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 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1226 di_flags |= XFS_DIFLAG_EXTSIZE;
1227 ip->i_d.di_extsize = pip->i_d.di_extsize;
1230 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1231 xfs_inherit_noatime)
1232 di_flags |= XFS_DIFLAG_NOATIME;
1233 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1235 di_flags |= XFS_DIFLAG_NODUMP;
1236 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1238 di_flags |= XFS_DIFLAG_SYNC;
1239 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1240 xfs_inherit_nosymlinks)
1241 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1242 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1243 di_flags |= XFS_DIFLAG_PROJINHERIT;
1244 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1245 xfs_inherit_nodefrag)
1246 di_flags |= XFS_DIFLAG_NODEFRAG;
1247 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
1248 di_flags |= XFS_DIFLAG_FILESTREAM;
1249 ip->i_d.di_flags |= di_flags;
1253 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1254 ip->i_df.if_flags = XFS_IFEXTENTS;
1255 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1256 ip->i_df.if_u1.if_extents = NULL;
1262 * Attribute fork settings for new inode.
1264 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1265 ip->i_d.di_anextents = 0;
1268 * Log the new values stuffed into the inode.
1270 xfs_trans_log_inode(tp, ip, flags);
1272 /* now that we have an i_mode we can setup inode ops and unlock */
1273 xfs_initialize_vnode(tp->t_mountp, vp, ip);
1280 * Check to make sure that there are no blocks allocated to the
1281 * file beyond the size of the file. We don't check this for
1282 * files with fixed size extents or real time extents, but we
1283 * at least do it for regular files.
1292 xfs_fileoff_t map_first;
1294 xfs_bmbt_irec_t imaps[2];
1296 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1299 if (ip->i_d.di_flags & (XFS_DIFLAG_REALTIME | XFS_DIFLAG_EXTSIZE))
1303 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1305 * The filesystem could be shutting down, so bmapi may return
1308 if (xfs_bmapi(NULL, ip, map_first,
1310 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1312 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1315 ASSERT(nimaps == 1);
1316 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1321 * Calculate the last possible buffered byte in a file. This must
1322 * include data that was buffered beyond the EOF by the write code.
1323 * This also needs to deal with overflowing the xfs_fsize_t type
1324 * which can happen for sizes near the limit.
1326 * We also need to take into account any blocks beyond the EOF. It
1327 * may be the case that they were buffered by a write which failed.
1328 * In that case the pages will still be in memory, but the inode size
1329 * will never have been updated.
1336 xfs_fsize_t last_byte;
1337 xfs_fileoff_t last_block;
1338 xfs_fileoff_t size_last_block;
1341 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE | MR_ACCESS));
1345 * Only check for blocks beyond the EOF if the extents have
1346 * been read in. This eliminates the need for the inode lock,
1347 * and it also saves us from looking when it really isn't
1350 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1351 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1359 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_size);
1360 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1362 last_byte = XFS_FSB_TO_B(mp, last_block);
1363 if (last_byte < 0) {
1364 return XFS_MAXIOFFSET(mp);
1366 last_byte += (1 << mp->m_writeio_log);
1367 if (last_byte < 0) {
1368 return XFS_MAXIOFFSET(mp);
1373 #if defined(XFS_RW_TRACE)
1379 xfs_fsize_t new_size,
1380 xfs_off_t toss_start,
1381 xfs_off_t toss_finish)
1383 if (ip->i_rwtrace == NULL) {
1387 ktrace_enter(ip->i_rwtrace,
1390 (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
1391 (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
1392 (void*)((long)flag),
1393 (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
1394 (void*)(unsigned long)(new_size & 0xffffffff),
1395 (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
1396 (void*)(unsigned long)(toss_start & 0xffffffff),
1397 (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
1398 (void*)(unsigned long)(toss_finish & 0xffffffff),
1399 (void*)(unsigned long)current_cpu(),
1400 (void*)(unsigned long)current_pid(),
1406 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1410 * Start the truncation of the file to new_size. The new size
1411 * must be smaller than the current size. This routine will
1412 * clear the buffer and page caches of file data in the removed
1413 * range, and xfs_itruncate_finish() will remove the underlying
1416 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1417 * must NOT have the inode lock held at all. This is because we're
1418 * calling into the buffer/page cache code and we can't hold the
1419 * inode lock when we do so.
1421 * We need to wait for any direct I/Os in flight to complete before we
1422 * proceed with the truncate. This is needed to prevent the extents
1423 * being read or written by the direct I/Os from being removed while the
1424 * I/O is in flight as there is no other method of synchronising
1425 * direct I/O with the truncate operation. Also, because we hold
1426 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1427 * started until the truncate completes and drops the lock. Essentially,
1428 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1429 * between direct I/Os and the truncate operation.
1431 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1432 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1433 * in the case that the caller is locking things out of order and
1434 * may not be able to call xfs_itruncate_finish() with the inode lock
1435 * held without dropping the I/O lock. If the caller must drop the
1436 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1437 * must be called again with all the same restrictions as the initial
1441 xfs_itruncate_start(
1444 xfs_fsize_t new_size)
1446 xfs_fsize_t last_byte;
1447 xfs_off_t toss_start;
1452 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1453 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1454 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1455 (flags == XFS_ITRUNC_MAYBE));
1460 /* wait for the completion of any pending DIOs */
1461 if (new_size < ip->i_size)
1465 * Call toss_pages or flushinval_pages to get rid of pages
1466 * overlapping the region being removed. We have to use
1467 * the less efficient flushinval_pages in the case that the
1468 * caller may not be able to finish the truncate without
1469 * dropping the inode's I/O lock. Make sure
1470 * to catch any pages brought in by buffers overlapping
1471 * the EOF by searching out beyond the isize by our
1472 * block size. We round new_size up to a block boundary
1473 * so that we don't toss things on the same block as
1474 * new_size but before it.
1476 * Before calling toss_page or flushinval_pages, make sure to
1477 * call remapf() over the same region if the file is mapped.
1478 * This frees up mapped file references to the pages in the
1479 * given range and for the flushinval_pages case it ensures
1480 * that we get the latest mapped changes flushed out.
1482 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1483 toss_start = XFS_FSB_TO_B(mp, toss_start);
1484 if (toss_start < 0) {
1486 * The place to start tossing is beyond our maximum
1487 * file size, so there is no way that the data extended
1492 last_byte = xfs_file_last_byte(ip);
1493 xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
1495 if (last_byte > toss_start) {
1496 if (flags & XFS_ITRUNC_DEFINITE) {
1497 xfs_tosspages(ip, toss_start,
1498 -1, FI_REMAPF_LOCKED);
1500 error = xfs_flushinval_pages(ip, toss_start,
1501 -1, FI_REMAPF_LOCKED);
1506 if (new_size == 0) {
1507 ASSERT(VN_CACHED(vp) == 0);
1514 * Shrink the file to the given new_size. The new
1515 * size must be smaller than the current size.
1516 * This will free up the underlying blocks
1517 * in the removed range after a call to xfs_itruncate_start()
1518 * or xfs_atruncate_start().
1520 * The transaction passed to this routine must have made
1521 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1522 * This routine may commit the given transaction and
1523 * start new ones, so make sure everything involved in
1524 * the transaction is tidy before calling here.
1525 * Some transaction will be returned to the caller to be
1526 * committed. The incoming transaction must already include
1527 * the inode, and both inode locks must be held exclusively.
1528 * The inode must also be "held" within the transaction. On
1529 * return the inode will be "held" within the returned transaction.
1530 * This routine does NOT require any disk space to be reserved
1531 * for it within the transaction.
1533 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1534 * and it indicates the fork which is to be truncated. For the
1535 * attribute fork we only support truncation to size 0.
1537 * We use the sync parameter to indicate whether or not the first
1538 * transaction we perform might have to be synchronous. For the attr fork,
1539 * it needs to be so if the unlink of the inode is not yet known to be
1540 * permanent in the log. This keeps us from freeing and reusing the
1541 * blocks of the attribute fork before the unlink of the inode becomes
1544 * For the data fork, we normally have to run synchronously if we're
1545 * being called out of the inactive path or we're being called
1546 * out of the create path where we're truncating an existing file.
1547 * Either way, the truncate needs to be sync so blocks don't reappear
1548 * in the file with altered data in case of a crash. wsync filesystems
1549 * can run the first case async because anything that shrinks the inode
1550 * has to run sync so by the time we're called here from inactive, the
1551 * inode size is permanently set to 0.
1553 * Calls from the truncate path always need to be sync unless we're
1554 * in a wsync filesystem and the file has already been unlinked.
1556 * The caller is responsible for correctly setting the sync parameter.
1557 * It gets too hard for us to guess here which path we're being called
1558 * out of just based on inode state.
1561 xfs_itruncate_finish(
1564 xfs_fsize_t new_size,
1568 xfs_fsblock_t first_block;
1569 xfs_fileoff_t first_unmap_block;
1570 xfs_fileoff_t last_block;
1571 xfs_filblks_t unmap_len=0;
1576 xfs_bmap_free_t free_list;
1579 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1580 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
1581 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1582 ASSERT(*tp != NULL);
1583 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1584 ASSERT(ip->i_transp == *tp);
1585 ASSERT(ip->i_itemp != NULL);
1586 ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
1590 mp = (ntp)->t_mountp;
1591 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1594 * We only support truncating the entire attribute fork.
1596 if (fork == XFS_ATTR_FORK) {
1599 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1600 xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
1602 * The first thing we do is set the size to new_size permanently
1603 * on disk. This way we don't have to worry about anyone ever
1604 * being able to look at the data being freed even in the face
1605 * of a crash. What we're getting around here is the case where
1606 * we free a block, it is allocated to another file, it is written
1607 * to, and then we crash. If the new data gets written to the
1608 * file but the log buffers containing the free and reallocation
1609 * don't, then we'd end up with garbage in the blocks being freed.
1610 * As long as we make the new_size permanent before actually
1611 * freeing any blocks it doesn't matter if they get writtten to.
1613 * The callers must signal into us whether or not the size
1614 * setting here must be synchronous. There are a few cases
1615 * where it doesn't have to be synchronous. Those cases
1616 * occur if the file is unlinked and we know the unlink is
1617 * permanent or if the blocks being truncated are guaranteed
1618 * to be beyond the inode eof (regardless of the link count)
1619 * and the eof value is permanent. Both of these cases occur
1620 * only on wsync-mounted filesystems. In those cases, we're
1621 * guaranteed that no user will ever see the data in the blocks
1622 * that are being truncated so the truncate can run async.
1623 * In the free beyond eof case, the file may wind up with
1624 * more blocks allocated to it than it needs if we crash
1625 * and that won't get fixed until the next time the file
1626 * is re-opened and closed but that's ok as that shouldn't
1627 * be too many blocks.
1629 * However, we can't just make all wsync xactions run async
1630 * because there's one call out of the create path that needs
1631 * to run sync where it's truncating an existing file to size
1632 * 0 whose size is > 0.
1634 * It's probably possible to come up with a test in this
1635 * routine that would correctly distinguish all the above
1636 * cases from the values of the function parameters and the
1637 * inode state but for sanity's sake, I've decided to let the
1638 * layers above just tell us. It's simpler to correctly figure
1639 * out in the layer above exactly under what conditions we
1640 * can run async and I think it's easier for others read and
1641 * follow the logic in case something has to be changed.
1642 * cscope is your friend -- rcc.
1644 * The attribute fork is much simpler.
1646 * For the attribute fork we allow the caller to tell us whether
1647 * the unlink of the inode that led to this call is yet permanent
1648 * in the on disk log. If it is not and we will be freeing extents
1649 * in this inode then we make the first transaction synchronous
1650 * to make sure that the unlink is permanent by the time we free
1653 if (fork == XFS_DATA_FORK) {
1654 if (ip->i_d.di_nextents > 0) {
1656 * If we are not changing the file size then do
1657 * not update the on-disk file size - we may be
1658 * called from xfs_inactive_free_eofblocks(). If we
1659 * update the on-disk file size and then the system
1660 * crashes before the contents of the file are
1661 * flushed to disk then the files may be full of
1662 * holes (ie NULL files bug).
1664 if (ip->i_size != new_size) {
1665 ip->i_d.di_size = new_size;
1666 ip->i_size = new_size;
1667 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1671 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1672 if (ip->i_d.di_anextents > 0)
1673 xfs_trans_set_sync(ntp);
1675 ASSERT(fork == XFS_DATA_FORK ||
1676 (fork == XFS_ATTR_FORK &&
1677 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1678 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1681 * Since it is possible for space to become allocated beyond
1682 * the end of the file (in a crash where the space is allocated
1683 * but the inode size is not yet updated), simply remove any
1684 * blocks which show up between the new EOF and the maximum
1685 * possible file size. If the first block to be removed is
1686 * beyond the maximum file size (ie it is the same as last_block),
1687 * then there is nothing to do.
1689 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1690 ASSERT(first_unmap_block <= last_block);
1692 if (last_block == first_unmap_block) {
1695 unmap_len = last_block - first_unmap_block + 1;
1699 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1700 * will tell us whether it freed the entire range or
1701 * not. If this is a synchronous mount (wsync),
1702 * then we can tell bunmapi to keep all the
1703 * transactions asynchronous since the unlink
1704 * transaction that made this inode inactive has
1705 * already hit the disk. There's no danger of
1706 * the freed blocks being reused, there being a
1707 * crash, and the reused blocks suddenly reappearing
1708 * in this file with garbage in them once recovery
1711 XFS_BMAP_INIT(&free_list, &first_block);
1712 error = xfs_bunmapi(ntp, ip,
1713 first_unmap_block, unmap_len,
1714 XFS_BMAPI_AFLAG(fork) |
1715 (sync ? 0 : XFS_BMAPI_ASYNC),
1716 XFS_ITRUNC_MAX_EXTENTS,
1717 &first_block, &free_list,
1721 * If the bunmapi call encounters an error,
1722 * return to the caller where the transaction
1723 * can be properly aborted. We just need to
1724 * make sure we're not holding any resources
1725 * that we were not when we came in.
1727 xfs_bmap_cancel(&free_list);
1732 * Duplicate the transaction that has the permanent
1733 * reservation and commit the old transaction.
1735 error = xfs_bmap_finish(tp, &free_list, &committed);
1739 * If the bmap finish call encounters an error,
1740 * return to the caller where the transaction
1741 * can be properly aborted. We just need to
1742 * make sure we're not holding any resources
1743 * that we were not when we came in.
1745 * Aborting from this point might lose some
1746 * blocks in the file system, but oh well.
1748 xfs_bmap_cancel(&free_list);
1751 * If the passed in transaction committed
1752 * in xfs_bmap_finish(), then we want to
1753 * add the inode to this one before returning.
1754 * This keeps things simple for the higher
1755 * level code, because it always knows that
1756 * the inode is locked and held in the
1757 * transaction that returns to it whether
1758 * errors occur or not. We don't mark the
1759 * inode dirty so that this transaction can
1760 * be easily aborted if possible.
1762 xfs_trans_ijoin(ntp, ip,
1763 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1764 xfs_trans_ihold(ntp, ip);
1771 * The first xact was committed,
1772 * so add the inode to the new one.
1773 * Mark it dirty so it will be logged
1774 * and moved forward in the log as
1775 * part of every commit.
1777 xfs_trans_ijoin(ntp, ip,
1778 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1779 xfs_trans_ihold(ntp, ip);
1780 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1782 ntp = xfs_trans_dup(ntp);
1783 (void) xfs_trans_commit(*tp, 0);
1785 error = xfs_trans_reserve(ntp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1786 XFS_TRANS_PERM_LOG_RES,
1787 XFS_ITRUNCATE_LOG_COUNT);
1789 * Add the inode being truncated to the next chained
1792 xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1793 xfs_trans_ihold(ntp, ip);
1798 * Only update the size in the case of the data fork, but
1799 * always re-log the inode so that our permanent transaction
1800 * can keep on rolling it forward in the log.
1802 if (fork == XFS_DATA_FORK) {
1803 xfs_isize_check(mp, ip, new_size);
1805 * If we are not changing the file size then do
1806 * not update the on-disk file size - we may be
1807 * called from xfs_inactive_free_eofblocks(). If we
1808 * update the on-disk file size and then the system
1809 * crashes before the contents of the file are
1810 * flushed to disk then the files may be full of
1811 * holes (ie NULL files bug).
1813 if (ip->i_size != new_size) {
1814 ip->i_d.di_size = new_size;
1815 ip->i_size = new_size;
1818 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1819 ASSERT((new_size != 0) ||
1820 (fork == XFS_ATTR_FORK) ||
1821 (ip->i_delayed_blks == 0));
1822 ASSERT((new_size != 0) ||
1823 (fork == XFS_ATTR_FORK) ||
1824 (ip->i_d.di_nextents == 0));
1825 xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
1833 * Do the first part of growing a file: zero any data in the last
1834 * block that is beyond the old EOF. We need to do this before
1835 * the inode is joined to the transaction to modify the i_size.
1836 * That way we can drop the inode lock and call into the buffer
1837 * cache to get the buffer mapping the EOF.
1842 xfs_fsize_t new_size,
1845 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1846 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1847 ASSERT(new_size > ip->i_size);
1850 * Zero any pages that may have been created by
1851 * xfs_write_file() beyond the end of the file
1852 * and any blocks between the old and new file sizes.
1854 return xfs_zero_eof(ip, new_size, ip->i_size);
1860 * This routine is called to extend the size of a file.
1861 * The inode must have both the iolock and the ilock locked
1862 * for update and it must be a part of the current transaction.
1863 * The xfs_igrow_start() function must have been called previously.
1864 * If the change_flag is not zero, the inode change timestamp will
1871 xfs_fsize_t new_size,
1874 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1875 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1876 ASSERT(ip->i_transp == tp);
1877 ASSERT(new_size > ip->i_size);
1880 * Update the file size. Update the inode change timestamp
1881 * if change_flag set.
1883 ip->i_d.di_size = new_size;
1884 ip->i_size = new_size;
1886 xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
1887 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1893 * This is called when the inode's link count goes to 0.
1894 * We place the on-disk inode on a list in the AGI. It
1895 * will be pulled from this list when the inode is freed.
1907 xfs_agnumber_t agno;
1908 xfs_daddr_t agdaddr;
1915 ASSERT(ip->i_d.di_nlink == 0);
1916 ASSERT(ip->i_d.di_mode != 0);
1917 ASSERT(ip->i_transp == tp);
1921 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1922 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1925 * Get the agi buffer first. It ensures lock ordering
1928 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1929 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1934 * Validate the magic number of the agi block.
1936 agi = XFS_BUF_TO_AGI(agibp);
1938 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1939 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
1940 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK,
1941 XFS_RANDOM_IUNLINK))) {
1942 XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi);
1943 xfs_trans_brelse(tp, agibp);
1944 return XFS_ERROR(EFSCORRUPTED);
1947 * Get the index into the agi hash table for the
1948 * list this inode will go on.
1950 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1952 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1953 ASSERT(agi->agi_unlinked[bucket_index]);
1954 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1956 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
1961 * Clear the on-disk di_nlink. This is to prevent xfs_bulkstat
1962 * from picking up this inode when it is reclaimed (its incore state
1963 * initialzed but not flushed to disk yet). The in-core di_nlink is
1964 * already cleared in xfs_droplink() and a corresponding transaction
1965 * logged. The hack here just synchronizes the in-core to on-disk
1966 * di_nlink value in advance before the actual inode sync to disk.
1967 * This is OK because the inode is already unlinked and would never
1968 * change its di_nlink again for this inode generation.
1969 * This is a temporary hack that would require a proper fix
1972 dip->di_core.di_nlink = 0;
1974 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
1976 * There is already another inode in the bucket we need
1977 * to add ourselves to. Add us at the front of the list.
1978 * Here we put the head pointer into our next pointer,
1979 * and then we fall through to point the head at us.
1981 ASSERT(be32_to_cpu(dip->di_next_unlinked) == NULLAGINO);
1982 /* both on-disk, don't endian flip twice */
1983 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1984 offset = ip->i_boffset +
1985 offsetof(xfs_dinode_t, di_next_unlinked);
1986 xfs_trans_inode_buf(tp, ibp);
1987 xfs_trans_log_buf(tp, ibp, offset,
1988 (offset + sizeof(xfs_agino_t) - 1));
1989 xfs_inobp_check(mp, ibp);
1993 * Point the bucket head pointer at the inode being inserted.
1996 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1997 offset = offsetof(xfs_agi_t, agi_unlinked) +
1998 (sizeof(xfs_agino_t) * bucket_index);
1999 xfs_trans_log_buf(tp, agibp, offset,
2000 (offset + sizeof(xfs_agino_t) - 1));
2005 * Pull the on-disk inode from the AGI unlinked list.
2018 xfs_agnumber_t agno;
2019 xfs_daddr_t agdaddr;
2021 xfs_agino_t next_agino;
2022 xfs_buf_t *last_ibp;
2023 xfs_dinode_t *last_dip = NULL;
2025 int offset, last_offset = 0;
2030 * First pull the on-disk inode from the AGI unlinked list.
2034 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2035 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
2038 * Get the agi buffer first. It ensures lock ordering
2041 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
2042 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
2045 "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
2046 error, mp->m_fsname);
2050 * Validate the magic number of the agi block.
2052 agi = XFS_BUF_TO_AGI(agibp);
2054 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
2055 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
2056 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE,
2057 XFS_RANDOM_IUNLINK_REMOVE))) {
2058 XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW,
2060 xfs_trans_brelse(tp, agibp);
2062 "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
2064 return XFS_ERROR(EFSCORRUPTED);
2067 * Get the index into the agi hash table for the
2068 * list this inode will go on.
2070 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2072 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2073 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
2074 ASSERT(agi->agi_unlinked[bucket_index]);
2076 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2078 * We're at the head of the list. Get the inode's
2079 * on-disk buffer to see if there is anyone after us
2080 * on the list. Only modify our next pointer if it
2081 * is not already NULLAGINO. This saves us the overhead
2082 * of dealing with the buffer when there is no need to
2085 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
2088 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2089 error, mp->m_fsname);
2092 next_agino = be32_to_cpu(dip->di_next_unlinked);
2093 ASSERT(next_agino != 0);
2094 if (next_agino != NULLAGINO) {
2095 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2096 offset = ip->i_boffset +
2097 offsetof(xfs_dinode_t, di_next_unlinked);
2098 xfs_trans_inode_buf(tp, ibp);
2099 xfs_trans_log_buf(tp, ibp, offset,
2100 (offset + sizeof(xfs_agino_t) - 1));
2101 xfs_inobp_check(mp, ibp);
2103 xfs_trans_brelse(tp, ibp);
2106 * Point the bucket head pointer at the next inode.
2108 ASSERT(next_agino != 0);
2109 ASSERT(next_agino != agino);
2110 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2111 offset = offsetof(xfs_agi_t, agi_unlinked) +
2112 (sizeof(xfs_agino_t) * bucket_index);
2113 xfs_trans_log_buf(tp, agibp, offset,
2114 (offset + sizeof(xfs_agino_t) - 1));
2117 * We need to search the list for the inode being freed.
2119 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2121 while (next_agino != agino) {
2123 * If the last inode wasn't the one pointing to
2124 * us, then release its buffer since we're not
2125 * going to do anything with it.
2127 if (last_ibp != NULL) {
2128 xfs_trans_brelse(tp, last_ibp);
2130 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2131 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
2132 &last_ibp, &last_offset);
2135 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
2136 error, mp->m_fsname);
2139 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2140 ASSERT(next_agino != NULLAGINO);
2141 ASSERT(next_agino != 0);
2144 * Now last_ibp points to the buffer previous to us on
2145 * the unlinked list. Pull us from the list.
2147 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
2150 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2151 error, mp->m_fsname);
2154 next_agino = be32_to_cpu(dip->di_next_unlinked);
2155 ASSERT(next_agino != 0);
2156 ASSERT(next_agino != agino);
2157 if (next_agino != NULLAGINO) {
2158 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2159 offset = ip->i_boffset +
2160 offsetof(xfs_dinode_t, di_next_unlinked);
2161 xfs_trans_inode_buf(tp, ibp);
2162 xfs_trans_log_buf(tp, ibp, offset,
2163 (offset + sizeof(xfs_agino_t) - 1));
2164 xfs_inobp_check(mp, ibp);
2166 xfs_trans_brelse(tp, ibp);
2169 * Point the previous inode on the list to the next inode.
2171 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2172 ASSERT(next_agino != 0);
2173 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2174 xfs_trans_inode_buf(tp, last_ibp);
2175 xfs_trans_log_buf(tp, last_ibp, offset,
2176 (offset + sizeof(xfs_agino_t) - 1));
2177 xfs_inobp_check(mp, last_ibp);
2182 STATIC_INLINE int xfs_inode_clean(xfs_inode_t *ip)
2184 return (((ip->i_itemp == NULL) ||
2185 !(ip->i_itemp->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
2186 (ip->i_update_core == 0));
2191 xfs_inode_t *free_ip,
2195 xfs_mount_t *mp = free_ip->i_mount;
2196 int blks_per_cluster;
2199 int i, j, found, pre_flushed;
2202 xfs_inode_t *ip, **ip_found;
2203 xfs_inode_log_item_t *iip;
2204 xfs_log_item_t *lip;
2205 xfs_perag_t *pag = xfs_get_perag(mp, inum);
2207 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
2208 blks_per_cluster = 1;
2209 ninodes = mp->m_sb.sb_inopblock;
2210 nbufs = XFS_IALLOC_BLOCKS(mp);
2212 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
2213 mp->m_sb.sb_blocksize;
2214 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
2215 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
2218 ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
2220 for (j = 0; j < nbufs; j++, inum += ninodes) {
2221 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2222 XFS_INO_TO_AGBNO(mp, inum));
2226 * Look for each inode in memory and attempt to lock it,
2227 * we can be racing with flush and tail pushing here.
2228 * any inode we get the locks on, add to an array of
2229 * inode items to process later.
2231 * The get the buffer lock, we could beat a flush
2232 * or tail pushing thread to the lock here, in which
2233 * case they will go looking for the inode buffer
2234 * and fail, we need some other form of interlock
2238 for (i = 0; i < ninodes; i++) {
2239 read_lock(&pag->pag_ici_lock);
2240 ip = radix_tree_lookup(&pag->pag_ici_root,
2241 XFS_INO_TO_AGINO(mp, (inum + i)));
2243 /* Inode not in memory or we found it already,
2246 if (!ip || xfs_iflags_test(ip, XFS_ISTALE)) {
2247 read_unlock(&pag->pag_ici_lock);
2251 if (xfs_inode_clean(ip)) {
2252 read_unlock(&pag->pag_ici_lock);
2256 /* If we can get the locks then add it to the
2257 * list, otherwise by the time we get the bp lock
2258 * below it will already be attached to the
2262 /* This inode will already be locked - by us, lets
2266 if (ip == free_ip) {
2267 if (xfs_iflock_nowait(ip)) {
2268 xfs_iflags_set(ip, XFS_ISTALE);
2269 if (xfs_inode_clean(ip)) {
2272 ip_found[found++] = ip;
2275 read_unlock(&pag->pag_ici_lock);
2279 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2280 if (xfs_iflock_nowait(ip)) {
2281 xfs_iflags_set(ip, XFS_ISTALE);
2283 if (xfs_inode_clean(ip)) {
2285 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2287 ip_found[found++] = ip;
2290 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2293 read_unlock(&pag->pag_ici_lock);
2296 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2297 mp->m_bsize * blks_per_cluster,
2301 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
2303 if (lip->li_type == XFS_LI_INODE) {
2304 iip = (xfs_inode_log_item_t *)lip;
2305 ASSERT(iip->ili_logged == 1);
2306 lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
2307 spin_lock(&mp->m_ail_lock);
2308 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2309 spin_unlock(&mp->m_ail_lock);
2310 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2313 lip = lip->li_bio_list;
2316 for (i = 0; i < found; i++) {
2321 ip->i_update_core = 0;
2323 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2327 iip->ili_last_fields = iip->ili_format.ilf_fields;
2328 iip->ili_format.ilf_fields = 0;
2329 iip->ili_logged = 1;
2330 spin_lock(&mp->m_ail_lock);
2331 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2332 spin_unlock(&mp->m_ail_lock);
2334 xfs_buf_attach_iodone(bp,
2335 (void(*)(xfs_buf_t*,xfs_log_item_t*))
2336 xfs_istale_done, (xfs_log_item_t *)iip);
2337 if (ip != free_ip) {
2338 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2342 if (found || pre_flushed)
2343 xfs_trans_stale_inode_buf(tp, bp);
2344 xfs_trans_binval(tp, bp);
2347 kmem_free(ip_found, ninodes * sizeof(xfs_inode_t *));
2348 xfs_put_perag(mp, pag);
2352 * This is called to return an inode to the inode free list.
2353 * The inode should already be truncated to 0 length and have
2354 * no pages associated with it. This routine also assumes that
2355 * the inode is already a part of the transaction.
2357 * The on-disk copy of the inode will have been added to the list
2358 * of unlinked inodes in the AGI. We need to remove the inode from
2359 * that list atomically with respect to freeing it here.
2365 xfs_bmap_free_t *flist)
2369 xfs_ino_t first_ino;
2371 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2372 ASSERT(ip->i_transp == tp);
2373 ASSERT(ip->i_d.di_nlink == 0);
2374 ASSERT(ip->i_d.di_nextents == 0);
2375 ASSERT(ip->i_d.di_anextents == 0);
2376 ASSERT((ip->i_d.di_size == 0 && ip->i_size == 0) ||
2377 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2378 ASSERT(ip->i_d.di_nblocks == 0);
2381 * Pull the on-disk inode from the AGI unlinked list.
2383 error = xfs_iunlink_remove(tp, ip);
2388 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2392 ip->i_d.di_mode = 0; /* mark incore inode as free */
2393 ip->i_d.di_flags = 0;
2394 ip->i_d.di_dmevmask = 0;
2395 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2396 ip->i_df.if_ext_max =
2397 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2398 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2399 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2401 * Bump the generation count so no one will be confused
2402 * by reincarnations of this inode.
2405 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2408 xfs_ifree_cluster(ip, tp, first_ino);
2415 * Reallocate the space for if_broot based on the number of records
2416 * being added or deleted as indicated in rec_diff. Move the records
2417 * and pointers in if_broot to fit the new size. When shrinking this
2418 * will eliminate holes between the records and pointers created by
2419 * the caller. When growing this will create holes to be filled in
2422 * The caller must not request to add more records than would fit in
2423 * the on-disk inode root. If the if_broot is currently NULL, then
2424 * if we adding records one will be allocated. The caller must also
2425 * not request that the number of records go below zero, although
2426 * it can go to zero.
2428 * ip -- the inode whose if_broot area is changing
2429 * ext_diff -- the change in the number of records, positive or negative,
2430 * requested for the if_broot array.
2440 xfs_bmbt_block_t *new_broot;
2447 * Handle the degenerate case quietly.
2449 if (rec_diff == 0) {
2453 ifp = XFS_IFORK_PTR(ip, whichfork);
2456 * If there wasn't any memory allocated before, just
2457 * allocate it now and get out.
2459 if (ifp->if_broot_bytes == 0) {
2460 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
2461 ifp->if_broot = (xfs_bmbt_block_t*)kmem_alloc(new_size,
2463 ifp->if_broot_bytes = (int)new_size;
2468 * If there is already an existing if_broot, then we need
2469 * to realloc() it and shift the pointers to their new
2470 * location. The records don't change location because
2471 * they are kept butted up against the btree block header.
2473 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2474 new_max = cur_max + rec_diff;
2475 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2476 ifp->if_broot = (xfs_bmbt_block_t *)
2477 kmem_realloc(ifp->if_broot,
2479 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2481 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2482 ifp->if_broot_bytes);
2483 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2485 ifp->if_broot_bytes = (int)new_size;
2486 ASSERT(ifp->if_broot_bytes <=
2487 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2488 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2493 * rec_diff is less than 0. In this case, we are shrinking the
2494 * if_broot buffer. It must already exist. If we go to zero
2495 * records, just get rid of the root and clear the status bit.
2497 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
2498 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2499 new_max = cur_max + rec_diff;
2500 ASSERT(new_max >= 0);
2502 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2506 new_broot = (xfs_bmbt_block_t *)kmem_alloc(new_size, KM_SLEEP);
2508 * First copy over the btree block header.
2510 memcpy(new_broot, ifp->if_broot, sizeof(xfs_bmbt_block_t));
2513 ifp->if_flags &= ~XFS_IFBROOT;
2517 * Only copy the records and pointers if there are any.
2521 * First copy the records.
2523 op = (char *)XFS_BMAP_BROOT_REC_ADDR(ifp->if_broot, 1,
2524 ifp->if_broot_bytes);
2525 np = (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot, 1,
2527 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2530 * Then copy the pointers.
2532 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2533 ifp->if_broot_bytes);
2534 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot, 1,
2536 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2538 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2539 ifp->if_broot = new_broot;
2540 ifp->if_broot_bytes = (int)new_size;
2541 ASSERT(ifp->if_broot_bytes <=
2542 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2548 * This is called when the amount of space needed for if_data
2549 * is increased or decreased. The change in size is indicated by
2550 * the number of bytes that need to be added or deleted in the
2551 * byte_diff parameter.
2553 * If the amount of space needed has decreased below the size of the
2554 * inline buffer, then switch to using the inline buffer. Otherwise,
2555 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2556 * to what is needed.
2558 * ip -- the inode whose if_data area is changing
2559 * byte_diff -- the change in the number of bytes, positive or negative,
2560 * requested for the if_data array.
2572 if (byte_diff == 0) {
2576 ifp = XFS_IFORK_PTR(ip, whichfork);
2577 new_size = (int)ifp->if_bytes + byte_diff;
2578 ASSERT(new_size >= 0);
2580 if (new_size == 0) {
2581 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2582 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2584 ifp->if_u1.if_data = NULL;
2586 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2588 * If the valid extents/data can fit in if_inline_ext/data,
2589 * copy them from the malloc'd vector and free it.
2591 if (ifp->if_u1.if_data == NULL) {
2592 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2593 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2594 ASSERT(ifp->if_real_bytes != 0);
2595 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2597 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2598 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2603 * Stuck with malloc/realloc.
2604 * For inline data, the underlying buffer must be
2605 * a multiple of 4 bytes in size so that it can be
2606 * logged and stay on word boundaries. We enforce
2609 real_size = roundup(new_size, 4);
2610 if (ifp->if_u1.if_data == NULL) {
2611 ASSERT(ifp->if_real_bytes == 0);
2612 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2613 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2615 * Only do the realloc if the underlying size
2616 * is really changing.
2618 if (ifp->if_real_bytes != real_size) {
2619 ifp->if_u1.if_data =
2620 kmem_realloc(ifp->if_u1.if_data,
2626 ASSERT(ifp->if_real_bytes == 0);
2627 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2628 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2632 ifp->if_real_bytes = real_size;
2633 ifp->if_bytes = new_size;
2634 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2641 * Map inode to disk block and offset.
2643 * mp -- the mount point structure for the current file system
2644 * tp -- the current transaction
2645 * ino -- the inode number of the inode to be located
2646 * imap -- this structure is filled in with the information necessary
2647 * to retrieve the given inode from disk
2648 * flags -- flags to pass to xfs_dilocate indicating whether or not
2649 * lookups in the inode btree were OK or not
2659 xfs_fsblock_t fsbno;
2664 fsbno = imap->im_blkno ?
2665 XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK;
2666 error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags);
2670 imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno);
2671 imap->im_len = XFS_FSB_TO_BB(mp, len);
2672 imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno);
2673 imap->im_ioffset = (ushort)off;
2674 imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog);
2685 ifp = XFS_IFORK_PTR(ip, whichfork);
2686 if (ifp->if_broot != NULL) {
2687 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2688 ifp->if_broot = NULL;
2692 * If the format is local, then we can't have an extents
2693 * array so just look for an inline data array. If we're
2694 * not local then we may or may not have an extents list,
2695 * so check and free it up if we do.
2697 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2698 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2699 (ifp->if_u1.if_data != NULL)) {
2700 ASSERT(ifp->if_real_bytes != 0);
2701 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2702 ifp->if_u1.if_data = NULL;
2703 ifp->if_real_bytes = 0;
2705 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2706 ((ifp->if_flags & XFS_IFEXTIREC) ||
2707 ((ifp->if_u1.if_extents != NULL) &&
2708 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
2709 ASSERT(ifp->if_real_bytes != 0);
2710 xfs_iext_destroy(ifp);
2712 ASSERT(ifp->if_u1.if_extents == NULL ||
2713 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2714 ASSERT(ifp->if_real_bytes == 0);
2715 if (whichfork == XFS_ATTR_FORK) {
2716 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2722 * This is called free all the memory associated with an inode.
2723 * It must free the inode itself and any buffers allocated for
2724 * if_extents/if_data and if_broot. It must also free the lock
2725 * associated with the inode.
2731 switch (ip->i_d.di_mode & S_IFMT) {
2735 xfs_idestroy_fork(ip, XFS_DATA_FORK);
2739 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
2740 mrfree(&ip->i_lock);
2741 mrfree(&ip->i_iolock);
2742 freesema(&ip->i_flock);
2744 #ifdef XFS_INODE_TRACE
2745 ktrace_free(ip->i_trace);
2747 #ifdef XFS_BMAP_TRACE
2748 ktrace_free(ip->i_xtrace);
2750 #ifdef XFS_BMBT_TRACE
2751 ktrace_free(ip->i_btrace);
2754 ktrace_free(ip->i_rwtrace);
2756 #ifdef XFS_ILOCK_TRACE
2757 ktrace_free(ip->i_lock_trace);
2759 #ifdef XFS_DIR2_TRACE
2760 ktrace_free(ip->i_dir_trace);
2764 * Only if we are shutting down the fs will we see an
2765 * inode still in the AIL. If it is there, we should remove
2766 * it to prevent a use-after-free from occurring.
2768 xfs_mount_t *mp = ip->i_mount;
2769 xfs_log_item_t *lip = &ip->i_itemp->ili_item;
2771 ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) ||
2772 XFS_FORCED_SHUTDOWN(ip->i_mount));
2773 if (lip->li_flags & XFS_LI_IN_AIL) {
2774 spin_lock(&mp->m_ail_lock);
2775 if (lip->li_flags & XFS_LI_IN_AIL)
2776 xfs_trans_delete_ail(mp, lip);
2778 spin_unlock(&mp->m_ail_lock);
2780 xfs_inode_item_destroy(ip);
2782 kmem_zone_free(xfs_inode_zone, ip);
2787 * Increment the pin count of the given buffer.
2788 * This value is protected by ipinlock spinlock in the mount structure.
2794 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2796 atomic_inc(&ip->i_pincount);
2800 * Decrement the pin count of the given inode, and wake up
2801 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2802 * inode must have been previously pinned with a call to xfs_ipin().
2808 ASSERT(atomic_read(&ip->i_pincount) > 0);
2810 if (atomic_dec_and_lock(&ip->i_pincount, &ip->i_flags_lock)) {
2813 * If the inode is currently being reclaimed, the link between
2814 * the bhv_vnode and the xfs_inode will be broken after the
2815 * XFS_IRECLAIM* flag is set. Hence, if these flags are not
2816 * set, then we can move forward and mark the linux inode dirty
2817 * knowing that it is still valid as it won't freed until after
2818 * the bhv_vnode<->xfs_inode link is broken in xfs_reclaim. The
2819 * i_flags_lock is used to synchronise the setting of the
2820 * XFS_IRECLAIM* flags and the breaking of the link, and so we
2821 * can execute atomically w.r.t to reclaim by holding this lock
2824 * However, we still need to issue the unpin wakeup call as the
2825 * inode reclaim may be blocked waiting for the inode to become
2829 if (!__xfs_iflags_test(ip, XFS_IRECLAIM|XFS_IRECLAIMABLE)) {
2830 bhv_vnode_t *vp = XFS_ITOV_NULL(ip);
2831 struct inode *inode = NULL;
2834 inode = vn_to_inode(vp);
2835 BUG_ON(inode->i_state & I_CLEAR);
2837 /* make sync come back and flush this inode */
2838 if (!(inode->i_state & (I_NEW|I_FREEING)))
2839 mark_inode_dirty_sync(inode);
2841 spin_unlock(&ip->i_flags_lock);
2842 wake_up(&ip->i_ipin_wait);
2847 * This is called to wait for the given inode to be unpinned.
2848 * It will sleep until this happens. The caller must have the
2849 * inode locked in at least shared mode so that the buffer cannot
2850 * be subsequently pinned once someone is waiting for it to be
2857 xfs_inode_log_item_t *iip;
2860 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE | MR_ACCESS));
2862 if (atomic_read(&ip->i_pincount) == 0) {
2867 if (iip && iip->ili_last_lsn) {
2868 lsn = iip->ili_last_lsn;
2874 * Give the log a push so we don't wait here too long.
2876 xfs_log_force(ip->i_mount, lsn, XFS_LOG_FORCE);
2878 wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
2883 * xfs_iextents_copy()
2885 * This is called to copy the REAL extents (as opposed to the delayed
2886 * allocation extents) from the inode into the given buffer. It
2887 * returns the number of bytes copied into the buffer.
2889 * If there are no delayed allocation extents, then we can just
2890 * memcpy() the extents into the buffer. Otherwise, we need to
2891 * examine each extent in turn and skip those which are delayed.
2903 xfs_fsblock_t start_block;
2905 ifp = XFS_IFORK_PTR(ip, whichfork);
2906 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
2907 ASSERT(ifp->if_bytes > 0);
2909 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2910 XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork);
2914 * There are some delayed allocation extents in the
2915 * inode, so copy the extents one at a time and skip
2916 * the delayed ones. There must be at least one
2917 * non-delayed extent.
2920 for (i = 0; i < nrecs; i++) {
2921 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
2922 start_block = xfs_bmbt_get_startblock(ep);
2923 if (ISNULLSTARTBLOCK(start_block)) {
2925 * It's a delayed allocation extent, so skip it.
2930 /* Translate to on disk format */
2931 put_unaligned(cpu_to_be64(ep->l0), &dp->l0);
2932 put_unaligned(cpu_to_be64(ep->l1), &dp->l1);
2936 ASSERT(copied != 0);
2937 xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip));
2939 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2943 * Each of the following cases stores data into the same region
2944 * of the on-disk inode, so only one of them can be valid at
2945 * any given time. While it is possible to have conflicting formats
2946 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2947 * in EXTENTS format, this can only happen when the fork has
2948 * changed formats after being modified but before being flushed.
2949 * In these cases, the format always takes precedence, because the
2950 * format indicates the current state of the fork.
2957 xfs_inode_log_item_t *iip,
2964 #ifdef XFS_TRANS_DEBUG
2967 static const short brootflag[2] =
2968 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2969 static const short dataflag[2] =
2970 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2971 static const short extflag[2] =
2972 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2976 ifp = XFS_IFORK_PTR(ip, whichfork);
2978 * This can happen if we gave up in iformat in an error path,
2979 * for the attribute fork.
2982 ASSERT(whichfork == XFS_ATTR_FORK);
2985 cp = XFS_DFORK_PTR(dip, whichfork);
2987 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2988 case XFS_DINODE_FMT_LOCAL:
2989 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2990 (ifp->if_bytes > 0)) {
2991 ASSERT(ifp->if_u1.if_data != NULL);
2992 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2993 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2997 case XFS_DINODE_FMT_EXTENTS:
2998 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2999 !(iip->ili_format.ilf_fields & extflag[whichfork]));
3000 ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
3001 (ifp->if_bytes == 0));
3002 ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
3003 (ifp->if_bytes > 0));
3004 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
3005 (ifp->if_bytes > 0)) {
3006 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
3007 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
3012 case XFS_DINODE_FMT_BTREE:
3013 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
3014 (ifp->if_broot_bytes > 0)) {
3015 ASSERT(ifp->if_broot != NULL);
3016 ASSERT(ifp->if_broot_bytes <=
3017 (XFS_IFORK_SIZE(ip, whichfork) +
3018 XFS_BROOT_SIZE_ADJ));
3019 xfs_bmbt_to_bmdr(ifp->if_broot, ifp->if_broot_bytes,
3020 (xfs_bmdr_block_t *)cp,
3021 XFS_DFORK_SIZE(dip, mp, whichfork));
3025 case XFS_DINODE_FMT_DEV:
3026 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
3027 ASSERT(whichfork == XFS_DATA_FORK);
3028 dip->di_u.di_dev = cpu_to_be32(ip->i_df.if_u2.if_rdev);
3032 case XFS_DINODE_FMT_UUID:
3033 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
3034 ASSERT(whichfork == XFS_DATA_FORK);
3035 memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid,
3049 * xfs_iflush() will write a modified inode's changes out to the
3050 * inode's on disk home. The caller must have the inode lock held
3051 * in at least shared mode and the inode flush semaphore must be
3052 * held as well. The inode lock will still be held upon return from
3053 * the call and the caller is free to unlock it.
3054 * The inode flush lock will be unlocked when the inode reaches the disk.
3055 * The flags indicate how the inode's buffer should be written out.
3062 xfs_inode_log_item_t *iip;
3069 int clcount; /* count of inodes clustered */
3071 struct hlist_node *entry;
3072 enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
3074 XFS_STATS_INC(xs_iflush_count);
3076 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
3077 ASSERT(issemalocked(&(ip->i_flock)));
3078 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3079 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3085 * If the inode isn't dirty, then just release the inode
3086 * flush lock and do nothing.
3088 if ((ip->i_update_core == 0) &&
3089 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3090 ASSERT((iip != NULL) ?
3091 !(iip->ili_item.li_flags & XFS_LI_IN_AIL) : 1);
3097 * We can't flush the inode until it is unpinned, so
3098 * wait for it. We know noone new can pin it, because
3099 * we are holding the inode lock shared and you need
3100 * to hold it exclusively to pin the inode.
3102 xfs_iunpin_wait(ip);
3105 * This may have been unpinned because the filesystem is shutting
3106 * down forcibly. If that's the case we must not write this inode
3107 * to disk, because the log record didn't make it to disk!
3109 if (XFS_FORCED_SHUTDOWN(mp)) {
3110 ip->i_update_core = 0;
3112 iip->ili_format.ilf_fields = 0;
3114 return XFS_ERROR(EIO);
3118 * Get the buffer containing the on-disk inode.
3120 error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0, 0);
3127 * Decide how buffer will be flushed out. This is done before
3128 * the call to xfs_iflush_int because this field is zeroed by it.
3130 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3132 * Flush out the inode buffer according to the directions
3133 * of the caller. In the cases where the caller has given
3134 * us a choice choose the non-delwri case. This is because
3135 * the inode is in the AIL and we need to get it out soon.
3138 case XFS_IFLUSH_SYNC:
3139 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3142 case XFS_IFLUSH_ASYNC:
3143 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3146 case XFS_IFLUSH_DELWRI:
3156 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3157 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3158 case XFS_IFLUSH_DELWRI:
3161 case XFS_IFLUSH_ASYNC:
3164 case XFS_IFLUSH_SYNC:
3175 * First flush out the inode that xfs_iflush was called with.
3177 error = xfs_iflush_int(ip, bp);
3184 * see if other inodes can be gathered into this write
3186 spin_lock(&ip->i_cluster->icl_lock);
3187 ip->i_cluster->icl_buf = bp;
3190 hlist_for_each_entry(iq, entry, &ip->i_cluster->icl_inodes, i_cnode) {
3195 * Do an un-protected check to see if the inode is dirty and
3196 * is a candidate for flushing. These checks will be repeated
3197 * later after the appropriate locks are acquired.
3200 if ((iq->i_update_core == 0) &&
3202 !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
3203 xfs_ipincount(iq) == 0) {
3208 * Try to get locks. If any are unavailable,
3209 * then this inode cannot be flushed and is skipped.
3212 /* get inode locks (just i_lock) */
3213 if (xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) {
3214 /* get inode flush lock */
3215 if (xfs_iflock_nowait(iq)) {
3216 /* check if pinned */
3217 if (xfs_ipincount(iq) == 0) {
3218 /* arriving here means that
3219 * this inode can be flushed.
3220 * first re-check that it's
3224 if ((iq->i_update_core != 0)||
3226 (iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3228 error = xfs_iflush_int(iq, bp);
3232 goto cluster_corrupt_out;
3241 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3244 spin_unlock(&ip->i_cluster->icl_lock);
3247 XFS_STATS_INC(xs_icluster_flushcnt);
3248 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
3252 * If the buffer is pinned then push on the log so we won't
3253 * get stuck waiting in the write for too long.
3255 if (XFS_BUF_ISPINNED(bp)){
3256 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
3259 if (flags & INT_DELWRI) {
3260 xfs_bdwrite(mp, bp);
3261 } else if (flags & INT_ASYNC) {
3262 xfs_bawrite(mp, bp);
3264 error = xfs_bwrite(mp, bp);
3270 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3271 xfs_iflush_abort(ip);
3273 * Unlocks the flush lock
3275 return XFS_ERROR(EFSCORRUPTED);
3277 cluster_corrupt_out:
3278 /* Corruption detected in the clustering loop. Invalidate the
3279 * inode buffer and shut down the filesystem.
3281 spin_unlock(&ip->i_cluster->icl_lock);
3284 * Clean up the buffer. If it was B_DELWRI, just release it --
3285 * brelse can handle it with no problems. If not, shut down the
3286 * filesystem before releasing the buffer.
3288 if ((bufwasdelwri= XFS_BUF_ISDELAYWRITE(bp))) {
3292 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3296 * Just like incore_relse: if we have b_iodone functions,
3297 * mark the buffer as an error and call them. Otherwise
3298 * mark it as stale and brelse.
3300 if (XFS_BUF_IODONE_FUNC(bp)) {
3301 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
3305 XFS_BUF_ERROR(bp,EIO);
3313 xfs_iflush_abort(iq);
3315 * Unlocks the flush lock
3317 return XFS_ERROR(EFSCORRUPTED);
3326 xfs_inode_log_item_t *iip;
3329 #ifdef XFS_TRANS_DEBUG
3333 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
3334 ASSERT(issemalocked(&(ip->i_flock)));
3335 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3336 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3343 * If the inode isn't dirty, then just release the inode
3344 * flush lock and do nothing.
3346 if ((ip->i_update_core == 0) &&
3347 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3352 /* set *dip = inode's place in the buffer */
3353 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset);
3356 * Clear i_update_core before copying out the data.
3357 * This is for coordination with our timestamp updates
3358 * that don't hold the inode lock. They will always
3359 * update the timestamps BEFORE setting i_update_core,
3360 * so if we clear i_update_core after they set it we
3361 * are guaranteed to see their updates to the timestamps.
3362 * I believe that this depends on strongly ordered memory
3363 * semantics, but we have that. We use the SYNCHRONIZE
3364 * macro to make sure that the compiler does not reorder
3365 * the i_update_core access below the data copy below.
3367 ip->i_update_core = 0;
3371 * Make sure to get the latest atime from the Linux inode.
3373 xfs_synchronize_atime(ip);
3375 if (XFS_TEST_ERROR(be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC,
3376 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3377 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3378 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3379 ip->i_ino, be16_to_cpu(dip->di_core.di_magic), dip);
3382 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3383 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3384 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3385 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3386 ip->i_ino, ip, ip->i_d.di_magic);
3389 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3391 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3392 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3393 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3394 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3395 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3399 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
3401 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3402 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3403 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3404 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3405 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3406 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3411 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3412 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3413 XFS_RANDOM_IFLUSH_5)) {
3414 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3415 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3417 ip->i_d.di_nextents + ip->i_d.di_anextents,
3422 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3423 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3424 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3425 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3426 ip->i_ino, ip->i_d.di_forkoff, ip);
3430 * bump the flush iteration count, used to detect flushes which
3431 * postdate a log record during recovery.
3434 ip->i_d.di_flushiter++;
3437 * Copy the dirty parts of the inode into the on-disk
3438 * inode. We always copy out the core of the inode,
3439 * because if the inode is dirty at all the core must
3442 xfs_dinode_to_disk(&dip->di_core, &ip->i_d);
3444 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3445 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3446 ip->i_d.di_flushiter = 0;
3449 * If this is really an old format inode and the superblock version
3450 * has not been updated to support only new format inodes, then
3451 * convert back to the old inode format. If the superblock version
3452 * has been updated, then make the conversion permanent.
3454 ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 ||
3455 XFS_SB_VERSION_HASNLINK(&mp->m_sb));
3456 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
3457 if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) {
3461 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3462 dip->di_core.di_onlink = cpu_to_be16(ip->i_d.di_nlink);
3465 * The superblock version has already been bumped,
3466 * so just make the conversion to the new inode
3469 ip->i_d.di_version = XFS_DINODE_VERSION_2;
3470 dip->di_core.di_version = XFS_DINODE_VERSION_2;
3471 ip->i_d.di_onlink = 0;
3472 dip->di_core.di_onlink = 0;
3473 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3474 memset(&(dip->di_core.di_pad[0]), 0,
3475 sizeof(dip->di_core.di_pad));
3476 ASSERT(ip->i_d.di_projid == 0);
3480 if (xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp) == EFSCORRUPTED) {
3484 if (XFS_IFORK_Q(ip)) {
3486 * The only error from xfs_iflush_fork is on the data fork.
3488 (void) xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3490 xfs_inobp_check(mp, bp);
3493 * We've recorded everything logged in the inode, so we'd
3494 * like to clear the ilf_fields bits so we don't log and
3495 * flush things unnecessarily. However, we can't stop
3496 * logging all this information until the data we've copied
3497 * into the disk buffer is written to disk. If we did we might
3498 * overwrite the copy of the inode in the log with all the
3499 * data after re-logging only part of it, and in the face of
3500 * a crash we wouldn't have all the data we need to recover.
3502 * What we do is move the bits to the ili_last_fields field.
3503 * When logging the inode, these bits are moved back to the
3504 * ilf_fields field. In the xfs_iflush_done() routine we
3505 * clear ili_last_fields, since we know that the information
3506 * those bits represent is permanently on disk. As long as
3507 * the flush completes before the inode is logged again, then
3508 * both ilf_fields and ili_last_fields will be cleared.
3510 * We can play with the ilf_fields bits here, because the inode
3511 * lock must be held exclusively in order to set bits there
3512 * and the flush lock protects the ili_last_fields bits.
3513 * Set ili_logged so the flush done
3514 * routine can tell whether or not to look in the AIL.
3515 * Also, store the current LSN of the inode so that we can tell
3516 * whether the item has moved in the AIL from xfs_iflush_done().
3517 * In order to read the lsn we need the AIL lock, because
3518 * it is a 64 bit value that cannot be read atomically.
3520 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3521 iip->ili_last_fields = iip->ili_format.ilf_fields;
3522 iip->ili_format.ilf_fields = 0;
3523 iip->ili_logged = 1;
3525 ASSERT(sizeof(xfs_lsn_t) == 8); /* don't lock if it shrinks */
3526 spin_lock(&mp->m_ail_lock);
3527 iip->ili_flush_lsn = iip->ili_item.li_lsn;
3528 spin_unlock(&mp->m_ail_lock);
3531 * Attach the function xfs_iflush_done to the inode's
3532 * buffer. This will remove the inode from the AIL
3533 * and unlock the inode's flush lock when the inode is
3534 * completely written to disk.
3536 xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
3537 xfs_iflush_done, (xfs_log_item_t *)iip);
3539 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3540 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3543 * We're flushing an inode which is not in the AIL and has
3544 * not been logged but has i_update_core set. For this
3545 * case we can use a B_DELWRI flush and immediately drop
3546 * the inode flush lock because we can avoid the whole
3547 * AIL state thing. It's OK to drop the flush lock now,
3548 * because we've already locked the buffer and to do anything
3549 * you really need both.
3552 ASSERT(iip->ili_logged == 0);
3553 ASSERT(iip->ili_last_fields == 0);
3554 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3562 return XFS_ERROR(EFSCORRUPTED);
3567 * Flush all inactive inodes in mp.
3577 XFS_MOUNT_ILOCK(mp);
3583 /* Make sure we skip markers inserted by sync */
3584 if (ip->i_mount == NULL) {
3589 vp = XFS_ITOV_NULL(ip);
3591 XFS_MOUNT_IUNLOCK(mp);
3592 xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC);
3596 ASSERT(vn_count(vp) == 0);
3599 } while (ip != mp->m_inodes);
3601 XFS_MOUNT_IUNLOCK(mp);
3605 * xfs_iaccess: check accessibility of inode for mode.
3614 mode_t orgmode = mode;
3615 struct inode *inode = vn_to_inode(XFS_ITOV(ip));
3617 if (mode & S_IWUSR) {
3618 umode_t imode = inode->i_mode;
3620 if (IS_RDONLY(inode) &&
3621 (S_ISREG(imode) || S_ISDIR(imode) || S_ISLNK(imode)))
3622 return XFS_ERROR(EROFS);
3624 if (IS_IMMUTABLE(inode))
3625 return XFS_ERROR(EACCES);
3629 * If there's an Access Control List it's used instead of
3632 if ((error = _ACL_XFS_IACCESS(ip, mode, cr)) != -1)
3633 return error ? XFS_ERROR(error) : 0;
3635 if (current_fsuid(cr) != ip->i_d.di_uid) {
3637 if (!in_group_p((gid_t)ip->i_d.di_gid))
3642 * If the DACs are ok we don't need any capability check.
3644 if ((ip->i_d.di_mode & mode) == mode)
3647 * Read/write DACs are always overridable.
3648 * Executable DACs are overridable if at least one exec bit is set.
3650 if (!(orgmode & S_IXUSR) ||
3651 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
3652 if (capable_cred(cr, CAP_DAC_OVERRIDE))
3655 if ((orgmode == S_IRUSR) ||
3656 (S_ISDIR(inode->i_mode) && (!(orgmode & S_IWUSR)))) {
3657 if (capable_cred(cr, CAP_DAC_READ_SEARCH))
3660 cmn_err(CE_NOTE, "Ick: mode=%o, orgmode=%o", mode, orgmode);
3662 return XFS_ERROR(EACCES);
3664 return XFS_ERROR(EACCES);
3668 * xfs_iroundup: round up argument to next power of two
3677 if ((v & (v - 1)) == 0)
3679 ASSERT((v & 0x80000000) == 0);
3680 if ((v & (v + 1)) == 0)
3682 for (i = 0, m = 1; i < 31; i++, m <<= 1) {
3686 if ((v & (v + 1)) == 0)
3693 #ifdef XFS_ILOCK_TRACE
3694 ktrace_t *xfs_ilock_trace_buf;
3697 xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
3699 ktrace_enter(ip->i_lock_trace,
3701 (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
3702 (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
3703 (void *)ra, /* caller of ilock */
3704 (void *)(unsigned long)current_cpu(),
3705 (void *)(unsigned long)current_pid(),
3706 NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL);
3711 * Return a pointer to the extent record at file index idx.
3713 xfs_bmbt_rec_host_t *
3715 xfs_ifork_t *ifp, /* inode fork pointer */
3716 xfs_extnum_t idx) /* index of target extent */
3719 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
3720 return ifp->if_u1.if_ext_irec->er_extbuf;
3721 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3722 xfs_ext_irec_t *erp; /* irec pointer */
3723 int erp_idx = 0; /* irec index */
3724 xfs_extnum_t page_idx = idx; /* ext index in target list */
3726 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3727 return &erp->er_extbuf[page_idx];
3728 } else if (ifp->if_bytes) {
3729 return &ifp->if_u1.if_extents[idx];
3736 * Insert new item(s) into the extent records for incore inode
3737 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3741 xfs_ifork_t *ifp, /* inode fork pointer */
3742 xfs_extnum_t idx, /* starting index of new items */
3743 xfs_extnum_t count, /* number of inserted items */
3744 xfs_bmbt_irec_t *new) /* items to insert */
3746 xfs_extnum_t i; /* extent record index */
3748 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3749 xfs_iext_add(ifp, idx, count);
3750 for (i = idx; i < idx + count; i++, new++)
3751 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new);
3755 * This is called when the amount of space required for incore file
3756 * extents needs to be increased. The ext_diff parameter stores the
3757 * number of new extents being added and the idx parameter contains
3758 * the extent index where the new extents will be added. If the new
3759 * extents are being appended, then we just need to (re)allocate and
3760 * initialize the space. Otherwise, if the new extents are being
3761 * inserted into the middle of the existing entries, a bit more work
3762 * is required to make room for the new extents to be inserted. The
3763 * caller is responsible for filling in the new extent entries upon
3768 xfs_ifork_t *ifp, /* inode fork pointer */
3769 xfs_extnum_t idx, /* index to begin adding exts */
3770 int ext_diff) /* number of extents to add */
3772 int byte_diff; /* new bytes being added */
3773 int new_size; /* size of extents after adding */
3774 xfs_extnum_t nextents; /* number of extents in file */
3776 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3777 ASSERT((idx >= 0) && (idx <= nextents));
3778 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
3779 new_size = ifp->if_bytes + byte_diff;
3781 * If the new number of extents (nextents + ext_diff)
3782 * fits inside the inode, then continue to use the inline
3785 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
3786 if (idx < nextents) {
3787 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
3788 &ifp->if_u2.if_inline_ext[idx],
3789 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3790 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
3792 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3793 ifp->if_real_bytes = 0;
3794 ifp->if_lastex = nextents + ext_diff;
3797 * Otherwise use a linear (direct) extent list.
3798 * If the extents are currently inside the inode,
3799 * xfs_iext_realloc_direct will switch us from
3800 * inline to direct extent allocation mode.
3802 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
3803 xfs_iext_realloc_direct(ifp, new_size);
3804 if (idx < nextents) {
3805 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
3806 &ifp->if_u1.if_extents[idx],
3807 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3808 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
3811 /* Indirection array */
3813 xfs_ext_irec_t *erp;
3817 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
3818 if (ifp->if_flags & XFS_IFEXTIREC) {
3819 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3821 xfs_iext_irec_init(ifp);
3822 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3823 erp = ifp->if_u1.if_ext_irec;
3825 /* Extents fit in target extent page */
3826 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3827 if (page_idx < erp->er_extcount) {
3828 memmove(&erp->er_extbuf[page_idx + ext_diff],
3829 &erp->er_extbuf[page_idx],
3830 (erp->er_extcount - page_idx) *
3831 sizeof(xfs_bmbt_rec_t));
3832 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3834 erp->er_extcount += ext_diff;
3835 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3837 /* Insert a new extent page */
3839 xfs_iext_add_indirect_multi(ifp,
3840 erp_idx, page_idx, ext_diff);
3843 * If extent(s) are being appended to the last page in
3844 * the indirection array and the new extent(s) don't fit
3845 * in the page, then erp is NULL and erp_idx is set to
3846 * the next index needed in the indirection array.
3849 int count = ext_diff;
3852 erp = xfs_iext_irec_new(ifp, erp_idx);
3853 erp->er_extcount = count;
3854 count -= MIN(count, (int)XFS_LINEAR_EXTS);
3861 ifp->if_bytes = new_size;
3865 * This is called when incore extents are being added to the indirection
3866 * array and the new extents do not fit in the target extent list. The
3867 * erp_idx parameter contains the irec index for the target extent list
3868 * in the indirection array, and the idx parameter contains the extent
3869 * index within the list. The number of extents being added is stored
3870 * in the count parameter.
3872 * |-------| |-------|
3873 * | | | | idx - number of extents before idx
3875 * | | | | count - number of extents being inserted at idx
3876 * |-------| |-------|
3877 * | count | | nex2 | nex2 - number of extents after idx + count
3878 * |-------| |-------|
3881 xfs_iext_add_indirect_multi(
3882 xfs_ifork_t *ifp, /* inode fork pointer */
3883 int erp_idx, /* target extent irec index */
3884 xfs_extnum_t idx, /* index within target list */
3885 int count) /* new extents being added */
3887 int byte_diff; /* new bytes being added */
3888 xfs_ext_irec_t *erp; /* pointer to irec entry */
3889 xfs_extnum_t ext_diff; /* number of extents to add */
3890 xfs_extnum_t ext_cnt; /* new extents still needed */
3891 xfs_extnum_t nex2; /* extents after idx + count */
3892 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3893 int nlists; /* number of irec's (lists) */
3895 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3896 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3897 nex2 = erp->er_extcount - idx;
3898 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3901 * Save second part of target extent list
3902 * (all extents past */
3904 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3905 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_SLEEP);
3906 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3907 erp->er_extcount -= nex2;
3908 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3909 memset(&erp->er_extbuf[idx], 0, byte_diff);
3913 * Add the new extents to the end of the target
3914 * list, then allocate new irec record(s) and
3915 * extent buffer(s) as needed to store the rest
3916 * of the new extents.
3919 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3921 erp->er_extcount += ext_diff;
3922 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3923 ext_cnt -= ext_diff;
3927 erp = xfs_iext_irec_new(ifp, erp_idx);
3928 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3929 erp->er_extcount = ext_diff;
3930 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3931 ext_cnt -= ext_diff;
3934 /* Add nex2 extents back to indirection array */
3936 xfs_extnum_t ext_avail;
3939 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3940 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3943 * If nex2 extents fit in the current page, append
3944 * nex2_ep after the new extents.
3946 if (nex2 <= ext_avail) {
3947 i = erp->er_extcount;
3950 * Otherwise, check if space is available in the
3953 else if ((erp_idx < nlists - 1) &&
3954 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3955 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3958 /* Create a hole for nex2 extents */
3959 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3960 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3963 * Final choice, create a new extent page for
3968 erp = xfs_iext_irec_new(ifp, erp_idx);
3970 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
3971 kmem_free(nex2_ep, byte_diff);
3972 erp->er_extcount += nex2;
3973 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3978 * This is called when the amount of space required for incore file
3979 * extents needs to be decreased. The ext_diff parameter stores the
3980 * number of extents to be removed and the idx parameter contains
3981 * the extent index where the extents will be removed from.
3983 * If the amount of space needed has decreased below the linear
3984 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3985 * extent array. Otherwise, use kmem_realloc() to adjust the
3986 * size to what is needed.
3990 xfs_ifork_t *ifp, /* inode fork pointer */
3991 xfs_extnum_t idx, /* index to begin removing exts */
3992 int ext_diff) /* number of extents to remove */
3994 xfs_extnum_t nextents; /* number of extents in file */
3995 int new_size; /* size of extents after removal */
3997 ASSERT(ext_diff > 0);
3998 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3999 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
4001 if (new_size == 0) {
4002 xfs_iext_destroy(ifp);
4003 } else if (ifp->if_flags & XFS_IFEXTIREC) {
4004 xfs_iext_remove_indirect(ifp, idx, ext_diff);
4005 } else if (ifp->if_real_bytes) {
4006 xfs_iext_remove_direct(ifp, idx, ext_diff);
4008 xfs_iext_remove_inline(ifp, idx, ext_diff);
4010 ifp->if_bytes = new_size;
4014 * This removes ext_diff extents from the inline buffer, beginning
4015 * at extent index idx.
4018 xfs_iext_remove_inline(
4019 xfs_ifork_t *ifp, /* inode fork pointer */
4020 xfs_extnum_t idx, /* index to begin removing exts */
4021 int ext_diff) /* number of extents to remove */
4023 int nextents; /* number of extents in file */
4025 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4026 ASSERT(idx < XFS_INLINE_EXTS);
4027 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4028 ASSERT(((nextents - ext_diff) > 0) &&
4029 (nextents - ext_diff) < XFS_INLINE_EXTS);
4031 if (idx + ext_diff < nextents) {
4032 memmove(&ifp->if_u2.if_inline_ext[idx],
4033 &ifp->if_u2.if_inline_ext[idx + ext_diff],
4034 (nextents - (idx + ext_diff)) *
4035 sizeof(xfs_bmbt_rec_t));
4036 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
4037 0, ext_diff * sizeof(xfs_bmbt_rec_t));
4039 memset(&ifp->if_u2.if_inline_ext[idx], 0,
4040 ext_diff * sizeof(xfs_bmbt_rec_t));
4045 * This removes ext_diff extents from a linear (direct) extent list,
4046 * beginning at extent index idx. If the extents are being removed
4047 * from the end of the list (ie. truncate) then we just need to re-
4048 * allocate the list to remove the extra space. Otherwise, if the
4049 * extents are being removed from the middle of the existing extent
4050 * entries, then we first need to move the extent records beginning
4051 * at idx + ext_diff up in the list to overwrite the records being
4052 * removed, then remove the extra space via kmem_realloc.
4055 xfs_iext_remove_direct(
4056 xfs_ifork_t *ifp, /* inode fork pointer */
4057 xfs_extnum_t idx, /* index to begin removing exts */
4058 int ext_diff) /* number of extents to remove */
4060 xfs_extnum_t nextents; /* number of extents in file */
4061 int new_size; /* size of extents after removal */
4063 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4064 new_size = ifp->if_bytes -
4065 (ext_diff * sizeof(xfs_bmbt_rec_t));
4066 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4068 if (new_size == 0) {
4069 xfs_iext_destroy(ifp);
4072 /* Move extents up in the list (if needed) */
4073 if (idx + ext_diff < nextents) {
4074 memmove(&ifp->if_u1.if_extents[idx],
4075 &ifp->if_u1.if_extents[idx + ext_diff],
4076 (nextents - (idx + ext_diff)) *
4077 sizeof(xfs_bmbt_rec_t));
4079 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
4080 0, ext_diff * sizeof(xfs_bmbt_rec_t));
4082 * Reallocate the direct extent list. If the extents
4083 * will fit inside the inode then xfs_iext_realloc_direct
4084 * will switch from direct to inline extent allocation
4087 xfs_iext_realloc_direct(ifp, new_size);
4088 ifp->if_bytes = new_size;
4092 * This is called when incore extents are being removed from the
4093 * indirection array and the extents being removed span multiple extent
4094 * buffers. The idx parameter contains the file extent index where we
4095 * want to begin removing extents, and the count parameter contains
4096 * how many extents need to be removed.
4098 * |-------| |-------|
4099 * | nex1 | | | nex1 - number of extents before idx
4100 * |-------| | count |
4101 * | | | | count - number of extents being removed at idx
4102 * | count | |-------|
4103 * | | | nex2 | nex2 - number of extents after idx + count
4104 * |-------| |-------|
4107 xfs_iext_remove_indirect(
4108 xfs_ifork_t *ifp, /* inode fork pointer */
4109 xfs_extnum_t idx, /* index to begin removing extents */
4110 int count) /* number of extents to remove */
4112 xfs_ext_irec_t *erp; /* indirection array pointer */
4113 int erp_idx = 0; /* indirection array index */
4114 xfs_extnum_t ext_cnt; /* extents left to remove */
4115 xfs_extnum_t ext_diff; /* extents to remove in current list */
4116 xfs_extnum_t nex1; /* number of extents before idx */
4117 xfs_extnum_t nex2; /* extents after idx + count */
4118 int nlists; /* entries in indirection array */
4119 int page_idx = idx; /* index in target extent list */
4121 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4122 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
4123 ASSERT(erp != NULL);
4124 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4128 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
4129 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
4131 * Check for deletion of entire list;
4132 * xfs_iext_irec_remove() updates extent offsets.
4134 if (ext_diff == erp->er_extcount) {
4135 xfs_iext_irec_remove(ifp, erp_idx);
4136 ext_cnt -= ext_diff;
4139 ASSERT(erp_idx < ifp->if_real_bytes /
4141 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4148 /* Move extents up (if needed) */
4150 memmove(&erp->er_extbuf[nex1],
4151 &erp->er_extbuf[nex1 + ext_diff],
4152 nex2 * sizeof(xfs_bmbt_rec_t));
4154 /* Zero out rest of page */
4155 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
4156 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
4157 /* Update remaining counters */
4158 erp->er_extcount -= ext_diff;
4159 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
4160 ext_cnt -= ext_diff;
4165 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
4166 xfs_iext_irec_compact(ifp);
4170 * Create, destroy, or resize a linear (direct) block of extents.
4173 xfs_iext_realloc_direct(
4174 xfs_ifork_t *ifp, /* inode fork pointer */
4175 int new_size) /* new size of extents */
4177 int rnew_size; /* real new size of extents */
4179 rnew_size = new_size;
4181 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
4182 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
4183 (new_size != ifp->if_real_bytes)));
4185 /* Free extent records */
4186 if (new_size == 0) {
4187 xfs_iext_destroy(ifp);
4189 /* Resize direct extent list and zero any new bytes */
4190 else if (ifp->if_real_bytes) {
4191 /* Check if extents will fit inside the inode */
4192 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
4193 xfs_iext_direct_to_inline(ifp, new_size /
4194 (uint)sizeof(xfs_bmbt_rec_t));
4195 ifp->if_bytes = new_size;
4198 if (!is_power_of_2(new_size)){
4199 rnew_size = xfs_iroundup(new_size);
4201 if (rnew_size != ifp->if_real_bytes) {
4202 ifp->if_u1.if_extents =
4203 kmem_realloc(ifp->if_u1.if_extents,
4208 if (rnew_size > ifp->if_real_bytes) {
4209 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
4210 (uint)sizeof(xfs_bmbt_rec_t)], 0,
4211 rnew_size - ifp->if_real_bytes);
4215 * Switch from the inline extent buffer to a direct
4216 * extent list. Be sure to include the inline extent
4217 * bytes in new_size.
4220 new_size += ifp->if_bytes;
4221 if (!is_power_of_2(new_size)) {
4222 rnew_size = xfs_iroundup(new_size);
4224 xfs_iext_inline_to_direct(ifp, rnew_size);
4226 ifp->if_real_bytes = rnew_size;
4227 ifp->if_bytes = new_size;
4231 * Switch from linear (direct) extent records to inline buffer.
4234 xfs_iext_direct_to_inline(
4235 xfs_ifork_t *ifp, /* inode fork pointer */
4236 xfs_extnum_t nextents) /* number of extents in file */
4238 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4239 ASSERT(nextents <= XFS_INLINE_EXTS);
4241 * The inline buffer was zeroed when we switched
4242 * from inline to direct extent allocation mode,
4243 * so we don't need to clear it here.
4245 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
4246 nextents * sizeof(xfs_bmbt_rec_t));
4247 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
4248 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
4249 ifp->if_real_bytes = 0;
4253 * Switch from inline buffer to linear (direct) extent records.
4254 * new_size should already be rounded up to the next power of 2
4255 * by the caller (when appropriate), so use new_size as it is.
4256 * However, since new_size may be rounded up, we can't update
4257 * if_bytes here. It is the caller's responsibility to update
4258 * if_bytes upon return.
4261 xfs_iext_inline_to_direct(
4262 xfs_ifork_t *ifp, /* inode fork pointer */
4263 int new_size) /* number of extents in file */
4265 ifp->if_u1.if_extents = kmem_alloc(new_size, KM_SLEEP);
4266 memset(ifp->if_u1.if_extents, 0, new_size);
4267 if (ifp->if_bytes) {
4268 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
4270 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4271 sizeof(xfs_bmbt_rec_t));
4273 ifp->if_real_bytes = new_size;
4277 * Resize an extent indirection array to new_size bytes.
4280 xfs_iext_realloc_indirect(
4281 xfs_ifork_t *ifp, /* inode fork pointer */
4282 int new_size) /* new indirection array size */
4284 int nlists; /* number of irec's (ex lists) */
4285 int size; /* current indirection array size */
4287 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4288 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4289 size = nlists * sizeof(xfs_ext_irec_t);
4290 ASSERT(ifp->if_real_bytes);
4291 ASSERT((new_size >= 0) && (new_size != size));
4292 if (new_size == 0) {
4293 xfs_iext_destroy(ifp);
4295 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
4296 kmem_realloc(ifp->if_u1.if_ext_irec,
4297 new_size, size, KM_SLEEP);
4302 * Switch from indirection array to linear (direct) extent allocations.
4305 xfs_iext_indirect_to_direct(
4306 xfs_ifork_t *ifp) /* inode fork pointer */
4308 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
4309 xfs_extnum_t nextents; /* number of extents in file */
4310 int size; /* size of file extents */
4312 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4313 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4314 ASSERT(nextents <= XFS_LINEAR_EXTS);
4315 size = nextents * sizeof(xfs_bmbt_rec_t);
4317 xfs_iext_irec_compact_full(ifp);
4318 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
4320 ep = ifp->if_u1.if_ext_irec->er_extbuf;
4321 kmem_free(ifp->if_u1.if_ext_irec, sizeof(xfs_ext_irec_t));
4322 ifp->if_flags &= ~XFS_IFEXTIREC;
4323 ifp->if_u1.if_extents = ep;
4324 ifp->if_bytes = size;
4325 if (nextents < XFS_LINEAR_EXTS) {
4326 xfs_iext_realloc_direct(ifp, size);
4331 * Free incore file extents.
4335 xfs_ifork_t *ifp) /* inode fork pointer */
4337 if (ifp->if_flags & XFS_IFEXTIREC) {
4341 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4342 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
4343 xfs_iext_irec_remove(ifp, erp_idx);
4345 ifp->if_flags &= ~XFS_IFEXTIREC;
4346 } else if (ifp->if_real_bytes) {
4347 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
4348 } else if (ifp->if_bytes) {
4349 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4350 sizeof(xfs_bmbt_rec_t));
4352 ifp->if_u1.if_extents = NULL;
4353 ifp->if_real_bytes = 0;
4358 * Return a pointer to the extent record for file system block bno.
4360 xfs_bmbt_rec_host_t * /* pointer to found extent record */
4361 xfs_iext_bno_to_ext(
4362 xfs_ifork_t *ifp, /* inode fork pointer */
4363 xfs_fileoff_t bno, /* block number to search for */
4364 xfs_extnum_t *idxp) /* index of target extent */
4366 xfs_bmbt_rec_host_t *base; /* pointer to first extent */
4367 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
4368 xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */
4369 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4370 int high; /* upper boundary in search */
4371 xfs_extnum_t idx = 0; /* index of target extent */
4372 int low; /* lower boundary in search */
4373 xfs_extnum_t nextents; /* number of file extents */
4374 xfs_fileoff_t startoff = 0; /* start offset of extent */
4376 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4377 if (nextents == 0) {
4382 if (ifp->if_flags & XFS_IFEXTIREC) {
4383 /* Find target extent list */
4385 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
4386 base = erp->er_extbuf;
4387 high = erp->er_extcount - 1;
4389 base = ifp->if_u1.if_extents;
4390 high = nextents - 1;
4392 /* Binary search extent records */
4393 while (low <= high) {
4394 idx = (low + high) >> 1;
4396 startoff = xfs_bmbt_get_startoff(ep);
4397 blockcount = xfs_bmbt_get_blockcount(ep);
4398 if (bno < startoff) {
4400 } else if (bno >= startoff + blockcount) {
4403 /* Convert back to file-based extent index */
4404 if (ifp->if_flags & XFS_IFEXTIREC) {
4405 idx += erp->er_extoff;
4411 /* Convert back to file-based extent index */
4412 if (ifp->if_flags & XFS_IFEXTIREC) {
4413 idx += erp->er_extoff;
4415 if (bno >= startoff + blockcount) {
4416 if (++idx == nextents) {
4419 ep = xfs_iext_get_ext(ifp, idx);
4427 * Return a pointer to the indirection array entry containing the
4428 * extent record for filesystem block bno. Store the index of the
4429 * target irec in *erp_idxp.
4431 xfs_ext_irec_t * /* pointer to found extent record */
4432 xfs_iext_bno_to_irec(
4433 xfs_ifork_t *ifp, /* inode fork pointer */
4434 xfs_fileoff_t bno, /* block number to search for */
4435 int *erp_idxp) /* irec index of target ext list */
4437 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4438 xfs_ext_irec_t *erp_next; /* next indirection array entry */
4439 int erp_idx; /* indirection array index */
4440 int nlists; /* number of extent irec's (lists) */
4441 int high; /* binary search upper limit */
4442 int low; /* binary search lower limit */
4444 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4445 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4449 while (low <= high) {
4450 erp_idx = (low + high) >> 1;
4451 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4452 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
4453 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
4455 } else if (erp_next && bno >=
4456 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
4462 *erp_idxp = erp_idx;
4467 * Return a pointer to the indirection array entry containing the
4468 * extent record at file extent index *idxp. Store the index of the
4469 * target irec in *erp_idxp and store the page index of the target
4470 * extent record in *idxp.
4473 xfs_iext_idx_to_irec(
4474 xfs_ifork_t *ifp, /* inode fork pointer */
4475 xfs_extnum_t *idxp, /* extent index (file -> page) */
4476 int *erp_idxp, /* pointer to target irec */
4477 int realloc) /* new bytes were just added */
4479 xfs_ext_irec_t *prev; /* pointer to previous irec */
4480 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
4481 int erp_idx; /* indirection array index */
4482 int nlists; /* number of irec's (ex lists) */
4483 int high; /* binary search upper limit */
4484 int low; /* binary search lower limit */
4485 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
4487 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4488 ASSERT(page_idx >= 0 && page_idx <=
4489 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
4490 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4495 /* Binary search extent irec's */
4496 while (low <= high) {
4497 erp_idx = (low + high) >> 1;
4498 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4499 prev = erp_idx > 0 ? erp - 1 : NULL;
4500 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
4501 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
4503 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
4504 (page_idx == erp->er_extoff + erp->er_extcount &&
4507 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
4508 erp->er_extcount == XFS_LINEAR_EXTS) {
4512 erp = erp_idx < nlists ? erp + 1 : NULL;
4515 page_idx -= erp->er_extoff;
4520 *erp_idxp = erp_idx;
4525 * Allocate and initialize an indirection array once the space needed
4526 * for incore extents increases above XFS_IEXT_BUFSZ.
4530 xfs_ifork_t *ifp) /* inode fork pointer */
4532 xfs_ext_irec_t *erp; /* indirection array pointer */
4533 xfs_extnum_t nextents; /* number of extents in file */
4535 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4536 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4537 ASSERT(nextents <= XFS_LINEAR_EXTS);
4539 erp = (xfs_ext_irec_t *)
4540 kmem_alloc(sizeof(xfs_ext_irec_t), KM_SLEEP);
4542 if (nextents == 0) {
4543 ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
4544 } else if (!ifp->if_real_bytes) {
4545 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
4546 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
4547 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
4549 erp->er_extbuf = ifp->if_u1.if_extents;
4550 erp->er_extcount = nextents;
4553 ifp->if_flags |= XFS_IFEXTIREC;
4554 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
4555 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
4556 ifp->if_u1.if_ext_irec = erp;
4562 * Allocate and initialize a new entry in the indirection array.
4566 xfs_ifork_t *ifp, /* inode fork pointer */
4567 int erp_idx) /* index for new irec */
4569 xfs_ext_irec_t *erp; /* indirection array pointer */
4570 int i; /* loop counter */
4571 int nlists; /* number of irec's (ex lists) */
4573 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4574 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4576 /* Resize indirection array */
4577 xfs_iext_realloc_indirect(ifp, ++nlists *
4578 sizeof(xfs_ext_irec_t));
4580 * Move records down in the array so the
4581 * new page can use erp_idx.
4583 erp = ifp->if_u1.if_ext_irec;
4584 for (i = nlists - 1; i > erp_idx; i--) {
4585 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
4587 ASSERT(i == erp_idx);
4589 /* Initialize new extent record */
4590 erp = ifp->if_u1.if_ext_irec;
4591 erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
4592 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4593 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
4594 erp[erp_idx].er_extcount = 0;
4595 erp[erp_idx].er_extoff = erp_idx > 0 ?
4596 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
4597 return (&erp[erp_idx]);
4601 * Remove a record from the indirection array.
4604 xfs_iext_irec_remove(
4605 xfs_ifork_t *ifp, /* inode fork pointer */
4606 int erp_idx) /* irec index to remove */
4608 xfs_ext_irec_t *erp; /* indirection array pointer */
4609 int i; /* loop counter */
4610 int nlists; /* number of irec's (ex lists) */
4612 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4613 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4614 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4615 if (erp->er_extbuf) {
4616 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
4618 kmem_free(erp->er_extbuf, XFS_IEXT_BUFSZ);
4620 /* Compact extent records */
4621 erp = ifp->if_u1.if_ext_irec;
4622 for (i = erp_idx; i < nlists - 1; i++) {
4623 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
4626 * Manually free the last extent record from the indirection
4627 * array. A call to xfs_iext_realloc_indirect() with a size
4628 * of zero would result in a call to xfs_iext_destroy() which
4629 * would in turn call this function again, creating a nasty
4633 xfs_iext_realloc_indirect(ifp,
4634 nlists * sizeof(xfs_ext_irec_t));
4636 kmem_free(ifp->if_u1.if_ext_irec,
4637 sizeof(xfs_ext_irec_t));
4639 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4643 * This is called to clean up large amounts of unused memory allocated
4644 * by the indirection array. Before compacting anything though, verify
4645 * that the indirection array is still needed and switch back to the
4646 * linear extent list (or even the inline buffer) if possible. The
4647 * compaction policy is as follows:
4649 * Full Compaction: Extents fit into a single page (or inline buffer)
4650 * Full Compaction: Extents occupy less than 10% of allocated space
4651 * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
4652 * No Compaction: Extents occupy at least 50% of allocated space
4655 xfs_iext_irec_compact(
4656 xfs_ifork_t *ifp) /* inode fork pointer */
4658 xfs_extnum_t nextents; /* number of extents in file */
4659 int nlists; /* number of irec's (ex lists) */
4661 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4662 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4663 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4665 if (nextents == 0) {
4666 xfs_iext_destroy(ifp);
4667 } else if (nextents <= XFS_INLINE_EXTS) {
4668 xfs_iext_indirect_to_direct(ifp);
4669 xfs_iext_direct_to_inline(ifp, nextents);
4670 } else if (nextents <= XFS_LINEAR_EXTS) {
4671 xfs_iext_indirect_to_direct(ifp);
4672 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 3) {
4673 xfs_iext_irec_compact_full(ifp);
4674 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
4675 xfs_iext_irec_compact_pages(ifp);
4680 * Combine extents from neighboring extent pages.
4683 xfs_iext_irec_compact_pages(
4684 xfs_ifork_t *ifp) /* inode fork pointer */
4686 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
4687 int erp_idx = 0; /* indirection array index */
4688 int nlists; /* number of irec's (ex lists) */
4690 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4691 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4692 while (erp_idx < nlists - 1) {
4693 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4695 if (erp_next->er_extcount <=
4696 (XFS_LINEAR_EXTS - erp->er_extcount)) {
4697 memmove(&erp->er_extbuf[erp->er_extcount],
4698 erp_next->er_extbuf, erp_next->er_extcount *
4699 sizeof(xfs_bmbt_rec_t));
4700 erp->er_extcount += erp_next->er_extcount;
4702 * Free page before removing extent record
4703 * so er_extoffs don't get modified in
4704 * xfs_iext_irec_remove.
4706 kmem_free(erp_next->er_extbuf, XFS_IEXT_BUFSZ);
4707 erp_next->er_extbuf = NULL;
4708 xfs_iext_irec_remove(ifp, erp_idx + 1);
4709 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4717 * Fully compact the extent records managed by the indirection array.
4720 xfs_iext_irec_compact_full(
4721 xfs_ifork_t *ifp) /* inode fork pointer */
4723 xfs_bmbt_rec_host_t *ep, *ep_next; /* extent record pointers */
4724 xfs_ext_irec_t *erp, *erp_next; /* extent irec pointers */
4725 int erp_idx = 0; /* extent irec index */
4726 int ext_avail; /* empty entries in ex list */
4727 int ext_diff; /* number of exts to add */
4728 int nlists; /* number of irec's (ex lists) */
4730 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4731 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4732 erp = ifp->if_u1.if_ext_irec;
4733 ep = &erp->er_extbuf[erp->er_extcount];
4735 ep_next = erp_next->er_extbuf;
4736 while (erp_idx < nlists - 1) {
4737 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
4738 ext_diff = MIN(ext_avail, erp_next->er_extcount);
4739 memcpy(ep, ep_next, ext_diff * sizeof(xfs_bmbt_rec_t));
4740 erp->er_extcount += ext_diff;
4741 erp_next->er_extcount -= ext_diff;
4742 /* Remove next page */
4743 if (erp_next->er_extcount == 0) {
4745 * Free page before removing extent record
4746 * so er_extoffs don't get modified in
4747 * xfs_iext_irec_remove.
4749 kmem_free(erp_next->er_extbuf,
4750 erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
4751 erp_next->er_extbuf = NULL;
4752 xfs_iext_irec_remove(ifp, erp_idx + 1);
4753 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4754 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4755 /* Update next page */
4757 /* Move rest of page up to become next new page */
4758 memmove(erp_next->er_extbuf, ep_next,
4759 erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
4760 ep_next = erp_next->er_extbuf;
4761 memset(&ep_next[erp_next->er_extcount], 0,
4762 (XFS_LINEAR_EXTS - erp_next->er_extcount) *
4763 sizeof(xfs_bmbt_rec_t));
4765 if (erp->er_extcount == XFS_LINEAR_EXTS) {
4767 if (erp_idx < nlists)
4768 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4772 ep = &erp->er_extbuf[erp->er_extcount];
4774 ep_next = erp_next->er_extbuf;
4779 * This is called to update the er_extoff field in the indirection
4780 * array when extents have been added or removed from one of the
4781 * extent lists. erp_idx contains the irec index to begin updating
4782 * at and ext_diff contains the number of extents that were added
4786 xfs_iext_irec_update_extoffs(
4787 xfs_ifork_t *ifp, /* inode fork pointer */
4788 int erp_idx, /* irec index to update */
4789 int ext_diff) /* number of new extents */
4791 int i; /* loop counter */
4792 int nlists; /* number of irec's (ex lists */
4794 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4795 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4796 for (i = erp_idx; i < nlists; i++) {
4797 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;