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"
52 #include "xfs_vnodeops.h"
54 kmem_zone_t *xfs_ifork_zone;
55 kmem_zone_t *xfs_inode_zone;
56 kmem_zone_t *xfs_icluster_zone;
59 * Used in xfs_itruncate(). This is the maximum number of extents
60 * freed from a file in a single transaction.
62 #define XFS_ITRUNC_MAX_EXTENTS 2
64 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
65 STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
66 STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
67 STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
71 * Make sure that the extents in the given memory buffer
81 xfs_bmbt_rec_host_t rec;
84 for (i = 0; i < nrecs; i++) {
85 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
86 rec.l0 = get_unaligned(&ep->l0);
87 rec.l1 = get_unaligned(&ep->l1);
88 xfs_bmbt_get_all(&rec, &irec);
89 if (fmt == XFS_EXTFMT_NOSTATE)
90 ASSERT(irec.br_state == XFS_EXT_NORM);
94 #define xfs_validate_extents(ifp, nrecs, fmt)
98 * Check that none of the inode's in the buffer have a next
99 * unlinked field of 0.
111 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
113 for (i = 0; i < j; i++) {
114 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
115 i * mp->m_sb.sb_inodesize);
116 if (!dip->di_next_unlinked) {
117 xfs_fs_cmn_err(CE_ALERT, mp,
118 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
120 ASSERT(dip->di_next_unlinked);
127 * This routine is called to map an inode number within a file
128 * system to the buffer containing the on-disk version of the
129 * inode. It returns a pointer to the buffer containing the
130 * on-disk inode in the bpp parameter, and in the dip parameter
131 * it returns a pointer to the on-disk inode within that buffer.
133 * If a non-zero error is returned, then the contents of bpp and
134 * dipp are undefined.
136 * Use xfs_imap() to determine the size and location of the
137 * buffer to read from disk.
155 * Call the space management code to find the location of the
159 error = xfs_imap(mp, tp, ino, &imap, XFS_IMAP_LOOKUP);
162 "xfs_inotobp: xfs_imap() returned an "
163 "error %d on %s. Returning error.", error, mp->m_fsname);
168 * If the inode number maps to a block outside the bounds of the
169 * file system then return NULL rather than calling read_buf
170 * and panicing when we get an error from the driver.
172 if ((imap.im_blkno + imap.im_len) >
173 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
175 "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
176 "of the file system %s. Returning EINVAL.",
177 (unsigned long long)imap.im_blkno,
178 imap.im_len, mp->m_fsname);
179 return XFS_ERROR(EINVAL);
183 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
184 * default to just a read_buf() call.
186 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
187 (int)imap.im_len, XFS_BUF_LOCK, &bp);
191 "xfs_inotobp: xfs_trans_read_buf() returned an "
192 "error %d on %s. Returning error.", error, mp->m_fsname);
195 dip = (xfs_dinode_t *)xfs_buf_offset(bp, 0);
197 be16_to_cpu(dip->di_core.di_magic) == XFS_DINODE_MAGIC &&
198 XFS_DINODE_GOOD_VERSION(dip->di_core.di_version);
199 if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP,
200 XFS_RANDOM_ITOBP_INOTOBP))) {
201 XFS_CORRUPTION_ERROR("xfs_inotobp", XFS_ERRLEVEL_LOW, mp, dip);
202 xfs_trans_brelse(tp, bp);
204 "xfs_inotobp: XFS_TEST_ERROR() returned an "
205 "error on %s. Returning EFSCORRUPTED.", mp->m_fsname);
206 return XFS_ERROR(EFSCORRUPTED);
209 xfs_inobp_check(mp, bp);
212 * Set *dipp to point to the on-disk inode in the buffer.
214 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
216 *offset = imap.im_boffset;
222 * This routine is called to map an inode to the buffer containing
223 * the on-disk version of the inode. It returns a pointer to the
224 * buffer containing the on-disk inode in the bpp parameter, and in
225 * the dip parameter it returns a pointer to the on-disk inode within
228 * If a non-zero error is returned, then the contents of bpp and
229 * dipp are undefined.
231 * If the inode is new and has not yet been initialized, use xfs_imap()
232 * to determine the size and location of the buffer to read from disk.
233 * If the inode has already been mapped to its buffer and read in once,
234 * then use the mapping information stored in the inode rather than
235 * calling xfs_imap(). This allows us to avoid the overhead of looking
236 * at the inode btree for small block file systems (see xfs_dilocate()).
237 * We can tell whether the inode has been mapped in before by comparing
238 * its disk block address to 0. Only uninitialized inodes will have
239 * 0 for the disk block address.
257 if (ip->i_blkno == (xfs_daddr_t)0) {
259 * Call the space management code to find the location of the
263 if ((error = xfs_imap(mp, tp, ip->i_ino, &imap,
264 XFS_IMAP_LOOKUP | imap_flags)))
268 * If the inode number maps to a block outside the bounds
269 * of the file system then return NULL rather than calling
270 * read_buf and panicing when we get an error from the
273 if ((imap.im_blkno + imap.im_len) >
274 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
276 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
277 "(imap.im_blkno (0x%llx) "
278 "+ imap.im_len (0x%llx)) > "
279 " XFS_FSB_TO_BB(mp, "
280 "mp->m_sb.sb_dblocks) (0x%llx)",
281 (unsigned long long) imap.im_blkno,
282 (unsigned long long) imap.im_len,
283 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
285 return XFS_ERROR(EINVAL);
289 * Fill in the fields in the inode that will be used to
290 * map the inode to its buffer from now on.
292 ip->i_blkno = imap.im_blkno;
293 ip->i_len = imap.im_len;
294 ip->i_boffset = imap.im_boffset;
297 * We've already mapped the inode once, so just use the
298 * mapping that we saved the first time.
300 imap.im_blkno = ip->i_blkno;
301 imap.im_len = ip->i_len;
302 imap.im_boffset = ip->i_boffset;
304 ASSERT(bno == 0 || bno == imap.im_blkno);
307 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
308 * default to just a read_buf() call.
310 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
311 (int)imap.im_len, XFS_BUF_LOCK, &bp);
314 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
315 "xfs_trans_read_buf() returned error %d, "
316 "imap.im_blkno 0x%llx, imap.im_len 0x%llx",
317 error, (unsigned long long) imap.im_blkno,
318 (unsigned long long) imap.im_len);
324 * Validate the magic number and version of every inode in the buffer
325 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
326 * No validation is done here in userspace (xfs_repair).
328 #if !defined(__KERNEL__)
331 ni = BBTOB(imap.im_len) >> mp->m_sb.sb_inodelog;
332 #else /* usual case */
336 for (i = 0; i < ni; i++) {
340 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
341 (i << mp->m_sb.sb_inodelog));
342 di_ok = be16_to_cpu(dip->di_core.di_magic) == XFS_DINODE_MAGIC &&
343 XFS_DINODE_GOOD_VERSION(dip->di_core.di_version);
344 if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
345 XFS_ERRTAG_ITOBP_INOTOBP,
346 XFS_RANDOM_ITOBP_INOTOBP))) {
347 if (imap_flags & XFS_IMAP_BULKSTAT) {
348 xfs_trans_brelse(tp, bp);
349 return XFS_ERROR(EINVAL);
353 "Device %s - bad inode magic/vsn "
354 "daddr %lld #%d (magic=%x)",
355 XFS_BUFTARG_NAME(mp->m_ddev_targp),
356 (unsigned long long)imap.im_blkno, i,
357 be16_to_cpu(dip->di_core.di_magic));
359 XFS_CORRUPTION_ERROR("xfs_itobp", XFS_ERRLEVEL_HIGH,
361 xfs_trans_brelse(tp, bp);
362 return XFS_ERROR(EFSCORRUPTED);
366 xfs_inobp_check(mp, bp);
369 * Mark the buffer as an inode buffer now that it looks good
371 XFS_BUF_SET_VTYPE(bp, B_FS_INO);
374 * Set *dipp to point to the on-disk inode in the buffer.
376 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
382 * Move inode type and inode format specific information from the
383 * on-disk inode to the in-core inode. For fifos, devs, and sockets
384 * this means set if_rdev to the proper value. For files, directories,
385 * and symlinks this means to bring in the in-line data or extent
386 * pointers. For a file in B-tree format, only the root is immediately
387 * brought in-core. The rest will be in-lined in if_extents when it
388 * is first referenced (see xfs_iread_extents()).
395 xfs_attr_shortform_t *atp;
399 ip->i_df.if_ext_max =
400 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
403 if (unlikely(be32_to_cpu(dip->di_core.di_nextents) +
404 be16_to_cpu(dip->di_core.di_anextents) >
405 be64_to_cpu(dip->di_core.di_nblocks))) {
406 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
407 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
408 (unsigned long long)ip->i_ino,
409 (int)(be32_to_cpu(dip->di_core.di_nextents) +
410 be16_to_cpu(dip->di_core.di_anextents)),
412 be64_to_cpu(dip->di_core.di_nblocks));
413 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
415 return XFS_ERROR(EFSCORRUPTED);
418 if (unlikely(dip->di_core.di_forkoff > ip->i_mount->m_sb.sb_inodesize)) {
419 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
420 "corrupt dinode %Lu, forkoff = 0x%x.",
421 (unsigned long long)ip->i_ino,
422 dip->di_core.di_forkoff);
423 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
425 return XFS_ERROR(EFSCORRUPTED);
428 switch (ip->i_d.di_mode & S_IFMT) {
433 if (unlikely(dip->di_core.di_format != XFS_DINODE_FMT_DEV)) {
434 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
436 return XFS_ERROR(EFSCORRUPTED);
440 ip->i_df.if_u2.if_rdev = be32_to_cpu(dip->di_u.di_dev);
446 switch (dip->di_core.di_format) {
447 case XFS_DINODE_FMT_LOCAL:
449 * no local regular files yet
451 if (unlikely((be16_to_cpu(dip->di_core.di_mode) & S_IFMT) == S_IFREG)) {
452 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
454 "(local format for regular file).",
455 (unsigned long long) ip->i_ino);
456 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
459 return XFS_ERROR(EFSCORRUPTED);
462 di_size = be64_to_cpu(dip->di_core.di_size);
463 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
464 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
466 "(bad size %Ld for local inode).",
467 (unsigned long long) ip->i_ino,
468 (long long) di_size);
469 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
472 return XFS_ERROR(EFSCORRUPTED);
476 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
478 case XFS_DINODE_FMT_EXTENTS:
479 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
481 case XFS_DINODE_FMT_BTREE:
482 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
485 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
487 return XFS_ERROR(EFSCORRUPTED);
492 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
493 return XFS_ERROR(EFSCORRUPTED);
498 if (!XFS_DFORK_Q(dip))
500 ASSERT(ip->i_afp == NULL);
501 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
502 ip->i_afp->if_ext_max =
503 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
504 switch (dip->di_core.di_aformat) {
505 case XFS_DINODE_FMT_LOCAL:
506 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
507 size = be16_to_cpu(atp->hdr.totsize);
508 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
510 case XFS_DINODE_FMT_EXTENTS:
511 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
513 case XFS_DINODE_FMT_BTREE:
514 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
517 error = XFS_ERROR(EFSCORRUPTED);
521 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
523 xfs_idestroy_fork(ip, XFS_DATA_FORK);
529 * The file is in-lined in the on-disk inode.
530 * If it fits into if_inline_data, then copy
531 * it there, otherwise allocate a buffer for it
532 * and copy the data there. Either way, set
533 * if_data to point at the data.
534 * If we allocate a buffer for the data, make
535 * sure that its size is a multiple of 4 and
536 * record the real size in i_real_bytes.
549 * If the size is unreasonable, then something
550 * is wrong and we just bail out rather than crash in
551 * kmem_alloc() or memcpy() below.
553 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
554 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
556 "(bad size %d for local fork, size = %d).",
557 (unsigned long long) ip->i_ino, size,
558 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
559 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
561 return XFS_ERROR(EFSCORRUPTED);
563 ifp = XFS_IFORK_PTR(ip, whichfork);
566 ifp->if_u1.if_data = NULL;
567 else if (size <= sizeof(ifp->if_u2.if_inline_data))
568 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
570 real_size = roundup(size, 4);
571 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
573 ifp->if_bytes = size;
574 ifp->if_real_bytes = real_size;
576 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
577 ifp->if_flags &= ~XFS_IFEXTENTS;
578 ifp->if_flags |= XFS_IFINLINE;
583 * The file consists of a set of extents all
584 * of which fit into the on-disk inode.
585 * If there are few enough extents to fit into
586 * the if_inline_ext, then copy them there.
587 * Otherwise allocate a buffer for them and copy
588 * them into it. Either way, set if_extents
589 * to point at the extents.
603 ifp = XFS_IFORK_PTR(ip, whichfork);
604 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
605 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
608 * If the number of extents is unreasonable, then something
609 * is wrong and we just bail out rather than crash in
610 * kmem_alloc() or memcpy() below.
612 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
613 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
614 "corrupt inode %Lu ((a)extents = %d).",
615 (unsigned long long) ip->i_ino, nex);
616 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
618 return XFS_ERROR(EFSCORRUPTED);
621 ifp->if_real_bytes = 0;
623 ifp->if_u1.if_extents = NULL;
624 else if (nex <= XFS_INLINE_EXTS)
625 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
627 xfs_iext_add(ifp, 0, nex);
629 ifp->if_bytes = size;
631 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
632 xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip));
633 for (i = 0; i < nex; i++, dp++) {
634 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
635 ep->l0 = be64_to_cpu(get_unaligned(&dp->l0));
636 ep->l1 = be64_to_cpu(get_unaligned(&dp->l1));
638 XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork);
639 if (whichfork != XFS_DATA_FORK ||
640 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
641 if (unlikely(xfs_check_nostate_extents(
643 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
646 return XFS_ERROR(EFSCORRUPTED);
649 ifp->if_flags |= XFS_IFEXTENTS;
654 * The file has too many extents to fit into
655 * the inode, so they are in B-tree format.
656 * Allocate a buffer for the root of the B-tree
657 * and copy the root into it. The i_extents
658 * field will remain NULL until all of the
659 * extents are read in (when they are needed).
667 xfs_bmdr_block_t *dfp;
673 ifp = XFS_IFORK_PTR(ip, whichfork);
674 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
675 size = XFS_BMAP_BROOT_SPACE(dfp);
676 nrecs = XFS_BMAP_BROOT_NUMRECS(dfp);
679 * blow out if -- fork has less extents than can fit in
680 * fork (fork shouldn't be a btree format), root btree
681 * block has more records than can fit into the fork,
682 * or the number of extents is greater than the number of
685 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
686 || XFS_BMDR_SPACE_CALC(nrecs) >
687 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
688 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
689 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
690 "corrupt inode %Lu (btree).",
691 (unsigned long long) ip->i_ino);
692 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
694 return XFS_ERROR(EFSCORRUPTED);
697 ifp->if_broot_bytes = size;
698 ifp->if_broot = kmem_alloc(size, KM_SLEEP);
699 ASSERT(ifp->if_broot != NULL);
701 * Copy and convert from the on-disk structure
702 * to the in-memory structure.
704 xfs_bmdr_to_bmbt(dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
705 ifp->if_broot, size);
706 ifp->if_flags &= ~XFS_IFEXTENTS;
707 ifp->if_flags |= XFS_IFBROOT;
713 xfs_dinode_from_disk(
715 xfs_dinode_core_t *from)
717 to->di_magic = be16_to_cpu(from->di_magic);
718 to->di_mode = be16_to_cpu(from->di_mode);
719 to->di_version = from ->di_version;
720 to->di_format = from->di_format;
721 to->di_onlink = be16_to_cpu(from->di_onlink);
722 to->di_uid = be32_to_cpu(from->di_uid);
723 to->di_gid = be32_to_cpu(from->di_gid);
724 to->di_nlink = be32_to_cpu(from->di_nlink);
725 to->di_projid = be16_to_cpu(from->di_projid);
726 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
727 to->di_flushiter = be16_to_cpu(from->di_flushiter);
728 to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec);
729 to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec);
730 to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec);
731 to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec);
732 to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec);
733 to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec);
734 to->di_size = be64_to_cpu(from->di_size);
735 to->di_nblocks = be64_to_cpu(from->di_nblocks);
736 to->di_extsize = be32_to_cpu(from->di_extsize);
737 to->di_nextents = be32_to_cpu(from->di_nextents);
738 to->di_anextents = be16_to_cpu(from->di_anextents);
739 to->di_forkoff = from->di_forkoff;
740 to->di_aformat = from->di_aformat;
741 to->di_dmevmask = be32_to_cpu(from->di_dmevmask);
742 to->di_dmstate = be16_to_cpu(from->di_dmstate);
743 to->di_flags = be16_to_cpu(from->di_flags);
744 to->di_gen = be32_to_cpu(from->di_gen);
749 xfs_dinode_core_t *to,
750 xfs_icdinode_t *from)
752 to->di_magic = cpu_to_be16(from->di_magic);
753 to->di_mode = cpu_to_be16(from->di_mode);
754 to->di_version = from ->di_version;
755 to->di_format = from->di_format;
756 to->di_onlink = cpu_to_be16(from->di_onlink);
757 to->di_uid = cpu_to_be32(from->di_uid);
758 to->di_gid = cpu_to_be32(from->di_gid);
759 to->di_nlink = cpu_to_be32(from->di_nlink);
760 to->di_projid = cpu_to_be16(from->di_projid);
761 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
762 to->di_flushiter = cpu_to_be16(from->di_flushiter);
763 to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec);
764 to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec);
765 to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec);
766 to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec);
767 to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec);
768 to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec);
769 to->di_size = cpu_to_be64(from->di_size);
770 to->di_nblocks = cpu_to_be64(from->di_nblocks);
771 to->di_extsize = cpu_to_be32(from->di_extsize);
772 to->di_nextents = cpu_to_be32(from->di_nextents);
773 to->di_anextents = cpu_to_be16(from->di_anextents);
774 to->di_forkoff = from->di_forkoff;
775 to->di_aformat = from->di_aformat;
776 to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
777 to->di_dmstate = cpu_to_be16(from->di_dmstate);
778 to->di_flags = cpu_to_be16(from->di_flags);
779 to->di_gen = cpu_to_be32(from->di_gen);
788 if (di_flags & XFS_DIFLAG_ANY) {
789 if (di_flags & XFS_DIFLAG_REALTIME)
790 flags |= XFS_XFLAG_REALTIME;
791 if (di_flags & XFS_DIFLAG_PREALLOC)
792 flags |= XFS_XFLAG_PREALLOC;
793 if (di_flags & XFS_DIFLAG_IMMUTABLE)
794 flags |= XFS_XFLAG_IMMUTABLE;
795 if (di_flags & XFS_DIFLAG_APPEND)
796 flags |= XFS_XFLAG_APPEND;
797 if (di_flags & XFS_DIFLAG_SYNC)
798 flags |= XFS_XFLAG_SYNC;
799 if (di_flags & XFS_DIFLAG_NOATIME)
800 flags |= XFS_XFLAG_NOATIME;
801 if (di_flags & XFS_DIFLAG_NODUMP)
802 flags |= XFS_XFLAG_NODUMP;
803 if (di_flags & XFS_DIFLAG_RTINHERIT)
804 flags |= XFS_XFLAG_RTINHERIT;
805 if (di_flags & XFS_DIFLAG_PROJINHERIT)
806 flags |= XFS_XFLAG_PROJINHERIT;
807 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
808 flags |= XFS_XFLAG_NOSYMLINKS;
809 if (di_flags & XFS_DIFLAG_EXTSIZE)
810 flags |= XFS_XFLAG_EXTSIZE;
811 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
812 flags |= XFS_XFLAG_EXTSZINHERIT;
813 if (di_flags & XFS_DIFLAG_NODEFRAG)
814 flags |= XFS_XFLAG_NODEFRAG;
815 if (di_flags & XFS_DIFLAG_FILESTREAM)
816 flags |= XFS_XFLAG_FILESTREAM;
826 xfs_icdinode_t *dic = &ip->i_d;
828 return _xfs_dic2xflags(dic->di_flags) |
829 (XFS_CFORK_Q(dic) ? XFS_XFLAG_HASATTR : 0);
834 xfs_dinode_core_t *dic)
836 return _xfs_dic2xflags(be16_to_cpu(dic->di_flags)) |
837 (XFS_CFORK_Q_DISK(dic) ? XFS_XFLAG_HASATTR : 0);
841 * Given a mount structure and an inode number, return a pointer
842 * to a newly allocated in-core inode corresponding to the given
845 * Initialize the inode's attributes and extent pointers if it
846 * already has them (it will not if the inode has no links).
862 ASSERT(xfs_inode_zone != NULL);
864 ip = kmem_zone_zalloc(xfs_inode_zone, KM_SLEEP);
867 atomic_set(&ip->i_iocount, 0);
868 spin_lock_init(&ip->i_flags_lock);
871 * Get pointer's to the on-disk inode and the buffer containing it.
872 * If the inode number refers to a block outside the file system
873 * then xfs_itobp() will return NULL. In this case we should
874 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
875 * know that this is a new incore inode.
877 error = xfs_itobp(mp, tp, ip, &dip, &bp, bno, imap_flags);
879 kmem_zone_free(xfs_inode_zone, ip);
884 * Initialize inode's trace buffers.
885 * Do this before xfs_iformat in case it adds entries.
887 #ifdef XFS_INODE_TRACE
888 ip->i_trace = ktrace_alloc(INODE_TRACE_SIZE, KM_SLEEP);
890 #ifdef XFS_BMAP_TRACE
891 ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_SLEEP);
893 #ifdef XFS_BMBT_TRACE
894 ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_SLEEP);
897 ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_SLEEP);
899 #ifdef XFS_ILOCK_TRACE
900 ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_SLEEP);
902 #ifdef XFS_DIR2_TRACE
903 ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_SLEEP);
907 * If we got something that isn't an inode it means someone
908 * (nfs or dmi) has a stale handle.
910 if (be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC) {
911 kmem_zone_free(xfs_inode_zone, ip);
912 xfs_trans_brelse(tp, bp);
914 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
915 "dip->di_core.di_magic (0x%x) != "
916 "XFS_DINODE_MAGIC (0x%x)",
917 be16_to_cpu(dip->di_core.di_magic),
920 return XFS_ERROR(EINVAL);
924 * If the on-disk inode is already linked to a directory
925 * entry, copy all of the inode into the in-core inode.
926 * xfs_iformat() handles copying in the inode format
927 * specific information.
928 * Otherwise, just get the truly permanent information.
930 if (dip->di_core.di_mode) {
931 xfs_dinode_from_disk(&ip->i_d, &dip->di_core);
932 error = xfs_iformat(ip, dip);
934 kmem_zone_free(xfs_inode_zone, ip);
935 xfs_trans_brelse(tp, bp);
937 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
938 "xfs_iformat() returned error %d",
944 ip->i_d.di_magic = be16_to_cpu(dip->di_core.di_magic);
945 ip->i_d.di_version = dip->di_core.di_version;
946 ip->i_d.di_gen = be32_to_cpu(dip->di_core.di_gen);
947 ip->i_d.di_flushiter = be16_to_cpu(dip->di_core.di_flushiter);
949 * Make sure to pull in the mode here as well in
950 * case the inode is released without being used.
951 * This ensures that xfs_inactive() will see that
952 * the inode is already free and not try to mess
953 * with the uninitialized part of it.
957 * Initialize the per-fork minima and maxima for a new
958 * inode here. xfs_iformat will do it for old inodes.
960 ip->i_df.if_ext_max =
961 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
964 INIT_LIST_HEAD(&ip->i_reclaim);
967 * The inode format changed when we moved the link count and
968 * made it 32 bits long. If this is an old format inode,
969 * convert it in memory to look like a new one. If it gets
970 * flushed to disk we will convert back before flushing or
971 * logging it. We zero out the new projid field and the old link
972 * count field. We'll handle clearing the pad field (the remains
973 * of the old uuid field) when we actually convert the inode to
974 * the new format. We don't change the version number so that we
975 * can distinguish this from a real new format inode.
977 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
978 ip->i_d.di_nlink = ip->i_d.di_onlink;
979 ip->i_d.di_onlink = 0;
980 ip->i_d.di_projid = 0;
983 ip->i_delayed_blks = 0;
984 ip->i_size = ip->i_d.di_size;
987 * Mark the buffer containing the inode as something to keep
988 * around for a while. This helps to keep recently accessed
989 * meta-data in-core longer.
991 XFS_BUF_SET_REF(bp, XFS_INO_REF);
994 * Use xfs_trans_brelse() to release the buffer containing the
995 * on-disk inode, because it was acquired with xfs_trans_read_buf()
996 * in xfs_itobp() above. If tp is NULL, this is just a normal
997 * brelse(). If we're within a transaction, then xfs_trans_brelse()
998 * will only release the buffer if it is not dirty within the
999 * transaction. It will be OK to release the buffer in this case,
1000 * because inodes on disk are never destroyed and we will be
1001 * locking the new in-core inode before putting it in the hash
1002 * table where other processes can find it. Thus we don't have
1003 * to worry about the inode being changed just because we released
1006 xfs_trans_brelse(tp, bp);
1012 * Read in extents from a btree-format inode.
1013 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1023 xfs_extnum_t nextents;
1026 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
1027 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
1029 return XFS_ERROR(EFSCORRUPTED);
1031 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
1032 size = nextents * sizeof(xfs_bmbt_rec_t);
1033 ifp = XFS_IFORK_PTR(ip, whichfork);
1036 * We know that the size is valid (it's checked in iformat_btree)
1038 ifp->if_lastex = NULLEXTNUM;
1039 ifp->if_bytes = ifp->if_real_bytes = 0;
1040 ifp->if_flags |= XFS_IFEXTENTS;
1041 xfs_iext_add(ifp, 0, nextents);
1042 error = xfs_bmap_read_extents(tp, ip, whichfork);
1044 xfs_iext_destroy(ifp);
1045 ifp->if_flags &= ~XFS_IFEXTENTS;
1048 xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip));
1053 * Allocate an inode on disk and return a copy of its in-core version.
1054 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1055 * appropriately within the inode. The uid and gid for the inode are
1056 * set according to the contents of the given cred structure.
1058 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1059 * has a free inode available, call xfs_iget()
1060 * to obtain the in-core version of the allocated inode. Finally,
1061 * fill in the inode and log its initial contents. In this case,
1062 * ialloc_context would be set to NULL and call_again set to false.
1064 * If xfs_dialloc() does not have an available inode,
1065 * it will replenish its supply by doing an allocation. Since we can
1066 * only do one allocation within a transaction without deadlocks, we
1067 * must commit the current transaction before returning the inode itself.
1068 * In this case, therefore, we will set call_again to true and return.
1069 * The caller should then commit the current transaction, start a new
1070 * transaction, and call xfs_ialloc() again to actually get the inode.
1072 * To ensure that some other process does not grab the inode that
1073 * was allocated during the first call to xfs_ialloc(), this routine
1074 * also returns the [locked] bp pointing to the head of the freelist
1075 * as ialloc_context. The caller should hold this buffer across
1076 * the commit and pass it back into this routine on the second call.
1078 * If we are allocating quota inodes, we do not have a parent inode
1079 * to attach to or associate with (i.e. pip == NULL) because they
1080 * are not linked into the directory structure - they are attached
1081 * directly to the superblock - and so have no parent.
1093 xfs_buf_t **ialloc_context,
1094 boolean_t *call_again,
1104 * Call the space management code to pick
1105 * the on-disk inode to be allocated.
1107 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
1108 ialloc_context, call_again, &ino);
1112 if (*call_again || ino == NULLFSINO) {
1116 ASSERT(*ialloc_context == NULL);
1119 * Get the in-core inode with the lock held exclusively.
1120 * This is because we're setting fields here we need
1121 * to prevent others from looking at until we're done.
1123 error = xfs_trans_iget(tp->t_mountp, tp, ino,
1124 XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
1131 ip->i_d.di_mode = (__uint16_t)mode;
1132 ip->i_d.di_onlink = 0;
1133 ip->i_d.di_nlink = nlink;
1134 ASSERT(ip->i_d.di_nlink == nlink);
1135 ip->i_d.di_uid = current_fsuid(cr);
1136 ip->i_d.di_gid = current_fsgid(cr);
1137 ip->i_d.di_projid = prid;
1138 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1141 * If the superblock version is up to where we support new format
1142 * inodes and this is currently an old format inode, then change
1143 * the inode version number now. This way we only do the conversion
1144 * here rather than here and in the flush/logging code.
1146 if (XFS_SB_VERSION_HASNLINK(&tp->t_mountp->m_sb) &&
1147 ip->i_d.di_version == XFS_DINODE_VERSION_1) {
1148 ip->i_d.di_version = XFS_DINODE_VERSION_2;
1150 * We've already zeroed the old link count, the projid field,
1151 * and the pad field.
1156 * Project ids won't be stored on disk if we are using a version 1 inode.
1158 if ((prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1))
1159 xfs_bump_ino_vers2(tp, ip);
1161 if (pip && XFS_INHERIT_GID(pip)) {
1162 ip->i_d.di_gid = pip->i_d.di_gid;
1163 if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1164 ip->i_d.di_mode |= S_ISGID;
1169 * If the group ID of the new file does not match the effective group
1170 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1171 * (and only if the irix_sgid_inherit compatibility variable is set).
1173 if ((irix_sgid_inherit) &&
1174 (ip->i_d.di_mode & S_ISGID) &&
1175 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1176 ip->i_d.di_mode &= ~S_ISGID;
1179 ip->i_d.di_size = 0;
1181 ip->i_d.di_nextents = 0;
1182 ASSERT(ip->i_d.di_nblocks == 0);
1183 xfs_ichgtime(ip, XFS_ICHGTIME_CHG|XFS_ICHGTIME_ACC|XFS_ICHGTIME_MOD);
1185 * di_gen will have been taken care of in xfs_iread.
1187 ip->i_d.di_extsize = 0;
1188 ip->i_d.di_dmevmask = 0;
1189 ip->i_d.di_dmstate = 0;
1190 ip->i_d.di_flags = 0;
1191 flags = XFS_ILOG_CORE;
1192 switch (mode & S_IFMT) {
1197 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1198 ip->i_df.if_u2.if_rdev = rdev;
1199 ip->i_df.if_flags = 0;
1200 flags |= XFS_ILOG_DEV;
1203 if (pip && xfs_inode_is_filestream(pip)) {
1204 error = xfs_filestream_associate(pip, ip);
1208 xfs_iflags_set(ip, XFS_IFILESTREAM);
1212 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
1215 if ((mode & S_IFMT) == S_IFDIR) {
1216 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1217 di_flags |= XFS_DIFLAG_RTINHERIT;
1218 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1219 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1220 ip->i_d.di_extsize = pip->i_d.di_extsize;
1222 } else if ((mode & S_IFMT) == S_IFREG) {
1223 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) {
1224 di_flags |= XFS_DIFLAG_REALTIME;
1225 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
1227 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1228 di_flags |= XFS_DIFLAG_EXTSIZE;
1229 ip->i_d.di_extsize = pip->i_d.di_extsize;
1232 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1233 xfs_inherit_noatime)
1234 di_flags |= XFS_DIFLAG_NOATIME;
1235 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1237 di_flags |= XFS_DIFLAG_NODUMP;
1238 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1240 di_flags |= XFS_DIFLAG_SYNC;
1241 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1242 xfs_inherit_nosymlinks)
1243 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1244 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1245 di_flags |= XFS_DIFLAG_PROJINHERIT;
1246 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1247 xfs_inherit_nodefrag)
1248 di_flags |= XFS_DIFLAG_NODEFRAG;
1249 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
1250 di_flags |= XFS_DIFLAG_FILESTREAM;
1251 ip->i_d.di_flags |= di_flags;
1255 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1256 ip->i_df.if_flags = XFS_IFEXTENTS;
1257 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1258 ip->i_df.if_u1.if_extents = NULL;
1264 * Attribute fork settings for new inode.
1266 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1267 ip->i_d.di_anextents = 0;
1270 * Log the new values stuffed into the inode.
1272 xfs_trans_log_inode(tp, ip, flags);
1274 /* now that we have an i_mode we can setup inode ops and unlock */
1275 xfs_initialize_vnode(tp->t_mountp, vp, ip);
1282 * Check to make sure that there are no blocks allocated to the
1283 * file beyond the size of the file. We don't check this for
1284 * files with fixed size extents or real time extents, but we
1285 * at least do it for regular files.
1294 xfs_fileoff_t map_first;
1296 xfs_bmbt_irec_t imaps[2];
1298 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1301 if (ip->i_d.di_flags & (XFS_DIFLAG_REALTIME | XFS_DIFLAG_EXTSIZE))
1305 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1307 * The filesystem could be shutting down, so bmapi may return
1310 if (xfs_bmapi(NULL, ip, map_first,
1312 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1314 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1317 ASSERT(nimaps == 1);
1318 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1323 * Calculate the last possible buffered byte in a file. This must
1324 * include data that was buffered beyond the EOF by the write code.
1325 * This also needs to deal with overflowing the xfs_fsize_t type
1326 * which can happen for sizes near the limit.
1328 * We also need to take into account any blocks beyond the EOF. It
1329 * may be the case that they were buffered by a write which failed.
1330 * In that case the pages will still be in memory, but the inode size
1331 * will never have been updated.
1338 xfs_fsize_t last_byte;
1339 xfs_fileoff_t last_block;
1340 xfs_fileoff_t size_last_block;
1343 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE | MR_ACCESS));
1347 * Only check for blocks beyond the EOF if the extents have
1348 * been read in. This eliminates the need for the inode lock,
1349 * and it also saves us from looking when it really isn't
1352 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1353 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1361 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_size);
1362 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1364 last_byte = XFS_FSB_TO_B(mp, last_block);
1365 if (last_byte < 0) {
1366 return XFS_MAXIOFFSET(mp);
1368 last_byte += (1 << mp->m_writeio_log);
1369 if (last_byte < 0) {
1370 return XFS_MAXIOFFSET(mp);
1375 #if defined(XFS_RW_TRACE)
1381 xfs_fsize_t new_size,
1382 xfs_off_t toss_start,
1383 xfs_off_t toss_finish)
1385 if (ip->i_rwtrace == NULL) {
1389 ktrace_enter(ip->i_rwtrace,
1392 (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
1393 (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
1394 (void*)((long)flag),
1395 (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
1396 (void*)(unsigned long)(new_size & 0xffffffff),
1397 (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
1398 (void*)(unsigned long)(toss_start & 0xffffffff),
1399 (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
1400 (void*)(unsigned long)(toss_finish & 0xffffffff),
1401 (void*)(unsigned long)current_cpu(),
1402 (void*)(unsigned long)current_pid(),
1408 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1412 * Start the truncation of the file to new_size. The new size
1413 * must be smaller than the current size. This routine will
1414 * clear the buffer and page caches of file data in the removed
1415 * range, and xfs_itruncate_finish() will remove the underlying
1418 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1419 * must NOT have the inode lock held at all. This is because we're
1420 * calling into the buffer/page cache code and we can't hold the
1421 * inode lock when we do so.
1423 * We need to wait for any direct I/Os in flight to complete before we
1424 * proceed with the truncate. This is needed to prevent the extents
1425 * being read or written by the direct I/Os from being removed while the
1426 * I/O is in flight as there is no other method of synchronising
1427 * direct I/O with the truncate operation. Also, because we hold
1428 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1429 * started until the truncate completes and drops the lock. Essentially,
1430 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1431 * between direct I/Os and the truncate operation.
1433 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1434 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1435 * in the case that the caller is locking things out of order and
1436 * may not be able to call xfs_itruncate_finish() with the inode lock
1437 * held without dropping the I/O lock. If the caller must drop the
1438 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1439 * must be called again with all the same restrictions as the initial
1443 xfs_itruncate_start(
1446 xfs_fsize_t new_size)
1448 xfs_fsize_t last_byte;
1449 xfs_off_t toss_start;
1454 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1455 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1456 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1457 (flags == XFS_ITRUNC_MAYBE));
1462 /* wait for the completion of any pending DIOs */
1463 if (new_size < ip->i_size)
1467 * Call toss_pages or flushinval_pages to get rid of pages
1468 * overlapping the region being removed. We have to use
1469 * the less efficient flushinval_pages in the case that the
1470 * caller may not be able to finish the truncate without
1471 * dropping the inode's I/O lock. Make sure
1472 * to catch any pages brought in by buffers overlapping
1473 * the EOF by searching out beyond the isize by our
1474 * block size. We round new_size up to a block boundary
1475 * so that we don't toss things on the same block as
1476 * new_size but before it.
1478 * Before calling toss_page or flushinval_pages, make sure to
1479 * call remapf() over the same region if the file is mapped.
1480 * This frees up mapped file references to the pages in the
1481 * given range and for the flushinval_pages case it ensures
1482 * that we get the latest mapped changes flushed out.
1484 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1485 toss_start = XFS_FSB_TO_B(mp, toss_start);
1486 if (toss_start < 0) {
1488 * The place to start tossing is beyond our maximum
1489 * file size, so there is no way that the data extended
1494 last_byte = xfs_file_last_byte(ip);
1495 xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
1497 if (last_byte > toss_start) {
1498 if (flags & XFS_ITRUNC_DEFINITE) {
1499 xfs_tosspages(ip, toss_start,
1500 -1, FI_REMAPF_LOCKED);
1502 error = xfs_flushinval_pages(ip, toss_start,
1503 -1, FI_REMAPF_LOCKED);
1508 if (new_size == 0) {
1509 ASSERT(VN_CACHED(vp) == 0);
1516 * Shrink the file to the given new_size. The new
1517 * size must be smaller than the current size.
1518 * This will free up the underlying blocks
1519 * in the removed range after a call to xfs_itruncate_start()
1520 * or xfs_atruncate_start().
1522 * The transaction passed to this routine must have made
1523 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1524 * This routine may commit the given transaction and
1525 * start new ones, so make sure everything involved in
1526 * the transaction is tidy before calling here.
1527 * Some transaction will be returned to the caller to be
1528 * committed. The incoming transaction must already include
1529 * the inode, and both inode locks must be held exclusively.
1530 * The inode must also be "held" within the transaction. On
1531 * return the inode will be "held" within the returned transaction.
1532 * This routine does NOT require any disk space to be reserved
1533 * for it within the transaction.
1535 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1536 * and it indicates the fork which is to be truncated. For the
1537 * attribute fork we only support truncation to size 0.
1539 * We use the sync parameter to indicate whether or not the first
1540 * transaction we perform might have to be synchronous. For the attr fork,
1541 * it needs to be so if the unlink of the inode is not yet known to be
1542 * permanent in the log. This keeps us from freeing and reusing the
1543 * blocks of the attribute fork before the unlink of the inode becomes
1546 * For the data fork, we normally have to run synchronously if we're
1547 * being called out of the inactive path or we're being called
1548 * out of the create path where we're truncating an existing file.
1549 * Either way, the truncate needs to be sync so blocks don't reappear
1550 * in the file with altered data in case of a crash. wsync filesystems
1551 * can run the first case async because anything that shrinks the inode
1552 * has to run sync so by the time we're called here from inactive, the
1553 * inode size is permanently set to 0.
1555 * Calls from the truncate path always need to be sync unless we're
1556 * in a wsync filesystem and the file has already been unlinked.
1558 * The caller is responsible for correctly setting the sync parameter.
1559 * It gets too hard for us to guess here which path we're being called
1560 * out of just based on inode state.
1563 xfs_itruncate_finish(
1566 xfs_fsize_t new_size,
1570 xfs_fsblock_t first_block;
1571 xfs_fileoff_t first_unmap_block;
1572 xfs_fileoff_t last_block;
1573 xfs_filblks_t unmap_len=0;
1578 xfs_bmap_free_t free_list;
1581 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1582 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
1583 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1584 ASSERT(*tp != NULL);
1585 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1586 ASSERT(ip->i_transp == *tp);
1587 ASSERT(ip->i_itemp != NULL);
1588 ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
1592 mp = (ntp)->t_mountp;
1593 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1596 * We only support truncating the entire attribute fork.
1598 if (fork == XFS_ATTR_FORK) {
1601 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1602 xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
1604 * The first thing we do is set the size to new_size permanently
1605 * on disk. This way we don't have to worry about anyone ever
1606 * being able to look at the data being freed even in the face
1607 * of a crash. What we're getting around here is the case where
1608 * we free a block, it is allocated to another file, it is written
1609 * to, and then we crash. If the new data gets written to the
1610 * file but the log buffers containing the free and reallocation
1611 * don't, then we'd end up with garbage in the blocks being freed.
1612 * As long as we make the new_size permanent before actually
1613 * freeing any blocks it doesn't matter if they get writtten to.
1615 * The callers must signal into us whether or not the size
1616 * setting here must be synchronous. There are a few cases
1617 * where it doesn't have to be synchronous. Those cases
1618 * occur if the file is unlinked and we know the unlink is
1619 * permanent or if the blocks being truncated are guaranteed
1620 * to be beyond the inode eof (regardless of the link count)
1621 * and the eof value is permanent. Both of these cases occur
1622 * only on wsync-mounted filesystems. In those cases, we're
1623 * guaranteed that no user will ever see the data in the blocks
1624 * that are being truncated so the truncate can run async.
1625 * In the free beyond eof case, the file may wind up with
1626 * more blocks allocated to it than it needs if we crash
1627 * and that won't get fixed until the next time the file
1628 * is re-opened and closed but that's ok as that shouldn't
1629 * be too many blocks.
1631 * However, we can't just make all wsync xactions run async
1632 * because there's one call out of the create path that needs
1633 * to run sync where it's truncating an existing file to size
1634 * 0 whose size is > 0.
1636 * It's probably possible to come up with a test in this
1637 * routine that would correctly distinguish all the above
1638 * cases from the values of the function parameters and the
1639 * inode state but for sanity's sake, I've decided to let the
1640 * layers above just tell us. It's simpler to correctly figure
1641 * out in the layer above exactly under what conditions we
1642 * can run async and I think it's easier for others read and
1643 * follow the logic in case something has to be changed.
1644 * cscope is your friend -- rcc.
1646 * The attribute fork is much simpler.
1648 * For the attribute fork we allow the caller to tell us whether
1649 * the unlink of the inode that led to this call is yet permanent
1650 * in the on disk log. If it is not and we will be freeing extents
1651 * in this inode then we make the first transaction synchronous
1652 * to make sure that the unlink is permanent by the time we free
1655 if (fork == XFS_DATA_FORK) {
1656 if (ip->i_d.di_nextents > 0) {
1658 * If we are not changing the file size then do
1659 * not update the on-disk file size - we may be
1660 * called from xfs_inactive_free_eofblocks(). If we
1661 * update the on-disk file size and then the system
1662 * crashes before the contents of the file are
1663 * flushed to disk then the files may be full of
1664 * holes (ie NULL files bug).
1666 if (ip->i_size != new_size) {
1667 ip->i_d.di_size = new_size;
1668 ip->i_size = new_size;
1669 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1673 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1674 if (ip->i_d.di_anextents > 0)
1675 xfs_trans_set_sync(ntp);
1677 ASSERT(fork == XFS_DATA_FORK ||
1678 (fork == XFS_ATTR_FORK &&
1679 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1680 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1683 * Since it is possible for space to become allocated beyond
1684 * the end of the file (in a crash where the space is allocated
1685 * but the inode size is not yet updated), simply remove any
1686 * blocks which show up between the new EOF and the maximum
1687 * possible file size. If the first block to be removed is
1688 * beyond the maximum file size (ie it is the same as last_block),
1689 * then there is nothing to do.
1691 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1692 ASSERT(first_unmap_block <= last_block);
1694 if (last_block == first_unmap_block) {
1697 unmap_len = last_block - first_unmap_block + 1;
1701 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1702 * will tell us whether it freed the entire range or
1703 * not. If this is a synchronous mount (wsync),
1704 * then we can tell bunmapi to keep all the
1705 * transactions asynchronous since the unlink
1706 * transaction that made this inode inactive has
1707 * already hit the disk. There's no danger of
1708 * the freed blocks being reused, there being a
1709 * crash, and the reused blocks suddenly reappearing
1710 * in this file with garbage in them once recovery
1713 XFS_BMAP_INIT(&free_list, &first_block);
1714 error = xfs_bunmapi(ntp, ip,
1715 first_unmap_block, unmap_len,
1716 XFS_BMAPI_AFLAG(fork) |
1717 (sync ? 0 : XFS_BMAPI_ASYNC),
1718 XFS_ITRUNC_MAX_EXTENTS,
1719 &first_block, &free_list,
1723 * If the bunmapi call encounters an error,
1724 * return to the caller where the transaction
1725 * can be properly aborted. We just need to
1726 * make sure we're not holding any resources
1727 * that we were not when we came in.
1729 xfs_bmap_cancel(&free_list);
1734 * Duplicate the transaction that has the permanent
1735 * reservation and commit the old transaction.
1737 error = xfs_bmap_finish(tp, &free_list, &committed);
1741 * If the bmap finish call encounters an error,
1742 * return to the caller where the transaction
1743 * can be properly aborted. We just need to
1744 * make sure we're not holding any resources
1745 * that we were not when we came in.
1747 * Aborting from this point might lose some
1748 * blocks in the file system, but oh well.
1750 xfs_bmap_cancel(&free_list);
1753 * If the passed in transaction committed
1754 * in xfs_bmap_finish(), then we want to
1755 * add the inode to this one before returning.
1756 * This keeps things simple for the higher
1757 * level code, because it always knows that
1758 * the inode is locked and held in the
1759 * transaction that returns to it whether
1760 * errors occur or not. We don't mark the
1761 * inode dirty so that this transaction can
1762 * be easily aborted if possible.
1764 xfs_trans_ijoin(ntp, ip,
1765 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1766 xfs_trans_ihold(ntp, ip);
1773 * The first xact was committed,
1774 * so add the inode to the new one.
1775 * Mark it dirty so it will be logged
1776 * and moved forward in the log as
1777 * part of every commit.
1779 xfs_trans_ijoin(ntp, ip,
1780 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1781 xfs_trans_ihold(ntp, ip);
1782 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1784 ntp = xfs_trans_dup(ntp);
1785 (void) xfs_trans_commit(*tp, 0);
1787 error = xfs_trans_reserve(ntp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1788 XFS_TRANS_PERM_LOG_RES,
1789 XFS_ITRUNCATE_LOG_COUNT);
1791 * Add the inode being truncated to the next chained
1794 xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1795 xfs_trans_ihold(ntp, ip);
1800 * Only update the size in the case of the data fork, but
1801 * always re-log the inode so that our permanent transaction
1802 * can keep on rolling it forward in the log.
1804 if (fork == XFS_DATA_FORK) {
1805 xfs_isize_check(mp, ip, new_size);
1807 * If we are not changing the file size then do
1808 * not update the on-disk file size - we may be
1809 * called from xfs_inactive_free_eofblocks(). If we
1810 * update the on-disk file size and then the system
1811 * crashes before the contents of the file are
1812 * flushed to disk then the files may be full of
1813 * holes (ie NULL files bug).
1815 if (ip->i_size != new_size) {
1816 ip->i_d.di_size = new_size;
1817 ip->i_size = new_size;
1820 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1821 ASSERT((new_size != 0) ||
1822 (fork == XFS_ATTR_FORK) ||
1823 (ip->i_delayed_blks == 0));
1824 ASSERT((new_size != 0) ||
1825 (fork == XFS_ATTR_FORK) ||
1826 (ip->i_d.di_nextents == 0));
1827 xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
1835 * Do the first part of growing a file: zero any data in the last
1836 * block that is beyond the old EOF. We need to do this before
1837 * the inode is joined to the transaction to modify the i_size.
1838 * That way we can drop the inode lock and call into the buffer
1839 * cache to get the buffer mapping the EOF.
1844 xfs_fsize_t new_size,
1847 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1848 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1849 ASSERT(new_size > ip->i_size);
1852 * Zero any pages that may have been created by
1853 * xfs_write_file() beyond the end of the file
1854 * and any blocks between the old and new file sizes.
1856 return xfs_zero_eof(ip, new_size, ip->i_size);
1862 * This routine is called to extend the size of a file.
1863 * The inode must have both the iolock and the ilock locked
1864 * for update and it must be a part of the current transaction.
1865 * The xfs_igrow_start() function must have been called previously.
1866 * If the change_flag is not zero, the inode change timestamp will
1873 xfs_fsize_t new_size,
1876 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1877 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1878 ASSERT(ip->i_transp == tp);
1879 ASSERT(new_size > ip->i_size);
1882 * Update the file size. Update the inode change timestamp
1883 * if change_flag set.
1885 ip->i_d.di_size = new_size;
1886 ip->i_size = new_size;
1888 xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
1889 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1895 * This is called when the inode's link count goes to 0.
1896 * We place the on-disk inode on a list in the AGI. It
1897 * will be pulled from this list when the inode is freed.
1909 xfs_agnumber_t agno;
1910 xfs_daddr_t agdaddr;
1917 ASSERT(ip->i_d.di_nlink == 0);
1918 ASSERT(ip->i_d.di_mode != 0);
1919 ASSERT(ip->i_transp == tp);
1923 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1924 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1927 * Get the agi buffer first. It ensures lock ordering
1930 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1931 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1936 * Validate the magic number of the agi block.
1938 agi = XFS_BUF_TO_AGI(agibp);
1940 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1941 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
1942 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK,
1943 XFS_RANDOM_IUNLINK))) {
1944 XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi);
1945 xfs_trans_brelse(tp, agibp);
1946 return XFS_ERROR(EFSCORRUPTED);
1949 * Get the index into the agi hash table for the
1950 * list this inode will go on.
1952 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1954 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1955 ASSERT(agi->agi_unlinked[bucket_index]);
1956 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1958 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
1963 * Clear the on-disk di_nlink. This is to prevent xfs_bulkstat
1964 * from picking up this inode when it is reclaimed (its incore state
1965 * initialzed but not flushed to disk yet). The in-core di_nlink is
1966 * already cleared in xfs_droplink() and a corresponding transaction
1967 * logged. The hack here just synchronizes the in-core to on-disk
1968 * di_nlink value in advance before the actual inode sync to disk.
1969 * This is OK because the inode is already unlinked and would never
1970 * change its di_nlink again for this inode generation.
1971 * This is a temporary hack that would require a proper fix
1974 dip->di_core.di_nlink = 0;
1976 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
1978 * There is already another inode in the bucket we need
1979 * to add ourselves to. Add us at the front of the list.
1980 * Here we put the head pointer into our next pointer,
1981 * and then we fall through to point the head at us.
1983 ASSERT(be32_to_cpu(dip->di_next_unlinked) == NULLAGINO);
1984 /* both on-disk, don't endian flip twice */
1985 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1986 offset = ip->i_boffset +
1987 offsetof(xfs_dinode_t, di_next_unlinked);
1988 xfs_trans_inode_buf(tp, ibp);
1989 xfs_trans_log_buf(tp, ibp, offset,
1990 (offset + sizeof(xfs_agino_t) - 1));
1991 xfs_inobp_check(mp, ibp);
1995 * Point the bucket head pointer at the inode being inserted.
1998 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1999 offset = offsetof(xfs_agi_t, agi_unlinked) +
2000 (sizeof(xfs_agino_t) * bucket_index);
2001 xfs_trans_log_buf(tp, agibp, offset,
2002 (offset + sizeof(xfs_agino_t) - 1));
2007 * Pull the on-disk inode from the AGI unlinked list.
2020 xfs_agnumber_t agno;
2021 xfs_daddr_t agdaddr;
2023 xfs_agino_t next_agino;
2024 xfs_buf_t *last_ibp;
2025 xfs_dinode_t *last_dip = NULL;
2027 int offset, last_offset = 0;
2032 * First pull the on-disk inode from the AGI unlinked list.
2036 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2037 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
2040 * Get the agi buffer first. It ensures lock ordering
2043 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
2044 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
2047 "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
2048 error, mp->m_fsname);
2052 * Validate the magic number of the agi block.
2054 agi = XFS_BUF_TO_AGI(agibp);
2056 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
2057 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
2058 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE,
2059 XFS_RANDOM_IUNLINK_REMOVE))) {
2060 XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW,
2062 xfs_trans_brelse(tp, agibp);
2064 "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
2066 return XFS_ERROR(EFSCORRUPTED);
2069 * Get the index into the agi hash table for the
2070 * list this inode will go on.
2072 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2074 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2075 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
2076 ASSERT(agi->agi_unlinked[bucket_index]);
2078 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2080 * We're at the head of the list. Get the inode's
2081 * on-disk buffer to see if there is anyone after us
2082 * on the list. Only modify our next pointer if it
2083 * is not already NULLAGINO. This saves us the overhead
2084 * of dealing with the buffer when there is no need to
2087 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
2090 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2091 error, mp->m_fsname);
2094 next_agino = be32_to_cpu(dip->di_next_unlinked);
2095 ASSERT(next_agino != 0);
2096 if (next_agino != NULLAGINO) {
2097 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2098 offset = ip->i_boffset +
2099 offsetof(xfs_dinode_t, di_next_unlinked);
2100 xfs_trans_inode_buf(tp, ibp);
2101 xfs_trans_log_buf(tp, ibp, offset,
2102 (offset + sizeof(xfs_agino_t) - 1));
2103 xfs_inobp_check(mp, ibp);
2105 xfs_trans_brelse(tp, ibp);
2108 * Point the bucket head pointer at the next inode.
2110 ASSERT(next_agino != 0);
2111 ASSERT(next_agino != agino);
2112 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2113 offset = offsetof(xfs_agi_t, agi_unlinked) +
2114 (sizeof(xfs_agino_t) * bucket_index);
2115 xfs_trans_log_buf(tp, agibp, offset,
2116 (offset + sizeof(xfs_agino_t) - 1));
2119 * We need to search the list for the inode being freed.
2121 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2123 while (next_agino != agino) {
2125 * If the last inode wasn't the one pointing to
2126 * us, then release its buffer since we're not
2127 * going to do anything with it.
2129 if (last_ibp != NULL) {
2130 xfs_trans_brelse(tp, last_ibp);
2132 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2133 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
2134 &last_ibp, &last_offset);
2137 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
2138 error, mp->m_fsname);
2141 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2142 ASSERT(next_agino != NULLAGINO);
2143 ASSERT(next_agino != 0);
2146 * Now last_ibp points to the buffer previous to us on
2147 * the unlinked list. Pull us from the list.
2149 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
2152 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2153 error, mp->m_fsname);
2156 next_agino = be32_to_cpu(dip->di_next_unlinked);
2157 ASSERT(next_agino != 0);
2158 ASSERT(next_agino != agino);
2159 if (next_agino != NULLAGINO) {
2160 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2161 offset = ip->i_boffset +
2162 offsetof(xfs_dinode_t, di_next_unlinked);
2163 xfs_trans_inode_buf(tp, ibp);
2164 xfs_trans_log_buf(tp, ibp, offset,
2165 (offset + sizeof(xfs_agino_t) - 1));
2166 xfs_inobp_check(mp, ibp);
2168 xfs_trans_brelse(tp, ibp);
2171 * Point the previous inode on the list to the next inode.
2173 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2174 ASSERT(next_agino != 0);
2175 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2176 xfs_trans_inode_buf(tp, last_ibp);
2177 xfs_trans_log_buf(tp, last_ibp, offset,
2178 (offset + sizeof(xfs_agino_t) - 1));
2179 xfs_inobp_check(mp, last_ibp);
2184 STATIC_INLINE int xfs_inode_clean(xfs_inode_t *ip)
2186 return (((ip->i_itemp == NULL) ||
2187 !(ip->i_itemp->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
2188 (ip->i_update_core == 0));
2193 xfs_inode_t *free_ip,
2197 xfs_mount_t *mp = free_ip->i_mount;
2198 int blks_per_cluster;
2201 int i, j, found, pre_flushed;
2204 xfs_inode_t *ip, **ip_found;
2205 xfs_inode_log_item_t *iip;
2206 xfs_log_item_t *lip;
2207 xfs_perag_t *pag = xfs_get_perag(mp, inum);
2209 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
2210 blks_per_cluster = 1;
2211 ninodes = mp->m_sb.sb_inopblock;
2212 nbufs = XFS_IALLOC_BLOCKS(mp);
2214 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
2215 mp->m_sb.sb_blocksize;
2216 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
2217 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
2220 ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
2222 for (j = 0; j < nbufs; j++, inum += ninodes) {
2223 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2224 XFS_INO_TO_AGBNO(mp, inum));
2228 * Look for each inode in memory and attempt to lock it,
2229 * we can be racing with flush and tail pushing here.
2230 * any inode we get the locks on, add to an array of
2231 * inode items to process later.
2233 * The get the buffer lock, we could beat a flush
2234 * or tail pushing thread to the lock here, in which
2235 * case they will go looking for the inode buffer
2236 * and fail, we need some other form of interlock
2240 for (i = 0; i < ninodes; i++) {
2241 read_lock(&pag->pag_ici_lock);
2242 ip = radix_tree_lookup(&pag->pag_ici_root,
2243 XFS_INO_TO_AGINO(mp, (inum + i)));
2245 /* Inode not in memory or we found it already,
2248 if (!ip || xfs_iflags_test(ip, XFS_ISTALE)) {
2249 read_unlock(&pag->pag_ici_lock);
2253 if (xfs_inode_clean(ip)) {
2254 read_unlock(&pag->pag_ici_lock);
2258 /* If we can get the locks then add it to the
2259 * list, otherwise by the time we get the bp lock
2260 * below it will already be attached to the
2264 /* This inode will already be locked - by us, lets
2268 if (ip == free_ip) {
2269 if (xfs_iflock_nowait(ip)) {
2270 xfs_iflags_set(ip, XFS_ISTALE);
2271 if (xfs_inode_clean(ip)) {
2274 ip_found[found++] = ip;
2277 read_unlock(&pag->pag_ici_lock);
2281 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2282 if (xfs_iflock_nowait(ip)) {
2283 xfs_iflags_set(ip, XFS_ISTALE);
2285 if (xfs_inode_clean(ip)) {
2287 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2289 ip_found[found++] = ip;
2292 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2295 read_unlock(&pag->pag_ici_lock);
2298 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2299 mp->m_bsize * blks_per_cluster,
2303 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
2305 if (lip->li_type == XFS_LI_INODE) {
2306 iip = (xfs_inode_log_item_t *)lip;
2307 ASSERT(iip->ili_logged == 1);
2308 lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
2309 spin_lock(&mp->m_ail_lock);
2310 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2311 spin_unlock(&mp->m_ail_lock);
2312 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2315 lip = lip->li_bio_list;
2318 for (i = 0; i < found; i++) {
2323 ip->i_update_core = 0;
2325 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2329 iip->ili_last_fields = iip->ili_format.ilf_fields;
2330 iip->ili_format.ilf_fields = 0;
2331 iip->ili_logged = 1;
2332 spin_lock(&mp->m_ail_lock);
2333 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2334 spin_unlock(&mp->m_ail_lock);
2336 xfs_buf_attach_iodone(bp,
2337 (void(*)(xfs_buf_t*,xfs_log_item_t*))
2338 xfs_istale_done, (xfs_log_item_t *)iip);
2339 if (ip != free_ip) {
2340 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2344 if (found || pre_flushed)
2345 xfs_trans_stale_inode_buf(tp, bp);
2346 xfs_trans_binval(tp, bp);
2349 kmem_free(ip_found, ninodes * sizeof(xfs_inode_t *));
2350 xfs_put_perag(mp, pag);
2354 * This is called to return an inode to the inode free list.
2355 * The inode should already be truncated to 0 length and have
2356 * no pages associated with it. This routine also assumes that
2357 * the inode is already a part of the transaction.
2359 * The on-disk copy of the inode will have been added to the list
2360 * of unlinked inodes in the AGI. We need to remove the inode from
2361 * that list atomically with respect to freeing it here.
2367 xfs_bmap_free_t *flist)
2371 xfs_ino_t first_ino;
2373 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2374 ASSERT(ip->i_transp == tp);
2375 ASSERT(ip->i_d.di_nlink == 0);
2376 ASSERT(ip->i_d.di_nextents == 0);
2377 ASSERT(ip->i_d.di_anextents == 0);
2378 ASSERT((ip->i_d.di_size == 0 && ip->i_size == 0) ||
2379 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2380 ASSERT(ip->i_d.di_nblocks == 0);
2383 * Pull the on-disk inode from the AGI unlinked list.
2385 error = xfs_iunlink_remove(tp, ip);
2390 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2394 ip->i_d.di_mode = 0; /* mark incore inode as free */
2395 ip->i_d.di_flags = 0;
2396 ip->i_d.di_dmevmask = 0;
2397 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2398 ip->i_df.if_ext_max =
2399 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2400 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2401 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2403 * Bump the generation count so no one will be confused
2404 * by reincarnations of this inode.
2407 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2410 xfs_ifree_cluster(ip, tp, first_ino);
2417 * Reallocate the space for if_broot based on the number of records
2418 * being added or deleted as indicated in rec_diff. Move the records
2419 * and pointers in if_broot to fit the new size. When shrinking this
2420 * will eliminate holes between the records and pointers created by
2421 * the caller. When growing this will create holes to be filled in
2424 * The caller must not request to add more records than would fit in
2425 * the on-disk inode root. If the if_broot is currently NULL, then
2426 * if we adding records one will be allocated. The caller must also
2427 * not request that the number of records go below zero, although
2428 * it can go to zero.
2430 * ip -- the inode whose if_broot area is changing
2431 * ext_diff -- the change in the number of records, positive or negative,
2432 * requested for the if_broot array.
2442 xfs_bmbt_block_t *new_broot;
2449 * Handle the degenerate case quietly.
2451 if (rec_diff == 0) {
2455 ifp = XFS_IFORK_PTR(ip, whichfork);
2458 * If there wasn't any memory allocated before, just
2459 * allocate it now and get out.
2461 if (ifp->if_broot_bytes == 0) {
2462 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
2463 ifp->if_broot = (xfs_bmbt_block_t*)kmem_alloc(new_size,
2465 ifp->if_broot_bytes = (int)new_size;
2470 * If there is already an existing if_broot, then we need
2471 * to realloc() it and shift the pointers to their new
2472 * location. The records don't change location because
2473 * they are kept butted up against the btree block header.
2475 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2476 new_max = cur_max + rec_diff;
2477 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2478 ifp->if_broot = (xfs_bmbt_block_t *)
2479 kmem_realloc(ifp->if_broot,
2481 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2483 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2484 ifp->if_broot_bytes);
2485 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2487 ifp->if_broot_bytes = (int)new_size;
2488 ASSERT(ifp->if_broot_bytes <=
2489 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2490 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2495 * rec_diff is less than 0. In this case, we are shrinking the
2496 * if_broot buffer. It must already exist. If we go to zero
2497 * records, just get rid of the root and clear the status bit.
2499 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
2500 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2501 new_max = cur_max + rec_diff;
2502 ASSERT(new_max >= 0);
2504 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2508 new_broot = (xfs_bmbt_block_t *)kmem_alloc(new_size, KM_SLEEP);
2510 * First copy over the btree block header.
2512 memcpy(new_broot, ifp->if_broot, sizeof(xfs_bmbt_block_t));
2515 ifp->if_flags &= ~XFS_IFBROOT;
2519 * Only copy the records and pointers if there are any.
2523 * First copy the records.
2525 op = (char *)XFS_BMAP_BROOT_REC_ADDR(ifp->if_broot, 1,
2526 ifp->if_broot_bytes);
2527 np = (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot, 1,
2529 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2532 * Then copy the pointers.
2534 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2535 ifp->if_broot_bytes);
2536 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot, 1,
2538 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2540 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2541 ifp->if_broot = new_broot;
2542 ifp->if_broot_bytes = (int)new_size;
2543 ASSERT(ifp->if_broot_bytes <=
2544 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2550 * This is called when the amount of space needed for if_data
2551 * is increased or decreased. The change in size is indicated by
2552 * the number of bytes that need to be added or deleted in the
2553 * byte_diff parameter.
2555 * If the amount of space needed has decreased below the size of the
2556 * inline buffer, then switch to using the inline buffer. Otherwise,
2557 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2558 * to what is needed.
2560 * ip -- the inode whose if_data area is changing
2561 * byte_diff -- the change in the number of bytes, positive or negative,
2562 * requested for the if_data array.
2574 if (byte_diff == 0) {
2578 ifp = XFS_IFORK_PTR(ip, whichfork);
2579 new_size = (int)ifp->if_bytes + byte_diff;
2580 ASSERT(new_size >= 0);
2582 if (new_size == 0) {
2583 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2584 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2586 ifp->if_u1.if_data = NULL;
2588 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2590 * If the valid extents/data can fit in if_inline_ext/data,
2591 * copy them from the malloc'd vector and free it.
2593 if (ifp->if_u1.if_data == NULL) {
2594 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2595 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2596 ASSERT(ifp->if_real_bytes != 0);
2597 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2599 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2600 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2605 * Stuck with malloc/realloc.
2606 * For inline data, the underlying buffer must be
2607 * a multiple of 4 bytes in size so that it can be
2608 * logged and stay on word boundaries. We enforce
2611 real_size = roundup(new_size, 4);
2612 if (ifp->if_u1.if_data == NULL) {
2613 ASSERT(ifp->if_real_bytes == 0);
2614 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2615 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2617 * Only do the realloc if the underlying size
2618 * is really changing.
2620 if (ifp->if_real_bytes != real_size) {
2621 ifp->if_u1.if_data =
2622 kmem_realloc(ifp->if_u1.if_data,
2628 ASSERT(ifp->if_real_bytes == 0);
2629 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2630 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2634 ifp->if_real_bytes = real_size;
2635 ifp->if_bytes = new_size;
2636 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2643 * Map inode to disk block and offset.
2645 * mp -- the mount point structure for the current file system
2646 * tp -- the current transaction
2647 * ino -- the inode number of the inode to be located
2648 * imap -- this structure is filled in with the information necessary
2649 * to retrieve the given inode from disk
2650 * flags -- flags to pass to xfs_dilocate indicating whether or not
2651 * lookups in the inode btree were OK or not
2661 xfs_fsblock_t fsbno;
2666 fsbno = imap->im_blkno ?
2667 XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK;
2668 error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags);
2672 imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno);
2673 imap->im_len = XFS_FSB_TO_BB(mp, len);
2674 imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno);
2675 imap->im_ioffset = (ushort)off;
2676 imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog);
2687 ifp = XFS_IFORK_PTR(ip, whichfork);
2688 if (ifp->if_broot != NULL) {
2689 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2690 ifp->if_broot = NULL;
2694 * If the format is local, then we can't have an extents
2695 * array so just look for an inline data array. If we're
2696 * not local then we may or may not have an extents list,
2697 * so check and free it up if we do.
2699 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2700 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2701 (ifp->if_u1.if_data != NULL)) {
2702 ASSERT(ifp->if_real_bytes != 0);
2703 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2704 ifp->if_u1.if_data = NULL;
2705 ifp->if_real_bytes = 0;
2707 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2708 ((ifp->if_flags & XFS_IFEXTIREC) ||
2709 ((ifp->if_u1.if_extents != NULL) &&
2710 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
2711 ASSERT(ifp->if_real_bytes != 0);
2712 xfs_iext_destroy(ifp);
2714 ASSERT(ifp->if_u1.if_extents == NULL ||
2715 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2716 ASSERT(ifp->if_real_bytes == 0);
2717 if (whichfork == XFS_ATTR_FORK) {
2718 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2724 * This is called free all the memory associated with an inode.
2725 * It must free the inode itself and any buffers allocated for
2726 * if_extents/if_data and if_broot. It must also free the lock
2727 * associated with the inode.
2733 switch (ip->i_d.di_mode & S_IFMT) {
2737 xfs_idestroy_fork(ip, XFS_DATA_FORK);
2741 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
2742 mrfree(&ip->i_lock);
2743 mrfree(&ip->i_iolock);
2744 freesema(&ip->i_flock);
2746 #ifdef XFS_INODE_TRACE
2747 ktrace_free(ip->i_trace);
2749 #ifdef XFS_BMAP_TRACE
2750 ktrace_free(ip->i_xtrace);
2752 #ifdef XFS_BMBT_TRACE
2753 ktrace_free(ip->i_btrace);
2756 ktrace_free(ip->i_rwtrace);
2758 #ifdef XFS_ILOCK_TRACE
2759 ktrace_free(ip->i_lock_trace);
2761 #ifdef XFS_DIR2_TRACE
2762 ktrace_free(ip->i_dir_trace);
2766 * Only if we are shutting down the fs will we see an
2767 * inode still in the AIL. If it is there, we should remove
2768 * it to prevent a use-after-free from occurring.
2770 xfs_mount_t *mp = ip->i_mount;
2771 xfs_log_item_t *lip = &ip->i_itemp->ili_item;
2773 ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) ||
2774 XFS_FORCED_SHUTDOWN(ip->i_mount));
2775 if (lip->li_flags & XFS_LI_IN_AIL) {
2776 spin_lock(&mp->m_ail_lock);
2777 if (lip->li_flags & XFS_LI_IN_AIL)
2778 xfs_trans_delete_ail(mp, lip);
2780 spin_unlock(&mp->m_ail_lock);
2782 xfs_inode_item_destroy(ip);
2784 kmem_zone_free(xfs_inode_zone, ip);
2789 * Increment the pin count of the given buffer.
2790 * This value is protected by ipinlock spinlock in the mount structure.
2796 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2798 atomic_inc(&ip->i_pincount);
2802 * Decrement the pin count of the given inode, and wake up
2803 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2804 * inode must have been previously pinned with a call to xfs_ipin().
2810 ASSERT(atomic_read(&ip->i_pincount) > 0);
2812 if (atomic_dec_and_lock(&ip->i_pincount, &ip->i_flags_lock)) {
2815 * If the inode is currently being reclaimed, the link between
2816 * the bhv_vnode and the xfs_inode will be broken after the
2817 * XFS_IRECLAIM* flag is set. Hence, if these flags are not
2818 * set, then we can move forward and mark the linux inode dirty
2819 * knowing that it is still valid as it won't freed until after
2820 * the bhv_vnode<->xfs_inode link is broken in xfs_reclaim. The
2821 * i_flags_lock is used to synchronise the setting of the
2822 * XFS_IRECLAIM* flags and the breaking of the link, and so we
2823 * can execute atomically w.r.t to reclaim by holding this lock
2826 * However, we still need to issue the unpin wakeup call as the
2827 * inode reclaim may be blocked waiting for the inode to become
2831 if (!__xfs_iflags_test(ip, XFS_IRECLAIM|XFS_IRECLAIMABLE)) {
2832 bhv_vnode_t *vp = XFS_ITOV_NULL(ip);
2833 struct inode *inode = NULL;
2836 inode = vn_to_inode(vp);
2837 BUG_ON(inode->i_state & I_CLEAR);
2839 /* make sync come back and flush this inode */
2840 if (!(inode->i_state & (I_NEW|I_FREEING)))
2841 mark_inode_dirty_sync(inode);
2843 spin_unlock(&ip->i_flags_lock);
2844 wake_up(&ip->i_ipin_wait);
2849 * This is called to wait for the given inode to be unpinned.
2850 * It will sleep until this happens. The caller must have the
2851 * inode locked in at least shared mode so that the buffer cannot
2852 * be subsequently pinned once someone is waiting for it to be
2859 xfs_inode_log_item_t *iip;
2862 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE | MR_ACCESS));
2864 if (atomic_read(&ip->i_pincount) == 0) {
2869 if (iip && iip->ili_last_lsn) {
2870 lsn = iip->ili_last_lsn;
2876 * Give the log a push so we don't wait here too long.
2878 xfs_log_force(ip->i_mount, lsn, XFS_LOG_FORCE);
2880 wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
2885 * xfs_iextents_copy()
2887 * This is called to copy the REAL extents (as opposed to the delayed
2888 * allocation extents) from the inode into the given buffer. It
2889 * returns the number of bytes copied into the buffer.
2891 * If there are no delayed allocation extents, then we can just
2892 * memcpy() the extents into the buffer. Otherwise, we need to
2893 * examine each extent in turn and skip those which are delayed.
2905 xfs_fsblock_t start_block;
2907 ifp = XFS_IFORK_PTR(ip, whichfork);
2908 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
2909 ASSERT(ifp->if_bytes > 0);
2911 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2912 XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork);
2916 * There are some delayed allocation extents in the
2917 * inode, so copy the extents one at a time and skip
2918 * the delayed ones. There must be at least one
2919 * non-delayed extent.
2922 for (i = 0; i < nrecs; i++) {
2923 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
2924 start_block = xfs_bmbt_get_startblock(ep);
2925 if (ISNULLSTARTBLOCK(start_block)) {
2927 * It's a delayed allocation extent, so skip it.
2932 /* Translate to on disk format */
2933 put_unaligned(cpu_to_be64(ep->l0), &dp->l0);
2934 put_unaligned(cpu_to_be64(ep->l1), &dp->l1);
2938 ASSERT(copied != 0);
2939 xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip));
2941 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2945 * Each of the following cases stores data into the same region
2946 * of the on-disk inode, so only one of them can be valid at
2947 * any given time. While it is possible to have conflicting formats
2948 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2949 * in EXTENTS format, this can only happen when the fork has
2950 * changed formats after being modified but before being flushed.
2951 * In these cases, the format always takes precedence, because the
2952 * format indicates the current state of the fork.
2959 xfs_inode_log_item_t *iip,
2966 #ifdef XFS_TRANS_DEBUG
2969 static const short brootflag[2] =
2970 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2971 static const short dataflag[2] =
2972 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2973 static const short extflag[2] =
2974 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2978 ifp = XFS_IFORK_PTR(ip, whichfork);
2980 * This can happen if we gave up in iformat in an error path,
2981 * for the attribute fork.
2984 ASSERT(whichfork == XFS_ATTR_FORK);
2987 cp = XFS_DFORK_PTR(dip, whichfork);
2989 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2990 case XFS_DINODE_FMT_LOCAL:
2991 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2992 (ifp->if_bytes > 0)) {
2993 ASSERT(ifp->if_u1.if_data != NULL);
2994 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2995 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2999 case XFS_DINODE_FMT_EXTENTS:
3000 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
3001 !(iip->ili_format.ilf_fields & extflag[whichfork]));
3002 ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
3003 (ifp->if_bytes == 0));
3004 ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
3005 (ifp->if_bytes > 0));
3006 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
3007 (ifp->if_bytes > 0)) {
3008 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
3009 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
3014 case XFS_DINODE_FMT_BTREE:
3015 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
3016 (ifp->if_broot_bytes > 0)) {
3017 ASSERT(ifp->if_broot != NULL);
3018 ASSERT(ifp->if_broot_bytes <=
3019 (XFS_IFORK_SIZE(ip, whichfork) +
3020 XFS_BROOT_SIZE_ADJ));
3021 xfs_bmbt_to_bmdr(ifp->if_broot, ifp->if_broot_bytes,
3022 (xfs_bmdr_block_t *)cp,
3023 XFS_DFORK_SIZE(dip, mp, whichfork));
3027 case XFS_DINODE_FMT_DEV:
3028 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
3029 ASSERT(whichfork == XFS_DATA_FORK);
3030 dip->di_u.di_dev = cpu_to_be32(ip->i_df.if_u2.if_rdev);
3034 case XFS_DINODE_FMT_UUID:
3035 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
3036 ASSERT(whichfork == XFS_DATA_FORK);
3037 memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid,
3051 * xfs_iflush() will write a modified inode's changes out to the
3052 * inode's on disk home. The caller must have the inode lock held
3053 * in at least shared mode and the inode flush semaphore must be
3054 * held as well. The inode lock will still be held upon return from
3055 * the call and the caller is free to unlock it.
3056 * The inode flush lock will be unlocked when the inode reaches the disk.
3057 * The flags indicate how the inode's buffer should be written out.
3064 xfs_inode_log_item_t *iip;
3071 int clcount; /* count of inodes clustered */
3073 struct hlist_node *entry;
3074 enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
3076 XFS_STATS_INC(xs_iflush_count);
3078 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
3079 ASSERT(issemalocked(&(ip->i_flock)));
3080 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3081 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3087 * If the inode isn't dirty, then just release the inode
3088 * flush lock and do nothing.
3090 if ((ip->i_update_core == 0) &&
3091 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3092 ASSERT((iip != NULL) ?
3093 !(iip->ili_item.li_flags & XFS_LI_IN_AIL) : 1);
3099 * We can't flush the inode until it is unpinned, so
3100 * wait for it. We know noone new can pin it, because
3101 * we are holding the inode lock shared and you need
3102 * to hold it exclusively to pin the inode.
3104 xfs_iunpin_wait(ip);
3107 * This may have been unpinned because the filesystem is shutting
3108 * down forcibly. If that's the case we must not write this inode
3109 * to disk, because the log record didn't make it to disk!
3111 if (XFS_FORCED_SHUTDOWN(mp)) {
3112 ip->i_update_core = 0;
3114 iip->ili_format.ilf_fields = 0;
3116 return XFS_ERROR(EIO);
3120 * Get the buffer containing the on-disk inode.
3122 error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0, 0);
3129 * Decide how buffer will be flushed out. This is done before
3130 * the call to xfs_iflush_int because this field is zeroed by it.
3132 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3134 * Flush out the inode buffer according to the directions
3135 * of the caller. In the cases where the caller has given
3136 * us a choice choose the non-delwri case. This is because
3137 * the inode is in the AIL and we need to get it out soon.
3140 case XFS_IFLUSH_SYNC:
3141 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3144 case XFS_IFLUSH_ASYNC:
3145 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3148 case XFS_IFLUSH_DELWRI:
3158 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3159 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3160 case XFS_IFLUSH_DELWRI:
3163 case XFS_IFLUSH_ASYNC:
3166 case XFS_IFLUSH_SYNC:
3177 * First flush out the inode that xfs_iflush was called with.
3179 error = xfs_iflush_int(ip, bp);
3186 * see if other inodes can be gathered into this write
3188 spin_lock(&ip->i_cluster->icl_lock);
3189 ip->i_cluster->icl_buf = bp;
3192 hlist_for_each_entry(iq, entry, &ip->i_cluster->icl_inodes, i_cnode) {
3197 * Do an un-protected check to see if the inode is dirty and
3198 * is a candidate for flushing. These checks will be repeated
3199 * later after the appropriate locks are acquired.
3202 if ((iq->i_update_core == 0) &&
3204 !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
3205 xfs_ipincount(iq) == 0) {
3210 * Try to get locks. If any are unavailable,
3211 * then this inode cannot be flushed and is skipped.
3214 /* get inode locks (just i_lock) */
3215 if (xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) {
3216 /* get inode flush lock */
3217 if (xfs_iflock_nowait(iq)) {
3218 /* check if pinned */
3219 if (xfs_ipincount(iq) == 0) {
3220 /* arriving here means that
3221 * this inode can be flushed.
3222 * first re-check that it's
3226 if ((iq->i_update_core != 0)||
3228 (iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3230 error = xfs_iflush_int(iq, bp);
3234 goto cluster_corrupt_out;
3243 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3246 spin_unlock(&ip->i_cluster->icl_lock);
3249 XFS_STATS_INC(xs_icluster_flushcnt);
3250 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
3254 * If the buffer is pinned then push on the log so we won't
3255 * get stuck waiting in the write for too long.
3257 if (XFS_BUF_ISPINNED(bp)){
3258 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
3261 if (flags & INT_DELWRI) {
3262 xfs_bdwrite(mp, bp);
3263 } else if (flags & INT_ASYNC) {
3264 xfs_bawrite(mp, bp);
3266 error = xfs_bwrite(mp, bp);
3272 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3273 xfs_iflush_abort(ip);
3275 * Unlocks the flush lock
3277 return XFS_ERROR(EFSCORRUPTED);
3279 cluster_corrupt_out:
3280 /* Corruption detected in the clustering loop. Invalidate the
3281 * inode buffer and shut down the filesystem.
3283 spin_unlock(&ip->i_cluster->icl_lock);
3286 * Clean up the buffer. If it was B_DELWRI, just release it --
3287 * brelse can handle it with no problems. If not, shut down the
3288 * filesystem before releasing the buffer.
3290 if ((bufwasdelwri= XFS_BUF_ISDELAYWRITE(bp))) {
3294 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3298 * Just like incore_relse: if we have b_iodone functions,
3299 * mark the buffer as an error and call them. Otherwise
3300 * mark it as stale and brelse.
3302 if (XFS_BUF_IODONE_FUNC(bp)) {
3303 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
3307 XFS_BUF_ERROR(bp,EIO);
3315 xfs_iflush_abort(iq);
3317 * Unlocks the flush lock
3319 return XFS_ERROR(EFSCORRUPTED);
3328 xfs_inode_log_item_t *iip;
3331 #ifdef XFS_TRANS_DEBUG
3335 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
3336 ASSERT(issemalocked(&(ip->i_flock)));
3337 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3338 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3345 * If the inode isn't dirty, then just release the inode
3346 * flush lock and do nothing.
3348 if ((ip->i_update_core == 0) &&
3349 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3354 /* set *dip = inode's place in the buffer */
3355 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset);
3358 * Clear i_update_core before copying out the data.
3359 * This is for coordination with our timestamp updates
3360 * that don't hold the inode lock. They will always
3361 * update the timestamps BEFORE setting i_update_core,
3362 * so if we clear i_update_core after they set it we
3363 * are guaranteed to see their updates to the timestamps.
3364 * I believe that this depends on strongly ordered memory
3365 * semantics, but we have that. We use the SYNCHRONIZE
3366 * macro to make sure that the compiler does not reorder
3367 * the i_update_core access below the data copy below.
3369 ip->i_update_core = 0;
3373 * Make sure to get the latest atime from the Linux inode.
3375 xfs_synchronize_atime(ip);
3377 if (XFS_TEST_ERROR(be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC,
3378 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3379 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3380 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3381 ip->i_ino, be16_to_cpu(dip->di_core.di_magic), dip);
3384 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3385 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3386 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3387 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3388 ip->i_ino, ip, ip->i_d.di_magic);
3391 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3393 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3394 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3395 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3396 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3397 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3401 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
3403 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3404 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3405 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3406 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3407 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3408 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3413 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3414 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3415 XFS_RANDOM_IFLUSH_5)) {
3416 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3417 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3419 ip->i_d.di_nextents + ip->i_d.di_anextents,
3424 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3425 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3426 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3427 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3428 ip->i_ino, ip->i_d.di_forkoff, ip);
3432 * bump the flush iteration count, used to detect flushes which
3433 * postdate a log record during recovery.
3436 ip->i_d.di_flushiter++;
3439 * Copy the dirty parts of the inode into the on-disk
3440 * inode. We always copy out the core of the inode,
3441 * because if the inode is dirty at all the core must
3444 xfs_dinode_to_disk(&dip->di_core, &ip->i_d);
3446 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3447 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3448 ip->i_d.di_flushiter = 0;
3451 * If this is really an old format inode and the superblock version
3452 * has not been updated to support only new format inodes, then
3453 * convert back to the old inode format. If the superblock version
3454 * has been updated, then make the conversion permanent.
3456 ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 ||
3457 XFS_SB_VERSION_HASNLINK(&mp->m_sb));
3458 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
3459 if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) {
3463 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3464 dip->di_core.di_onlink = cpu_to_be16(ip->i_d.di_nlink);
3467 * The superblock version has already been bumped,
3468 * so just make the conversion to the new inode
3471 ip->i_d.di_version = XFS_DINODE_VERSION_2;
3472 dip->di_core.di_version = XFS_DINODE_VERSION_2;
3473 ip->i_d.di_onlink = 0;
3474 dip->di_core.di_onlink = 0;
3475 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3476 memset(&(dip->di_core.di_pad[0]), 0,
3477 sizeof(dip->di_core.di_pad));
3478 ASSERT(ip->i_d.di_projid == 0);
3482 if (xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp) == EFSCORRUPTED) {
3486 if (XFS_IFORK_Q(ip)) {
3488 * The only error from xfs_iflush_fork is on the data fork.
3490 (void) xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3492 xfs_inobp_check(mp, bp);
3495 * We've recorded everything logged in the inode, so we'd
3496 * like to clear the ilf_fields bits so we don't log and
3497 * flush things unnecessarily. However, we can't stop
3498 * logging all this information until the data we've copied
3499 * into the disk buffer is written to disk. If we did we might
3500 * overwrite the copy of the inode in the log with all the
3501 * data after re-logging only part of it, and in the face of
3502 * a crash we wouldn't have all the data we need to recover.
3504 * What we do is move the bits to the ili_last_fields field.
3505 * When logging the inode, these bits are moved back to the
3506 * ilf_fields field. In the xfs_iflush_done() routine we
3507 * clear ili_last_fields, since we know that the information
3508 * those bits represent is permanently on disk. As long as
3509 * the flush completes before the inode is logged again, then
3510 * both ilf_fields and ili_last_fields will be cleared.
3512 * We can play with the ilf_fields bits here, because the inode
3513 * lock must be held exclusively in order to set bits there
3514 * and the flush lock protects the ili_last_fields bits.
3515 * Set ili_logged so the flush done
3516 * routine can tell whether or not to look in the AIL.
3517 * Also, store the current LSN of the inode so that we can tell
3518 * whether the item has moved in the AIL from xfs_iflush_done().
3519 * In order to read the lsn we need the AIL lock, because
3520 * it is a 64 bit value that cannot be read atomically.
3522 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3523 iip->ili_last_fields = iip->ili_format.ilf_fields;
3524 iip->ili_format.ilf_fields = 0;
3525 iip->ili_logged = 1;
3527 ASSERT(sizeof(xfs_lsn_t) == 8); /* don't lock if it shrinks */
3528 spin_lock(&mp->m_ail_lock);
3529 iip->ili_flush_lsn = iip->ili_item.li_lsn;
3530 spin_unlock(&mp->m_ail_lock);
3533 * Attach the function xfs_iflush_done to the inode's
3534 * buffer. This will remove the inode from the AIL
3535 * and unlock the inode's flush lock when the inode is
3536 * completely written to disk.
3538 xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
3539 xfs_iflush_done, (xfs_log_item_t *)iip);
3541 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3542 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3545 * We're flushing an inode which is not in the AIL and has
3546 * not been logged but has i_update_core set. For this
3547 * case we can use a B_DELWRI flush and immediately drop
3548 * the inode flush lock because we can avoid the whole
3549 * AIL state thing. It's OK to drop the flush lock now,
3550 * because we've already locked the buffer and to do anything
3551 * you really need both.
3554 ASSERT(iip->ili_logged == 0);
3555 ASSERT(iip->ili_last_fields == 0);
3556 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3564 return XFS_ERROR(EFSCORRUPTED);
3569 * Flush all inactive inodes in mp.
3579 XFS_MOUNT_ILOCK(mp);
3585 /* Make sure we skip markers inserted by sync */
3586 if (ip->i_mount == NULL) {
3591 vp = XFS_ITOV_NULL(ip);
3593 XFS_MOUNT_IUNLOCK(mp);
3594 xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC);
3598 ASSERT(vn_count(vp) == 0);
3601 } while (ip != mp->m_inodes);
3603 XFS_MOUNT_IUNLOCK(mp);
3607 * xfs_iaccess: check accessibility of inode for mode.
3616 mode_t orgmode = mode;
3617 struct inode *inode = vn_to_inode(XFS_ITOV(ip));
3619 if (mode & S_IWUSR) {
3620 umode_t imode = inode->i_mode;
3622 if (IS_RDONLY(inode) &&
3623 (S_ISREG(imode) || S_ISDIR(imode) || S_ISLNK(imode)))
3624 return XFS_ERROR(EROFS);
3626 if (IS_IMMUTABLE(inode))
3627 return XFS_ERROR(EACCES);
3631 * If there's an Access Control List it's used instead of
3634 if ((error = _ACL_XFS_IACCESS(ip, mode, cr)) != -1)
3635 return error ? XFS_ERROR(error) : 0;
3637 if (current_fsuid(cr) != ip->i_d.di_uid) {
3639 if (!in_group_p((gid_t)ip->i_d.di_gid))
3644 * If the DACs are ok we don't need any capability check.
3646 if ((ip->i_d.di_mode & mode) == mode)
3649 * Read/write DACs are always overridable.
3650 * Executable DACs are overridable if at least one exec bit is set.
3652 if (!(orgmode & S_IXUSR) ||
3653 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
3654 if (capable_cred(cr, CAP_DAC_OVERRIDE))
3657 if ((orgmode == S_IRUSR) ||
3658 (S_ISDIR(inode->i_mode) && (!(orgmode & S_IWUSR)))) {
3659 if (capable_cred(cr, CAP_DAC_READ_SEARCH))
3662 cmn_err(CE_NOTE, "Ick: mode=%o, orgmode=%o", mode, orgmode);
3664 return XFS_ERROR(EACCES);
3666 return XFS_ERROR(EACCES);
3670 * xfs_iroundup: round up argument to next power of two
3679 if ((v & (v - 1)) == 0)
3681 ASSERT((v & 0x80000000) == 0);
3682 if ((v & (v + 1)) == 0)
3684 for (i = 0, m = 1; i < 31; i++, m <<= 1) {
3688 if ((v & (v + 1)) == 0)
3695 #ifdef XFS_ILOCK_TRACE
3696 ktrace_t *xfs_ilock_trace_buf;
3699 xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
3701 ktrace_enter(ip->i_lock_trace,
3703 (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
3704 (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
3705 (void *)ra, /* caller of ilock */
3706 (void *)(unsigned long)current_cpu(),
3707 (void *)(unsigned long)current_pid(),
3708 NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL);
3713 * Return a pointer to the extent record at file index idx.
3715 xfs_bmbt_rec_host_t *
3717 xfs_ifork_t *ifp, /* inode fork pointer */
3718 xfs_extnum_t idx) /* index of target extent */
3721 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
3722 return ifp->if_u1.if_ext_irec->er_extbuf;
3723 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3724 xfs_ext_irec_t *erp; /* irec pointer */
3725 int erp_idx = 0; /* irec index */
3726 xfs_extnum_t page_idx = idx; /* ext index in target list */
3728 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3729 return &erp->er_extbuf[page_idx];
3730 } else if (ifp->if_bytes) {
3731 return &ifp->if_u1.if_extents[idx];
3738 * Insert new item(s) into the extent records for incore inode
3739 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3743 xfs_ifork_t *ifp, /* inode fork pointer */
3744 xfs_extnum_t idx, /* starting index of new items */
3745 xfs_extnum_t count, /* number of inserted items */
3746 xfs_bmbt_irec_t *new) /* items to insert */
3748 xfs_extnum_t i; /* extent record index */
3750 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3751 xfs_iext_add(ifp, idx, count);
3752 for (i = idx; i < idx + count; i++, new++)
3753 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new);
3757 * This is called when the amount of space required for incore file
3758 * extents needs to be increased. The ext_diff parameter stores the
3759 * number of new extents being added and the idx parameter contains
3760 * the extent index where the new extents will be added. If the new
3761 * extents are being appended, then we just need to (re)allocate and
3762 * initialize the space. Otherwise, if the new extents are being
3763 * inserted into the middle of the existing entries, a bit more work
3764 * is required to make room for the new extents to be inserted. The
3765 * caller is responsible for filling in the new extent entries upon
3770 xfs_ifork_t *ifp, /* inode fork pointer */
3771 xfs_extnum_t idx, /* index to begin adding exts */
3772 int ext_diff) /* number of extents to add */
3774 int byte_diff; /* new bytes being added */
3775 int new_size; /* size of extents after adding */
3776 xfs_extnum_t nextents; /* number of extents in file */
3778 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3779 ASSERT((idx >= 0) && (idx <= nextents));
3780 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
3781 new_size = ifp->if_bytes + byte_diff;
3783 * If the new number of extents (nextents + ext_diff)
3784 * fits inside the inode, then continue to use the inline
3787 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
3788 if (idx < nextents) {
3789 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
3790 &ifp->if_u2.if_inline_ext[idx],
3791 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3792 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
3794 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3795 ifp->if_real_bytes = 0;
3796 ifp->if_lastex = nextents + ext_diff;
3799 * Otherwise use a linear (direct) extent list.
3800 * If the extents are currently inside the inode,
3801 * xfs_iext_realloc_direct will switch us from
3802 * inline to direct extent allocation mode.
3804 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
3805 xfs_iext_realloc_direct(ifp, new_size);
3806 if (idx < nextents) {
3807 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
3808 &ifp->if_u1.if_extents[idx],
3809 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3810 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
3813 /* Indirection array */
3815 xfs_ext_irec_t *erp;
3819 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
3820 if (ifp->if_flags & XFS_IFEXTIREC) {
3821 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3823 xfs_iext_irec_init(ifp);
3824 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3825 erp = ifp->if_u1.if_ext_irec;
3827 /* Extents fit in target extent page */
3828 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3829 if (page_idx < erp->er_extcount) {
3830 memmove(&erp->er_extbuf[page_idx + ext_diff],
3831 &erp->er_extbuf[page_idx],
3832 (erp->er_extcount - page_idx) *
3833 sizeof(xfs_bmbt_rec_t));
3834 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3836 erp->er_extcount += ext_diff;
3837 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3839 /* Insert a new extent page */
3841 xfs_iext_add_indirect_multi(ifp,
3842 erp_idx, page_idx, ext_diff);
3845 * If extent(s) are being appended to the last page in
3846 * the indirection array and the new extent(s) don't fit
3847 * in the page, then erp is NULL and erp_idx is set to
3848 * the next index needed in the indirection array.
3851 int count = ext_diff;
3854 erp = xfs_iext_irec_new(ifp, erp_idx);
3855 erp->er_extcount = count;
3856 count -= MIN(count, (int)XFS_LINEAR_EXTS);
3863 ifp->if_bytes = new_size;
3867 * This is called when incore extents are being added to the indirection
3868 * array and the new extents do not fit in the target extent list. The
3869 * erp_idx parameter contains the irec index for the target extent list
3870 * in the indirection array, and the idx parameter contains the extent
3871 * index within the list. The number of extents being added is stored
3872 * in the count parameter.
3874 * |-------| |-------|
3875 * | | | | idx - number of extents before idx
3877 * | | | | count - number of extents being inserted at idx
3878 * |-------| |-------|
3879 * | count | | nex2 | nex2 - number of extents after idx + count
3880 * |-------| |-------|
3883 xfs_iext_add_indirect_multi(
3884 xfs_ifork_t *ifp, /* inode fork pointer */
3885 int erp_idx, /* target extent irec index */
3886 xfs_extnum_t idx, /* index within target list */
3887 int count) /* new extents being added */
3889 int byte_diff; /* new bytes being added */
3890 xfs_ext_irec_t *erp; /* pointer to irec entry */
3891 xfs_extnum_t ext_diff; /* number of extents to add */
3892 xfs_extnum_t ext_cnt; /* new extents still needed */
3893 xfs_extnum_t nex2; /* extents after idx + count */
3894 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3895 int nlists; /* number of irec's (lists) */
3897 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3898 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3899 nex2 = erp->er_extcount - idx;
3900 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3903 * Save second part of target extent list
3904 * (all extents past */
3906 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3907 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_SLEEP);
3908 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3909 erp->er_extcount -= nex2;
3910 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3911 memset(&erp->er_extbuf[idx], 0, byte_diff);
3915 * Add the new extents to the end of the target
3916 * list, then allocate new irec record(s) and
3917 * extent buffer(s) as needed to store the rest
3918 * of the new extents.
3921 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3923 erp->er_extcount += ext_diff;
3924 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3925 ext_cnt -= ext_diff;
3929 erp = xfs_iext_irec_new(ifp, erp_idx);
3930 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3931 erp->er_extcount = ext_diff;
3932 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3933 ext_cnt -= ext_diff;
3936 /* Add nex2 extents back to indirection array */
3938 xfs_extnum_t ext_avail;
3941 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3942 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3945 * If nex2 extents fit in the current page, append
3946 * nex2_ep after the new extents.
3948 if (nex2 <= ext_avail) {
3949 i = erp->er_extcount;
3952 * Otherwise, check if space is available in the
3955 else if ((erp_idx < nlists - 1) &&
3956 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3957 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3960 /* Create a hole for nex2 extents */
3961 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3962 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3965 * Final choice, create a new extent page for
3970 erp = xfs_iext_irec_new(ifp, erp_idx);
3972 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
3973 kmem_free(nex2_ep, byte_diff);
3974 erp->er_extcount += nex2;
3975 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3980 * This is called when the amount of space required for incore file
3981 * extents needs to be decreased. The ext_diff parameter stores the
3982 * number of extents to be removed and the idx parameter contains
3983 * the extent index where the extents will be removed from.
3985 * If the amount of space needed has decreased below the linear
3986 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3987 * extent array. Otherwise, use kmem_realloc() to adjust the
3988 * size to what is needed.
3992 xfs_ifork_t *ifp, /* inode fork pointer */
3993 xfs_extnum_t idx, /* index to begin removing exts */
3994 int ext_diff) /* number of extents to remove */
3996 xfs_extnum_t nextents; /* number of extents in file */
3997 int new_size; /* size of extents after removal */
3999 ASSERT(ext_diff > 0);
4000 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4001 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
4003 if (new_size == 0) {
4004 xfs_iext_destroy(ifp);
4005 } else if (ifp->if_flags & XFS_IFEXTIREC) {
4006 xfs_iext_remove_indirect(ifp, idx, ext_diff);
4007 } else if (ifp->if_real_bytes) {
4008 xfs_iext_remove_direct(ifp, idx, ext_diff);
4010 xfs_iext_remove_inline(ifp, idx, ext_diff);
4012 ifp->if_bytes = new_size;
4016 * This removes ext_diff extents from the inline buffer, beginning
4017 * at extent index idx.
4020 xfs_iext_remove_inline(
4021 xfs_ifork_t *ifp, /* inode fork pointer */
4022 xfs_extnum_t idx, /* index to begin removing exts */
4023 int ext_diff) /* number of extents to remove */
4025 int nextents; /* number of extents in file */
4027 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4028 ASSERT(idx < XFS_INLINE_EXTS);
4029 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4030 ASSERT(((nextents - ext_diff) > 0) &&
4031 (nextents - ext_diff) < XFS_INLINE_EXTS);
4033 if (idx + ext_diff < nextents) {
4034 memmove(&ifp->if_u2.if_inline_ext[idx],
4035 &ifp->if_u2.if_inline_ext[idx + ext_diff],
4036 (nextents - (idx + ext_diff)) *
4037 sizeof(xfs_bmbt_rec_t));
4038 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
4039 0, ext_diff * sizeof(xfs_bmbt_rec_t));
4041 memset(&ifp->if_u2.if_inline_ext[idx], 0,
4042 ext_diff * sizeof(xfs_bmbt_rec_t));
4047 * This removes ext_diff extents from a linear (direct) extent list,
4048 * beginning at extent index idx. If the extents are being removed
4049 * from the end of the list (ie. truncate) then we just need to re-
4050 * allocate the list to remove the extra space. Otherwise, if the
4051 * extents are being removed from the middle of the existing extent
4052 * entries, then we first need to move the extent records beginning
4053 * at idx + ext_diff up in the list to overwrite the records being
4054 * removed, then remove the extra space via kmem_realloc.
4057 xfs_iext_remove_direct(
4058 xfs_ifork_t *ifp, /* inode fork pointer */
4059 xfs_extnum_t idx, /* index to begin removing exts */
4060 int ext_diff) /* number of extents to remove */
4062 xfs_extnum_t nextents; /* number of extents in file */
4063 int new_size; /* size of extents after removal */
4065 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4066 new_size = ifp->if_bytes -
4067 (ext_diff * sizeof(xfs_bmbt_rec_t));
4068 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4070 if (new_size == 0) {
4071 xfs_iext_destroy(ifp);
4074 /* Move extents up in the list (if needed) */
4075 if (idx + ext_diff < nextents) {
4076 memmove(&ifp->if_u1.if_extents[idx],
4077 &ifp->if_u1.if_extents[idx + ext_diff],
4078 (nextents - (idx + ext_diff)) *
4079 sizeof(xfs_bmbt_rec_t));
4081 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
4082 0, ext_diff * sizeof(xfs_bmbt_rec_t));
4084 * Reallocate the direct extent list. If the extents
4085 * will fit inside the inode then xfs_iext_realloc_direct
4086 * will switch from direct to inline extent allocation
4089 xfs_iext_realloc_direct(ifp, new_size);
4090 ifp->if_bytes = new_size;
4094 * This is called when incore extents are being removed from the
4095 * indirection array and the extents being removed span multiple extent
4096 * buffers. The idx parameter contains the file extent index where we
4097 * want to begin removing extents, and the count parameter contains
4098 * how many extents need to be removed.
4100 * |-------| |-------|
4101 * | nex1 | | | nex1 - number of extents before idx
4102 * |-------| | count |
4103 * | | | | count - number of extents being removed at idx
4104 * | count | |-------|
4105 * | | | nex2 | nex2 - number of extents after idx + count
4106 * |-------| |-------|
4109 xfs_iext_remove_indirect(
4110 xfs_ifork_t *ifp, /* inode fork pointer */
4111 xfs_extnum_t idx, /* index to begin removing extents */
4112 int count) /* number of extents to remove */
4114 xfs_ext_irec_t *erp; /* indirection array pointer */
4115 int erp_idx = 0; /* indirection array index */
4116 xfs_extnum_t ext_cnt; /* extents left to remove */
4117 xfs_extnum_t ext_diff; /* extents to remove in current list */
4118 xfs_extnum_t nex1; /* number of extents before idx */
4119 xfs_extnum_t nex2; /* extents after idx + count */
4120 int nlists; /* entries in indirection array */
4121 int page_idx = idx; /* index in target extent list */
4123 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4124 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
4125 ASSERT(erp != NULL);
4126 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4130 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
4131 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
4133 * Check for deletion of entire list;
4134 * xfs_iext_irec_remove() updates extent offsets.
4136 if (ext_diff == erp->er_extcount) {
4137 xfs_iext_irec_remove(ifp, erp_idx);
4138 ext_cnt -= ext_diff;
4141 ASSERT(erp_idx < ifp->if_real_bytes /
4143 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4150 /* Move extents up (if needed) */
4152 memmove(&erp->er_extbuf[nex1],
4153 &erp->er_extbuf[nex1 + ext_diff],
4154 nex2 * sizeof(xfs_bmbt_rec_t));
4156 /* Zero out rest of page */
4157 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
4158 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
4159 /* Update remaining counters */
4160 erp->er_extcount -= ext_diff;
4161 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
4162 ext_cnt -= ext_diff;
4167 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
4168 xfs_iext_irec_compact(ifp);
4172 * Create, destroy, or resize a linear (direct) block of extents.
4175 xfs_iext_realloc_direct(
4176 xfs_ifork_t *ifp, /* inode fork pointer */
4177 int new_size) /* new size of extents */
4179 int rnew_size; /* real new size of extents */
4181 rnew_size = new_size;
4183 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
4184 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
4185 (new_size != ifp->if_real_bytes)));
4187 /* Free extent records */
4188 if (new_size == 0) {
4189 xfs_iext_destroy(ifp);
4191 /* Resize direct extent list and zero any new bytes */
4192 else if (ifp->if_real_bytes) {
4193 /* Check if extents will fit inside the inode */
4194 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
4195 xfs_iext_direct_to_inline(ifp, new_size /
4196 (uint)sizeof(xfs_bmbt_rec_t));
4197 ifp->if_bytes = new_size;
4200 if (!is_power_of_2(new_size)){
4201 rnew_size = xfs_iroundup(new_size);
4203 if (rnew_size != ifp->if_real_bytes) {
4204 ifp->if_u1.if_extents =
4205 kmem_realloc(ifp->if_u1.if_extents,
4210 if (rnew_size > ifp->if_real_bytes) {
4211 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
4212 (uint)sizeof(xfs_bmbt_rec_t)], 0,
4213 rnew_size - ifp->if_real_bytes);
4217 * Switch from the inline extent buffer to a direct
4218 * extent list. Be sure to include the inline extent
4219 * bytes in new_size.
4222 new_size += ifp->if_bytes;
4223 if (!is_power_of_2(new_size)) {
4224 rnew_size = xfs_iroundup(new_size);
4226 xfs_iext_inline_to_direct(ifp, rnew_size);
4228 ifp->if_real_bytes = rnew_size;
4229 ifp->if_bytes = new_size;
4233 * Switch from linear (direct) extent records to inline buffer.
4236 xfs_iext_direct_to_inline(
4237 xfs_ifork_t *ifp, /* inode fork pointer */
4238 xfs_extnum_t nextents) /* number of extents in file */
4240 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4241 ASSERT(nextents <= XFS_INLINE_EXTS);
4243 * The inline buffer was zeroed when we switched
4244 * from inline to direct extent allocation mode,
4245 * so we don't need to clear it here.
4247 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
4248 nextents * sizeof(xfs_bmbt_rec_t));
4249 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
4250 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
4251 ifp->if_real_bytes = 0;
4255 * Switch from inline buffer to linear (direct) extent records.
4256 * new_size should already be rounded up to the next power of 2
4257 * by the caller (when appropriate), so use new_size as it is.
4258 * However, since new_size may be rounded up, we can't update
4259 * if_bytes here. It is the caller's responsibility to update
4260 * if_bytes upon return.
4263 xfs_iext_inline_to_direct(
4264 xfs_ifork_t *ifp, /* inode fork pointer */
4265 int new_size) /* number of extents in file */
4267 ifp->if_u1.if_extents = kmem_alloc(new_size, KM_SLEEP);
4268 memset(ifp->if_u1.if_extents, 0, new_size);
4269 if (ifp->if_bytes) {
4270 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
4272 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4273 sizeof(xfs_bmbt_rec_t));
4275 ifp->if_real_bytes = new_size;
4279 * Resize an extent indirection array to new_size bytes.
4282 xfs_iext_realloc_indirect(
4283 xfs_ifork_t *ifp, /* inode fork pointer */
4284 int new_size) /* new indirection array size */
4286 int nlists; /* number of irec's (ex lists) */
4287 int size; /* current indirection array size */
4289 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4290 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4291 size = nlists * sizeof(xfs_ext_irec_t);
4292 ASSERT(ifp->if_real_bytes);
4293 ASSERT((new_size >= 0) && (new_size != size));
4294 if (new_size == 0) {
4295 xfs_iext_destroy(ifp);
4297 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
4298 kmem_realloc(ifp->if_u1.if_ext_irec,
4299 new_size, size, KM_SLEEP);
4304 * Switch from indirection array to linear (direct) extent allocations.
4307 xfs_iext_indirect_to_direct(
4308 xfs_ifork_t *ifp) /* inode fork pointer */
4310 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
4311 xfs_extnum_t nextents; /* number of extents in file */
4312 int size; /* size of file extents */
4314 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4315 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4316 ASSERT(nextents <= XFS_LINEAR_EXTS);
4317 size = nextents * sizeof(xfs_bmbt_rec_t);
4319 xfs_iext_irec_compact_full(ifp);
4320 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
4322 ep = ifp->if_u1.if_ext_irec->er_extbuf;
4323 kmem_free(ifp->if_u1.if_ext_irec, sizeof(xfs_ext_irec_t));
4324 ifp->if_flags &= ~XFS_IFEXTIREC;
4325 ifp->if_u1.if_extents = ep;
4326 ifp->if_bytes = size;
4327 if (nextents < XFS_LINEAR_EXTS) {
4328 xfs_iext_realloc_direct(ifp, size);
4333 * Free incore file extents.
4337 xfs_ifork_t *ifp) /* inode fork pointer */
4339 if (ifp->if_flags & XFS_IFEXTIREC) {
4343 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4344 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
4345 xfs_iext_irec_remove(ifp, erp_idx);
4347 ifp->if_flags &= ~XFS_IFEXTIREC;
4348 } else if (ifp->if_real_bytes) {
4349 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
4350 } else if (ifp->if_bytes) {
4351 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4352 sizeof(xfs_bmbt_rec_t));
4354 ifp->if_u1.if_extents = NULL;
4355 ifp->if_real_bytes = 0;
4360 * Return a pointer to the extent record for file system block bno.
4362 xfs_bmbt_rec_host_t * /* pointer to found extent record */
4363 xfs_iext_bno_to_ext(
4364 xfs_ifork_t *ifp, /* inode fork pointer */
4365 xfs_fileoff_t bno, /* block number to search for */
4366 xfs_extnum_t *idxp) /* index of target extent */
4368 xfs_bmbt_rec_host_t *base; /* pointer to first extent */
4369 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
4370 xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */
4371 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4372 int high; /* upper boundary in search */
4373 xfs_extnum_t idx = 0; /* index of target extent */
4374 int low; /* lower boundary in search */
4375 xfs_extnum_t nextents; /* number of file extents */
4376 xfs_fileoff_t startoff = 0; /* start offset of extent */
4378 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4379 if (nextents == 0) {
4384 if (ifp->if_flags & XFS_IFEXTIREC) {
4385 /* Find target extent list */
4387 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
4388 base = erp->er_extbuf;
4389 high = erp->er_extcount - 1;
4391 base = ifp->if_u1.if_extents;
4392 high = nextents - 1;
4394 /* Binary search extent records */
4395 while (low <= high) {
4396 idx = (low + high) >> 1;
4398 startoff = xfs_bmbt_get_startoff(ep);
4399 blockcount = xfs_bmbt_get_blockcount(ep);
4400 if (bno < startoff) {
4402 } else if (bno >= startoff + blockcount) {
4405 /* Convert back to file-based extent index */
4406 if (ifp->if_flags & XFS_IFEXTIREC) {
4407 idx += erp->er_extoff;
4413 /* Convert back to file-based extent index */
4414 if (ifp->if_flags & XFS_IFEXTIREC) {
4415 idx += erp->er_extoff;
4417 if (bno >= startoff + blockcount) {
4418 if (++idx == nextents) {
4421 ep = xfs_iext_get_ext(ifp, idx);
4429 * Return a pointer to the indirection array entry containing the
4430 * extent record for filesystem block bno. Store the index of the
4431 * target irec in *erp_idxp.
4433 xfs_ext_irec_t * /* pointer to found extent record */
4434 xfs_iext_bno_to_irec(
4435 xfs_ifork_t *ifp, /* inode fork pointer */
4436 xfs_fileoff_t bno, /* block number to search for */
4437 int *erp_idxp) /* irec index of target ext list */
4439 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4440 xfs_ext_irec_t *erp_next; /* next indirection array entry */
4441 int erp_idx; /* indirection array index */
4442 int nlists; /* number of extent irec's (lists) */
4443 int high; /* binary search upper limit */
4444 int low; /* binary search lower limit */
4446 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4447 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4451 while (low <= high) {
4452 erp_idx = (low + high) >> 1;
4453 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4454 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
4455 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
4457 } else if (erp_next && bno >=
4458 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
4464 *erp_idxp = erp_idx;
4469 * Return a pointer to the indirection array entry containing the
4470 * extent record at file extent index *idxp. Store the index of the
4471 * target irec in *erp_idxp and store the page index of the target
4472 * extent record in *idxp.
4475 xfs_iext_idx_to_irec(
4476 xfs_ifork_t *ifp, /* inode fork pointer */
4477 xfs_extnum_t *idxp, /* extent index (file -> page) */
4478 int *erp_idxp, /* pointer to target irec */
4479 int realloc) /* new bytes were just added */
4481 xfs_ext_irec_t *prev; /* pointer to previous irec */
4482 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
4483 int erp_idx; /* indirection array index */
4484 int nlists; /* number of irec's (ex lists) */
4485 int high; /* binary search upper limit */
4486 int low; /* binary search lower limit */
4487 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
4489 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4490 ASSERT(page_idx >= 0 && page_idx <=
4491 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
4492 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4497 /* Binary search extent irec's */
4498 while (low <= high) {
4499 erp_idx = (low + high) >> 1;
4500 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4501 prev = erp_idx > 0 ? erp - 1 : NULL;
4502 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
4503 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
4505 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
4506 (page_idx == erp->er_extoff + erp->er_extcount &&
4509 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
4510 erp->er_extcount == XFS_LINEAR_EXTS) {
4514 erp = erp_idx < nlists ? erp + 1 : NULL;
4517 page_idx -= erp->er_extoff;
4522 *erp_idxp = erp_idx;
4527 * Allocate and initialize an indirection array once the space needed
4528 * for incore extents increases above XFS_IEXT_BUFSZ.
4532 xfs_ifork_t *ifp) /* inode fork pointer */
4534 xfs_ext_irec_t *erp; /* indirection array pointer */
4535 xfs_extnum_t nextents; /* number of extents in file */
4537 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4538 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4539 ASSERT(nextents <= XFS_LINEAR_EXTS);
4541 erp = (xfs_ext_irec_t *)
4542 kmem_alloc(sizeof(xfs_ext_irec_t), KM_SLEEP);
4544 if (nextents == 0) {
4545 ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
4546 } else if (!ifp->if_real_bytes) {
4547 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
4548 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
4549 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
4551 erp->er_extbuf = ifp->if_u1.if_extents;
4552 erp->er_extcount = nextents;
4555 ifp->if_flags |= XFS_IFEXTIREC;
4556 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
4557 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
4558 ifp->if_u1.if_ext_irec = erp;
4564 * Allocate and initialize a new entry in the indirection array.
4568 xfs_ifork_t *ifp, /* inode fork pointer */
4569 int erp_idx) /* index for new irec */
4571 xfs_ext_irec_t *erp; /* indirection array pointer */
4572 int i; /* loop counter */
4573 int nlists; /* number of irec's (ex lists) */
4575 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4576 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4578 /* Resize indirection array */
4579 xfs_iext_realloc_indirect(ifp, ++nlists *
4580 sizeof(xfs_ext_irec_t));
4582 * Move records down in the array so the
4583 * new page can use erp_idx.
4585 erp = ifp->if_u1.if_ext_irec;
4586 for (i = nlists - 1; i > erp_idx; i--) {
4587 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
4589 ASSERT(i == erp_idx);
4591 /* Initialize new extent record */
4592 erp = ifp->if_u1.if_ext_irec;
4593 erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
4594 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4595 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
4596 erp[erp_idx].er_extcount = 0;
4597 erp[erp_idx].er_extoff = erp_idx > 0 ?
4598 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
4599 return (&erp[erp_idx]);
4603 * Remove a record from the indirection array.
4606 xfs_iext_irec_remove(
4607 xfs_ifork_t *ifp, /* inode fork pointer */
4608 int erp_idx) /* irec index to remove */
4610 xfs_ext_irec_t *erp; /* indirection array pointer */
4611 int i; /* loop counter */
4612 int nlists; /* number of irec's (ex lists) */
4614 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4615 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4616 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4617 if (erp->er_extbuf) {
4618 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
4620 kmem_free(erp->er_extbuf, XFS_IEXT_BUFSZ);
4622 /* Compact extent records */
4623 erp = ifp->if_u1.if_ext_irec;
4624 for (i = erp_idx; i < nlists - 1; i++) {
4625 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
4628 * Manually free the last extent record from the indirection
4629 * array. A call to xfs_iext_realloc_indirect() with a size
4630 * of zero would result in a call to xfs_iext_destroy() which
4631 * would in turn call this function again, creating a nasty
4635 xfs_iext_realloc_indirect(ifp,
4636 nlists * sizeof(xfs_ext_irec_t));
4638 kmem_free(ifp->if_u1.if_ext_irec,
4639 sizeof(xfs_ext_irec_t));
4641 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4645 * This is called to clean up large amounts of unused memory allocated
4646 * by the indirection array. Before compacting anything though, verify
4647 * that the indirection array is still needed and switch back to the
4648 * linear extent list (or even the inline buffer) if possible. The
4649 * compaction policy is as follows:
4651 * Full Compaction: Extents fit into a single page (or inline buffer)
4652 * Full Compaction: Extents occupy less than 10% of allocated space
4653 * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
4654 * No Compaction: Extents occupy at least 50% of allocated space
4657 xfs_iext_irec_compact(
4658 xfs_ifork_t *ifp) /* inode fork pointer */
4660 xfs_extnum_t nextents; /* number of extents in file */
4661 int nlists; /* number of irec's (ex lists) */
4663 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4664 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4665 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4667 if (nextents == 0) {
4668 xfs_iext_destroy(ifp);
4669 } else if (nextents <= XFS_INLINE_EXTS) {
4670 xfs_iext_indirect_to_direct(ifp);
4671 xfs_iext_direct_to_inline(ifp, nextents);
4672 } else if (nextents <= XFS_LINEAR_EXTS) {
4673 xfs_iext_indirect_to_direct(ifp);
4674 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 3) {
4675 xfs_iext_irec_compact_full(ifp);
4676 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
4677 xfs_iext_irec_compact_pages(ifp);
4682 * Combine extents from neighboring extent pages.
4685 xfs_iext_irec_compact_pages(
4686 xfs_ifork_t *ifp) /* inode fork pointer */
4688 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
4689 int erp_idx = 0; /* indirection array index */
4690 int nlists; /* number of irec's (ex lists) */
4692 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4693 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4694 while (erp_idx < nlists - 1) {
4695 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4697 if (erp_next->er_extcount <=
4698 (XFS_LINEAR_EXTS - erp->er_extcount)) {
4699 memmove(&erp->er_extbuf[erp->er_extcount],
4700 erp_next->er_extbuf, erp_next->er_extcount *
4701 sizeof(xfs_bmbt_rec_t));
4702 erp->er_extcount += erp_next->er_extcount;
4704 * Free page before removing extent record
4705 * so er_extoffs don't get modified in
4706 * xfs_iext_irec_remove.
4708 kmem_free(erp_next->er_extbuf, XFS_IEXT_BUFSZ);
4709 erp_next->er_extbuf = NULL;
4710 xfs_iext_irec_remove(ifp, erp_idx + 1);
4711 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4719 * Fully compact the extent records managed by the indirection array.
4722 xfs_iext_irec_compact_full(
4723 xfs_ifork_t *ifp) /* inode fork pointer */
4725 xfs_bmbt_rec_host_t *ep, *ep_next; /* extent record pointers */
4726 xfs_ext_irec_t *erp, *erp_next; /* extent irec pointers */
4727 int erp_idx = 0; /* extent irec index */
4728 int ext_avail; /* empty entries in ex list */
4729 int ext_diff; /* number of exts to add */
4730 int nlists; /* number of irec's (ex lists) */
4732 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4733 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4734 erp = ifp->if_u1.if_ext_irec;
4735 ep = &erp->er_extbuf[erp->er_extcount];
4737 ep_next = erp_next->er_extbuf;
4738 while (erp_idx < nlists - 1) {
4739 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
4740 ext_diff = MIN(ext_avail, erp_next->er_extcount);
4741 memcpy(ep, ep_next, ext_diff * sizeof(xfs_bmbt_rec_t));
4742 erp->er_extcount += ext_diff;
4743 erp_next->er_extcount -= ext_diff;
4744 /* Remove next page */
4745 if (erp_next->er_extcount == 0) {
4747 * Free page before removing extent record
4748 * so er_extoffs don't get modified in
4749 * xfs_iext_irec_remove.
4751 kmem_free(erp_next->er_extbuf,
4752 erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
4753 erp_next->er_extbuf = NULL;
4754 xfs_iext_irec_remove(ifp, erp_idx + 1);
4755 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4756 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4757 /* Update next page */
4759 /* Move rest of page up to become next new page */
4760 memmove(erp_next->er_extbuf, ep_next,
4761 erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
4762 ep_next = erp_next->er_extbuf;
4763 memset(&ep_next[erp_next->er_extcount], 0,
4764 (XFS_LINEAR_EXTS - erp_next->er_extcount) *
4765 sizeof(xfs_bmbt_rec_t));
4767 if (erp->er_extcount == XFS_LINEAR_EXTS) {
4769 if (erp_idx < nlists)
4770 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4774 ep = &erp->er_extbuf[erp->er_extcount];
4776 ep_next = erp_next->er_extbuf;
4781 * This is called to update the er_extoff field in the indirection
4782 * array when extents have been added or removed from one of the
4783 * extent lists. erp_idx contains the irec index to begin updating
4784 * at and ext_diff contains the number of extents that were added
4788 xfs_iext_irec_update_extoffs(
4789 xfs_ifork_t *ifp, /* inode fork pointer */
4790 int erp_idx, /* irec index to update */
4791 int ext_diff) /* number of new extents */
4793 int i; /* loop counter */
4794 int nlists; /* number of irec's (ex lists */
4796 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4797 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4798 for (i = erp_idx; i < nlists; i++) {
4799 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;