 #include "xfs_alloc.h"
43 #include "xfs_ialloc.h"
46 #include "xfs_error.h"
47 #include "xfs_utils.h"
48 #include "xfs_dir2_trace.h"
49 #include "xfs_quota.h"
51 #include "xfs_filestream.h"
53 kmem_zone_t *xfs_ifork_zone;
54 kmem_zone_t *xfs_inode_zone;
55 kmem_zone_t *xfs_icluster_zone;
58 * Used in xfs_itruncate(). This is the maximum number of extents
59 * freed from a file in a single transaction.
61 #define XFS_ITRUNC_MAX_EXTENTS 2
63 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
64 STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
65 STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
66 STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
70 * Make sure that the extents in the given memory buffer
80 xfs_bmbt_rec_host_t rec;
83 for (i = 0; i < nrecs; i++) {
84 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
85 rec.l0 = get_unaligned(&ep->l0);
86 rec.l1 = get_unaligned(&ep->l1);
87 xfs_bmbt_get_all(&rec, &irec);
88 if (fmt == XFS_EXTFMT_NOSTATE)
89 ASSERT(irec.br_state == XFS_EXT_NORM);
93 #define xfs_validate_extents(ifp, nrecs, fmt)
97 * Check that none of the inode's in the buffer have a next
98 * unlinked field of 0.
110 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
112 for (i = 0; i < j; i++) {
113 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
114 i * mp->m_sb.sb_inodesize);
115 if (!dip->di_next_unlinked) {
116 xfs_fs_cmn_err(CE_ALERT, mp,
117 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
119 ASSERT(dip->di_next_unlinked);
126 * This routine is called to map an inode number within a file
127 * system to the buffer containing the on-disk version of the
128 * inode. It returns a pointer to the buffer containing the
129 * on-disk inode in the bpp parameter, and in the dip parameter
130 * it returns a pointer to the on-disk inode within that buffer.
132 * If a non-zero error is returned, then the contents of bpp and
133 * dipp are undefined.
135 * Use xfs_imap() to determine the size and location of the
136 * buffer to read from disk.
154 * Call the space management code to find the location of the
158 error = xfs_imap(mp, tp, ino, &imap, XFS_IMAP_LOOKUP);
161 "xfs_inotobp: xfs_imap() returned an "
162 "error %d on %s. Returning error.", error, mp->m_fsname);
167 * If the inode number maps to a block outside the bounds of the
168 * file system then return NULL rather than calling read_buf
169 * and panicing when we get an error from the driver.
171 if ((imap.im_blkno + imap.im_len) >
172 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
174 "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
175 "of the file system %s. Returning EINVAL.",
176 (unsigned long long)imap.im_blkno,
177 imap.im_len, mp->m_fsname);
178 return XFS_ERROR(EINVAL);
182 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
183 * default to just a read_buf() call.
185 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
186 (int)imap.im_len, XFS_BUF_LOCK, &bp);
190 "xfs_inotobp: xfs_trans_read_buf() returned an "
191 "error %d on %s. Returning error.", error, mp->m_fsname);
194 dip = (xfs_dinode_t *)xfs_buf_offset(bp, 0);
196 be16_to_cpu(dip->di_core.di_magic) == XFS_DINODE_MAGIC &&
197 XFS_DINODE_GOOD_VERSION(dip->di_core.di_version);
198 if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP,
199 XFS_RANDOM_ITOBP_INOTOBP))) {
200 XFS_CORRUPTION_ERROR("xfs_inotobp", XFS_ERRLEVEL_LOW, mp, dip);
201 xfs_trans_brelse(tp, bp);
203 "xfs_inotobp: XFS_TEST_ERROR() returned an "
204 "error on %s. Returning EFSCORRUPTED.", mp->m_fsname);
205 return XFS_ERROR(EFSCORRUPTED);
208 xfs_inobp_check(mp, bp);
211 * Set *dipp to point to the on-disk inode in the buffer.
213 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
215 *offset = imap.im_boffset;
221 * This routine is called to map an inode to the buffer containing
222 * the on-disk version of the inode. It returns a pointer to the
223 * buffer containing the on-disk inode in the bpp parameter, and in
224 * the dip parameter it returns a pointer to the on-disk inode within
227 * If a non-zero error is returned, then the contents of bpp and
228 * dipp are undefined.
230 * If the inode is new and has not yet been initialized, use xfs_imap()
231 * to determine the size and location of the buffer to read from disk.
232 * If the inode has already been mapped to its buffer and read in once,
233 * then use the mapping information stored in the inode rather than
234 * calling xfs_imap(). This allows us to avoid the overhead of looking
235 * at the inode btree for small block file systems (see xfs_dilocate()).
236 * We can tell whether the inode has been mapped in before by comparing
237 * its disk block address to 0. Only uninitialized inodes will have
238 * 0 for the disk block address.
256 if (ip->i_blkno == (xfs_daddr_t)0) {
258 * Call the space management code to find the location of the
262 if ((error = xfs_imap(mp, tp, ip->i_ino, &imap,
263 XFS_IMAP_LOOKUP | imap_flags)))
267 * If the inode number maps to a block outside the bounds
268 * of the file system then return NULL rather than calling
269 * read_buf and panicing when we get an error from the
272 if ((imap.im_blkno + imap.im_len) >
273 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
275 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
276 "(imap.im_blkno (0x%llx) "
277 "+ imap.im_len (0x%llx)) > "
278 " XFS_FSB_TO_BB(mp, "
279 "mp->m_sb.sb_dblocks) (0x%llx)",
280 (unsigned long long) imap.im_blkno,
281 (unsigned long long) imap.im_len,
282 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
284 return XFS_ERROR(EINVAL);
288 * Fill in the fields in the inode that will be used to
289 * map the inode to its buffer from now on.
291 ip->i_blkno = imap.im_blkno;
292 ip->i_len = imap.im_len;
293 ip->i_boffset = imap.im_boffset;
296 * We've already mapped the inode once, so just use the
297 * mapping that we saved the first time.
299 imap.im_blkno = ip->i_blkno;
300 imap.im_len = ip->i_len;
301 imap.im_boffset = ip->i_boffset;
303 ASSERT(bno == 0 || bno == imap.im_blkno);
306 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
307 * default to just a read_buf() call.
309 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
310 (int)imap.im_len, XFS_BUF_LOCK, &bp);
313 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
314 "xfs_trans_read_buf() returned error %d, "
315 "imap.im_blkno 0x%llx, imap.im_len 0x%llx",
316 error, (unsigned long long) imap.im_blkno,
317 (unsigned long long) imap.im_len);
323 * Validate the magic number and version of every inode in the buffer
324 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
325 * No validation is done here in userspace (xfs_repair).
327 #if !defined(__KERNEL__)
330 ni = BBTOB(imap.im_len) >> mp->m_sb.sb_inodelog;
331 #else /* usual case */
335 for (i = 0; i < ni; i++) {
339 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
340 (i << mp->m_sb.sb_inodelog));
341 di_ok = be16_to_cpu(dip->di_core.di_magic) == XFS_DINODE_MAGIC &&
342 XFS_DINODE_GOOD_VERSION(dip->di_core.di_version);
343 if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
344 XFS_ERRTAG_ITOBP_INOTOBP,
345 XFS_RANDOM_ITOBP_INOTOBP))) {
346 if (imap_flags & XFS_IMAP_BULKSTAT) {
347 xfs_trans_brelse(tp, bp);
348 return XFS_ERROR(EINVAL);
352 "Device %s - bad inode magic/vsn "
353 "daddr %lld #%d (magic=%x)",
354 XFS_BUFTARG_NAME(mp->m_ddev_targp),
355 (unsigned long long)imap.im_blkno, i,
356 be16_to_cpu(dip->di_core.di_magic));
358 XFS_CORRUPTION_ERROR("xfs_itobp", XFS_ERRLEVEL_HIGH,
360 xfs_trans_brelse(tp, bp);
361 return XFS_ERROR(EFSCORRUPTED);
365 xfs_inobp_check(mp, bp);
368 * Mark the buffer as an inode buffer now that it looks good
370 XFS_BUF_SET_VTYPE(bp, B_FS_INO);
373 * Set *dipp to point to the on-disk inode in the buffer.
375 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
381 * Move inode type and inode format specific information from the
382 * on-disk inode to the in-core inode. For fifos, devs, and sockets
383 * this means set if_rdev to the proper value. For files, directories,
384 * and symlinks this means to bring in the in-line data or extent
385 * pointers. For a file in B-tree format, only the root is immediately
386 * brought in-core. The rest will be in-lined in if_extents when it
387 * is first referenced (see xfs_iread_extents()).
394 xfs_attr_shortform_t *atp;
398 ip->i_df.if_ext_max =
399 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
402 if (unlikely(be32_to_cpu(dip->di_core.di_nextents) +
403 be16_to_cpu(dip->di_core.di_anextents) >
404 be64_to_cpu(dip->di_core.di_nblocks))) {
405 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
406 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
407 (unsigned long long)ip->i_ino,
408 (int)(be32_to_cpu(dip->di_core.di_nextents) +
409 be16_to_cpu(dip->di_core.di_anextents)),
411 be64_to_cpu(dip->di_core.di_nblocks));
412 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
414 return XFS_ERROR(EFSCORRUPTED);
417 if (unlikely(dip->di_core.di_forkoff > ip->i_mount->m_sb.sb_inodesize)) {
418 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
419 "corrupt dinode %Lu, forkoff = 0x%x.",
420 (unsigned long long)ip->i_ino,
421 dip->di_core.di_forkoff);
422 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
424 return XFS_ERROR(EFSCORRUPTED);
427 switch (ip->i_d.di_mode & S_IFMT) {
432 if (unlikely(dip->di_core.di_format != XFS_DINODE_FMT_DEV)) {
433 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
435 return XFS_ERROR(EFSCORRUPTED);
439 ip->i_df.if_u2.if_rdev = be32_to_cpu(dip->di_u.di_dev);
445 switch (dip->di_core.di_format) {
446 case XFS_DINODE_FMT_LOCAL:
448 * no local regular files yet
450 if (unlikely((be16_to_cpu(dip->di_core.di_mode) & S_IFMT) == S_IFREG)) {
451 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
453 "(local format for regular file).",
454 (unsigned long long) ip->i_ino);
455 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
458 return XFS_ERROR(EFSCORRUPTED);
461 di_size = be64_to_cpu(dip->di_core.di_size);
462 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
463 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
465 "(bad size %Ld for local inode).",
466 (unsigned long long) ip->i_ino,
467 (long long) di_size);
468 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
471 return XFS_ERROR(EFSCORRUPTED);
475 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
477 case XFS_DINODE_FMT_EXTENTS:
478 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
480 case XFS_DINODE_FMT_BTREE:
481 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
484 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
486 return XFS_ERROR(EFSCORRUPTED);
491 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
492 return XFS_ERROR(EFSCORRUPTED);
497 if (!XFS_DFORK_Q(dip))
499 ASSERT(ip->i_afp == NULL);
500 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
501 ip->i_afp->if_ext_max =
502 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
503 switch (dip->di_core.di_aformat) {
504 case XFS_DINODE_FMT_LOCAL:
505 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
506 size = be16_to_cpu(atp->hdr.totsize);
507 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
509 case XFS_DINODE_FMT_EXTENTS:
510 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
512 case XFS_DINODE_FMT_BTREE:
513 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
516 error = XFS_ERROR(EFSCORRUPTED);
520 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
522 xfs_idestroy_fork(ip, XFS_DATA_FORK);
528 * The file is in-lined in the on-disk inode.
529 * If it fits into if_inline_data, then copy
530 * it there, otherwise allocate a buffer for it
531 * and copy the data there. Either way, set
532 * if_data to point at the data.
533 * If we allocate a buffer for the data, make
534 * sure that its size is a multiple of 4 and
535 * record the real size in i_real_bytes.
548 * If the size is unreasonable, then something
549 * is wrong and we just bail out rather than crash in
550 * kmem_alloc() or memcpy() below.
552 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
553 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
555 "(bad size %d for local fork, size = %d).",
556 (unsigned long long) ip->i_ino, size,
557 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
558 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
560 return XFS_ERROR(EFSCORRUPTED);
562 ifp = XFS_IFORK_PTR(ip, whichfork);
565 ifp->if_u1.if_data = NULL;
566 else if (size <= sizeof(ifp->if_u2.if_inline_data))
567 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
569 real_size = roundup(size, 4);
570 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
572 ifp->if_bytes = size;
573 ifp->if_real_bytes = real_size;
575 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
576 ifp->if_flags &= ~XFS_IFEXTENTS;
577 ifp->if_flags |= XFS_IFINLINE;
582 * The file consists of a set of extents all
583 * of which fit into the on-disk inode.
584 * If there are few enough extents to fit into
585 * the if_inline_ext, then copy them there.
586 * Otherwise allocate a buffer for them and copy
587 * them into it. Either way, set if_extents
588 * to point at the extents.
602 ifp = XFS_IFORK_PTR(ip, whichfork);
603 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
604 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
607 * If the number of extents is unreasonable, then something
608 * is wrong and we just bail out rather than crash in
609 * kmem_alloc() or memcpy() below.
611 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
612 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
613 "corrupt inode %Lu ((a)extents = %d).",
614 (unsigned long long) ip->i_ino, nex);
615 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
617 return XFS_ERROR(EFSCORRUPTED);
620 ifp->if_real_bytes = 0;
622 ifp->if_u1.if_extents = NULL;
623 else if (nex <= XFS_INLINE_EXTS)
624 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
626 xfs_iext_add(ifp, 0, nex);
628 ifp->if_bytes = size;
630 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
631 xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip));
632 for (i = 0; i < nex; i++, dp++) {
633 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
634 ep->l0 = be64_to_cpu(get_unaligned(&dp->l0));
635 ep->l1 = be64_to_cpu(get_unaligned(&dp->l1));
637 XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork);
638 if (whichfork != XFS_DATA_FORK ||
639 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
640 if (unlikely(xfs_check_nostate_extents(
642 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
645 return XFS_ERROR(EFSCORRUPTED);
648 ifp->if_flags |= XFS_IFEXTENTS;
653 * The file has too many extents to fit into
654 * the inode, so they are in B-tree format.
655 * Allocate a buffer for the root of the B-tree
656 * and copy the root into it. The i_extents
657 * field will remain NULL until all of the
658 * extents are read in (when they are needed).
666 xfs_bmdr_block_t *dfp;
672 ifp = XFS_IFORK_PTR(ip, whichfork);
673 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
674 size = XFS_BMAP_BROOT_SPACE(dfp);
675 nrecs = XFS_BMAP_BROOT_NUMRECS(dfp);
678 * blow out if -- fork has less extents than can fit in
679 * fork (fork shouldn't be a btree format), root btree
680 * block has more records than can fit into the fork,
681 * or the number of extents is greater than the number of
684 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
685 || XFS_BMDR_SPACE_CALC(nrecs) >
686 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
687 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
688 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
689 "corrupt inode %Lu (btree).",
690 (unsigned long long) ip->i_ino);
691 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
693 return XFS_ERROR(EFSCORRUPTED);
696 ifp->if_broot_bytes = size;
697 ifp->if_broot = kmem_alloc(size, KM_SLEEP);
698 ASSERT(ifp->if_broot != NULL);
700 * Copy and convert from the on-disk structure
701 * to the in-memory structure.
703 xfs_bmdr_to_bmbt(dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
704 ifp->if_broot, size);
705 ifp->if_flags &= ~XFS_IFEXTENTS;
706 ifp->if_flags |= XFS_IFBROOT;
712 xfs_dinode_from_disk(
714 xfs_dinode_core_t *from)
716 to->di_magic = be16_to_cpu(from->di_magic);
717 to->di_mode = be16_to_cpu(from->di_mode);
718 to->di_version = from ->di_version;
719 to->di_format = from->di_format;
720 to->di_onlink = be16_to_cpu(from->di_onlink);
721 to->di_uid = be32_to_cpu(from->di_uid);
722 to->di_gid = be32_to_cpu(from->di_gid);
723 to->di_nlink = be32_to_cpu(from->di_nlink);
724 to->di_projid = be16_to_cpu(from->di_projid);
725 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
726 to->di_flushiter = be16_to_cpu(from->di_flushiter);
727 to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec);
728 to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec);
729 to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec);
730 to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec);
731 to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec);
732 to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec);
733 to->di_size = be64_to_cpu(from->di_size);
734 to->di_nblocks = be64_to_cpu(from->di_nblocks);
735 to->di_extsize = be32_to_cpu(from->di_extsize);
736 to->di_nextents = be32_to_cpu(from->di_nextents);
737 to->di_anextents = be16_to_cpu(from->di_anextents);
738 to->di_forkoff = from->di_forkoff;
739 to->di_aformat = from->di_aformat;
740 to->di_dmevmask = be32_to_cpu(from->di_dmevmask);
741 to->di_dmstate = be16_to_cpu(from->di_dmstate);
742 to->di_flags = be16_to_cpu(from->di_flags);
743 to->di_gen = be32_to_cpu(from->di_gen);
748 xfs_dinode_core_t *to,
749 xfs_icdinode_t *from)
751 to->di_magic = cpu_to_be16(from->di_magic);
752 to->di_mode = cpu_to_be16(from->di_mode);
753 to->di_version = from ->di_version;
754 to->di_format = from->di_format;
755 to->di_onlink = cpu_to_be16(from->di_onlink);
756 to->di_uid = cpu_to_be32(from->di_uid);
757 to->di_gid = cpu_to_be32(from->di_gid);
758 to->di_nlink = cpu_to_be32(from->di_nlink);
759 to->di_projid = cpu_to_be16(from->di_projid);
760 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
761 to->di_flushiter = cpu_to_be16(from->di_flushiter);
762 to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec);
763 to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec);
764 to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec);
765 to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec);
766 to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec);
767 to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec);
768 to->di_size = cpu_to_be64(from->di_size);
769 to->di_nblocks = cpu_to_be64(from->di_nblocks);
770 to->di_extsize = cpu_to_be32(from->di_extsize);
771 to->di_nextents = cpu_to_be32(from->di_nextents);
772 to->di_anextents = cpu_to_be16(from->di_anextents);
773 to->di_forkoff = from->di_forkoff;
774 to->di_aformat = from->di_aformat;
775 to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
776 to->di_dmstate = cpu_to_be16(from->di_dmstate);
777 to->di_flags = cpu_to_be16(from->di_flags);
778 to->di_gen = cpu_to_be32(from->di_gen);
787 if (di_flags & XFS_DIFLAG_ANY) {
788 if (di_flags & XFS_DIFLAG_REALTIME)
789 flags |= XFS_XFLAG_REALTIME;
790 if (di_flags & XFS_DIFLAG_PREALLOC)
791 flags |= XFS_XFLAG_PREALLOC;
792 if (di_flags & XFS_DIFLAG_IMMUTABLE)
793 flags |= XFS_XFLAG_IMMUTABLE;
794 if (di_flags & XFS_DIFLAG_APPEND)
795 flags |= XFS_XFLAG_APPEND;
796 if (di_flags & XFS_DIFLAG_SYNC)
797 flags |= XFS_XFLAG_SYNC;
798 if (di_flags & XFS_DIFLAG_NOATIME)
799 flags |= XFS_XFLAG_NOATIME;
800 if (di_flags & XFS_DIFLAG_NODUMP)
801 flags |= XFS_XFLAG_NODUMP;
802 if (di_flags & XFS_DIFLAG_RTINHERIT)
803 flags |= XFS_XFLAG_RTINHERIT;
804 if (di_flags & XFS_DIFLAG_PROJINHERIT)
805 flags |= XFS_XFLAG_PROJINHERIT;
806 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
807 flags |= XFS_XFLAG_NOSYMLINKS;
808 if (di_flags & XFS_DIFLAG_EXTSIZE)
809 flags |= XFS_XFLAG_EXTSIZE;
810 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
811 flags |= XFS_XFLAG_EXTSZINHERIT;
812 if (di_flags & XFS_DIFLAG_NODEFRAG)
813 flags |= XFS_XFLAG_NODEFRAG;
814 if (di_flags & XFS_DIFLAG_FILESTREAM)
815 flags |= XFS_XFLAG_FILESTREAM;
825 xfs_icdinode_t *dic = &ip->i_d;
827 return _xfs_dic2xflags(dic->di_flags) |
828 (XFS_CFORK_Q(dic) ? XFS_XFLAG_HASATTR : 0);
833 xfs_dinode_core_t *dic)
835 return _xfs_dic2xflags(be16_to_cpu(dic->di_flags)) |
836 (XFS_CFORK_Q_DISK(dic) ? XFS_XFLAG_HASATTR : 0);
840 * Given a mount structure and an inode number, return a pointer
841 * to a newly allocated in-core inode corresponding to the given
844 * Initialize the inode's attributes and extent pointers if it
845 * already has them (it will not if the inode has no links).
861 ASSERT(xfs_inode_zone != NULL);
863 ip = kmem_zone_zalloc(xfs_inode_zone, KM_SLEEP);
866 spin_lock_init(&ip->i_flags_lock);
869 * Get pointer's to the on-disk inode and the buffer containing it.
870 * If the inode number refers to a block outside the file system
871 * then xfs_itobp() will return NULL. In this case we should
872 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
873 * know that this is a new incore inode.
875 error = xfs_itobp(mp, tp, ip, &dip, &bp, bno, imap_flags);
877 kmem_zone_free(xfs_inode_zone, ip);
882 * Initialize inode's trace buffers.
883 * Do this before xfs_iformat in case it adds entries.
885 #ifdef XFS_BMAP_TRACE
886 ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_SLEEP);
888 #ifdef XFS_BMBT_TRACE
889 ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_SLEEP);
892 ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_SLEEP);
894 #ifdef XFS_ILOCK_TRACE
895 ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_SLEEP);
897 #ifdef XFS_DIR2_TRACE
898 ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_SLEEP);
902 * If we got something that isn't an inode it means someone
903 * (nfs or dmi) has a stale handle.
905 if (be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC) {
906 kmem_zone_free(xfs_inode_zone, ip);
907 xfs_trans_brelse(tp, bp);
909 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
910 "dip->di_core.di_magic (0x%x) != "
911 "XFS_DINODE_MAGIC (0x%x)",
912 be16_to_cpu(dip->di_core.di_magic),
915 return 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.
925 if (dip->di_core.di_mode) {
926 xfs_dinode_from_disk(&ip->i_d, &dip->di_core);
927 error = xfs_iformat(ip, dip);
929 kmem_zone_free(xfs_inode_zone, ip);
930 xfs_trans_brelse(tp, bp);
932 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
933 "xfs_iformat() returned error %d",
939 ip->i_d.di_magic = be16_to_cpu(dip->di_core.di_magic);
940 ip->i_d.di_version = dip->di_core.di_version;
941 ip->i_d.di_gen = be32_to_cpu(dip->di_core.di_gen);
942 ip->i_d.di_flushiter = be16_to_cpu(dip->di_core.di_flushiter);
944 * Make sure to pull in the mode here as well in
945 * case the inode is released without being used.
946 * This ensures that xfs_inactive() will see that
947 * the inode is already free and not try to mess
948 * with the uninitialized part of it.
952 * Initialize the per-fork minima and maxima for a new
953 * inode here. xfs_iformat will do it for old inodes.
955 ip->i_df.if_ext_max =
956 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
959 INIT_LIST_HEAD(&ip->i_reclaim);
962 * The inode format changed when we moved the link count and
963 * made it 32 bits long. If this is an old format inode,
964 * convert it in memory to look like a new one. If it gets
965 * flushed to disk we will convert back before flushing or
966 * logging it. We zero out the new projid field and the old link
967 * count field. We'll handle clearing the pad field (the remains
968 * of the old uuid field) when we actually convert the inode to
969 * the new format. We don't change the version number so that we
970 * can distinguish this from a real new format inode.
972 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
973 ip->i_d.di_nlink = ip->i_d.di_onlink;
974 ip->i_d.di_onlink = 0;
975 ip->i_d.di_projid = 0;
978 ip->i_delayed_blks = 0;
979 ip->i_size = ip->i_d.di_size;
982 * Mark the buffer containing the inode as something to keep
983 * around for a while. This helps to keep recently accessed
984 * meta-data in-core longer.
986 XFS_BUF_SET_REF(bp, XFS_INO_REF);
989 * Use xfs_trans_brelse() to release the buffer containing the
990 * on-disk inode, because it was acquired with xfs_trans_read_buf()
991 * in xfs_itobp() above. If tp is NULL, this is just a normal
992 * brelse(). If we're within a transaction, then xfs_trans_brelse()
993 * will only release the buffer if it is not dirty within the
994 * transaction. It will be OK to release the buffer in this case,
995 * because inodes on disk are never destroyed and we will be
996 * locking the new in-core inode before putting it in the hash
997 * table where other processes can find it. Thus we don't have
998 * to worry about the inode being changed just because we released
1001 xfs_trans_brelse(tp, bp);
1007 * Read in extents from a btree-format inode.
1008 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1018 xfs_extnum_t nextents;
1021 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
1022 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
1024 return XFS_ERROR(EFSCORRUPTED);
1026 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
1027 size = nextents * sizeof(xfs_bmbt_rec_t);
1028 ifp = XFS_IFORK_PTR(ip, whichfork);
1031 * We know that the size is valid (it's checked in iformat_btree)
1033 ifp->if_lastex = NULLEXTNUM;
1034 ifp->if_bytes = ifp->if_real_bytes = 0;
1035 ifp->if_flags |= XFS_IFEXTENTS;
1036 xfs_iext_add(ifp, 0, nextents);
1037 error = xfs_bmap_read_extents(tp, ip, whichfork);
1039 xfs_iext_destroy(ifp);
1040 ifp->if_flags &= ~XFS_IFEXTENTS;
1043 xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip));
1048 * Allocate an inode on disk and return a copy of its in-core version.
1049 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1050 * appropriately within the inode. The uid and gid for the inode are
1051 * set according to the contents of the given cred structure.
1053 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1054 * has a free inode available, call xfs_iget()
1055 * to obtain the in-core version of the allocated inode. Finally,
1056 * fill in the inode and log its initial contents. In this case,
1057 * ialloc_context would be set to NULL and call_again set to false.
1059 * If xfs_dialloc() does not have an available inode,
1060 * it will replenish its supply by doing an allocation. Since we can
1061 * only do one allocation within a transaction without deadlocks, we
1062 * must commit the current transaction before returning the inode itself.
1063 * In this case, therefore, we will set call_again to true and return.
1064 * The caller should then commit the current transaction, start a new
1065 * transaction, and call xfs_ialloc() again to actually get the inode.
1067 * To ensure that some other process does not grab the inode that
1068 * was allocated during the first call to xfs_ialloc(), this routine
1069 * also returns the [locked] bp pointing to the head of the freelist
1070 * as ialloc_context. The caller should hold this buffer across
1071 * the commit and pass it back into this routine on the second call.
1073 * If we are allocating quota inodes, we do not have a parent inode
1074 * to attach to or associate with (i.e. pip == NULL) because they
1075 * are not linked into the directory structure - they are attached
1076 * directly to the superblock - and so have no parent.
1088 xfs_buf_t **ialloc_context,
1089 boolean_t *call_again,
1099 * Call the space management code to pick
1100 * the on-disk inode to be allocated.
1102 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
1103 ialloc_context, call_again, &ino);
1107 if (*call_again || ino == NULLFSINO) {
1111 ASSERT(*ialloc_context == NULL);
1114 * Get the in-core inode with the lock held exclusively.
1115 * This is because we're setting fields here we need
1116 * to prevent others from looking at until we're done.
1118 error = xfs_trans_iget(tp->t_mountp, tp, ino,
1119 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(cr);
1131 ip->i_d.di_gid = current_fsgid(cr);
1132 ip->i_d.di_projid = prid;
1133 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1136 * If the superblock version is up to where we support new format
1137 * inodes and this is currently an old format inode, then change
1138 * the inode version number now. This way we only do the conversion
1139 * here rather than here and in the flush/logging code.
1141 if (XFS_SB_VERSION_HASNLINK(&tp->t_mountp->m_sb) &&
1142 ip->i_d.di_version == XFS_DINODE_VERSION_1) {
1143 ip->i_d.di_version = XFS_DINODE_VERSION_2;
1145 * We've already zeroed the old link count, the projid field,
1146 * and the pad field.
1151 * Project ids won't be stored on disk if we are using a version 1 inode.
1153 if ((prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1))
1154 xfs_bump_ino_vers2(tp, ip);
1156 if (pip && XFS_INHERIT_GID(pip, vp->v_vfsp)) {
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);
1178 xfs_ichgtime(ip, XFS_ICHGTIME_CHG|XFS_ICHGTIME_ACC|XFS_ICHGTIME_MOD);
1180 * di_gen will have been taken care of in xfs_iread.
1182 ip->i_d.di_extsize = 0;
1183 ip->i_d.di_dmevmask = 0;
1184 ip->i_d.di_dmstate = 0;
1185 ip->i_d.di_flags = 0;
1186 flags = XFS_ILOG_CORE;
1187 switch (mode & S_IFMT) {
1192 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1193 ip->i_df.if_u2.if_rdev = rdev;
1194 ip->i_df.if_flags = 0;
1195 flags |= XFS_ILOG_DEV;
1198 if (pip && xfs_inode_is_filestream(pip)) {
1199 error = xfs_filestream_associate(pip, ip);
1203 xfs_iflags_set(ip, XFS_IFILESTREAM);
1207 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
1210 if ((mode & S_IFMT) == S_IFDIR) {
1211 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1212 di_flags |= XFS_DIFLAG_RTINHERIT;
1213 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1214 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1215 ip->i_d.di_extsize = pip->i_d.di_extsize;
1217 } else if ((mode & S_IFMT) == S_IFREG) {
1218 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) {
1219 di_flags |= XFS_DIFLAG_REALTIME;
1220 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
1222 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1223 di_flags |= XFS_DIFLAG_EXTSIZE;
1224 ip->i_d.di_extsize = pip->i_d.di_extsize;
1227 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1228 xfs_inherit_noatime)
1229 di_flags |= XFS_DIFLAG_NOATIME;
1230 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1232 di_flags |= XFS_DIFLAG_NODUMP;
1233 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1235 di_flags |= XFS_DIFLAG_SYNC;
1236 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1237 xfs_inherit_nosymlinks)
1238 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1239 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1240 di_flags |= XFS_DIFLAG_PROJINHERIT;
1241 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1242 xfs_inherit_nodefrag)
1243 di_flags |= XFS_DIFLAG_NODEFRAG;
1244 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
1245 di_flags |= XFS_DIFLAG_FILESTREAM;
1246 ip->i_d.di_flags |= di_flags;
1250 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1251 ip->i_df.if_flags = XFS_IFEXTENTS;
1252 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1253 ip->i_df.if_u1.if_extents = NULL;
1259 * Attribute fork settings for new inode.
1261 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1262 ip->i_d.di_anextents = 0;
1265 * Log the new values stuffed into the inode.
1267 xfs_trans_log_inode(tp, ip, flags);
1269 /* now that we have an i_mode we can setup inode ops and unlock */
1270 bhv_vfs_init_vnode(XFS_MTOVFS(tp->t_mountp), vp, XFS_ITOBHV(ip), 1);
1277 * Check to make sure that there are no blocks allocated to the
1278 * file beyond the size of the file. We don't check this for
1279 * files with fixed size extents or real time extents, but we
1280 * at least do it for regular files.
1289 xfs_fileoff_t map_first;
1291 xfs_bmbt_irec_t imaps[2];
1293 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1296 if (ip->i_d.di_flags & (XFS_DIFLAG_REALTIME | XFS_DIFLAG_EXTSIZE))
1300 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1302 * The filesystem could be shutting down, so bmapi may return
1305 if (xfs_bmapi(NULL, ip, map_first,
1307 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1309 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1312 ASSERT(nimaps == 1);
1313 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1318 * Calculate the last possible buffered byte in a file. This must
1319 * include data that was buffered beyond the EOF by the write code.
1320 * This also needs to deal with overflowing the xfs_fsize_t type
1321 * which can happen for sizes near the limit.
1323 * We also need to take into account any blocks beyond the EOF. It
1324 * may be the case that they were buffered by a write which failed.
1325 * In that case the pages will still be in memory, but the inode size
1326 * will never have been updated.
1333 xfs_fsize_t last_byte;
1334 xfs_fileoff_t last_block;
1335 xfs_fileoff_t size_last_block;
1338 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE | MR_ACCESS));
1342 * Only check for blocks beyond the EOF if the extents have
1343 * been read in. This eliminates the need for the inode lock,
1344 * and it also saves us from looking when it really isn't
1347 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1348 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1356 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_size);
1357 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1359 last_byte = XFS_FSB_TO_B(mp, last_block);
1360 if (last_byte < 0) {
1361 return XFS_MAXIOFFSET(mp);
1363 last_byte += (1 << mp->m_writeio_log);
1364 if (last_byte < 0) {
1365 return XFS_MAXIOFFSET(mp);
1370 #if defined(XFS_RW_TRACE)
1376 xfs_fsize_t new_size,
1377 xfs_off_t toss_start,
1378 xfs_off_t toss_finish)
1380 if (ip->i_rwtrace == NULL) {
1384 ktrace_enter(ip->i_rwtrace,
1387 (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
1388 (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
1389 (void*)((long)flag),
1390 (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
1391 (void*)(unsigned long)(new_size & 0xffffffff),
1392 (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
1393 (void*)(unsigned long)(toss_start & 0xffffffff),
1394 (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
1395 (void*)(unsigned long)(toss_finish & 0xffffffff),
1396 (void*)(unsigned long)current_cpu(),
1397 (void*)(unsigned long)current_pid(),
1403 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1407 * Start the truncation of the file to new_size. The new size
1408 * must be smaller than the current size. This routine will
1409 * clear the buffer and page caches of file data in the removed
1410 * range, and xfs_itruncate_finish() will remove the underlying
1413 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1414 * must NOT have the inode lock held at all. This is because we're
1415 * calling into the buffer/page cache code and we can't hold the
1416 * inode lock when we do so.
1418 * We need to wait for any direct I/Os in flight to complete before we
1419 * proceed with the truncate. This is needed to prevent the extents
1420 * being read or written by the direct I/Os from being removed while the
1421 * I/O is in flight as there is no other method of synchronising
1422 * direct I/O with the truncate operation. Also, because we hold
1423 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1424 * started until the truncate completes and drops the lock. Essentially,
1425 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1426 * between direct I/Os and the truncate operation.
1428 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1429 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1430 * in the case that the caller is locking things out of order and
1431 * may not be able to call xfs_itruncate_finish() with the inode lock
1432 * held without dropping the I/O lock. If the caller must drop the
1433 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1434 * must be called again with all the same restrictions as the initial
1438 xfs_itruncate_start(
1441 xfs_fsize_t new_size)
1443 xfs_fsize_t last_byte;
1444 xfs_off_t toss_start;
1449 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1450 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1451 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1452 (flags == XFS_ITRUNC_MAYBE));
1457 vn_iowait(vp); /* wait for the completion of any pending DIOs */
1460 * Call toss_pages or flushinval_pages to get rid of pages
1461 * overlapping the region being removed. We have to use
1462 * the less efficient flushinval_pages in the case that the
1463 * caller may not be able to finish the truncate without
1464 * dropping the inode's I/O lock. Make sure
1465 * to catch any pages brought in by buffers overlapping
1466 * the EOF by searching out beyond the isize by our
1467 * block size. We round new_size up to a block boundary
1468 * so that we don't toss things on the same block as
1469 * new_size but before it.
1471 * Before calling toss_page or flushinval_pages, make sure to
1472 * call remapf() over the same region if the file is mapped.
1473 * This frees up mapped file references to the pages in the
1474 * given range and for the flushinval_pages case it ensures
1475 * that we get the latest mapped changes flushed out.
1477 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1478 toss_start = XFS_FSB_TO_B(mp, toss_start);
1479 if (toss_start < 0) {
1481 * The place to start tossing is beyond our maximum
1482 * file size, so there is no way that the data extended
1487 last_byte = xfs_file_last_byte(ip);
1488 xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
1490 if (last_byte > toss_start) {
1491 if (flags & XFS_ITRUNC_DEFINITE) {
1492 bhv_vop_toss_pages(vp, toss_start, -1, FI_REMAPF_LOCKED);
1494 error = bhv_vop_flushinval_pages(vp, toss_start, -1, FI_REMAPF_LOCKED);
1499 if (new_size == 0) {
1500 ASSERT(VN_CACHED(vp) == 0);
1507 * Shrink the file to the given new_size. The new
1508 * size must be smaller than the current size.
1509 * This will free up the underlying blocks
1510 * in the removed range after a call to xfs_itruncate_start()
1511 * or xfs_atruncate_start().
1513 * The transaction passed to this routine must have made
1514 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1515 * This routine may commit the given transaction and
1516 * start new ones, so make sure everything involved in
1517 * the transaction is tidy before calling here.
1518 * Some transaction will be returned to the caller to be
1519 * committed. The incoming transaction must already include
1520 * the inode, and both inode locks must be held exclusively.
1521 * The inode must also be "held" within the transaction. On
1522 * return the inode will be "held" within the returned transaction.
1523 * This routine does NOT require any disk space to be reserved
1524 * for it within the transaction.
1526 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1527 * and it indicates the fork which is to be truncated. For the
1528 * attribute fork we only support truncation to size 0.
1530 * We use the sync parameter to indicate whether or not the first
1531 * transaction we perform might have to be synchronous. For the attr fork,
1532 * it needs to be so if the unlink of the inode is not yet known to be
1533 * permanent in the log. This keeps us from freeing and reusing the
1534 * blocks of the attribute fork before the unlink of the inode becomes
1537 * For the data fork, we normally have to run synchronously if we're
1538 * being called out of the inactive path or we're being called
1539 * out of the create path where we're truncating an existing file.
1540 * Either way, the truncate needs to be sync so blocks don't reappear
1541 * in the file with altered data in case of a crash. wsync filesystems
1542 * can run the first case async because anything that shrinks the inode
1543 * has to run sync so by the time we're called here from inactive, the
1544 * inode size is permanently set to 0.
1546 * Calls from the truncate path always need to be sync unless we're
1547 * in a wsync filesystem and the file has already been unlinked.
1549 * The caller is responsible for correctly setting the sync parameter.
1550 * It gets too hard for us to guess here which path we're being called
1551 * out of just based on inode state.
1554 xfs_itruncate_finish(
1557 xfs_fsize_t new_size,
1561 xfs_fsblock_t first_block;
1562 xfs_fileoff_t first_unmap_block;
1563 xfs_fileoff_t last_block;
1564 xfs_filblks_t unmap_len=0;
1569 xfs_bmap_free_t free_list;
1572 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1573 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
1574 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1575 ASSERT(*tp != NULL);
1576 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1577 ASSERT(ip->i_transp == *tp);
1578 ASSERT(ip->i_itemp != NULL);
1579 ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
1583 mp = (ntp)->t_mountp;
1584 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1587 * We only support truncating the entire attribute fork.
1589 if (fork == XFS_ATTR_FORK) {
1592 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1593 xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
1595 * The first thing we do is set the size to new_size permanently
1596 * on disk. This way we don't have to worry about anyone ever
1597 * being able to look at the data being freed even in the face
1598 * of a crash. What we're getting around here is the case where
1599 * we free a block, it is allocated to another file, it is written
1600 * to, and then we crash. If the new data gets written to the
1601 * file but the log buffers containing the free and reallocation
1602 * don't, then we'd end up with garbage in the blocks being freed.
1603 * As long as we make the new_size permanent before actually
1604 * freeing any blocks it doesn't matter if they get writtten to.
1606 * The callers must signal into us whether or not the size
1607 * setting here must be synchronous. There are a few cases
1608 * where it doesn't have to be synchronous. Those cases
1609 * occur if the file is unlinked and we know the unlink is
1610 * permanent or if the blocks being truncated are guaranteed
1611 * to be beyond the inode eof (regardless of the link count)
1612 * and the eof value is permanent. Both of these cases occur
1613 * only on wsync-mounted filesystems. In those cases, we're
1614 * guaranteed that no user will ever see the data in the blocks
1615 * that are being truncated so the truncate can run async.
1616 * In the free beyond eof case, the file may wind up with
1617 * more blocks allocated to it than it needs if we crash
1618 * and that won't get fixed until the next time the file
1619 * is re-opened and closed but that's ok as that shouldn't
1620 * be too many blocks.
1622 * However, we can't just make all wsync xactions run async
1623 * because there's one call out of the create path that needs
1624 * to run sync where it's truncating an existing file to size
1625 * 0 whose size is > 0.
1627 * It's probably possible to come up with a test in this
1628 * routine that would correctly distinguish all the above
1629 * cases from the values of the function parameters and the
1630 * inode state but for sanity's sake, I've decided to let the
1631 * layers above just tell us. It's simpler to correctly figure
1632 * out in the layer above exactly under what conditions we
1633 * can run async and I think it's easier for others read and
1634 * follow the logic in case something has to be changed.
1635 * cscope is your friend -- rcc.
1637 * The attribute fork is much simpler.
1639 * For the attribute fork we allow the caller to tell us whether
1640 * the unlink of the inode that led to this call is yet permanent
1641 * in the on disk log. If it is not and we will be freeing extents
1642 * in this inode then we make the first transaction synchronous
1643 * to make sure that the unlink is permanent by the time we free
1646 if (fork == XFS_DATA_FORK) {
1647 if (ip->i_d.di_nextents > 0) {
1649 * If we are not changing the file size then do
1650 * not update the on-disk file size - we may be
1651 * called from xfs_inactive_free_eofblocks(). If we
1652 * update the on-disk file size and then the system
1653 * crashes before the contents of the file are
1654 * flushed to disk then the files may be full of
1655 * holes (ie NULL files bug).
1657 if (ip->i_size != new_size) {
1658 ip->i_d.di_size = new_size;
1659 ip->i_size = new_size;
1660 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1664 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1665 if (ip->i_d.di_anextents > 0)
1666 xfs_trans_set_sync(ntp);
1668 ASSERT(fork == XFS_DATA_FORK ||
1669 (fork == XFS_ATTR_FORK &&
1670 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1671 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1674 * Since it is possible for space to become allocated beyond
1675 * the end of the file (in a crash where the space is allocated
1676 * but the inode size is not yet updated), simply remove any
1677 * blocks which show up between the new EOF and the maximum
1678 * possible file size. If the first block to be removed is
1679 * beyond the maximum file size (ie it is the same as last_block),
1680 * then there is nothing to do.
1682 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1683 ASSERT(first_unmap_block <= last_block);
1685 if (last_block == first_unmap_block) {
1688 unmap_len = last_block - first_unmap_block + 1;
1692 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1693 * will tell us whether it freed the entire range or
1694 * not. If this is a synchronous mount (wsync),
1695 * then we can tell bunmapi to keep all the
1696 * transactions asynchronous since the unlink
1697 * transaction that made this inode inactive has
1698 * already hit the disk. There's no danger of
1699 * the freed blocks being reused, there being a
1700 * crash, and the reused blocks suddenly reappearing
1701 * in this file with garbage in them once recovery
1704 XFS_BMAP_INIT(&free_list, &first_block);
1705 error = XFS_BUNMAPI(mp, ntp, &ip->i_iocore,
1706 first_unmap_block, unmap_len,
1707 XFS_BMAPI_AFLAG(fork) |
1708 (sync ? 0 : XFS_BMAPI_ASYNC),
1709 XFS_ITRUNC_MAX_EXTENTS,
1710 &first_block, &free_list,
1714 * If the bunmapi call encounters an error,
1715 * return to the caller where the transaction
1716 * can be properly aborted. We just need to
1717 * make sure we're not holding any resources
1718 * that we were not when we came in.
1720 xfs_bmap_cancel(&free_list);
1725 * Duplicate the transaction that has the permanent
1726 * reservation and commit the old transaction.
1728 error = xfs_bmap_finish(tp, &free_list, &committed);
1732 * If the bmap finish call encounters an error,
1733 * return to the caller where the transaction
1734 * can be properly aborted. We just need to
1735 * make sure we're not holding any resources
1736 * that we were not when we came in.
1738 * Aborting from this point might lose some
1739 * blocks in the file system, but oh well.
1741 xfs_bmap_cancel(&free_list);
1744 * If the passed in transaction committed
1745 * in xfs_bmap_finish(), then we want to
1746 * add the inode to this one before returning.
1747 * This keeps things simple for the higher
1748 * level code, because it always knows that
1749 * the inode is locked and held in the
1750 * transaction that returns to it whether
1751 * errors occur or not. We don't mark the
1752 * inode dirty so that this transaction can
1753 * be easily aborted if possible.
1755 xfs_trans_ijoin(ntp, ip,
1756 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1757 xfs_trans_ihold(ntp, ip);
1764 * The first xact was committed,
1765 * so add the inode to the new one.
1766 * Mark it dirty so it will be logged
1767 * and moved forward in the log as
1768 * part of every commit.
1770 xfs_trans_ijoin(ntp, ip,
1771 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1772 xfs_trans_ihold(ntp, ip);
1773 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1775 ntp = xfs_trans_dup(ntp);
1776 (void) xfs_trans_commit(*tp, 0);
1778 error = xfs_trans_reserve(ntp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1779 XFS_TRANS_PERM_LOG_RES,
1780 XFS_ITRUNCATE_LOG_COUNT);
1782 * Add the inode being truncated to the next chained
1785 xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1786 xfs_trans_ihold(ntp, ip);
1791 * Only update the size in the case of the data fork, but
1792 * always re-log the inode so that our permanent transaction
1793 * can keep on rolling it forward in the log.
1795 if (fork == XFS_DATA_FORK) {
1796 xfs_isize_check(mp, ip, new_size);
1798 * If we are not changing the file size then do
1799 * not update the on-disk file size - we may be
1800 * called from xfs_inactive_free_eofblocks(). If we
1801 * update the on-disk file size and then the system
1802 * crashes before the contents of the file are
1803 * flushed to disk then the files may be full of
1804 * holes (ie NULL files bug).
1806 if (ip->i_size != new_size) {
1807 ip->i_d.di_size = new_size;
1808 ip->i_size = new_size;
1811 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1812 ASSERT((new_size != 0) ||
1813 (fork == XFS_ATTR_FORK) ||
1814 (ip->i_delayed_blks == 0));
1815 ASSERT((new_size != 0) ||
1816 (fork == XFS_ATTR_FORK) ||
1817 (ip->i_d.di_nextents == 0));
1818 xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
1826 * Do the first part of growing a file: zero any data in the last
1827 * block that is beyond the old EOF. We need to do this before
1828 * the inode is joined to the transaction to modify the i_size.
1829 * That way we can drop the inode lock and call into the buffer
1830 * cache to get the buffer mapping the EOF.
1835 xfs_fsize_t new_size,
1840 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1841 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1842 ASSERT(new_size > ip->i_size);
1845 * Zero any pages that may have been created by
1846 * xfs_write_file() beyond the end of the file
1847 * and any blocks between the old and new file sizes.
1849 error = xfs_zero_eof(XFS_ITOV(ip), &ip->i_iocore, new_size,
1857 * This routine is called to extend the size of a file.
1858 * The inode must have both the iolock and the ilock locked
1859 * for update and it must be a part of the current transaction.
1860 * The xfs_igrow_start() function must have been called previously.
1861 * If the change_flag is not zero, the inode change timestamp will
1868 xfs_fsize_t new_size,
1871 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1872 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1873 ASSERT(ip->i_transp == tp);
1874 ASSERT(new_size > ip->i_size);
1877 * Update the file size. Update the inode change timestamp
1878 * if change_flag set.
1880 ip->i_d.di_size = new_size;
1881 ip->i_size = new_size;
1883 xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
1884 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1890 * This is called when the inode's link count goes to 0.
1891 * We place the on-disk inode on a list in the AGI. It
1892 * will be pulled from this list when the inode is freed.
1904 xfs_agnumber_t agno;
1905 xfs_daddr_t agdaddr;
1912 ASSERT(ip->i_d.di_nlink == 0);
1913 ASSERT(ip->i_d.di_mode != 0);
1914 ASSERT(ip->i_transp == tp);
1918 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1919 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1922 * Get the agi buffer first. It ensures lock ordering
1925 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1926 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1931 * Validate the magic number of the agi block.
1933 agi = XFS_BUF_TO_AGI(agibp);
1935 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1936 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
1937 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK,
1938 XFS_RANDOM_IUNLINK))) {
1939 XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi);
1940 xfs_trans_brelse(tp, agibp);
1941 return XFS_ERROR(EFSCORRUPTED);
1944 * Get the index into the agi hash table for the
1945 * list this inode will go on.
1947 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1949 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1950 ASSERT(agi->agi_unlinked[bucket_index]);
1951 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1953 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
1955 * There is already another inode in the bucket we need
1956 * to add ourselves to. Add us at the front of the list.
1957 * Here we put the head pointer into our next pointer,
1958 * and then we fall through to point the head at us.
1960 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
1964 ASSERT(be32_to_cpu(dip->di_next_unlinked) == NULLAGINO);
1965 /* both on-disk, don't endian flip twice */
1966 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1967 offset = ip->i_boffset +
1968 offsetof(xfs_dinode_t, di_next_unlinked);
1969 xfs_trans_inode_buf(tp, ibp);
1970 xfs_trans_log_buf(tp, ibp, offset,
1971 (offset + sizeof(xfs_agino_t) - 1));
1972 xfs_inobp_check(mp, ibp);
1976 * Point the bucket head pointer at the inode being inserted.
1979 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1980 offset = offsetof(xfs_agi_t, agi_unlinked) +
1981 (sizeof(xfs_agino_t) * bucket_index);
1982 xfs_trans_log_buf(tp, agibp, offset,
1983 (offset + sizeof(xfs_agino_t) - 1));
1988 * Pull the on-disk inode from the AGI unlinked list.
2001 xfs_agnumber_t agno;
2002 xfs_daddr_t agdaddr;
2004 xfs_agino_t next_agino;
2005 xfs_buf_t *last_ibp;
2006 xfs_dinode_t *last_dip = NULL;
2008 int offset, last_offset = 0;
2013 * First pull the on-disk inode from the AGI unlinked list.
2017 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2018 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
2021 * Get the agi buffer first. It ensures lock ordering
2024 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
2025 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
2028 "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
2029 error, mp->m_fsname);
2033 * Validate the magic number of the agi block.
2035 agi = XFS_BUF_TO_AGI(agibp);
2037 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
2038 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
2039 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE,
2040 XFS_RANDOM_IUNLINK_REMOVE))) {
2041 XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW,
2043 xfs_trans_brelse(tp, agibp);
2045 "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
2047 return XFS_ERROR(EFSCORRUPTED);
2050 * Get the index into the agi hash table for the
2051 * list this inode will go on.
2053 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2055 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2056 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
2057 ASSERT(agi->agi_unlinked[bucket_index]);
2059 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2061 * We're at the head of the list. Get the inode's
2062 * on-disk buffer to see if there is anyone after us
2063 * on the list. Only modify our next pointer if it
2064 * is not already NULLAGINO. This saves us the overhead
2065 * of dealing with the buffer when there is no need to
2068 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
2071 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2072 error, mp->m_fsname);
2075 next_agino = be32_to_cpu(dip->di_next_unlinked);
2076 ASSERT(next_agino != 0);
2077 if (next_agino != NULLAGINO) {
2078 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2079 offset = ip->i_boffset +
2080 offsetof(xfs_dinode_t, di_next_unlinked);
2081 xfs_trans_inode_buf(tp, ibp);
2082 xfs_trans_log_buf(tp, ibp, offset,
2083 (offset + sizeof(xfs_agino_t) - 1));
2084 xfs_inobp_check(mp, ibp);
2086 xfs_trans_brelse(tp, ibp);
2089 * Point the bucket head pointer at the next inode.
2091 ASSERT(next_agino != 0);
2092 ASSERT(next_agino != agino);
2093 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2094 offset = offsetof(xfs_agi_t, agi_unlinked) +
2095 (sizeof(xfs_agino_t) * bucket_index);
2096 xfs_trans_log_buf(tp, agibp, offset,
2097 (offset + sizeof(xfs_agino_t) - 1));
2100 * We need to search the list for the inode being freed.
2102 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2104 while (next_agino != agino) {
2106 * If the last inode wasn't the one pointing to
2107 * us, then release its buffer since we're not
2108 * going to do anything with it.
2110 if (last_ibp != NULL) {
2111 xfs_trans_brelse(tp, last_ibp);
2113 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2114 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
2115 &last_ibp, &last_offset);
2118 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
2119 error, mp->m_fsname);
2122 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2123 ASSERT(next_agino != NULLAGINO);
2124 ASSERT(next_agino != 0);
2127 * Now last_ibp points to the buffer previous to us on
2128 * the unlinked list. Pull us from the list.
2130 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
2133 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2134 error, mp->m_fsname);
2137 next_agino = be32_to_cpu(dip->di_next_unlinked);
2138 ASSERT(next_agino != 0);
2139 ASSERT(next_agino != agino);
2140 if (next_agino != NULLAGINO) {
2141 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2142 offset = ip->i_boffset +
2143 offsetof(xfs_dinode_t, di_next_unlinked);
2144 xfs_trans_inode_buf(tp, ibp);
2145 xfs_trans_log_buf(tp, ibp, offset,
2146 (offset + sizeof(xfs_agino_t) - 1));
2147 xfs_inobp_check(mp, ibp);
2149 xfs_trans_brelse(tp, ibp);
2152 * Point the previous inode on the list to the next inode.
2154 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2155 ASSERT(next_agino != 0);
2156 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2157 xfs_trans_inode_buf(tp, last_ibp);
2158 xfs_trans_log_buf(tp, last_ibp, offset,
2159 (offset + sizeof(xfs_agino_t) - 1));
2160 xfs_inobp_check(mp, last_ibp);
2165 STATIC_INLINE int xfs_inode_clean(xfs_inode_t *ip)
2167 return (((ip->i_itemp == NULL) ||
2168 !(ip->i_itemp->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
2169 (ip->i_update_core == 0));
2174 xfs_inode_t *free_ip,
2178 xfs_mount_t *mp = free_ip->i_mount;
2179 int blks_per_cluster;
2182 int i, j, found, pre_flushed;
2185 xfs_inode_t *ip, **ip_found;
2186 xfs_inode_log_item_t *iip;
2187 xfs_log_item_t *lip;
2188 xfs_perag_t *pag = xfs_get_perag(mp, inum);
2191 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
2192 blks_per_cluster = 1;
2193 ninodes = mp->m_sb.sb_inopblock;
2194 nbufs = XFS_IALLOC_BLOCKS(mp);
2196 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
2197 mp->m_sb.sb_blocksize;
2198 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
2199 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
2202 ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
2204 for (j = 0; j < nbufs; j++, inum += ninodes) {
2205 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2206 XFS_INO_TO_AGBNO(mp, inum));
2210 * Look for each inode in memory and attempt to lock it,
2211 * we can be racing with flush and tail pushing here.
2212 * any inode we get the locks on, add to an array of
2213 * inode items to process later.
2215 * The get the buffer lock, we could beat a flush
2216 * or tail pushing thread to the lock here, in which
2217 * case they will go looking for the inode buffer
2218 * and fail, we need some other form of interlock
2222 for (i = 0; i < ninodes; i++) {
2223 read_lock(&pag->pag_ici_lock);
2224 ip = radix_tree_lookup(&pag->pag_ici_root,
2225 XFS_INO_TO_AGINO(mp, (inum + i)));
2227 /* Inode not in memory or we found it already,
2230 if (!ip || xfs_iflags_test(ip, XFS_ISTALE)) {
2231 read_unlock(&pag->pag_ici_lock);
2235 if (xfs_inode_clean(ip)) {
2236 read_unlock(&pag->pag_ici_lock);
2240 /* If we can get the locks then add it to the
2241 * list, otherwise by the time we get the bp lock
2242 * below it will already be attached to the
2246 /* This inode will already be locked - by us, lets
2250 if (ip == free_ip) {
2251 if (xfs_iflock_nowait(ip)) {
2252 xfs_iflags_set(ip, XFS_ISTALE);
2253 if (xfs_inode_clean(ip)) {
2256 ip_found[found++] = ip;
2259 read_unlock(&pag->pag_ici_lock);
2263 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2264 if (xfs_iflock_nowait(ip)) {
2265 xfs_iflags_set(ip, XFS_ISTALE);
2267 if (xfs_inode_clean(ip)) {
2269 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2271 ip_found[found++] = ip;
2274 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2277 read_unlock(&pag->pag_ici_lock);
2280 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2281 mp->m_bsize * blks_per_cluster,
2285 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
2287 if (lip->li_type == XFS_LI_INODE) {
2288 iip = (xfs_inode_log_item_t *)lip;
2289 ASSERT(iip->ili_logged == 1);
2290 lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
2292 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2294 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2297 lip = lip->li_bio_list;
2300 for (i = 0; i < found; i++) {
2305 ip->i_update_core = 0;
2307 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2311 iip->ili_last_fields = iip->ili_format.ilf_fields;
2312 iip->ili_format.ilf_fields = 0;
2313 iip->ili_logged = 1;
2315 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2318 xfs_buf_attach_iodone(bp,
2319 (void(*)(xfs_buf_t*,xfs_log_item_t*))
2320 xfs_istale_done, (xfs_log_item_t *)iip);
2321 if (ip != free_ip) {
2322 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2326 if (found || pre_flushed)
2327 xfs_trans_stale_inode_buf(tp, bp);
2328 xfs_trans_binval(tp, bp);
2331 kmem_free(ip_found, ninodes * sizeof(xfs_inode_t *));
2332 xfs_put_perag(mp, pag);
2336 * This is called to return an inode to the inode free list.
2337 * The inode should already be truncated to 0 length and have
2338 * no pages associated with it. This routine also assumes that
2339 * the inode is already a part of the transaction.
2341 * The on-disk copy of the inode will have been added to the list
2342 * of unlinked inodes in the AGI. We need to remove the inode from
2343 * that list atomically with respect to freeing it here.
2349 xfs_bmap_free_t *flist)
2353 xfs_ino_t first_ino;
2355 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2356 ASSERT(ip->i_transp == tp);
2357 ASSERT(ip->i_d.di_nlink == 0);
2358 ASSERT(ip->i_d.di_nextents == 0);
2359 ASSERT(ip->i_d.di_anextents == 0);
2360 ASSERT((ip->i_d.di_size == 0 && ip->i_size == 0) ||
2361 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2362 ASSERT(ip->i_d.di_nblocks == 0);
2365 * Pull the on-disk inode from the AGI unlinked list.
2367 error = xfs_iunlink_remove(tp, ip);
2372 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2376 ip->i_d.di_mode = 0; /* mark incore inode as free */
2377 ip->i_d.di_flags = 0;
2378 ip->i_d.di_dmevmask = 0;
2379 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2380 ip->i_df.if_ext_max =
2381 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2382 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2383 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2385 * Bump the generation count so no one will be confused
2386 * by reincarnations of this inode.
2389 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2392 xfs_ifree_cluster(ip, tp, first_ino);
2399 * Reallocate the space for if_broot based on the number of records
2400 * being added or deleted as indicated in rec_diff. Move the records
2401 * and pointers in if_broot to fit the new size. When shrinking this
2402 * will eliminate holes between the records and pointers created by
2403 * the caller. When growing this will create holes to be filled in
2406 * The caller must not request to add more records than would fit in
2407 * the on-disk inode root. If the if_broot is currently NULL, then
2408 * if we adding records one will be allocated. The caller must also
2409 * not request that the number of records go below zero, although
2410 * it can go to zero.
2412 * ip -- the inode whose if_broot area is changing
2413 * ext_diff -- the change in the number of records, positive or negative,
2414 * requested for the if_broot array.
2424 xfs_bmbt_block_t *new_broot;
2431 * Handle the degenerate case quietly.
2433 if (rec_diff == 0) {
2437 ifp = XFS_IFORK_PTR(ip, whichfork);
2440 * If there wasn't any memory allocated before, just
2441 * allocate it now and get out.
2443 if (ifp->if_broot_bytes == 0) {
2444 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
2445 ifp->if_broot = (xfs_bmbt_block_t*)kmem_alloc(new_size,
2447 ifp->if_broot_bytes = (int)new_size;
2452 * If there is already an existing if_broot, then we need
2453 * to realloc() it and shift the pointers to their new
2454 * location. The records don't change location because
2455 * they are kept butted up against the btree block header.
2457 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2458 new_max = cur_max + rec_diff;
2459 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2460 ifp->if_broot = (xfs_bmbt_block_t *)
2461 kmem_realloc(ifp->if_broot,
2463 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2465 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2466 ifp->if_broot_bytes);
2467 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2469 ifp->if_broot_bytes = (int)new_size;
2470 ASSERT(ifp->if_broot_bytes <=
2471 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2472 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2477 * rec_diff is less than 0. In this case, we are shrinking the
2478 * if_broot buffer. It must already exist. If we go to zero
2479 * records, just get rid of the root and clear the status bit.
2481 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
2482 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2483 new_max = cur_max + rec_diff;
2484 ASSERT(new_max >= 0);
2486 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2490 new_broot = (xfs_bmbt_block_t *)kmem_alloc(new_size, KM_SLEEP);
2492 * First copy over the btree block header.
2494 memcpy(new_broot, ifp->if_broot, sizeof(xfs_bmbt_block_t));
2497 ifp->if_flags &= ~XFS_IFBROOT;
2501 * Only copy the records and pointers if there are any.
2505 * First copy the records.
2507 op = (char *)XFS_BMAP_BROOT_REC_ADDR(ifp->if_broot, 1,
2508 ifp->if_broot_bytes);
2509 np = (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot, 1,
2511 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2514 * Then copy the pointers.
2516 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2517 ifp->if_broot_bytes);
2518 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot, 1,
2520 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2522 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2523 ifp->if_broot = new_broot;
2524 ifp->if_broot_bytes = (int)new_size;
2525 ASSERT(ifp->if_broot_bytes <=
2526 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2532 * This is called when the amount of space needed for if_data
2533 * is increased or decreased. The change in size is indicated by
2534 * the number of bytes that need to be added or deleted in the
2535 * byte_diff parameter.
2537 * If the amount of space needed has decreased below the size of the
2538 * inline buffer, then switch to using the inline buffer. Otherwise,
2539 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2540 * to what is needed.
2542 * ip -- the inode whose if_data area is changing
2543 * byte_diff -- the change in the number of bytes, positive or negative,
2544 * requested for the if_data array.
2556 if (byte_diff == 0) {
2560 ifp = XFS_IFORK_PTR(ip, whichfork);
2561 new_size = (int)ifp->if_bytes + byte_diff;
2562 ASSERT(new_size >= 0);
2564 if (new_size == 0) {
2565 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2566 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2568 ifp->if_u1.if_data = NULL;
2570 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2572 * If the valid extents/data can fit in if_inline_ext/data,
2573 * copy them from the malloc'd vector and free it.
2575 if (ifp->if_u1.if_data == NULL) {
2576 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2577 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2578 ASSERT(ifp->if_real_bytes != 0);
2579 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2581 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2582 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2587 * Stuck with malloc/realloc.
2588 * For inline data, the underlying buffer must be
2589 * a multiple of 4 bytes in size so that it can be
2590 * logged and stay on word boundaries. We enforce
2593 real_size = roundup(new_size, 4);
2594 if (ifp->if_u1.if_data == NULL) {
2595 ASSERT(ifp->if_real_bytes == 0);
2596 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2597 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2599 * Only do the realloc if the underlying size
2600 * is really changing.
2602 if (ifp->if_real_bytes != real_size) {
2603 ifp->if_u1.if_data =
2604 kmem_realloc(ifp->if_u1.if_data,
2610 ASSERT(ifp->if_real_bytes == 0);
2611 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2612 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2616 ifp->if_real_bytes = real_size;
2617 ifp->if_bytes = new_size;
2618 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2625 * Map inode to disk block and offset.
2627 * mp -- the mount point structure for the current file system
2628 * tp -- the current transaction
2629 * ino -- the inode number of the inode to be located
2630 * imap -- this structure is filled in with the information necessary
2631 * to retrieve the given inode from disk
2632 * flags -- flags to pass to xfs_dilocate indicating whether or not
2633 * lookups in the inode btree were OK or not
2643 xfs_fsblock_t fsbno;
2648 fsbno = imap->im_blkno ?
2649 XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK;
2650 error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags);
2654 imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno);
2655 imap->im_len = XFS_FSB_TO_BB(mp, len);
2656 imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno);
2657 imap->im_ioffset = (ushort)off;
2658 imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog);
2669 ifp = XFS_IFORK_PTR(ip, whichfork);
2670 if (ifp->if_broot != NULL) {
2671 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2672 ifp->if_broot = NULL;
2676 * If the format is local, then we can't have an extents
2677 * array so just look for an inline data array. If we're
2678 * not local then we may or may not have an extents list,
2679 * so check and free it up if we do.
2681 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2682 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2683 (ifp->if_u1.if_data != NULL)) {
2684 ASSERT(ifp->if_real_bytes != 0);
2685 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2686 ifp->if_u1.if_data = NULL;
2687 ifp->if_real_bytes = 0;
2689 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2690 ((ifp->if_flags & XFS_IFEXTIREC) ||
2691 ((ifp->if_u1.if_extents != NULL) &&
2692 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
2693 ASSERT(ifp->if_real_bytes != 0);
2694 xfs_iext_destroy(ifp);
2696 ASSERT(ifp->if_u1.if_extents == NULL ||
2697 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2698 ASSERT(ifp->if_real_bytes == 0);
2699 if (whichfork == XFS_ATTR_FORK) {
2700 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2706 * This is called free all the memory associated with an inode.
2707 * It must free the inode itself and any buffers allocated for
2708 * if_extents/if_data and if_broot. It must also free the lock
2709 * associated with the inode.
2716 switch (ip->i_d.di_mode & S_IFMT) {
2720 xfs_idestroy_fork(ip, XFS_DATA_FORK);
2724 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
2725 mrfree(&ip->i_lock);
2726 mrfree(&ip->i_iolock);
2727 freesema(&ip->i_flock);
2728 #ifdef XFS_BMAP_TRACE
2729 ktrace_free(ip->i_xtrace);
2731 #ifdef XFS_BMBT_TRACE
2732 ktrace_free(ip->i_btrace);
2735 ktrace_free(ip->i_rwtrace);
2737 #ifdef XFS_ILOCK_TRACE
2738 ktrace_free(ip->i_lock_trace);
2740 #ifdef XFS_DIR2_TRACE
2741 ktrace_free(ip->i_dir_trace);
2745 * Only if we are shutting down the fs will we see an
2746 * inode still in the AIL. If it is there, we should remove
2747 * it to prevent a use-after-free from occurring.
2749 xfs_mount_t *mp = ip->i_mount;
2750 xfs_log_item_t *lip = &ip->i_itemp->ili_item;
2753 ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) ||
2754 XFS_FORCED_SHUTDOWN(ip->i_mount));
2755 if (lip->li_flags & XFS_LI_IN_AIL) {
2757 if (lip->li_flags & XFS_LI_IN_AIL)
2758 xfs_trans_delete_ail(mp, lip, s);
2762 xfs_inode_item_destroy(ip);
2764 kmem_zone_free(xfs_inode_zone, ip);
2769 * Increment the pin count of the given buffer.
2770 * This value is protected by ipinlock spinlock in the mount structure.
2776 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2778 atomic_inc(&ip->i_pincount);
2782 * Decrement the pin count of the given inode, and wake up
2783 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2784 * inode must have been previously pinned with a call to xfs_ipin().
2790 ASSERT(atomic_read(&ip->i_pincount) > 0);
2792 if (atomic_dec_and_lock(&ip->i_pincount, &ip->i_flags_lock)) {
2795 * If the inode is currently being reclaimed, the link between
2796 * the bhv_vnode and the xfs_inode will be broken after the
2797 * XFS_IRECLAIM* flag is set. Hence, if these flags are not
2798 * set, then we can move forward and mark the linux inode dirty
2799 * knowing that it is still valid as it won't freed until after
2800 * the bhv_vnode<->xfs_inode link is broken in xfs_reclaim. The
2801 * i_flags_lock is used to synchronise the setting of the
2802 * XFS_IRECLAIM* flags and the breaking of the link, and so we
2803 * can execute atomically w.r.t to reclaim by holding this lock
2806 * However, we still need to issue the unpin wakeup call as the
2807 * inode reclaim may be blocked waiting for the inode to become
2811 if (!__xfs_iflags_test(ip, XFS_IRECLAIM|XFS_IRECLAIMABLE)) {
2812 bhv_vnode_t *vp = XFS_ITOV_NULL(ip);
2813 struct inode *inode = NULL;
2816 inode = vn_to_inode(vp);
2817 BUG_ON(inode->i_state & I_CLEAR);
2819 /* make sync come back and flush this inode */
2820 if (!(inode->i_state & (I_NEW|I_FREEING)))
2821 mark_inode_dirty_sync(inode);
2823 spin_unlock(&ip->i_flags_lock);
2824 wake_up(&ip->i_ipin_wait);
2829 * This is called to wait for the given inode to be unpinned.
2830 * It will sleep until this happens. The caller must have the
2831 * inode locked in at least shared mode so that the buffer cannot
2832 * be subsequently pinned once someone is waiting for it to be
2839 xfs_inode_log_item_t *iip;
2842 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE | MR_ACCESS));
2844 if (atomic_read(&ip->i_pincount) == 0) {
2849 if (iip && iip->ili_last_lsn) {
2850 lsn = iip->ili_last_lsn;
2856 * Give the log a push so we don't wait here too long.
2858 xfs_log_force(ip->i_mount, lsn, XFS_LOG_FORCE);
2860 wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
2865 * xfs_iextents_copy()
2867 * This is called to copy the REAL extents (as opposed to the delayed
2868 * allocation extents) from the inode into the given buffer. It
2869 * returns the number of bytes copied into the buffer.
2871 * If there are no delayed allocation extents, then we can just
2872 * memcpy() the extents into the buffer. Otherwise, we need to
2873 * examine each extent in turn and skip those which are delayed.
2885 xfs_fsblock_t start_block;
2887 ifp = XFS_IFORK_PTR(ip, whichfork);
2888 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
2889 ASSERT(ifp->if_bytes > 0);
2891 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2892 XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork);
2896 * There are some delayed allocation extents in the
2897 * inode, so copy the extents one at a time and skip
2898 * the delayed ones. There must be at least one
2899 * non-delayed extent.
2902 for (i = 0; i < nrecs; i++) {
2903 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
2904 start_block = xfs_bmbt_get_startblock(ep);
2905 if (ISNULLSTARTBLOCK(start_block)) {
2907 * It's a delayed allocation extent, so skip it.
2912 /* Translate to on disk format */
2913 put_unaligned(cpu_to_be64(ep->l0), &dp->l0);
2914 put_unaligned(cpu_to_be64(ep->l1), &dp->l1);
2918 ASSERT(copied != 0);
2919 xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip));
2921 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2925 * Each of the following cases stores data into the same region
2926 * of the on-disk inode, so only one of them can be valid at
2927 * any given time. While it is possible to have conflicting formats
2928 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2929 * in EXTENTS format, this can only happen when the fork has
2930 * changed formats after being modified but before being flushed.
2931 * In these cases, the format always takes precedence, because the
2932 * format indicates the current state of the fork.
2939 xfs_inode_log_item_t *iip,
2946 #ifdef XFS_TRANS_DEBUG
2949 static const short brootflag[2] =
2950 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2951 static const short dataflag[2] =
2952 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2953 static const short extflag[2] =
2954 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2958 ifp = XFS_IFORK_PTR(ip, whichfork);
2960 * This can happen if we gave up in iformat in an error path,
2961 * for the attribute fork.
2964 ASSERT(whichfork == XFS_ATTR_FORK);
2967 cp = XFS_DFORK_PTR(dip, whichfork);
2969 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2970 case XFS_DINODE_FMT_LOCAL:
2971 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2972 (ifp->if_bytes > 0)) {
2973 ASSERT(ifp->if_u1.if_data != NULL);
2974 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2975 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2979 case XFS_DINODE_FMT_EXTENTS:
2980 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2981 !(iip->ili_format.ilf_fields & extflag[whichfork]));
2982 ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
2983 (ifp->if_bytes == 0));
2984 ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
2985 (ifp->if_bytes > 0));
2986 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
2987 (ifp->if_bytes > 0)) {
2988 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2989 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2994 case XFS_DINODE_FMT_BTREE:
2995 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
2996 (ifp->if_broot_bytes > 0)) {
2997 ASSERT(ifp->if_broot != NULL);
2998 ASSERT(ifp->if_broot_bytes <=
2999 (XFS_IFORK_SIZE(ip, whichfork) +
3000 XFS_BROOT_SIZE_ADJ));
3001 xfs_bmbt_to_bmdr(ifp->if_broot, ifp->if_broot_bytes,
3002 (xfs_bmdr_block_t *)cp,
3003 XFS_DFORK_SIZE(dip, mp, whichfork));
3007 case XFS_DINODE_FMT_DEV:
3008 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
3009 ASSERT(whichfork == XFS_DATA_FORK);
3010 dip->di_u.di_dev = cpu_to_be32(ip->i_df.if_u2.if_rdev);
3014 case XFS_DINODE_FMT_UUID:
3015 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
3016 ASSERT(whichfork == XFS_DATA_FORK);
3017 memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid,
3031 * xfs_iflush() will write a modified inode's changes out to the
3032 * inode's on disk home. The caller must have the inode lock held
3033 * in at least shared mode and the inode flush semaphore must be
3034 * held as well. The inode lock will still be held upon return from
3035 * the call and the caller is free to unlock it.
3036 * The inode flush lock will be unlocked when the inode reaches the disk.
3037 * The flags indicate how the inode's buffer should be written out.
3044 xfs_inode_log_item_t *iip;
3051 int clcount; /* count of inodes clustered */
3053 struct hlist_node *entry;
3054 enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
3056 XFS_STATS_INC(xs_iflush_count);
3058 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
3059 ASSERT(issemalocked(&(ip->i_flock)));
3060 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3061 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3067 * If the inode isn't dirty, then just release the inode
3068 * flush lock and do nothing.
3070 if ((ip->i_update_core == 0) &&
3071 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3072 ASSERT((iip != NULL) ?
3073 !(iip->ili_item.li_flags & XFS_LI_IN_AIL) : 1);
3079 * We can't flush the inode until it is unpinned, so
3080 * wait for it. We know noone new can pin it, because
3081 * we are holding the inode lock shared and you need
3082 * to hold it exclusively to pin the inode.
3084 xfs_iunpin_wait(ip);
3087 * This may have been unpinned because the filesystem is shutting
3088 * down forcibly. If that's the case we must not write this inode
3089 * to disk, because the log record didn't make it to disk!
3091 if (XFS_FORCED_SHUTDOWN(mp)) {
3092 ip->i_update_core = 0;
3094 iip->ili_format.ilf_fields = 0;
3096 return XFS_ERROR(EIO);
3100 * Get the buffer containing the on-disk inode.
3102 error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0, 0);
3109 * Decide how buffer will be flushed out. This is done before
3110 * the call to xfs_iflush_int because this field is zeroed by it.
3112 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3114 * Flush out the inode buffer according to the directions
3115 * of the caller. In the cases where the caller has given
3116 * us a choice choose the non-delwri case. This is because
3117 * the inode is in the AIL and we need to get it out soon.
3120 case XFS_IFLUSH_SYNC:
3121 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3124 case XFS_IFLUSH_ASYNC:
3125 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3128 case XFS_IFLUSH_DELWRI:
3138 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3139 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3140 case XFS_IFLUSH_DELWRI:
3143 case XFS_IFLUSH_ASYNC:
3146 case XFS_IFLUSH_SYNC:
3157 * First flush out the inode that xfs_iflush was called with.
3159 error = xfs_iflush_int(ip, bp);
3166 * see if other inodes can be gathered into this write
3168 spin_lock(&ip->i_cluster->icl_lock);
3169 ip->i_cluster->icl_buf = bp;
3172 hlist_for_each_entry(iq, entry, &ip->i_cluster->icl_inodes, i_cnode) {
3177 * Do an un-protected check to see if the inode is dirty and
3178 * is a candidate for flushing. These checks will be repeated
3179 * later after the appropriate locks are acquired.
3182 if ((iq->i_update_core == 0) &&
3184 !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
3185 xfs_ipincount(iq) == 0) {
3190 * Try to get locks. If any are unavailable,
3191 * then this inode cannot be flushed and is skipped.
3194 /* get inode locks (just i_lock) */
3195 if (xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) {
3196 /* get inode flush lock */
3197 if (xfs_iflock_nowait(iq)) {
3198 /* check if pinned */
3199 if (xfs_ipincount(iq) == 0) {
3200 /* arriving here means that
3201 * this inode can be flushed.
3202 * first re-check that it's
3206 if ((iq->i_update_core != 0)||
3208 (iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3210 error = xfs_iflush_int(iq, bp);
3214 goto cluster_corrupt_out;
3223 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3226 spin_unlock(&ip->i_cluster->icl_lock);
3229 XFS_STATS_INC(xs_icluster_flushcnt);
3230 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
3234 * If the buffer is pinned then push on the log so we won't
3235 * get stuck waiting in the write for too long.
3237 if (XFS_BUF_ISPINNED(bp)){
3238 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
3241 if (flags & INT_DELWRI) {
3242 xfs_bdwrite(mp, bp);
3243 } else if (flags & INT_ASYNC) {
3244 xfs_bawrite(mp, bp);
3246 error = xfs_bwrite(mp, bp);
3252 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3253 xfs_iflush_abort(ip);
3255 * Unlocks the flush lock
3257 return XFS_ERROR(EFSCORRUPTED);
3259 cluster_corrupt_out:
3260 /* Corruption detected in the clustering loop. Invalidate the
3261 * inode buffer and shut down the filesystem.
3263 spin_unlock(&ip->i_cluster->icl_lock);
3266 * Clean up the buffer. If it was B_DELWRI, just release it --
3267 * brelse can handle it with no problems. If not, shut down the
3268 * filesystem before releasing the buffer.
3270 if ((bufwasdelwri= XFS_BUF_ISDELAYWRITE(bp))) {
3274 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3278 * Just like incore_relse: if we have b_iodone functions,
3279 * mark the buffer as an error and call them. Otherwise
3280 * mark it as stale and brelse.
3282 if (XFS_BUF_IODONE_FUNC(bp)) {
3283 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
3287 XFS_BUF_ERROR(bp,EIO);
3295 xfs_iflush_abort(iq);
3297 * Unlocks the flush lock
3299 return XFS_ERROR(EFSCORRUPTED);
3308 xfs_inode_log_item_t *iip;
3311 #ifdef XFS_TRANS_DEBUG
3316 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
3317 ASSERT(issemalocked(&(ip->i_flock)));
3318 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3319 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3326 * If the inode isn't dirty, then just release the inode
3327 * flush lock and do nothing.
3329 if ((ip->i_update_core == 0) &&
3330 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3335 /* set *dip = inode's place in the buffer */
3336 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset);
3339 * Clear i_update_core before copying out the data.
3340 * This is for coordination with our timestamp updates
3341 * that don't hold the inode lock. They will always
3342 * update the timestamps BEFORE setting i_update_core,
3343 * so if we clear i_update_core after they set it we
3344 * are guaranteed to see their updates to the timestamps.
3345 * I believe that this depends on strongly ordered memory
3346 * semantics, but we have that. We use the SYNCHRONIZE
3347 * macro to make sure that the compiler does not reorder
3348 * the i_update_core access below the data copy below.
3350 ip->i_update_core = 0;
3354 * Make sure to get the latest atime from the Linux inode.
3356 xfs_synchronize_atime(ip);
3358 if (XFS_TEST_ERROR(be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC,
3359 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3360 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3361 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3362 ip->i_ino, be16_to_cpu(dip->di_core.di_magic), dip);
3365 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3366 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3367 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3368 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3369 ip->i_ino, ip, ip->i_d.di_magic);
3372 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3374 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3375 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3376 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3377 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3378 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3382 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
3384 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3385 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3386 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3387 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3388 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3389 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3394 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3395 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3396 XFS_RANDOM_IFLUSH_5)) {
3397 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3398 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3400 ip->i_d.di_nextents + ip->i_d.di_anextents,
3405 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3406 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3407 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3408 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3409 ip->i_ino, ip->i_d.di_forkoff, ip);
3413 * bump the flush iteration count, used to detect flushes which
3414 * postdate a log record during recovery.
3417 ip->i_d.di_flushiter++;
3420 * Copy the dirty parts of the inode into the on-disk
3421 * inode. We always copy out the core of the inode,
3422 * because if the inode is dirty at all the core must
3425 xfs_dinode_to_disk(&dip->di_core, &ip->i_d);
3427 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3428 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3429 ip->i_d.di_flushiter = 0;
3432 * If this is really an old format inode and the superblock version
3433 * has not been updated to support only new format inodes, then
3434 * convert back to the old inode format. If the superblock version
3435 * has been updated, then make the conversion permanent.
3437 ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 ||
3438 XFS_SB_VERSION_HASNLINK(&mp->m_sb));
3439 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
3440 if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) {
3444 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3445 dip->di_core.di_onlink = cpu_to_be16(ip->i_d.di_nlink);
3448 * The superblock version has already been bumped,
3449 * so just make the conversion to the new inode
3452 ip->i_d.di_version = XFS_DINODE_VERSION_2;
3453 dip->di_core.di_version = XFS_DINODE_VERSION_2;
3454 ip->i_d.di_onlink = 0;
3455 dip->di_core.di_onlink = 0;
3456 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3457 memset(&(dip->di_core.di_pad[0]), 0,
3458 sizeof(dip->di_core.di_pad));
3459 ASSERT(ip->i_d.di_projid == 0);
3463 if (xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp) == EFSCORRUPTED) {
3467 if (XFS_IFORK_Q(ip)) {
3469 * The only error from xfs_iflush_fork is on the data fork.
3471 (void) xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3473 xfs_inobp_check(mp, bp);
3476 * We've recorded everything logged in the inode, so we'd
3477 * like to clear the ilf_fields bits so we don't log and
3478 * flush things unnecessarily. However, we can't stop
3479 * logging all this information until the data we've copied
3480 * into the disk buffer is written to disk. If we did we might
3481 * overwrite the copy of the inode in the log with all the
3482 * data after re-logging only part of it, and in the face of
3483 * a crash we wouldn't have all the data we need to recover.
3485 * What we do is move the bits to the ili_last_fields field.
3486 * When logging the inode, these bits are moved back to the
3487 * ilf_fields field. In the xfs_iflush_done() routine we
3488 * clear ili_last_fields, since we know that the information
3489 * those bits represent is permanently on disk. As long as
3490 * the flush completes before the inode is logged again, then
3491 * both ilf_fields and ili_last_fields will be cleared.
3493 * We can play with the ilf_fields bits here, because the inode
3494 * lock must be held exclusively in order to set bits there
3495 * and the flush lock protects the ili_last_fields bits.
3496 * Set ili_logged so the flush done
3497 * routine can tell whether or not to look in the AIL.
3498 * Also, store the current LSN of the inode so that we can tell
3499 * whether the item has moved in the AIL from xfs_iflush_done().
3500 * In order to read the lsn we need the AIL lock, because
3501 * it is a 64 bit value that cannot be read atomically.
3503 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3504 iip->ili_last_fields = iip->ili_format.ilf_fields;
3505 iip->ili_format.ilf_fields = 0;
3506 iip->ili_logged = 1;
3508 ASSERT(sizeof(xfs_lsn_t) == 8); /* don't lock if it shrinks */
3510 iip->ili_flush_lsn = iip->ili_item.li_lsn;
3514 * Attach the function xfs_iflush_done to the inode's
3515 * buffer. This will remove the inode from the AIL
3516 * and unlock the inode's flush lock when the inode is
3517 * completely written to disk.
3519 xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
3520 xfs_iflush_done, (xfs_log_item_t *)iip);
3522 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3523 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3526 * We're flushing an inode which is not in the AIL and has
3527 * not been logged but has i_update_core set. For this
3528 * case we can use a B_DELWRI flush and immediately drop
3529 * the inode flush lock because we can avoid the whole
3530 * AIL state thing. It's OK to drop the flush lock now,
3531 * because we've already locked the buffer and to do anything
3532 * you really need both.
3535 ASSERT(iip->ili_logged == 0);
3536 ASSERT(iip->ili_last_fields == 0);
3537 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3545 return XFS_ERROR(EFSCORRUPTED);
3550 * Flush all inactive inodes in mp.
3560 XFS_MOUNT_ILOCK(mp);
3566 /* Make sure we skip markers inserted by sync */
3567 if (ip->i_mount == NULL) {
3572 vp = XFS_ITOV_NULL(ip);
3574 XFS_MOUNT_IUNLOCK(mp);
3575 xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC);
3579 ASSERT(vn_count(vp) == 0);
3582 } while (ip != mp->m_inodes);
3584 XFS_MOUNT_IUNLOCK(mp);
3588 * xfs_iaccess: check accessibility of inode for mode.
3597 mode_t orgmode = mode;
3598 struct inode *inode = vn_to_inode(XFS_ITOV(ip));
3600 if (mode & S_IWUSR) {
3601 umode_t imode = inode->i_mode;
3603 if (IS_RDONLY(inode) &&
3604 (S_ISREG(imode) || S_ISDIR(imode) || S_ISLNK(imode)))
3605 return XFS_ERROR(EROFS);
3607 if (IS_IMMUTABLE(inode))
3608 return XFS_ERROR(EACCES);
3612 * If there's an Access Control List it's used instead of
3615 if ((error = _ACL_XFS_IACCESS(ip, mode, cr)) != -1)
3616 return error ? XFS_ERROR(error) : 0;
3618 if (current_fsuid(cr) != ip->i_d.di_uid) {
3620 if (!in_group_p((gid_t)ip->i_d.di_gid))
3625 * If the DACs are ok we don't need any capability check.
3627 if ((ip->i_d.di_mode & mode) == mode)
3630 * Read/write DACs are always overridable.
3631 * Executable DACs are overridable if at least one exec bit is set.
3633 if (!(orgmode & S_IXUSR) ||
3634 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
3635 if (capable_cred(cr, CAP_DAC_OVERRIDE))
3638 if ((orgmode == S_IRUSR) ||
3639 (S_ISDIR(inode->i_mode) && (!(orgmode & S_IWUSR)))) {
3640 if (capable_cred(cr, CAP_DAC_READ_SEARCH))
3643 cmn_err(CE_NOTE, "Ick: mode=%o, orgmode=%o", mode, orgmode);
3645 return XFS_ERROR(EACCES);
3647 return XFS_ERROR(EACCES);
3651 * xfs_iroundup: round up argument to next power of two
3660 if ((v & (v - 1)) == 0)
3662 ASSERT((v & 0x80000000) == 0);
3663 if ((v & (v + 1)) == 0)
3665 for (i = 0, m = 1; i < 31; i++, m <<= 1) {
3669 if ((v & (v + 1)) == 0)
3676 #ifdef XFS_ILOCK_TRACE
3677 ktrace_t *xfs_ilock_trace_buf;
3680 xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
3682 ktrace_enter(ip->i_lock_trace,
3684 (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
3685 (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
3686 (void *)ra, /* caller of ilock */
3687 (void *)(unsigned long)current_cpu(),
3688 (void *)(unsigned long)current_pid(),
3689 NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL);
3694 * Return a pointer to the extent record at file index idx.
3696 xfs_bmbt_rec_host_t *
3698 xfs_ifork_t *ifp, /* inode fork pointer */
3699 xfs_extnum_t idx) /* index of target extent */
3702 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
3703 return ifp->if_u1.if_ext_irec->er_extbuf;
3704 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3705 xfs_ext_irec_t *erp; /* irec pointer */
3706 int erp_idx = 0; /* irec index */
3707 xfs_extnum_t page_idx = idx; /* ext index in target list */
3709 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3710 return &erp->er_extbuf[page_idx];
3711 } else if (ifp->if_bytes) {
3712 return &ifp->if_u1.if_extents[idx];
3719 * Insert new item(s) into the extent records for incore inode
3720 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3724 xfs_ifork_t *ifp, /* inode fork pointer */
3725 xfs_extnum_t idx, /* starting index of new items */
3726 xfs_extnum_t count, /* number of inserted items */
3727 xfs_bmbt_irec_t *new) /* items to insert */
3729 xfs_extnum_t i; /* extent record index */
3731 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3732 xfs_iext_add(ifp, idx, count);
3733 for (i = idx; i < idx + count; i++, new++)
3734 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new);
3738 * This is called when the amount of space required for incore file
3739 * extents needs to be increased. The ext_diff parameter stores the
3740 * number of new extents being added and the idx parameter contains
3741 * the extent index where the new extents will be added. If the new
3742 * extents are being appended, then we just need to (re)allocate and
3743 * initialize the space. Otherwise, if the new extents are being
3744 * inserted into the middle of the existing entries, a bit more work
3745 * is required to make room for the new extents to be inserted. The
3746 * caller is responsible for filling in the new extent entries upon
3751 xfs_ifork_t *ifp, /* inode fork pointer */
3752 xfs_extnum_t idx, /* index to begin adding exts */
3753 int ext_diff) /* number of extents to add */
3755 int byte_diff; /* new bytes being added */
3756 int new_size; /* size of extents after adding */
3757 xfs_extnum_t nextents; /* number of extents in file */
3759 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3760 ASSERT((idx >= 0) && (idx <= nextents));
3761 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
3762 new_size = ifp->if_bytes + byte_diff;
3764 * If the new number of extents (nextents + ext_diff)
3765 * fits inside the inode, then continue to use the inline
3768 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
3769 if (idx < nextents) {
3770 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
3771 &ifp->if_u2.if_inline_ext[idx],
3772 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3773 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
3775 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3776 ifp->if_real_bytes = 0;
3777 ifp->if_lastex = nextents + ext_diff;
3780 * Otherwise use a linear (direct) extent list.
3781 * If the extents are currently inside the inode,
3782 * xfs_iext_realloc_direct will switch us from
3783 * inline to direct extent allocation mode.
3785 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
3786 xfs_iext_realloc_direct(ifp, new_size);
3787 if (idx < nextents) {
3788 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
3789 &ifp->if_u1.if_extents[idx],
3790 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3791 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
3794 /* Indirection array */
3796 xfs_ext_irec_t *erp;
3800 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
3801 if (ifp->if_flags & XFS_IFEXTIREC) {
3802 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3804 xfs_iext_irec_init(ifp);
3805 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3806 erp = ifp->if_u1.if_ext_irec;
3808 /* Extents fit in target extent page */
3809 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3810 if (page_idx < erp->er_extcount) {
3811 memmove(&erp->er_extbuf[page_idx + ext_diff],
3812 &erp->er_extbuf[page_idx],
3813 (erp->er_extcount - page_idx) *
3814 sizeof(xfs_bmbt_rec_t));
3815 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3817 erp->er_extcount += ext_diff;
3818 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3820 /* Insert a new extent page */
3822 xfs_iext_add_indirect_multi(ifp,
3823 erp_idx, page_idx, ext_diff);
3826 * If extent(s) are being appended to the last page in
3827 * the indirection array and the new extent(s) don't fit
3828 * in the page, then erp is NULL and erp_idx is set to
3829 * the next index needed in the indirection array.
3832 int count = ext_diff;
3835 erp = xfs_iext_irec_new(ifp, erp_idx);
3836 erp->er_extcount = count;
3837 count -= MIN(count, (int)XFS_LINEAR_EXTS);
3844 ifp->if_bytes = new_size;
3848 * This is called when incore extents are being added to the indirection
3849 * array and the new extents do not fit in the target extent list. The
3850 * erp_idx parameter contains the irec index for the target extent list
3851 * in the indirection array, and the idx parameter contains the extent
3852 * index within the list. The number of extents being added is stored
3853 * in the count parameter.
3855 * |-------| |-------|
3856 * | | | | idx - number of extents before idx
3858 * | | | | count - number of extents being inserted at idx
3859 * |-------| |-------|
3860 * | count | | nex2 | nex2 - number of extents after idx + count
3861 * |-------| |-------|
3864 xfs_iext_add_indirect_multi(
3865 xfs_ifork_t *ifp, /* inode fork pointer */
3866 int erp_idx, /* target extent irec index */
3867 xfs_extnum_t idx, /* index within target list */
3868 int count) /* new extents being added */
3870 int byte_diff; /* new bytes being added */
3871 xfs_ext_irec_t *erp; /* pointer to irec entry */
3872 xfs_extnum_t ext_diff; /* number of extents to add */
3873 xfs_extnum_t ext_cnt; /* new extents still needed */
3874 xfs_extnum_t nex2; /* extents after idx + count */
3875 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3876 int nlists; /* number of irec's (lists) */
3878 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3879 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3880 nex2 = erp->er_extcount - idx;
3881 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3884 * Save second part of target extent list
3885 * (all extents past */
3887 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3888 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_SLEEP);
3889 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3890 erp->er_extcount -= nex2;
3891 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3892 memset(&erp->er_extbuf[idx], 0, byte_diff);
3896 * Add the new extents to the end of the target
3897 * list, then allocate new irec record(s) and
3898 * extent buffer(s) as needed to store the rest
3899 * of the new extents.
3902 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3904 erp->er_extcount += ext_diff;
3905 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3906 ext_cnt -= ext_diff;
3910 erp = xfs_iext_irec_new(ifp, erp_idx);
3911 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3912 erp->er_extcount = ext_diff;
3913 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3914 ext_cnt -= ext_diff;
3917 /* Add nex2 extents back to indirection array */
3919 xfs_extnum_t ext_avail;
3922 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3923 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3926 * If nex2 extents fit in the current page, append
3927 * nex2_ep after the new extents.
3929 if (nex2 <= ext_avail) {
3930 i = erp->er_extcount;
3933 * Otherwise, check if space is available in the
3936 else if ((erp_idx < nlists - 1) &&
3937 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3938 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3941 /* Create a hole for nex2 extents */
3942 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3943 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3946 * Final choice, create a new extent page for
3951 erp = xfs_iext_irec_new(ifp, erp_idx);
3953 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
3954 kmem_free(nex2_ep, byte_diff);
3955 erp->er_extcount += nex2;
3956 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3961 * This is called when the amount of space required for incore file
3962 * extents needs to be decreased. The ext_diff parameter stores the
3963 * number of extents to be removed and the idx parameter contains
3964 * the extent index where the extents will be removed from.
3966 * If the amount of space needed has decreased below the linear
3967 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3968 * extent array. Otherwise, use kmem_realloc() to adjust the
3969 * size to what is needed.
3973 xfs_ifork_t *ifp, /* inode fork pointer */
3974 xfs_extnum_t idx, /* index to begin removing exts */
3975 int ext_diff) /* number of extents to remove */
3977 xfs_extnum_t nextents; /* number of extents in file */
3978 int new_size; /* size of extents after removal */
3980 ASSERT(ext_diff > 0);
3981 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3982 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3984 if (new_size == 0) {
3985 xfs_iext_destroy(ifp);
3986 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3987 xfs_iext_remove_indirect(ifp, idx, ext_diff);
3988 } else if (ifp->if_real_bytes) {
3989 xfs_iext_remove_direct(ifp, idx, ext_diff);
3991 xfs_iext_remove_inline(ifp, idx, ext_diff);
3993 ifp->if_bytes = new_size;
3997 * This removes ext_diff extents from the inline buffer, beginning
3998 * at extent index idx.
4001 xfs_iext_remove_inline(
4002 xfs_ifork_t *ifp, /* inode fork pointer */
4003 xfs_extnum_t idx, /* index to begin removing exts */
4004 int ext_diff) /* number of extents to remove */
4006 int nextents; /* number of extents in file */
4008 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4009 ASSERT(idx < XFS_INLINE_EXTS);
4010 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4011 ASSERT(((nextents - ext_diff) > 0) &&
4012 (nextents - ext_diff) < XFS_INLINE_EXTS);
4014 if (idx + ext_diff < nextents) {
4015 memmove(&ifp->if_u2.if_inline_ext[idx],
4016 &ifp->if_u2.if_inline_ext[idx + ext_diff],
4017 (nextents - (idx + ext_diff)) *
4018 sizeof(xfs_bmbt_rec_t));
4019 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
4020 0, ext_diff * sizeof(xfs_bmbt_rec_t));
4022 memset(&ifp->if_u2.if_inline_ext[idx], 0,
4023 ext_diff * sizeof(xfs_bmbt_rec_t));
4028 * This removes ext_diff extents from a linear (direct) extent list,
4029 * beginning at extent index idx. If the extents are being removed
4030 * from the end of the list (ie. truncate) then we just need to re-
4031 * allocate the list to remove the extra space. Otherwise, if the
4032 * extents are being removed from the middle of the existing extent
4033 * entries, then we first need to move the extent records beginning
4034 * at idx + ext_diff up in the list to overwrite the records being
4035 * removed, then remove the extra space via kmem_realloc.
4038 xfs_iext_remove_direct(
4039 xfs_ifork_t *ifp, /* inode fork pointer */
4040 xfs_extnum_t idx, /* index to begin removing exts */
4041 int ext_diff) /* number of extents to remove */
4043 xfs_extnum_t nextents; /* number of extents in file */
4044 int new_size; /* size of extents after removal */
4046 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4047 new_size = ifp->if_bytes -
4048 (ext_diff * sizeof(xfs_bmbt_rec_t));
4049 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4051 if (new_size == 0) {
4052 xfs_iext_destroy(ifp);
4055 /* Move extents up in the list (if needed) */
4056 if (idx + ext_diff < nextents) {
4057 memmove(&ifp->if_u1.if_extents[idx],
4058 &ifp->if_u1.if_extents[idx + ext_diff],
4059 (nextents - (idx + ext_diff)) *
4060 sizeof(xfs_bmbt_rec_t));
4062 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
4063 0, ext_diff * sizeof(xfs_bmbt_rec_t));
4065 * Reallocate the direct extent list. If the extents
4066 * will fit inside the inode then xfs_iext_realloc_direct
4067 * will switch from direct to inline extent allocation
4070 xfs_iext_realloc_direct(ifp, new_size);
4071 ifp->if_bytes = new_size;
4075 * This is called when incore extents are being removed from the
4076 * indirection array and the extents being removed span multiple extent
4077 * buffers. The idx parameter contains the file extent index where we
4078 * want to begin removing extents, and the count parameter contains
4079 * how many extents need to be removed.
4081 * |-------| |-------|
4082 * | nex1 | | | nex1 - number of extents before idx
4083 * |-------| | count |
4084 * | | | | count - number of extents being removed at idx
4085 * | count | |-------|
4086 * | | | nex2 | nex2 - number of extents after idx + count
4087 * |-------| |-------|
4090 xfs_iext_remove_indirect(
4091 xfs_ifork_t *ifp, /* inode fork pointer */
4092 xfs_extnum_t idx, /* index to begin removing extents */
4093 int count) /* number of extents to remove */
4095 xfs_ext_irec_t *erp; /* indirection array pointer */
4096 int erp_idx = 0; /* indirection array index */
4097 xfs_extnum_t ext_cnt; /* extents left to remove */
4098 xfs_extnum_t ext_diff; /* extents to remove in current list */
4099 xfs_extnum_t nex1; /* number of extents before idx */
4100 xfs_extnum_t nex2; /* extents after idx + count */
4101 int nlists; /* entries in indirection array */
4102 int page_idx = idx; /* index in target extent list */
4104 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4105 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
4106 ASSERT(erp != NULL);
4107 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4111 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
4112 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
4114 * Check for deletion of entire list;
4115 * xfs_iext_irec_remove() updates extent offsets.
4117 if (ext_diff == erp->er_extcount) {
4118 xfs_iext_irec_remove(ifp, erp_idx);
4119 ext_cnt -= ext_diff;
4122 ASSERT(erp_idx < ifp->if_real_bytes /
4124 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4131 /* Move extents up (if needed) */
4133 memmove(&erp->er_extbuf[nex1],
4134 &erp->er_extbuf[nex1 + ext_diff],
4135 nex2 * sizeof(xfs_bmbt_rec_t));
4137 /* Zero out rest of page */
4138 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
4139 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
4140 /* Update remaining counters */
4141 erp->er_extcount -= ext_diff;
4142 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
4143 ext_cnt -= ext_diff;
4148 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
4149 xfs_iext_irec_compact(ifp);
4153 * Create, destroy, or resize a linear (direct) block of extents.
4156 xfs_iext_realloc_direct(
4157 xfs_ifork_t *ifp, /* inode fork pointer */
4158 int new_size) /* new size of extents */
4160 int rnew_size; /* real new size of extents */
4162 rnew_size = new_size;
4164 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
4165 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
4166 (new_size != ifp->if_real_bytes)));
4168 /* Free extent records */
4169 if (new_size == 0) {
4170 xfs_iext_destroy(ifp);
4172 /* Resize direct extent list and zero any new bytes */
4173 else if (ifp->if_real_bytes) {
4174 /* Check if extents will fit inside the inode */
4175 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
4176 xfs_iext_direct_to_inline(ifp, new_size /
4177 (uint)sizeof(xfs_bmbt_rec_t));
4178 ifp->if_bytes = new_size;
4181 if (!is_power_of_2(new_size)){
4182 rnew_size = xfs_iroundup(new_size);
4184 if (rnew_size != ifp->if_real_bytes) {
4185 ifp->if_u1.if_extents =
4186 kmem_realloc(ifp->if_u1.if_extents,
4191 if (rnew_size > ifp->if_real_bytes) {
4192 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
4193 (uint)sizeof(xfs_bmbt_rec_t)], 0,
4194 rnew_size - ifp->if_real_bytes);
4198 * Switch from the inline extent buffer to a direct
4199 * extent list. Be sure to include the inline extent
4200 * bytes in new_size.
4203 new_size += ifp->if_bytes;
4204 if (!is_power_of_2(new_size)) {
4205 rnew_size = xfs_iroundup(new_size);
4207 xfs_iext_inline_to_direct(ifp, rnew_size);
4209 ifp->if_real_bytes = rnew_size;
4210 ifp->if_bytes = new_size;
4214 * Switch from linear (direct) extent records to inline buffer.
4217 xfs_iext_direct_to_inline(
4218 xfs_ifork_t *ifp, /* inode fork pointer */
4219 xfs_extnum_t nextents) /* number of extents in file */
4221 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4222 ASSERT(nextents <= XFS_INLINE_EXTS);
4224 * The inline buffer was zeroed when we switched
4225 * from inline to direct extent allocation mode,
4226 * so we don't need to clear it here.
4228 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
4229 nextents * sizeof(xfs_bmbt_rec_t));
4230 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
4231 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
4232 ifp->if_real_bytes = 0;
4236 * Switch from inline buffer to linear (direct) extent records.
4237 * new_size should already be rounded up to the next power of 2
4238 * by the caller (when appropriate), so use new_size as it is.
4239 * However, since new_size may be rounded up, we can't update
4240 * if_bytes here. It is the caller's responsibility to update
4241 * if_bytes upon return.
4244 xfs_iext_inline_to_direct(
4245 xfs_ifork_t *ifp, /* inode fork pointer */
4246 int new_size) /* number of extents in file */
4248 ifp->if_u1.if_extents = kmem_alloc(new_size, KM_SLEEP);
4249 memset(ifp->if_u1.if_extents, 0, new_size);
4250 if (ifp->if_bytes) {
4251 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
4253 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4254 sizeof(xfs_bmbt_rec_t));
4256 ifp->if_real_bytes = new_size;
4260 * Resize an extent indirection array to new_size bytes.
4263 xfs_iext_realloc_indirect(
4264 xfs_ifork_t *ifp, /* inode fork pointer */
4265 int new_size) /* new indirection array size */
4267 int nlists; /* number of irec's (ex lists) */
4268 int size; /* current indirection array size */
4270 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4271 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4272 size = nlists * sizeof(xfs_ext_irec_t);
4273 ASSERT(ifp->if_real_bytes);
4274 ASSERT((new_size >= 0) && (new_size != size));
4275 if (new_size == 0) {
4276 xfs_iext_destroy(ifp);
4278 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
4279 kmem_realloc(ifp->if_u1.if_ext_irec,
4280 new_size, size, KM_SLEEP);
4285 * Switch from indirection array to linear (direct) extent allocations.
4288 xfs_iext_indirect_to_direct(
4289 xfs_ifork_t *ifp) /* inode fork pointer */
4291 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
4292 xfs_extnum_t nextents; /* number of extents in file */
4293 int size; /* size of file extents */
4295 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4296 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4297 ASSERT(nextents <= XFS_LINEAR_EXTS);
4298 size = nextents * sizeof(xfs_bmbt_rec_t);
4300 xfs_iext_irec_compact_full(ifp);
4301 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
4303 ep = ifp->if_u1.if_ext_irec->er_extbuf;
4304 kmem_free(ifp->if_u1.if_ext_irec, sizeof(xfs_ext_irec_t));
4305 ifp->if_flags &= ~XFS_IFEXTIREC;
4306 ifp->if_u1.if_extents = ep;
4307 ifp->if_bytes = size;
4308 if (nextents < XFS_LINEAR_EXTS) {
4309 xfs_iext_realloc_direct(ifp, size);
4314 * Free incore file extents.
4318 xfs_ifork_t *ifp) /* inode fork pointer */
4320 if (ifp->if_flags & XFS_IFEXTIREC) {
4324 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4325 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
4326 xfs_iext_irec_remove(ifp, erp_idx);
4328 ifp->if_flags &= ~XFS_IFEXTIREC;
4329 } else if (ifp->if_real_bytes) {
4330 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
4331 } else if (ifp->if_bytes) {
4332 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4333 sizeof(xfs_bmbt_rec_t));
4335 ifp->if_u1.if_extents = NULL;
4336 ifp->if_real_bytes = 0;
4341 * Return a pointer to the extent record for file system block bno.
4343 xfs_bmbt_rec_host_t * /* pointer to found extent record */
4344 xfs_iext_bno_to_ext(
4345 xfs_ifork_t *ifp, /* inode fork pointer */
4346 xfs_fileoff_t bno, /* block number to search for */
4347 xfs_extnum_t *idxp) /* index of target extent */
4349 xfs_bmbt_rec_host_t *base; /* pointer to first extent */
4350 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
4351 xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */
4352 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4353 int high; /* upper boundary in search */
4354 xfs_extnum_t idx = 0; /* index of target extent */
4355 int low; /* lower boundary in search */
4356 xfs_extnum_t nextents; /* number of file extents */
4357 xfs_fileoff_t startoff = 0; /* start offset of extent */
4359 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4360 if (nextents == 0) {
4365 if (ifp->if_flags & XFS_IFEXTIREC) {
4366 /* Find target extent list */
4368 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
4369 base = erp->er_extbuf;
4370 high = erp->er_extcount - 1;
4372 base = ifp->if_u1.if_extents;
4373 high = nextents - 1;
4375 /* Binary search extent records */
4376 while (low <= high) {
4377 idx = (low + high) >> 1;
4379 startoff = xfs_bmbt_get_startoff(ep);
4380 blockcount = xfs_bmbt_get_blockcount(ep);
4381 if (bno < startoff) {
4383 } else if (bno >= startoff + blockcount) {
4386 /* Convert back to file-based extent index */
4387 if (ifp->if_flags & XFS_IFEXTIREC) {
4388 idx += erp->er_extoff;
4394 /* Convert back to file-based extent index */
4395 if (ifp->if_flags & XFS_IFEXTIREC) {
4396 idx += erp->er_extoff;
4398 if (bno >= startoff + blockcount) {
4399 if (++idx == nextents) {
4402 ep = xfs_iext_get_ext(ifp, idx);
4410 * Return a pointer to the indirection array entry containing the
4411 * extent record for filesystem block bno. Store the index of the
4412 * target irec in *erp_idxp.
4414 xfs_ext_irec_t * /* pointer to found extent record */
4415 xfs_iext_bno_to_irec(
4416 xfs_ifork_t *ifp, /* inode fork pointer */
4417 xfs_fileoff_t bno, /* block number to search for */
4418 int *erp_idxp) /* irec index of target ext list */
4420 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4421 xfs_ext_irec_t *erp_next; /* next indirection array entry */
4422 int erp_idx; /* indirection array index */
4423 int nlists; /* number of extent irec's (lists) */
4424 int high; /* binary search upper limit */
4425 int low; /* binary search lower limit */
4427 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4428 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4432 while (low <= high) {
4433 erp_idx = (low + high) >> 1;
4434 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4435 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
4436 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
4438 } else if (erp_next && bno >=
4439 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
4445 *erp_idxp = erp_idx;
4450 * Return a pointer to the indirection array entry containing the
4451 * extent record at file extent index *idxp. Store the index of the
4452 * target irec in *erp_idxp and store the page index of the target
4453 * extent record in *idxp.
4456 xfs_iext_idx_to_irec(
4457 xfs_ifork_t *ifp, /* inode fork pointer */
4458 xfs_extnum_t *idxp, /* extent index (file -> page) */
4459 int *erp_idxp, /* pointer to target irec */
4460 int realloc) /* new bytes were just added */
4462 xfs_ext_irec_t *prev; /* pointer to previous irec */
4463 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
4464 int erp_idx; /* indirection array index */
4465 int nlists; /* number of irec's (ex lists) */
4466 int high; /* binary search upper limit */
4467 int low; /* binary search lower limit */
4468 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
4470 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4471 ASSERT(page_idx >= 0 && page_idx <=
4472 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
4473 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4478 /* Binary search extent irec's */
4479 while (low <= high) {
4480 erp_idx = (low + high) >> 1;
4481 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4482 prev = erp_idx > 0 ? erp - 1 : NULL;
4483 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
4484 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
4486 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
4487 (page_idx == erp->er_extoff + erp->er_extcount &&
4490 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
4491 erp->er_extcount == XFS_LINEAR_EXTS) {
4495 erp = erp_idx < nlists ? erp + 1 : NULL;
4498 page_idx -= erp->er_extoff;
4503 *erp_idxp = erp_idx;
4508 * Allocate and initialize an indirection array once the space needed
4509 * for incore extents increases above XFS_IEXT_BUFSZ.
4513 xfs_ifork_t *ifp) /* inode fork pointer */
4515 xfs_ext_irec_t *erp; /* indirection array pointer */
4516 xfs_extnum_t nextents; /* number of extents in file */
4518 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4519 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4520 ASSERT(nextents <= XFS_LINEAR_EXTS);
4522 erp = (xfs_ext_irec_t *)
4523 kmem_alloc(sizeof(xfs_ext_irec_t), KM_SLEEP);
4525 if (nextents == 0) {
4526 ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
4527 } else if (!ifp->if_real_bytes) {
4528 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
4529 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
4530 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
4532 erp->er_extbuf = ifp->if_u1.if_extents;
4533 erp->er_extcount = nextents;
4536 ifp->if_flags |= XFS_IFEXTIREC;
4537 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
4538 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
4539 ifp->if_u1.if_ext_irec = erp;
4545 * Allocate and initialize a new entry in the indirection array.
4549 xfs_ifork_t *ifp, /* inode fork pointer */
4550 int erp_idx) /* index for new irec */
4552 xfs_ext_irec_t *erp; /* indirection array pointer */
4553 int i; /* loop counter */
4554 int nlists; /* number of irec's (ex lists) */
4556 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4557 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4559 /* Resize indirection array */
4560 xfs_iext_realloc_indirect(ifp, ++nlists *
4561 sizeof(xfs_ext_irec_t));
4563 * Move records down in the array so the
4564 * new page can use erp_idx.
4566 erp = ifp->if_u1.if_ext_irec;
4567 for (i = nlists - 1; i > erp_idx; i--) {
4568 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
4570 ASSERT(i == erp_idx);
4572 /* Initialize new extent record */
4573 erp = ifp->if_u1.if_ext_irec;
4574 erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
4575 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4576 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
4577 erp[erp_idx].er_extcount = 0;
4578 erp[erp_idx].er_extoff = erp_idx > 0 ?
4579 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
4580 return (&erp[erp_idx]);
4584 * Remove a record from the indirection array.
4587 xfs_iext_irec_remove(
4588 xfs_ifork_t *ifp, /* inode fork pointer */
4589 int erp_idx) /* irec index to remove */
4591 xfs_ext_irec_t *erp; /* indirection array pointer */
4592 int i; /* loop counter */
4593 int nlists; /* number of irec's (ex lists) */
4595 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4596 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4597 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4598 if (erp->er_extbuf) {
4599 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
4601 kmem_free(erp->er_extbuf, XFS_IEXT_BUFSZ);
4603 /* Compact extent records */
4604 erp = ifp->if_u1.if_ext_irec;
4605 for (i = erp_idx; i < nlists - 1; i++) {
4606 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
4609 * Manually free the last extent record from the indirection
4610 * array. A call to xfs_iext_realloc_indirect() with a size
4611 * of zero would result in a call to xfs_iext_destroy() which
4612 * would in turn call this function again, creating a nasty
4616 xfs_iext_realloc_indirect(ifp,
4617 nlists * sizeof(xfs_ext_irec_t));
4619 kmem_free(ifp->if_u1.if_ext_irec,
4620 sizeof(xfs_ext_irec_t));
4622 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4626 * This is called to clean up large amounts of unused memory allocated
4627 * by the indirection array. Before compacting anything though, verify
4628 * that the indirection array is still needed and switch back to the
4629 * linear extent list (or even the inline buffer) if possible. The
4630 * compaction policy is as follows:
4632 * Full Compaction: Extents fit into a single page (or inline buffer)
4633 * Full Compaction: Extents occupy less than 10% of allocated space
4634 * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
4635 * No Compaction: Extents occupy at least 50% of allocated space
4638 xfs_iext_irec_compact(
4639 xfs_ifork_t *ifp) /* inode fork pointer */
4641 xfs_extnum_t nextents; /* number of extents in file */
4642 int nlists; /* number of irec's (ex lists) */
4644 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4645 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4646 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4648 if (nextents == 0) {
4649 xfs_iext_destroy(ifp);
4650 } else if (nextents <= XFS_INLINE_EXTS) {
4651 xfs_iext_indirect_to_direct(ifp);
4652 xfs_iext_direct_to_inline(ifp, nextents);
4653 } else if (nextents <= XFS_LINEAR_EXTS) {
4654 xfs_iext_indirect_to_direct(ifp);
4655 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 3) {
4656 xfs_iext_irec_compact_full(ifp);
4657 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
4658 xfs_iext_irec_compact_pages(ifp);
4663 * Combine extents from neighboring extent pages.
4666 xfs_iext_irec_compact_pages(
4667 xfs_ifork_t *ifp) /* inode fork pointer */
4669 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
4670 int erp_idx = 0; /* indirection array index */
4671 int nlists; /* number of irec's (ex lists) */
4673 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4674 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4675 while (erp_idx < nlists - 1) {
4676 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4678 if (erp_next->er_extcount <=
4679 (XFS_LINEAR_EXTS - erp->er_extcount)) {
4680 memmove(&erp->er_extbuf[erp->er_extcount],
4681 erp_next->er_extbuf, erp_next->er_extcount *
4682 sizeof(xfs_bmbt_rec_t));
4683 erp->er_extcount += erp_next->er_extcount;
4685 * Free page before removing extent record
4686 * so er_extoffs don't get modified in
4687 * xfs_iext_irec_remove.
4689 kmem_free(erp_next->er_extbuf, XFS_IEXT_BUFSZ);
4690 erp_next->er_extbuf = NULL;
4691 xfs_iext_irec_remove(ifp, erp_idx + 1);
4692 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4700 * Fully compact the extent records managed by the indirection array.
4703 xfs_iext_irec_compact_full(
4704 xfs_ifork_t *ifp) /* inode fork pointer */
4706 xfs_bmbt_rec_host_t *ep, *ep_next; /* extent record pointers */
4707 xfs_ext_irec_t *erp, *erp_next; /* extent irec pointers */
4708 int erp_idx = 0; /* extent irec index */
4709 int ext_avail; /* empty entries in ex list */
4710 int ext_diff; /* number of exts to add */
4711 int nlists; /* number of irec's (ex lists) */
4713 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4714 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4715 erp = ifp->if_u1.if_ext_irec;
4716 ep = &erp->er_extbuf[erp->er_extcount];
4718 ep_next = erp_next->er_extbuf;
4719 while (erp_idx < nlists - 1) {
4720 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
4721 ext_diff = MIN(ext_avail, erp_next->er_extcount);
4722 memcpy(ep, ep_next, ext_diff * sizeof(xfs_bmbt_rec_t));
4723 erp->er_extcount += ext_diff;
4724 erp_next->er_extcount -= ext_diff;
4725 /* Remove next page */
4726 if (erp_next->er_extcount == 0) {
4728 * Free page before removing extent record
4729 * so er_extoffs don't get modified in
4730 * xfs_iext_irec_remove.
4732 kmem_free(erp_next->er_extbuf,
4733 erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
4734 erp_next->er_extbuf = NULL;
4735 xfs_iext_irec_remove(ifp, erp_idx + 1);
4736 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4737 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4738 /* Update next page */
4740 /* Move rest of page up to become next new page */
4741 memmove(erp_next->er_extbuf, ep_next,
4742 erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
4743 ep_next = erp_next->er_extbuf;
4744 memset(&ep_next[erp_next->er_extcount], 0,
4745 (XFS_LINEAR_EXTS - erp_next->er_extcount) *
4746 sizeof(xfs_bmbt_rec_t));
4748 if (erp->er_extcount == XFS_LINEAR_EXTS) {
4750 if (erp_idx < nlists)
4751 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4755 ep = &erp->er_extbuf[erp->er_extcount];
4757 ep_next = erp_next->er_extbuf;
4762 * This is called to update the er_extoff field in the indirection
4763 * array when extents have been added or removed from one of the
4764 * extent lists. erp_idx contains the irec index to begin updating
4765 * at and ext_diff contains the number of extents that were added
4769 xfs_iext_irec_update_extoffs(
4770 xfs_ifork_t *ifp, /* inode fork pointer */
4771 int erp_idx, /* irec index to update */
4772 int ext_diff) /* number of new extents */
4774 int i; /* loop counter */
4775 int nlists; /* number of irec's (ex lists */
4777 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4778 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4779 for (i = erp_idx; i < nlists; i++) {
4780 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;