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"
31 #include "xfs_dmapi.h"
32 #include "xfs_mount.h"
33 #include "xfs_bmap_btree.h"
34 #include "xfs_alloc_btree.h"
35 #include "xfs_ialloc_btree.h"
36 #include "xfs_dir_sf.h"
37 #include "xfs_dir2_sf.h"
38 #include "xfs_attr_sf.h"
39 #include "xfs_dinode.h"
40 #include "xfs_inode.h"
41 #include "xfs_buf_item.h"
42 #include "xfs_inode_item.h"
43 #include "xfs_btree.h"
44 #include "xfs_alloc.h"
45 #include "xfs_ialloc.h"
48 #include "xfs_error.h"
49 #include "xfs_utils.h"
50 #include "xfs_dir2_trace.h"
51 #include "xfs_quota.h"
56 kmem_zone_t *xfs_ifork_zone;
57 kmem_zone_t *xfs_inode_zone;
58 kmem_zone_t *xfs_chashlist_zone;
61 * Used in xfs_itruncate(). This is the maximum number of extents
62 * freed from a file in a single transaction.
64 #define XFS_ITRUNC_MAX_EXTENTS 2
66 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
67 STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
68 STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
69 STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
74 * Make sure that the extents in the given memory buffer
89 for (i = 0; i < nrecs; i++) {
90 ep = xfs_iext_get_ext(ifp, i);
91 rec.l0 = get_unaligned((__uint64_t*)&ep->l0);
92 rec.l1 = get_unaligned((__uint64_t*)&ep->l1);
94 xfs_bmbt_disk_get_all(&rec, &irec);
96 xfs_bmbt_get_all(&rec, &irec);
97 if (fmt == XFS_EXTFMT_NOSTATE)
98 ASSERT(irec.br_state == XFS_EXT_NORM);
102 #define xfs_validate_extents(ifp, nrecs, disk, fmt)
106 * Check that none of the inode's in the buffer have a next
107 * unlinked field of 0.
119 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
121 for (i = 0; i < j; i++) {
122 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
123 i * mp->m_sb.sb_inodesize);
124 if (!dip->di_next_unlinked) {
125 xfs_fs_cmn_err(CE_ALERT, mp,
126 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
128 ASSERT(dip->di_next_unlinked);
135 * This routine is called to map an inode number within a file
136 * system to the buffer containing the on-disk version of the
137 * inode. It returns a pointer to the buffer containing the
138 * on-disk inode in the bpp parameter, and in the dip parameter
139 * it returns a pointer to the on-disk inode within that buffer.
141 * If a non-zero error is returned, then the contents of bpp and
142 * dipp are undefined.
144 * Use xfs_imap() to determine the size and location of the
145 * buffer to read from disk.
163 * Call the space management code to find the location of the
167 error = xfs_imap(mp, tp, ino, &imap, XFS_IMAP_LOOKUP);
170 "xfs_inotobp: xfs_imap() returned an "
171 "error %d on %s. Returning error.", error, mp->m_fsname);
176 * If the inode number maps to a block outside the bounds of the
177 * file system then return NULL rather than calling read_buf
178 * and panicing when we get an error from the driver.
180 if ((imap.im_blkno + imap.im_len) >
181 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
183 "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
184 "of the file system %s. Returning EINVAL.",
185 (unsigned long long)imap.im_blkno,
186 imap.im_len, mp->m_fsname);
187 return XFS_ERROR(EINVAL);
191 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
192 * default to just a read_buf() call.
194 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
195 (int)imap.im_len, XFS_BUF_LOCK, &bp);
199 "xfs_inotobp: xfs_trans_read_buf() returned an "
200 "error %d on %s. Returning error.", error, mp->m_fsname);
203 dip = (xfs_dinode_t *)xfs_buf_offset(bp, 0);
205 INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
206 XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
207 if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP,
208 XFS_RANDOM_ITOBP_INOTOBP))) {
209 XFS_CORRUPTION_ERROR("xfs_inotobp", XFS_ERRLEVEL_LOW, mp, dip);
210 xfs_trans_brelse(tp, bp);
212 "xfs_inotobp: XFS_TEST_ERROR() returned an "
213 "error on %s. Returning EFSCORRUPTED.", mp->m_fsname);
214 return XFS_ERROR(EFSCORRUPTED);
217 xfs_inobp_check(mp, bp);
220 * Set *dipp to point to the on-disk inode in the buffer.
222 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
224 *offset = imap.im_boffset;
230 * This routine is called to map an inode to the buffer containing
231 * the on-disk version of the inode. It returns a pointer to the
232 * buffer containing the on-disk inode in the bpp parameter, and in
233 * the dip parameter it returns a pointer to the on-disk inode within
236 * If a non-zero error is returned, then the contents of bpp and
237 * dipp are undefined.
239 * If the inode is new and has not yet been initialized, use xfs_imap()
240 * to determine the size and location of the buffer to read from disk.
241 * If the inode has already been mapped to its buffer and read in once,
242 * then use the mapping information stored in the inode rather than
243 * calling xfs_imap(). This allows us to avoid the overhead of looking
244 * at the inode btree for small block file systems (see xfs_dilocate()).
245 * We can tell whether the inode has been mapped in before by comparing
246 * its disk block address to 0. Only uninitialized inodes will have
247 * 0 for the disk block address.
265 if (ip->i_blkno == (xfs_daddr_t)0) {
267 * Call the space management code to find the location of the
271 if ((error = xfs_imap(mp, tp, ip->i_ino, &imap,
272 XFS_IMAP_LOOKUP | imap_flags)))
276 * If the inode number maps to a block outside the bounds
277 * of the file system then return NULL rather than calling
278 * read_buf and panicing when we get an error from the
281 if ((imap.im_blkno + imap.im_len) >
282 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
284 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
285 "(imap.im_blkno (0x%llx) "
286 "+ imap.im_len (0x%llx)) > "
287 " XFS_FSB_TO_BB(mp, "
288 "mp->m_sb.sb_dblocks) (0x%llx)",
289 (unsigned long long) imap.im_blkno,
290 (unsigned long long) imap.im_len,
291 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
293 return XFS_ERROR(EINVAL);
297 * Fill in the fields in the inode that will be used to
298 * map the inode to its buffer from now on.
300 ip->i_blkno = imap.im_blkno;
301 ip->i_len = imap.im_len;
302 ip->i_boffset = imap.im_boffset;
305 * We've already mapped the inode once, so just use the
306 * mapping that we saved the first time.
308 imap.im_blkno = ip->i_blkno;
309 imap.im_len = ip->i_len;
310 imap.im_boffset = ip->i_boffset;
312 ASSERT(bno == 0 || bno == imap.im_blkno);
315 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
316 * default to just a read_buf() call.
318 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
319 (int)imap.im_len, XFS_BUF_LOCK, &bp);
322 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
323 "xfs_trans_read_buf() returned error %d, "
324 "imap.im_blkno 0x%llx, imap.im_len 0x%llx",
325 error, (unsigned long long) imap.im_blkno,
326 (unsigned long long) imap.im_len);
332 * Validate the magic number and version of every inode in the buffer
333 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
334 * No validation is done here in userspace (xfs_repair).
336 #if !defined(__KERNEL__)
339 ni = (imap_flags & XFS_IMAP_BULKSTAT) ? 0 :
340 (BBTOB(imap.im_len) >> mp->m_sb.sb_inodelog);
341 #else /* usual case */
342 ni = (imap_flags & XFS_IMAP_BULKSTAT) ? 0 : 1;
345 for (i = 0; i < ni; i++) {
349 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
350 (i << mp->m_sb.sb_inodelog));
351 di_ok = INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
352 XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
353 if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP,
354 XFS_RANDOM_ITOBP_INOTOBP))) {
356 if (!(imap_flags & XFS_IMAP_BULKSTAT))
358 "Device %s - bad inode magic/vsn "
359 "daddr %lld #%d (magic=%x)",
360 XFS_BUFTARG_NAME(mp->m_ddev_targp),
361 (unsigned long long)imap.im_blkno, i,
362 INT_GET(dip->di_core.di_magic, ARCH_CONVERT));
364 XFS_CORRUPTION_ERROR("xfs_itobp", XFS_ERRLEVEL_HIGH,
366 xfs_trans_brelse(tp, bp);
367 return XFS_ERROR(EFSCORRUPTED);
371 xfs_inobp_check(mp, bp);
374 * Mark the buffer as an inode buffer now that it looks good
376 XFS_BUF_SET_VTYPE(bp, B_FS_INO);
379 * Set *dipp to point to the on-disk inode in the buffer.
381 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
387 * Move inode type and inode format specific information from the
388 * on-disk inode to the in-core inode. For fifos, devs, and sockets
389 * this means set if_rdev to the proper value. For files, directories,
390 * and symlinks this means to bring in the in-line data or extent
391 * pointers. For a file in B-tree format, only the root is immediately
392 * brought in-core. The rest will be in-lined in if_extents when it
393 * is first referenced (see xfs_iread_extents()).
400 xfs_attr_shortform_t *atp;
404 ip->i_df.if_ext_max =
405 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
409 INT_GET(dip->di_core.di_nextents, ARCH_CONVERT) +
410 INT_GET(dip->di_core.di_anextents, ARCH_CONVERT) >
411 INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT))) {
412 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
413 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
414 (unsigned long long)ip->i_ino,
415 (int)(INT_GET(dip->di_core.di_nextents, ARCH_CONVERT)
416 + INT_GET(dip->di_core.di_anextents, ARCH_CONVERT)),
418 INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT));
419 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
421 return XFS_ERROR(EFSCORRUPTED);
424 if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) {
425 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
426 "corrupt dinode %Lu, forkoff = 0x%x.",
427 (unsigned long long)ip->i_ino,
428 (int)(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT)));
429 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
431 return XFS_ERROR(EFSCORRUPTED);
434 switch (ip->i_d.di_mode & S_IFMT) {
439 if (unlikely(INT_GET(dip->di_core.di_format, ARCH_CONVERT) != XFS_DINODE_FMT_DEV)) {
440 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
442 return XFS_ERROR(EFSCORRUPTED);
445 ip->i_df.if_u2.if_rdev = INT_GET(dip->di_u.di_dev, ARCH_CONVERT);
451 switch (INT_GET(dip->di_core.di_format, ARCH_CONVERT)) {
452 case XFS_DINODE_FMT_LOCAL:
454 * no local regular files yet
456 if (unlikely((INT_GET(dip->di_core.di_mode, ARCH_CONVERT) & S_IFMT) == S_IFREG)) {
457 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
459 "(local format for regular file).",
460 (unsigned long long) ip->i_ino);
461 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
464 return XFS_ERROR(EFSCORRUPTED);
467 di_size = INT_GET(dip->di_core.di_size, ARCH_CONVERT);
468 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
469 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
471 "(bad size %Ld for local inode).",
472 (unsigned long long) ip->i_ino,
473 (long long) di_size);
474 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
477 return XFS_ERROR(EFSCORRUPTED);
481 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
483 case XFS_DINODE_FMT_EXTENTS:
484 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
486 case XFS_DINODE_FMT_BTREE:
487 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
490 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
492 return XFS_ERROR(EFSCORRUPTED);
497 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
498 return XFS_ERROR(EFSCORRUPTED);
503 if (!XFS_DFORK_Q(dip))
505 ASSERT(ip->i_afp == NULL);
506 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
507 ip->i_afp->if_ext_max =
508 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
509 switch (INT_GET(dip->di_core.di_aformat, ARCH_CONVERT)) {
510 case XFS_DINODE_FMT_LOCAL:
511 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
512 size = be16_to_cpu(atp->hdr.totsize);
513 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
515 case XFS_DINODE_FMT_EXTENTS:
516 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
518 case XFS_DINODE_FMT_BTREE:
519 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
522 error = XFS_ERROR(EFSCORRUPTED);
526 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
528 xfs_idestroy_fork(ip, XFS_DATA_FORK);
534 * The file is in-lined in the on-disk inode.
535 * If it fits into if_inline_data, then copy
536 * it there, otherwise allocate a buffer for it
537 * and copy the data there. Either way, set
538 * if_data to point at the data.
539 * If we allocate a buffer for the data, make
540 * sure that its size is a multiple of 4 and
541 * record the real size in i_real_bytes.
554 * If the size is unreasonable, then something
555 * is wrong and we just bail out rather than crash in
556 * kmem_alloc() or memcpy() below.
558 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
559 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
561 "(bad size %d for local fork, size = %d).",
562 (unsigned long long) ip->i_ino, size,
563 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
564 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
566 return XFS_ERROR(EFSCORRUPTED);
568 ifp = XFS_IFORK_PTR(ip, whichfork);
571 ifp->if_u1.if_data = NULL;
572 else if (size <= sizeof(ifp->if_u2.if_inline_data))
573 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
575 real_size = roundup(size, 4);
576 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
578 ifp->if_bytes = size;
579 ifp->if_real_bytes = real_size;
581 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
582 ifp->if_flags &= ~XFS_IFEXTENTS;
583 ifp->if_flags |= XFS_IFINLINE;
588 * The file consists of a set of extents all
589 * of which fit into the on-disk inode.
590 * If there are few enough extents to fit into
591 * the if_inline_ext, then copy them there.
592 * Otherwise allocate a buffer for them and copy
593 * them into it. Either way, set if_extents
594 * to point at the extents.
602 xfs_bmbt_rec_t *ep, *dp;
608 ifp = XFS_IFORK_PTR(ip, whichfork);
609 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
610 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
613 * If the number of extents is unreasonable, then something
614 * is wrong and we just bail out rather than crash in
615 * kmem_alloc() or memcpy() below.
617 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
618 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
619 "corrupt inode %Lu ((a)extents = %d).",
620 (unsigned long long) ip->i_ino, nex);
621 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
623 return XFS_ERROR(EFSCORRUPTED);
626 ifp->if_real_bytes = 0;
628 ifp->if_u1.if_extents = NULL;
629 else if (nex <= XFS_INLINE_EXTS)
630 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
632 xfs_iext_add(ifp, 0, nex);
634 ifp->if_bytes = size;
636 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
637 xfs_validate_extents(ifp, nex, 1, XFS_EXTFMT_INODE(ip));
638 for (i = 0; i < nex; i++, dp++) {
639 ep = xfs_iext_get_ext(ifp, i);
640 ep->l0 = INT_GET(get_unaligned((__uint64_t*)&dp->l0),
642 ep->l1 = INT_GET(get_unaligned((__uint64_t*)&dp->l1),
645 xfs_bmap_trace_exlist("xfs_iformat_extents", ip, nex,
647 if (whichfork != XFS_DATA_FORK ||
648 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
649 if (unlikely(xfs_check_nostate_extents(
651 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
654 return XFS_ERROR(EFSCORRUPTED);
657 ifp->if_flags |= XFS_IFEXTENTS;
662 * The file has too many extents to fit into
663 * the inode, so they are in B-tree format.
664 * Allocate a buffer for the root of the B-tree
665 * and copy the root into it. The i_extents
666 * field will remain NULL until all of the
667 * extents are read in (when they are needed).
675 xfs_bmdr_block_t *dfp;
681 ifp = XFS_IFORK_PTR(ip, whichfork);
682 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
683 size = XFS_BMAP_BROOT_SPACE(dfp);
684 nrecs = XFS_BMAP_BROOT_NUMRECS(dfp);
687 * blow out if -- fork has less extents than can fit in
688 * fork (fork shouldn't be a btree format), root btree
689 * block has more records than can fit into the fork,
690 * or the number of extents is greater than the number of
693 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
694 || XFS_BMDR_SPACE_CALC(nrecs) >
695 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
696 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
697 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
698 "corrupt inode %Lu (btree).",
699 (unsigned long long) ip->i_ino);
700 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
702 return XFS_ERROR(EFSCORRUPTED);
705 ifp->if_broot_bytes = size;
706 ifp->if_broot = kmem_alloc(size, KM_SLEEP);
707 ASSERT(ifp->if_broot != NULL);
709 * Copy and convert from the on-disk structure
710 * to the in-memory structure.
712 xfs_bmdr_to_bmbt(dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
713 ifp->if_broot, size);
714 ifp->if_flags &= ~XFS_IFEXTENTS;
715 ifp->if_flags |= XFS_IFBROOT;
721 * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
724 * buf = on-disk representation
725 * dip = native representation
726 * dir = direction - +ve -> disk to native
727 * -ve -> native to disk
730 xfs_xlate_dinode_core(
732 xfs_dinode_core_t *dip,
735 xfs_dinode_core_t *buf_core = (xfs_dinode_core_t *)buf;
736 xfs_dinode_core_t *mem_core = (xfs_dinode_core_t *)dip;
737 xfs_arch_t arch = ARCH_CONVERT;
741 INT_XLATE(buf_core->di_magic, mem_core->di_magic, dir, arch);
742 INT_XLATE(buf_core->di_mode, mem_core->di_mode, dir, arch);
743 INT_XLATE(buf_core->di_version, mem_core->di_version, dir, arch);
744 INT_XLATE(buf_core->di_format, mem_core->di_format, dir, arch);
745 INT_XLATE(buf_core->di_onlink, mem_core->di_onlink, dir, arch);
746 INT_XLATE(buf_core->di_uid, mem_core->di_uid, dir, arch);
747 INT_XLATE(buf_core->di_gid, mem_core->di_gid, dir, arch);
748 INT_XLATE(buf_core->di_nlink, mem_core->di_nlink, dir, arch);
749 INT_XLATE(buf_core->di_projid, mem_core->di_projid, dir, arch);
752 memcpy(mem_core->di_pad, buf_core->di_pad,
753 sizeof(buf_core->di_pad));
755 memcpy(buf_core->di_pad, mem_core->di_pad,
756 sizeof(buf_core->di_pad));
759 INT_XLATE(buf_core->di_flushiter, mem_core->di_flushiter, dir, arch);
761 INT_XLATE(buf_core->di_atime.t_sec, mem_core->di_atime.t_sec,
763 INT_XLATE(buf_core->di_atime.t_nsec, mem_core->di_atime.t_nsec,
765 INT_XLATE(buf_core->di_mtime.t_sec, mem_core->di_mtime.t_sec,
767 INT_XLATE(buf_core->di_mtime.t_nsec, mem_core->di_mtime.t_nsec,
769 INT_XLATE(buf_core->di_ctime.t_sec, mem_core->di_ctime.t_sec,
771 INT_XLATE(buf_core->di_ctime.t_nsec, mem_core->di_ctime.t_nsec,
773 INT_XLATE(buf_core->di_size, mem_core->di_size, dir, arch);
774 INT_XLATE(buf_core->di_nblocks, mem_core->di_nblocks, dir, arch);
775 INT_XLATE(buf_core->di_extsize, mem_core->di_extsize, dir, arch);
776 INT_XLATE(buf_core->di_nextents, mem_core->di_nextents, dir, arch);
777 INT_XLATE(buf_core->di_anextents, mem_core->di_anextents, dir, arch);
778 INT_XLATE(buf_core->di_forkoff, mem_core->di_forkoff, dir, arch);
779 INT_XLATE(buf_core->di_aformat, mem_core->di_aformat, dir, arch);
780 INT_XLATE(buf_core->di_dmevmask, mem_core->di_dmevmask, dir, arch);
781 INT_XLATE(buf_core->di_dmstate, mem_core->di_dmstate, dir, arch);
782 INT_XLATE(buf_core->di_flags, mem_core->di_flags, dir, arch);
783 INT_XLATE(buf_core->di_gen, mem_core->di_gen, dir, arch);
792 if (di_flags & XFS_DIFLAG_ANY) {
793 if (di_flags & XFS_DIFLAG_REALTIME)
794 flags |= XFS_XFLAG_REALTIME;
795 if (di_flags & XFS_DIFLAG_PREALLOC)
796 flags |= XFS_XFLAG_PREALLOC;
797 if (di_flags & XFS_DIFLAG_IMMUTABLE)
798 flags |= XFS_XFLAG_IMMUTABLE;
799 if (di_flags & XFS_DIFLAG_APPEND)
800 flags |= XFS_XFLAG_APPEND;
801 if (di_flags & XFS_DIFLAG_SYNC)
802 flags |= XFS_XFLAG_SYNC;
803 if (di_flags & XFS_DIFLAG_NOATIME)
804 flags |= XFS_XFLAG_NOATIME;
805 if (di_flags & XFS_DIFLAG_NODUMP)
806 flags |= XFS_XFLAG_NODUMP;
807 if (di_flags & XFS_DIFLAG_RTINHERIT)
808 flags |= XFS_XFLAG_RTINHERIT;
809 if (di_flags & XFS_DIFLAG_PROJINHERIT)
810 flags |= XFS_XFLAG_PROJINHERIT;
811 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
812 flags |= XFS_XFLAG_NOSYMLINKS;
813 if (di_flags & XFS_DIFLAG_EXTSIZE)
814 flags |= XFS_XFLAG_EXTSIZE;
815 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
816 flags |= XFS_XFLAG_EXTSZINHERIT;
817 if (di_flags & XFS_DIFLAG_NODEFRAG)
818 flags |= XFS_XFLAG_NODEFRAG;
828 xfs_dinode_core_t *dic = &ip->i_d;
830 return _xfs_dic2xflags(dic->di_flags) |
831 (XFS_CFORK_Q(dic) ? XFS_XFLAG_HASATTR : 0);
836 xfs_dinode_core_t *dic)
838 return _xfs_dic2xflags(INT_GET(dic->di_flags, ARCH_CONVERT)) |
839 (XFS_CFORK_Q_DISK(dic) ? XFS_XFLAG_HASATTR : 0);
843 * Given a mount structure and an inode number, return a pointer
844 * to a newly allocated in-core inode corresponding to the given
847 * Initialize the inode's attributes and extent pointers if it
848 * already has them (it will not if the inode has no links).
863 ASSERT(xfs_inode_zone != NULL);
865 ip = kmem_zone_zalloc(xfs_inode_zone, KM_SLEEP);
870 * Get pointer's to the on-disk inode and the buffer containing it.
871 * If the inode number refers to a block outside the file system
872 * then xfs_itobp() will return NULL. In this case we should
873 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
874 * know that this is a new incore inode.
876 error = xfs_itobp(mp, tp, ip, &dip, &bp, bno, 0);
878 kmem_zone_free(xfs_inode_zone, ip);
883 * Initialize inode's trace buffers.
884 * Do this before xfs_iformat in case it adds entries.
886 #ifdef XFS_BMAP_TRACE
887 ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_SLEEP);
889 #ifdef XFS_BMBT_TRACE
890 ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_SLEEP);
893 ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_SLEEP);
895 #ifdef XFS_ILOCK_TRACE
896 ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_SLEEP);
898 #ifdef XFS_DIR2_TRACE
899 ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_SLEEP);
903 * If we got something that isn't an inode it means someone
904 * (nfs or dmi) has a stale handle.
906 if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC) {
907 kmem_zone_free(xfs_inode_zone, ip);
908 xfs_trans_brelse(tp, bp);
910 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
911 "dip->di_core.di_magic (0x%x) != "
912 "XFS_DINODE_MAGIC (0x%x)",
913 INT_GET(dip->di_core.di_magic, ARCH_CONVERT),
916 return XFS_ERROR(EINVAL);
920 * If the on-disk inode is already linked to a directory
921 * entry, copy all of the inode into the in-core inode.
922 * xfs_iformat() handles copying in the inode format
923 * specific information.
924 * Otherwise, just get the truly permanent information.
926 if (dip->di_core.di_mode) {
927 xfs_xlate_dinode_core((xfs_caddr_t)&dip->di_core,
929 error = xfs_iformat(ip, dip);
931 kmem_zone_free(xfs_inode_zone, ip);
932 xfs_trans_brelse(tp, bp);
934 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
935 "xfs_iformat() returned error %d",
941 ip->i_d.di_magic = INT_GET(dip->di_core.di_magic, ARCH_CONVERT);
942 ip->i_d.di_version = INT_GET(dip->di_core.di_version, ARCH_CONVERT);
943 ip->i_d.di_gen = INT_GET(dip->di_core.di_gen, ARCH_CONVERT);
944 ip->i_d.di_flushiter = INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT);
946 * Make sure to pull in the mode here as well in
947 * case the inode is released without being used.
948 * This ensures that xfs_inactive() will see that
949 * the inode is already free and not try to mess
950 * with the uninitialized part of it.
954 * Initialize the per-fork minima and maxima for a new
955 * inode here. xfs_iformat will do it for old inodes.
957 ip->i_df.if_ext_max =
958 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
961 INIT_LIST_HEAD(&ip->i_reclaim);
964 * The inode format changed when we moved the link count and
965 * made it 32 bits long. If this is an old format inode,
966 * convert it in memory to look like a new one. If it gets
967 * flushed to disk we will convert back before flushing or
968 * logging it. We zero out the new projid field and the old link
969 * count field. We'll handle clearing the pad field (the remains
970 * of the old uuid field) when we actually convert the inode to
971 * the new format. We don't change the version number so that we
972 * can distinguish this from a real new format inode.
974 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
975 ip->i_d.di_nlink = ip->i_d.di_onlink;
976 ip->i_d.di_onlink = 0;
977 ip->i_d.di_projid = 0;
980 ip->i_delayed_blks = 0;
983 * Mark the buffer containing the inode as something to keep
984 * around for a while. This helps to keep recently accessed
985 * meta-data in-core longer.
987 XFS_BUF_SET_REF(bp, XFS_INO_REF);
990 * Use xfs_trans_brelse() to release the buffer containing the
991 * on-disk inode, because it was acquired with xfs_trans_read_buf()
992 * in xfs_itobp() above. If tp is NULL, this is just a normal
993 * brelse(). If we're within a transaction, then xfs_trans_brelse()
994 * will only release the buffer if it is not dirty within the
995 * transaction. It will be OK to release the buffer in this case,
996 * because inodes on disk are never destroyed and we will be
997 * locking the new in-core inode before putting it in the hash
998 * table where other processes can find it. Thus we don't have
999 * to worry about the inode being changed just because we released
1002 xfs_trans_brelse(tp, bp);
1008 * Read in extents from a btree-format inode.
1009 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1019 xfs_extnum_t nextents;
1022 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
1023 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
1025 return XFS_ERROR(EFSCORRUPTED);
1027 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
1028 size = nextents * sizeof(xfs_bmbt_rec_t);
1029 ifp = XFS_IFORK_PTR(ip, whichfork);
1032 * We know that the size is valid (it's checked in iformat_btree)
1034 ifp->if_lastex = NULLEXTNUM;
1035 ifp->if_bytes = ifp->if_real_bytes = 0;
1036 ifp->if_flags |= XFS_IFEXTENTS;
1037 xfs_iext_add(ifp, 0, nextents);
1038 error = xfs_bmap_read_extents(tp, ip, whichfork);
1040 xfs_iext_destroy(ifp);
1041 ifp->if_flags &= ~XFS_IFEXTENTS;
1044 xfs_validate_extents(ifp, nextents, 0, XFS_EXTFMT_INODE(ip));
1049 * Allocate an inode on disk and return a copy of its in-core version.
1050 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1051 * appropriately within the inode. The uid and gid for the inode are
1052 * set according to the contents of the given cred structure.
1054 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1055 * has a free inode available, call xfs_iget()
1056 * to obtain the in-core version of the allocated inode. Finally,
1057 * fill in the inode and log its initial contents. In this case,
1058 * ialloc_context would be set to NULL and call_again set to false.
1060 * If xfs_dialloc() does not have an available inode,
1061 * it will replenish its supply by doing an allocation. Since we can
1062 * only do one allocation within a transaction without deadlocks, we
1063 * must commit the current transaction before returning the inode itself.
1064 * In this case, therefore, we will set call_again to true and return.
1065 * The caller should then commit the current transaction, start a new
1066 * transaction, and call xfs_ialloc() again to actually get the inode.
1068 * To ensure that some other process does not grab the inode that
1069 * was allocated during the first call to xfs_ialloc(), this routine
1070 * also returns the [locked] bp pointing to the head of the freelist
1071 * as ialloc_context. The caller should hold this buffer across
1072 * the commit and pass it back into this routine on the second call.
1084 xfs_buf_t **ialloc_context,
1085 boolean_t *call_again,
1095 * Call the space management code to pick
1096 * the on-disk inode to be allocated.
1098 error = xfs_dialloc(tp, pip->i_ino, mode, okalloc,
1099 ialloc_context, call_again, &ino);
1103 if (*call_again || ino == NULLFSINO) {
1107 ASSERT(*ialloc_context == NULL);
1110 * Get the in-core inode with the lock held exclusively.
1111 * This is because we're setting fields here we need
1112 * to prevent others from looking at until we're done.
1114 error = xfs_trans_iget(tp->t_mountp, tp, ino,
1115 IGET_CREATE, XFS_ILOCK_EXCL, &ip);
1122 ip->i_d.di_mode = (__uint16_t)mode;
1123 ip->i_d.di_onlink = 0;
1124 ip->i_d.di_nlink = nlink;
1125 ASSERT(ip->i_d.di_nlink == nlink);
1126 ip->i_d.di_uid = current_fsuid(cr);
1127 ip->i_d.di_gid = current_fsgid(cr);
1128 ip->i_d.di_projid = prid;
1129 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1132 * If the superblock version is up to where we support new format
1133 * inodes and this is currently an old format inode, then change
1134 * the inode version number now. This way we only do the conversion
1135 * here rather than here and in the flush/logging code.
1137 if (XFS_SB_VERSION_HASNLINK(&tp->t_mountp->m_sb) &&
1138 ip->i_d.di_version == XFS_DINODE_VERSION_1) {
1139 ip->i_d.di_version = XFS_DINODE_VERSION_2;
1141 * We've already zeroed the old link count, the projid field,
1142 * and the pad field.
1147 * Project ids won't be stored on disk if we are using a version 1 inode.
1149 if ( (prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1))
1150 xfs_bump_ino_vers2(tp, ip);
1152 if (XFS_INHERIT_GID(pip, vp->v_vfsp)) {
1153 ip->i_d.di_gid = pip->i_d.di_gid;
1154 if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1155 ip->i_d.di_mode |= S_ISGID;
1160 * If the group ID of the new file does not match the effective group
1161 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1162 * (and only if the irix_sgid_inherit compatibility variable is set).
1164 if ((irix_sgid_inherit) &&
1165 (ip->i_d.di_mode & S_ISGID) &&
1166 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1167 ip->i_d.di_mode &= ~S_ISGID;
1170 ip->i_d.di_size = 0;
1171 ip->i_d.di_nextents = 0;
1172 ASSERT(ip->i_d.di_nblocks == 0);
1173 xfs_ichgtime(ip, XFS_ICHGTIME_CHG|XFS_ICHGTIME_ACC|XFS_ICHGTIME_MOD);
1175 * di_gen will have been taken care of in xfs_iread.
1177 ip->i_d.di_extsize = 0;
1178 ip->i_d.di_dmevmask = 0;
1179 ip->i_d.di_dmstate = 0;
1180 ip->i_d.di_flags = 0;
1181 flags = XFS_ILOG_CORE;
1182 switch (mode & S_IFMT) {
1187 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1188 ip->i_df.if_u2.if_rdev = rdev;
1189 ip->i_df.if_flags = 0;
1190 flags |= XFS_ILOG_DEV;
1194 if (unlikely(pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
1197 if ((mode & S_IFMT) == S_IFDIR) {
1198 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1199 di_flags |= XFS_DIFLAG_RTINHERIT;
1200 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1201 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1202 ip->i_d.di_extsize = pip->i_d.di_extsize;
1204 } else if ((mode & S_IFMT) == S_IFREG) {
1205 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) {
1206 di_flags |= XFS_DIFLAG_REALTIME;
1207 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
1209 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1210 di_flags |= XFS_DIFLAG_EXTSIZE;
1211 ip->i_d.di_extsize = pip->i_d.di_extsize;
1214 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1215 xfs_inherit_noatime)
1216 di_flags |= XFS_DIFLAG_NOATIME;
1217 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1219 di_flags |= XFS_DIFLAG_NODUMP;
1220 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1222 di_flags |= XFS_DIFLAG_SYNC;
1223 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1224 xfs_inherit_nosymlinks)
1225 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1226 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1227 di_flags |= XFS_DIFLAG_PROJINHERIT;
1228 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1229 xfs_inherit_nodefrag)
1230 di_flags |= XFS_DIFLAG_NODEFRAG;
1231 ip->i_d.di_flags |= di_flags;
1235 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1236 ip->i_df.if_flags = XFS_IFEXTENTS;
1237 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1238 ip->i_df.if_u1.if_extents = NULL;
1244 * Attribute fork settings for new inode.
1246 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1247 ip->i_d.di_anextents = 0;
1250 * Log the new values stuffed into the inode.
1252 xfs_trans_log_inode(tp, ip, flags);
1254 /* now that we have an i_mode we can setup inode ops and unlock */
1255 bhv_vfs_init_vnode(XFS_MTOVFS(tp->t_mountp), vp, XFS_ITOBHV(ip), 1);
1262 * Check to make sure that there are no blocks allocated to the
1263 * file beyond the size of the file. We don't check this for
1264 * files with fixed size extents or real time extents, but we
1265 * at least do it for regular files.
1274 xfs_fileoff_t map_first;
1276 xfs_bmbt_irec_t imaps[2];
1278 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1281 if (ip->i_d.di_flags & (XFS_DIFLAG_REALTIME | XFS_DIFLAG_EXTSIZE))
1285 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1287 * The filesystem could be shutting down, so bmapi may return
1290 if (xfs_bmapi(NULL, ip, map_first,
1292 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1294 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1297 ASSERT(nimaps == 1);
1298 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1303 * Calculate the last possible buffered byte in a file. This must
1304 * include data that was buffered beyond the EOF by the write code.
1305 * This also needs to deal with overflowing the xfs_fsize_t type
1306 * which can happen for sizes near the limit.
1308 * We also need to take into account any blocks beyond the EOF. It
1309 * may be the case that they were buffered by a write which failed.
1310 * In that case the pages will still be in memory, but the inode size
1311 * will never have been updated.
1318 xfs_fsize_t last_byte;
1319 xfs_fileoff_t last_block;
1320 xfs_fileoff_t size_last_block;
1323 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE | MR_ACCESS));
1327 * Only check for blocks beyond the EOF if the extents have
1328 * been read in. This eliminates the need for the inode lock,
1329 * and it also saves us from looking when it really isn't
1332 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1333 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1341 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_d.di_size);
1342 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1344 last_byte = XFS_FSB_TO_B(mp, last_block);
1345 if (last_byte < 0) {
1346 return XFS_MAXIOFFSET(mp);
1348 last_byte += (1 << mp->m_writeio_log);
1349 if (last_byte < 0) {
1350 return XFS_MAXIOFFSET(mp);
1355 #if defined(XFS_RW_TRACE)
1361 xfs_fsize_t new_size,
1362 xfs_off_t toss_start,
1363 xfs_off_t toss_finish)
1365 if (ip->i_rwtrace == NULL) {
1369 ktrace_enter(ip->i_rwtrace,
1372 (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
1373 (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
1374 (void*)((long)flag),
1375 (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
1376 (void*)(unsigned long)(new_size & 0xffffffff),
1377 (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
1378 (void*)(unsigned long)(toss_start & 0xffffffff),
1379 (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
1380 (void*)(unsigned long)(toss_finish & 0xffffffff),
1381 (void*)(unsigned long)current_cpu(),
1382 (void*)(unsigned long)current_pid(),
1388 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1392 * Start the truncation of the file to new_size. The new size
1393 * must be smaller than the current size. This routine will
1394 * clear the buffer and page caches of file data in the removed
1395 * range, and xfs_itruncate_finish() will remove the underlying
1398 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1399 * must NOT have the inode lock held at all. This is because we're
1400 * calling into the buffer/page cache code and we can't hold the
1401 * inode lock when we do so.
1403 * We need to wait for any direct I/Os in flight to complete before we
1404 * proceed with the truncate. This is needed to prevent the extents
1405 * being read or written by the direct I/Os from being removed while the
1406 * I/O is in flight as there is no other method of synchronising
1407 * direct I/O with the truncate operation. Also, because we hold
1408 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1409 * started until the truncate completes and drops the lock. Essentially,
1410 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1411 * between direct I/Os and the truncate operation.
1413 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1414 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1415 * in the case that the caller is locking things out of order and
1416 * may not be able to call xfs_itruncate_finish() with the inode lock
1417 * held without dropping the I/O lock. If the caller must drop the
1418 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1419 * must be called again with all the same restrictions as the initial
1423 xfs_itruncate_start(
1426 xfs_fsize_t new_size)
1428 xfs_fsize_t last_byte;
1429 xfs_off_t toss_start;
1433 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1434 ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
1435 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1436 (flags == XFS_ITRUNC_MAYBE));
1441 vn_iowait(vp); /* wait for the completion of any pending DIOs */
1444 * Call toss_pages or flushinval_pages to get rid of pages
1445 * overlapping the region being removed. We have to use
1446 * the less efficient flushinval_pages in the case that the
1447 * caller may not be able to finish the truncate without
1448 * dropping the inode's I/O lock. Make sure
1449 * to catch any pages brought in by buffers overlapping
1450 * the EOF by searching out beyond the isize by our
1451 * block size. We round new_size up to a block boundary
1452 * so that we don't toss things on the same block as
1453 * new_size but before it.
1455 * Before calling toss_page or flushinval_pages, make sure to
1456 * call remapf() over the same region if the file is mapped.
1457 * This frees up mapped file references to the pages in the
1458 * given range and for the flushinval_pages case it ensures
1459 * that we get the latest mapped changes flushed out.
1461 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1462 toss_start = XFS_FSB_TO_B(mp, toss_start);
1463 if (toss_start < 0) {
1465 * The place to start tossing is beyond our maximum
1466 * file size, so there is no way that the data extended
1471 last_byte = xfs_file_last_byte(ip);
1472 xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
1474 if (last_byte > toss_start) {
1475 if (flags & XFS_ITRUNC_DEFINITE) {
1476 bhv_vop_toss_pages(vp, toss_start, -1, FI_REMAPF_LOCKED);
1478 bhv_vop_flushinval_pages(vp, toss_start, -1, FI_REMAPF_LOCKED);
1483 if (new_size == 0) {
1484 ASSERT(VN_CACHED(vp) == 0);
1490 * Shrink the file to the given new_size. The new
1491 * size must be smaller than the current size.
1492 * This will free up the underlying blocks
1493 * in the removed range after a call to xfs_itruncate_start()
1494 * or xfs_atruncate_start().
1496 * The transaction passed to this routine must have made
1497 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1498 * This routine may commit the given transaction and
1499 * start new ones, so make sure everything involved in
1500 * the transaction is tidy before calling here.
1501 * Some transaction will be returned to the caller to be
1502 * committed. The incoming transaction must already include
1503 * the inode, and both inode locks must be held exclusively.
1504 * The inode must also be "held" within the transaction. On
1505 * return the inode will be "held" within the returned transaction.
1506 * This routine does NOT require any disk space to be reserved
1507 * for it within the transaction.
1509 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1510 * and it indicates the fork which is to be truncated. For the
1511 * attribute fork we only support truncation to size 0.
1513 * We use the sync parameter to indicate whether or not the first
1514 * transaction we perform might have to be synchronous. For the attr fork,
1515 * it needs to be so if the unlink of the inode is not yet known to be
1516 * permanent in the log. This keeps us from freeing and reusing the
1517 * blocks of the attribute fork before the unlink of the inode becomes
1520 * For the data fork, we normally have to run synchronously if we're
1521 * being called out of the inactive path or we're being called
1522 * out of the create path where we're truncating an existing file.
1523 * Either way, the truncate needs to be sync so blocks don't reappear
1524 * in the file with altered data in case of a crash. wsync filesystems
1525 * can run the first case async because anything that shrinks the inode
1526 * has to run sync so by the time we're called here from inactive, the
1527 * inode size is permanently set to 0.
1529 * Calls from the truncate path always need to be sync unless we're
1530 * in a wsync filesystem and the file has already been unlinked.
1532 * The caller is responsible for correctly setting the sync parameter.
1533 * It gets too hard for us to guess here which path we're being called
1534 * out of just based on inode state.
1537 xfs_itruncate_finish(
1540 xfs_fsize_t new_size,
1544 xfs_fsblock_t first_block;
1545 xfs_fileoff_t first_unmap_block;
1546 xfs_fileoff_t last_block;
1547 xfs_filblks_t unmap_len=0;
1552 xfs_bmap_free_t free_list;
1555 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1556 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
1557 ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
1558 ASSERT(*tp != NULL);
1559 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1560 ASSERT(ip->i_transp == *tp);
1561 ASSERT(ip->i_itemp != NULL);
1562 ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
1566 mp = (ntp)->t_mountp;
1567 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1570 * We only support truncating the entire attribute fork.
1572 if (fork == XFS_ATTR_FORK) {
1575 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1576 xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
1578 * The first thing we do is set the size to new_size permanently
1579 * on disk. This way we don't have to worry about anyone ever
1580 * being able to look at the data being freed even in the face
1581 * of a crash. What we're getting around here is the case where
1582 * we free a block, it is allocated to another file, it is written
1583 * to, and then we crash. If the new data gets written to the
1584 * file but the log buffers containing the free and reallocation
1585 * don't, then we'd end up with garbage in the blocks being freed.
1586 * As long as we make the new_size permanent before actually
1587 * freeing any blocks it doesn't matter if they get writtten to.
1589 * The callers must signal into us whether or not the size
1590 * setting here must be synchronous. There are a few cases
1591 * where it doesn't have to be synchronous. Those cases
1592 * occur if the file is unlinked and we know the unlink is
1593 * permanent or if the blocks being truncated are guaranteed
1594 * to be beyond the inode eof (regardless of the link count)
1595 * and the eof value is permanent. Both of these cases occur
1596 * only on wsync-mounted filesystems. In those cases, we're
1597 * guaranteed that no user will ever see the data in the blocks
1598 * that are being truncated so the truncate can run async.
1599 * In the free beyond eof case, the file may wind up with
1600 * more blocks allocated to it than it needs if we crash
1601 * and that won't get fixed until the next time the file
1602 * is re-opened and closed but that's ok as that shouldn't
1603 * be too many blocks.
1605 * However, we can't just make all wsync xactions run async
1606 * because there's one call out of the create path that needs
1607 * to run sync where it's truncating an existing file to size
1608 * 0 whose size is > 0.
1610 * It's probably possible to come up with a test in this
1611 * routine that would correctly distinguish all the above
1612 * cases from the values of the function parameters and the
1613 * inode state but for sanity's sake, I've decided to let the
1614 * layers above just tell us. It's simpler to correctly figure
1615 * out in the layer above exactly under what conditions we
1616 * can run async and I think it's easier for others read and
1617 * follow the logic in case something has to be changed.
1618 * cscope is your friend -- rcc.
1620 * The attribute fork is much simpler.
1622 * For the attribute fork we allow the caller to tell us whether
1623 * the unlink of the inode that led to this call is yet permanent
1624 * in the on disk log. If it is not and we will be freeing extents
1625 * in this inode then we make the first transaction synchronous
1626 * to make sure that the unlink is permanent by the time we free
1629 if (fork == XFS_DATA_FORK) {
1630 if (ip->i_d.di_nextents > 0) {
1631 ip->i_d.di_size = new_size;
1632 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1635 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1636 if (ip->i_d.di_anextents > 0)
1637 xfs_trans_set_sync(ntp);
1639 ASSERT(fork == XFS_DATA_FORK ||
1640 (fork == XFS_ATTR_FORK &&
1641 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1642 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1645 * Since it is possible for space to become allocated beyond
1646 * the end of the file (in a crash where the space is allocated
1647 * but the inode size is not yet updated), simply remove any
1648 * blocks which show up between the new EOF and the maximum
1649 * possible file size. If the first block to be removed is
1650 * beyond the maximum file size (ie it is the same as last_block),
1651 * then there is nothing to do.
1653 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1654 ASSERT(first_unmap_block <= last_block);
1656 if (last_block == first_unmap_block) {
1659 unmap_len = last_block - first_unmap_block + 1;
1663 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1664 * will tell us whether it freed the entire range or
1665 * not. If this is a synchronous mount (wsync),
1666 * then we can tell bunmapi to keep all the
1667 * transactions asynchronous since the unlink
1668 * transaction that made this inode inactive has
1669 * already hit the disk. There's no danger of
1670 * the freed blocks being reused, there being a
1671 * crash, and the reused blocks suddenly reappearing
1672 * in this file with garbage in them once recovery
1675 XFS_BMAP_INIT(&free_list, &first_block);
1676 error = XFS_BUNMAPI(mp, ntp, &ip->i_iocore,
1677 first_unmap_block, unmap_len,
1678 XFS_BMAPI_AFLAG(fork) |
1679 (sync ? 0 : XFS_BMAPI_ASYNC),
1680 XFS_ITRUNC_MAX_EXTENTS,
1681 &first_block, &free_list,
1685 * If the bunmapi call encounters an error,
1686 * return to the caller where the transaction
1687 * can be properly aborted. We just need to
1688 * make sure we're not holding any resources
1689 * that we were not when we came in.
1691 xfs_bmap_cancel(&free_list);
1696 * Duplicate the transaction that has the permanent
1697 * reservation and commit the old transaction.
1699 error = xfs_bmap_finish(tp, &free_list, first_block,
1704 * If the bmap finish call encounters an error,
1705 * return to the caller where the transaction
1706 * can be properly aborted. We just need to
1707 * make sure we're not holding any resources
1708 * that we were not when we came in.
1710 * Aborting from this point might lose some
1711 * blocks in the file system, but oh well.
1713 xfs_bmap_cancel(&free_list);
1716 * If the passed in transaction committed
1717 * in xfs_bmap_finish(), then we want to
1718 * add the inode to this one before returning.
1719 * This keeps things simple for the higher
1720 * level code, because it always knows that
1721 * the inode is locked and held in the
1722 * transaction that returns to it whether
1723 * errors occur or not. We don't mark the
1724 * inode dirty so that this transaction can
1725 * be easily aborted if possible.
1727 xfs_trans_ijoin(ntp, ip,
1728 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1729 xfs_trans_ihold(ntp, ip);
1736 * The first xact was committed,
1737 * so add the inode to the new one.
1738 * Mark it dirty so it will be logged
1739 * and moved forward in the log as
1740 * part of every commit.
1742 xfs_trans_ijoin(ntp, ip,
1743 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1744 xfs_trans_ihold(ntp, ip);
1745 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1747 ntp = xfs_trans_dup(ntp);
1748 (void) xfs_trans_commit(*tp, 0, NULL);
1750 error = xfs_trans_reserve(ntp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1751 XFS_TRANS_PERM_LOG_RES,
1752 XFS_ITRUNCATE_LOG_COUNT);
1754 * Add the inode being truncated to the next chained
1757 xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1758 xfs_trans_ihold(ntp, ip);
1763 * Only update the size in the case of the data fork, but
1764 * always re-log the inode so that our permanent transaction
1765 * can keep on rolling it forward in the log.
1767 if (fork == XFS_DATA_FORK) {
1768 xfs_isize_check(mp, ip, new_size);
1769 ip->i_d.di_size = new_size;
1771 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1772 ASSERT((new_size != 0) ||
1773 (fork == XFS_ATTR_FORK) ||
1774 (ip->i_delayed_blks == 0));
1775 ASSERT((new_size != 0) ||
1776 (fork == XFS_ATTR_FORK) ||
1777 (ip->i_d.di_nextents == 0));
1778 xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
1786 * Do the first part of growing a file: zero any data in the last
1787 * block that is beyond the old EOF. We need to do this before
1788 * the inode is joined to the transaction to modify the i_size.
1789 * That way we can drop the inode lock and call into the buffer
1790 * cache to get the buffer mapping the EOF.
1795 xfs_fsize_t new_size,
1800 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1801 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1802 ASSERT(new_size > ip->i_d.di_size);
1805 * Zero any pages that may have been created by
1806 * xfs_write_file() beyond the end of the file
1807 * and any blocks between the old and new file sizes.
1809 error = xfs_zero_eof(XFS_ITOV(ip), &ip->i_iocore, new_size,
1810 ip->i_d.di_size, new_size);
1817 * This routine is called to extend the size of a file.
1818 * The inode must have both the iolock and the ilock locked
1819 * for update and it must be a part of the current transaction.
1820 * The xfs_igrow_start() function must have been called previously.
1821 * If the change_flag is not zero, the inode change timestamp will
1828 xfs_fsize_t new_size,
1831 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1832 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1833 ASSERT(ip->i_transp == tp);
1834 ASSERT(new_size > ip->i_d.di_size);
1837 * Update the file size. Update the inode change timestamp
1838 * if change_flag set.
1840 ip->i_d.di_size = new_size;
1842 xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
1843 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1849 * This is called when the inode's link count goes to 0.
1850 * We place the on-disk inode on a list in the AGI. It
1851 * will be pulled from this list when the inode is freed.
1863 xfs_agnumber_t agno;
1864 xfs_daddr_t agdaddr;
1871 ASSERT(ip->i_d.di_nlink == 0);
1872 ASSERT(ip->i_d.di_mode != 0);
1873 ASSERT(ip->i_transp == tp);
1877 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1878 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1881 * Get the agi buffer first. It ensures lock ordering
1884 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1885 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1890 * Validate the magic number of the agi block.
1892 agi = XFS_BUF_TO_AGI(agibp);
1894 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1895 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
1896 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK,
1897 XFS_RANDOM_IUNLINK))) {
1898 XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi);
1899 xfs_trans_brelse(tp, agibp);
1900 return XFS_ERROR(EFSCORRUPTED);
1903 * Get the index into the agi hash table for the
1904 * list this inode will go on.
1906 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1908 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1909 ASSERT(agi->agi_unlinked[bucket_index]);
1910 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1912 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
1914 * There is already another inode in the bucket we need
1915 * to add ourselves to. Add us at the front of the list.
1916 * Here we put the head pointer into our next pointer,
1917 * and then we fall through to point the head at us.
1919 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
1923 ASSERT(INT_GET(dip->di_next_unlinked, ARCH_CONVERT) == NULLAGINO);
1924 ASSERT(dip->di_next_unlinked);
1925 /* both on-disk, don't endian flip twice */
1926 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1927 offset = ip->i_boffset +
1928 offsetof(xfs_dinode_t, di_next_unlinked);
1929 xfs_trans_inode_buf(tp, ibp);
1930 xfs_trans_log_buf(tp, ibp, offset,
1931 (offset + sizeof(xfs_agino_t) - 1));
1932 xfs_inobp_check(mp, ibp);
1936 * Point the bucket head pointer at the inode being inserted.
1939 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1940 offset = offsetof(xfs_agi_t, agi_unlinked) +
1941 (sizeof(xfs_agino_t) * bucket_index);
1942 xfs_trans_log_buf(tp, agibp, offset,
1943 (offset + sizeof(xfs_agino_t) - 1));
1948 * Pull the on-disk inode from the AGI unlinked list.
1961 xfs_agnumber_t agno;
1962 xfs_daddr_t agdaddr;
1964 xfs_agino_t next_agino;
1965 xfs_buf_t *last_ibp;
1966 xfs_dinode_t *last_dip;
1968 int offset, last_offset;
1973 * First pull the on-disk inode from the AGI unlinked list.
1977 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1978 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1981 * Get the agi buffer first. It ensures lock ordering
1984 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1985 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1988 "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
1989 error, mp->m_fsname);
1993 * Validate the magic number of the agi block.
1995 agi = XFS_BUF_TO_AGI(agibp);
1997 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1998 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
1999 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE,
2000 XFS_RANDOM_IUNLINK_REMOVE))) {
2001 XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW,
2003 xfs_trans_brelse(tp, agibp);
2005 "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
2007 return XFS_ERROR(EFSCORRUPTED);
2010 * Get the index into the agi hash table for the
2011 * list this inode will go on.
2013 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2015 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2016 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
2017 ASSERT(agi->agi_unlinked[bucket_index]);
2019 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2021 * We're at the head of the list. Get the inode's
2022 * on-disk buffer to see if there is anyone after us
2023 * on the list. Only modify our next pointer if it
2024 * is not already NULLAGINO. This saves us the overhead
2025 * of dealing with the buffer when there is no need to
2028 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
2031 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2032 error, mp->m_fsname);
2035 next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
2036 ASSERT(next_agino != 0);
2037 if (next_agino != NULLAGINO) {
2038 INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
2039 offset = ip->i_boffset +
2040 offsetof(xfs_dinode_t, di_next_unlinked);
2041 xfs_trans_inode_buf(tp, ibp);
2042 xfs_trans_log_buf(tp, ibp, offset,
2043 (offset + sizeof(xfs_agino_t) - 1));
2044 xfs_inobp_check(mp, ibp);
2046 xfs_trans_brelse(tp, ibp);
2049 * Point the bucket head pointer at the next inode.
2051 ASSERT(next_agino != 0);
2052 ASSERT(next_agino != agino);
2053 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2054 offset = offsetof(xfs_agi_t, agi_unlinked) +
2055 (sizeof(xfs_agino_t) * bucket_index);
2056 xfs_trans_log_buf(tp, agibp, offset,
2057 (offset + sizeof(xfs_agino_t) - 1));
2060 * We need to search the list for the inode being freed.
2062 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2064 while (next_agino != agino) {
2066 * If the last inode wasn't the one pointing to
2067 * us, then release its buffer since we're not
2068 * going to do anything with it.
2070 if (last_ibp != NULL) {
2071 xfs_trans_brelse(tp, last_ibp);
2073 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2074 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
2075 &last_ibp, &last_offset);
2078 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
2079 error, mp->m_fsname);
2082 next_agino = INT_GET(last_dip->di_next_unlinked, ARCH_CONVERT);
2083 ASSERT(next_agino != NULLAGINO);
2084 ASSERT(next_agino != 0);
2087 * Now last_ibp points to the buffer previous to us on
2088 * the unlinked list. Pull us from the list.
2090 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
2093 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2094 error, mp->m_fsname);
2097 next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
2098 ASSERT(next_agino != 0);
2099 ASSERT(next_agino != agino);
2100 if (next_agino != NULLAGINO) {
2101 INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
2102 offset = ip->i_boffset +
2103 offsetof(xfs_dinode_t, di_next_unlinked);
2104 xfs_trans_inode_buf(tp, ibp);
2105 xfs_trans_log_buf(tp, ibp, offset,
2106 (offset + sizeof(xfs_agino_t) - 1));
2107 xfs_inobp_check(mp, ibp);
2109 xfs_trans_brelse(tp, ibp);
2112 * Point the previous inode on the list to the next inode.
2114 INT_SET(last_dip->di_next_unlinked, ARCH_CONVERT, next_agino);
2115 ASSERT(next_agino != 0);
2116 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2117 xfs_trans_inode_buf(tp, last_ibp);
2118 xfs_trans_log_buf(tp, last_ibp, offset,
2119 (offset + sizeof(xfs_agino_t) - 1));
2120 xfs_inobp_check(mp, last_ibp);
2125 static __inline__ int xfs_inode_clean(xfs_inode_t *ip)
2127 return (((ip->i_itemp == NULL) ||
2128 !(ip->i_itemp->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
2129 (ip->i_update_core == 0));
2134 xfs_inode_t *free_ip,
2138 xfs_mount_t *mp = free_ip->i_mount;
2139 int blks_per_cluster;
2142 int i, j, found, pre_flushed;
2146 xfs_inode_t *ip, **ip_found;
2147 xfs_inode_log_item_t *iip;
2148 xfs_log_item_t *lip;
2151 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
2152 blks_per_cluster = 1;
2153 ninodes = mp->m_sb.sb_inopblock;
2154 nbufs = XFS_IALLOC_BLOCKS(mp);
2156 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
2157 mp->m_sb.sb_blocksize;
2158 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
2159 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
2162 ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
2164 for (j = 0; j < nbufs; j++, inum += ninodes) {
2165 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2166 XFS_INO_TO_AGBNO(mp, inum));
2170 * Look for each inode in memory and attempt to lock it,
2171 * we can be racing with flush and tail pushing here.
2172 * any inode we get the locks on, add to an array of
2173 * inode items to process later.
2175 * The get the buffer lock, we could beat a flush
2176 * or tail pushing thread to the lock here, in which
2177 * case they will go looking for the inode buffer
2178 * and fail, we need some other form of interlock
2182 for (i = 0; i < ninodes; i++) {
2183 ih = XFS_IHASH(mp, inum + i);
2184 read_lock(&ih->ih_lock);
2185 for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
2186 if (ip->i_ino == inum + i)
2190 /* Inode not in memory or we found it already,
2193 if (!ip || (ip->i_flags & XFS_ISTALE)) {
2194 read_unlock(&ih->ih_lock);
2198 if (xfs_inode_clean(ip)) {
2199 read_unlock(&ih->ih_lock);
2203 /* If we can get the locks then add it to the
2204 * list, otherwise by the time we get the bp lock
2205 * below it will already be attached to the
2209 /* This inode will already be locked - by us, lets
2213 if (ip == free_ip) {
2214 if (xfs_iflock_nowait(ip)) {
2215 ip->i_flags |= XFS_ISTALE;
2217 if (xfs_inode_clean(ip)) {
2220 ip_found[found++] = ip;
2223 read_unlock(&ih->ih_lock);
2227 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2228 if (xfs_iflock_nowait(ip)) {
2229 ip->i_flags |= XFS_ISTALE;
2231 if (xfs_inode_clean(ip)) {
2233 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2235 ip_found[found++] = ip;
2238 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2242 read_unlock(&ih->ih_lock);
2245 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2246 mp->m_bsize * blks_per_cluster,
2250 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
2252 if (lip->li_type == XFS_LI_INODE) {
2253 iip = (xfs_inode_log_item_t *)lip;
2254 ASSERT(iip->ili_logged == 1);
2255 lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
2257 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2259 iip->ili_inode->i_flags |= XFS_ISTALE;
2262 lip = lip->li_bio_list;
2265 for (i = 0; i < found; i++) {
2270 ip->i_update_core = 0;
2272 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2276 iip->ili_last_fields = iip->ili_format.ilf_fields;
2277 iip->ili_format.ilf_fields = 0;
2278 iip->ili_logged = 1;
2280 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2283 xfs_buf_attach_iodone(bp,
2284 (void(*)(xfs_buf_t*,xfs_log_item_t*))
2285 xfs_istale_done, (xfs_log_item_t *)iip);
2286 if (ip != free_ip) {
2287 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2291 if (found || pre_flushed)
2292 xfs_trans_stale_inode_buf(tp, bp);
2293 xfs_trans_binval(tp, bp);
2296 kmem_free(ip_found, ninodes * sizeof(xfs_inode_t *));
2300 * This is called to return an inode to the inode free list.
2301 * The inode should already be truncated to 0 length and have
2302 * no pages associated with it. This routine also assumes that
2303 * the inode is already a part of the transaction.
2305 * The on-disk copy of the inode will have been added to the list
2306 * of unlinked inodes in the AGI. We need to remove the inode from
2307 * that list atomically with respect to freeing it here.
2313 xfs_bmap_free_t *flist)
2317 xfs_ino_t first_ino;
2319 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2320 ASSERT(ip->i_transp == tp);
2321 ASSERT(ip->i_d.di_nlink == 0);
2322 ASSERT(ip->i_d.di_nextents == 0);
2323 ASSERT(ip->i_d.di_anextents == 0);
2324 ASSERT((ip->i_d.di_size == 0) ||
2325 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2326 ASSERT(ip->i_d.di_nblocks == 0);
2329 * Pull the on-disk inode from the AGI unlinked list.
2331 error = xfs_iunlink_remove(tp, ip);
2336 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2340 ip->i_d.di_mode = 0; /* mark incore inode as free */
2341 ip->i_d.di_flags = 0;
2342 ip->i_d.di_dmevmask = 0;
2343 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2344 ip->i_df.if_ext_max =
2345 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2346 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2347 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2349 * Bump the generation count so no one will be confused
2350 * by reincarnations of this inode.
2353 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2356 xfs_ifree_cluster(ip, tp, first_ino);
2363 * Reallocate the space for if_broot based on the number of records
2364 * being added or deleted as indicated in rec_diff. Move the records
2365 * and pointers in if_broot to fit the new size. When shrinking this
2366 * will eliminate holes between the records and pointers created by
2367 * the caller. When growing this will create holes to be filled in
2370 * The caller must not request to add more records than would fit in
2371 * the on-disk inode root. If the if_broot is currently NULL, then
2372 * if we adding records one will be allocated. The caller must also
2373 * not request that the number of records go below zero, although
2374 * it can go to zero.
2376 * ip -- the inode whose if_broot area is changing
2377 * ext_diff -- the change in the number of records, positive or negative,
2378 * requested for the if_broot array.
2388 xfs_bmbt_block_t *new_broot;
2395 * Handle the degenerate case quietly.
2397 if (rec_diff == 0) {
2401 ifp = XFS_IFORK_PTR(ip, whichfork);
2404 * If there wasn't any memory allocated before, just
2405 * allocate it now and get out.
2407 if (ifp->if_broot_bytes == 0) {
2408 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
2409 ifp->if_broot = (xfs_bmbt_block_t*)kmem_alloc(new_size,
2411 ifp->if_broot_bytes = (int)new_size;
2416 * If there is already an existing if_broot, then we need
2417 * to realloc() it and shift the pointers to their new
2418 * location. The records don't change location because
2419 * they are kept butted up against the btree block header.
2421 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2422 new_max = cur_max + rec_diff;
2423 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2424 ifp->if_broot = (xfs_bmbt_block_t *)
2425 kmem_realloc(ifp->if_broot,
2427 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2429 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2430 ifp->if_broot_bytes);
2431 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2433 ifp->if_broot_bytes = (int)new_size;
2434 ASSERT(ifp->if_broot_bytes <=
2435 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2436 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2441 * rec_diff is less than 0. In this case, we are shrinking the
2442 * if_broot buffer. It must already exist. If we go to zero
2443 * records, just get rid of the root and clear the status bit.
2445 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
2446 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2447 new_max = cur_max + rec_diff;
2448 ASSERT(new_max >= 0);
2450 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2454 new_broot = (xfs_bmbt_block_t *)kmem_alloc(new_size, KM_SLEEP);
2456 * First copy over the btree block header.
2458 memcpy(new_broot, ifp->if_broot, sizeof(xfs_bmbt_block_t));
2461 ifp->if_flags &= ~XFS_IFBROOT;
2465 * Only copy the records and pointers if there are any.
2469 * First copy the records.
2471 op = (char *)XFS_BMAP_BROOT_REC_ADDR(ifp->if_broot, 1,
2472 ifp->if_broot_bytes);
2473 np = (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot, 1,
2475 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2478 * Then copy the pointers.
2480 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2481 ifp->if_broot_bytes);
2482 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot, 1,
2484 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2486 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2487 ifp->if_broot = new_broot;
2488 ifp->if_broot_bytes = (int)new_size;
2489 ASSERT(ifp->if_broot_bytes <=
2490 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2496 * This is called when the amount of space needed for if_data
2497 * is increased or decreased. The change in size is indicated by
2498 * the number of bytes that need to be added or deleted in the
2499 * byte_diff parameter.
2501 * If the amount of space needed has decreased below the size of the
2502 * inline buffer, then switch to using the inline buffer. Otherwise,
2503 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2504 * to what is needed.
2506 * ip -- the inode whose if_data area is changing
2507 * byte_diff -- the change in the number of bytes, positive or negative,
2508 * requested for the if_data array.
2520 if (byte_diff == 0) {
2524 ifp = XFS_IFORK_PTR(ip, whichfork);
2525 new_size = (int)ifp->if_bytes + byte_diff;
2526 ASSERT(new_size >= 0);
2528 if (new_size == 0) {
2529 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2530 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2532 ifp->if_u1.if_data = NULL;
2534 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2536 * If the valid extents/data can fit in if_inline_ext/data,
2537 * copy them from the malloc'd vector and free it.
2539 if (ifp->if_u1.if_data == NULL) {
2540 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2541 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2542 ASSERT(ifp->if_real_bytes != 0);
2543 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2545 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2546 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2551 * Stuck with malloc/realloc.
2552 * For inline data, the underlying buffer must be
2553 * a multiple of 4 bytes in size so that it can be
2554 * logged and stay on word boundaries. We enforce
2557 real_size = roundup(new_size, 4);
2558 if (ifp->if_u1.if_data == NULL) {
2559 ASSERT(ifp->if_real_bytes == 0);
2560 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2561 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2563 * Only do the realloc if the underlying size
2564 * is really changing.
2566 if (ifp->if_real_bytes != real_size) {
2567 ifp->if_u1.if_data =
2568 kmem_realloc(ifp->if_u1.if_data,
2574 ASSERT(ifp->if_real_bytes == 0);
2575 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2576 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2580 ifp->if_real_bytes = real_size;
2581 ifp->if_bytes = new_size;
2582 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2589 * Map inode to disk block and offset.
2591 * mp -- the mount point structure for the current file system
2592 * tp -- the current transaction
2593 * ino -- the inode number of the inode to be located
2594 * imap -- this structure is filled in with the information necessary
2595 * to retrieve the given inode from disk
2596 * flags -- flags to pass to xfs_dilocate indicating whether or not
2597 * lookups in the inode btree were OK or not
2607 xfs_fsblock_t fsbno;
2612 fsbno = imap->im_blkno ?
2613 XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK;
2614 error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags);
2618 imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno);
2619 imap->im_len = XFS_FSB_TO_BB(mp, len);
2620 imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno);
2621 imap->im_ioffset = (ushort)off;
2622 imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog);
2633 ifp = XFS_IFORK_PTR(ip, whichfork);
2634 if (ifp->if_broot != NULL) {
2635 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2636 ifp->if_broot = NULL;
2640 * If the format is local, then we can't have an extents
2641 * array so just look for an inline data array. If we're
2642 * not local then we may or may not have an extents list,
2643 * so check and free it up if we do.
2645 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2646 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2647 (ifp->if_u1.if_data != NULL)) {
2648 ASSERT(ifp->if_real_bytes != 0);
2649 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2650 ifp->if_u1.if_data = NULL;
2651 ifp->if_real_bytes = 0;
2653 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2654 ((ifp->if_flags & XFS_IFEXTIREC) ||
2655 ((ifp->if_u1.if_extents != NULL) &&
2656 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
2657 ASSERT(ifp->if_real_bytes != 0);
2658 xfs_iext_destroy(ifp);
2660 ASSERT(ifp->if_u1.if_extents == NULL ||
2661 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2662 ASSERT(ifp->if_real_bytes == 0);
2663 if (whichfork == XFS_ATTR_FORK) {
2664 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2670 * This is called free all the memory associated with an inode.
2671 * It must free the inode itself and any buffers allocated for
2672 * if_extents/if_data and if_broot. It must also free the lock
2673 * associated with the inode.
2680 switch (ip->i_d.di_mode & S_IFMT) {
2684 xfs_idestroy_fork(ip, XFS_DATA_FORK);
2688 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
2689 mrfree(&ip->i_lock);
2690 mrfree(&ip->i_iolock);
2691 freesema(&ip->i_flock);
2692 #ifdef XFS_BMAP_TRACE
2693 ktrace_free(ip->i_xtrace);
2695 #ifdef XFS_BMBT_TRACE
2696 ktrace_free(ip->i_btrace);
2699 ktrace_free(ip->i_rwtrace);
2701 #ifdef XFS_ILOCK_TRACE
2702 ktrace_free(ip->i_lock_trace);
2704 #ifdef XFS_DIR2_TRACE
2705 ktrace_free(ip->i_dir_trace);
2708 /* XXXdpd should be able to assert this but shutdown
2709 * is leaving the AIL behind. */
2710 ASSERT(((ip->i_itemp->ili_item.li_flags & XFS_LI_IN_AIL) == 0) ||
2711 XFS_FORCED_SHUTDOWN(ip->i_mount));
2712 xfs_inode_item_destroy(ip);
2714 kmem_zone_free(xfs_inode_zone, ip);
2719 * Increment the pin count of the given buffer.
2720 * This value is protected by ipinlock spinlock in the mount structure.
2726 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2728 atomic_inc(&ip->i_pincount);
2732 * Decrement the pin count of the given inode, and wake up
2733 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2734 * inode must have been previously pinned with a call to xfs_ipin().
2740 ASSERT(atomic_read(&ip->i_pincount) > 0);
2742 if (atomic_dec_and_test(&ip->i_pincount)) {
2744 * If the inode is currently being reclaimed, the
2745 * linux inode _and_ the xfs vnode may have been
2746 * freed so we cannot reference either of them safely.
2747 * Hence we should not try to do anything to them
2748 * if the xfs inode is currently in the reclaim
2751 * However, we still need to issue the unpin wakeup
2752 * call as the inode reclaim may be blocked waiting for
2753 * the inode to become unpinned.
2755 if (!(ip->i_flags & (XFS_IRECLAIM|XFS_IRECLAIMABLE))) {
2756 bhv_vnode_t *vp = XFS_ITOV_NULL(ip);
2758 /* make sync come back and flush this inode */
2760 struct inode *inode = vn_to_inode(vp);
2762 if (!(inode->i_state &
2763 (I_NEW|I_FREEING|I_CLEAR)))
2764 mark_inode_dirty_sync(inode);
2767 wake_up(&ip->i_ipin_wait);
2772 * This is called to wait for the given inode to be unpinned.
2773 * It will sleep until this happens. The caller must have the
2774 * inode locked in at least shared mode so that the buffer cannot
2775 * be subsequently pinned once someone is waiting for it to be
2782 xfs_inode_log_item_t *iip;
2785 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE | MR_ACCESS));
2787 if (atomic_read(&ip->i_pincount) == 0) {
2792 if (iip && iip->ili_last_lsn) {
2793 lsn = iip->ili_last_lsn;
2799 * Give the log a push so we don't wait here too long.
2801 xfs_log_force(ip->i_mount, lsn, XFS_LOG_FORCE);
2803 wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
2808 * xfs_iextents_copy()
2810 * This is called to copy the REAL extents (as opposed to the delayed
2811 * allocation extents) from the inode into the given buffer. It
2812 * returns the number of bytes copied into the buffer.
2814 * If there are no delayed allocation extents, then we can just
2815 * memcpy() the extents into the buffer. Otherwise, we need to
2816 * examine each extent in turn and skip those which are delayed.
2821 xfs_bmbt_rec_t *buffer,
2825 xfs_bmbt_rec_t *dest_ep;
2827 #ifdef XFS_BMAP_TRACE
2828 static char fname[] = "xfs_iextents_copy";
2833 xfs_fsblock_t start_block;
2835 ifp = XFS_IFORK_PTR(ip, whichfork);
2836 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
2837 ASSERT(ifp->if_bytes > 0);
2839 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2840 xfs_bmap_trace_exlist(fname, ip, nrecs, whichfork);
2844 * There are some delayed allocation extents in the
2845 * inode, so copy the extents one at a time and skip
2846 * the delayed ones. There must be at least one
2847 * non-delayed extent.
2851 for (i = 0; i < nrecs; i++) {
2852 ep = xfs_iext_get_ext(ifp, i);
2853 start_block = xfs_bmbt_get_startblock(ep);
2854 if (ISNULLSTARTBLOCK(start_block)) {
2856 * It's a delayed allocation extent, so skip it.
2861 /* Translate to on disk format */
2862 put_unaligned(INT_GET(ep->l0, ARCH_CONVERT),
2863 (__uint64_t*)&dest_ep->l0);
2864 put_unaligned(INT_GET(ep->l1, ARCH_CONVERT),
2865 (__uint64_t*)&dest_ep->l1);
2869 ASSERT(copied != 0);
2870 xfs_validate_extents(ifp, copied, 1, XFS_EXTFMT_INODE(ip));
2872 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2876 * Each of the following cases stores data into the same region
2877 * of the on-disk inode, so only one of them can be valid at
2878 * any given time. While it is possible to have conflicting formats
2879 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2880 * in EXTENTS format, this can only happen when the fork has
2881 * changed formats after being modified but before being flushed.
2882 * In these cases, the format always takes precedence, because the
2883 * format indicates the current state of the fork.
2890 xfs_inode_log_item_t *iip,
2897 #ifdef XFS_TRANS_DEBUG
2900 static const short brootflag[2] =
2901 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2902 static const short dataflag[2] =
2903 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2904 static const short extflag[2] =
2905 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2909 ifp = XFS_IFORK_PTR(ip, whichfork);
2911 * This can happen if we gave up in iformat in an error path,
2912 * for the attribute fork.
2915 ASSERT(whichfork == XFS_ATTR_FORK);
2918 cp = XFS_DFORK_PTR(dip, whichfork);
2920 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2921 case XFS_DINODE_FMT_LOCAL:
2922 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2923 (ifp->if_bytes > 0)) {
2924 ASSERT(ifp->if_u1.if_data != NULL);
2925 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2926 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2928 if (whichfork == XFS_DATA_FORK) {
2929 if (unlikely(XFS_DIR_SHORTFORM_VALIDATE_ONDISK(mp, dip))) {
2930 XFS_ERROR_REPORT("xfs_iflush_fork",
2931 XFS_ERRLEVEL_LOW, mp);
2932 return XFS_ERROR(EFSCORRUPTED);
2937 case XFS_DINODE_FMT_EXTENTS:
2938 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2939 !(iip->ili_format.ilf_fields & extflag[whichfork]));
2940 ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
2941 (ifp->if_bytes == 0));
2942 ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
2943 (ifp->if_bytes > 0));
2944 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
2945 (ifp->if_bytes > 0)) {
2946 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2947 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2952 case XFS_DINODE_FMT_BTREE:
2953 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
2954 (ifp->if_broot_bytes > 0)) {
2955 ASSERT(ifp->if_broot != NULL);
2956 ASSERT(ifp->if_broot_bytes <=
2957 (XFS_IFORK_SIZE(ip, whichfork) +
2958 XFS_BROOT_SIZE_ADJ));
2959 xfs_bmbt_to_bmdr(ifp->if_broot, ifp->if_broot_bytes,
2960 (xfs_bmdr_block_t *)cp,
2961 XFS_DFORK_SIZE(dip, mp, whichfork));
2965 case XFS_DINODE_FMT_DEV:
2966 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
2967 ASSERT(whichfork == XFS_DATA_FORK);
2968 INT_SET(dip->di_u.di_dev, ARCH_CONVERT, ip->i_df.if_u2.if_rdev);
2972 case XFS_DINODE_FMT_UUID:
2973 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
2974 ASSERT(whichfork == XFS_DATA_FORK);
2975 memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid,
2989 * xfs_iflush() will write a modified inode's changes out to the
2990 * inode's on disk home. The caller must have the inode lock held
2991 * in at least shared mode and the inode flush semaphore must be
2992 * held as well. The inode lock will still be held upon return from
2993 * the call and the caller is free to unlock it.
2994 * The inode flush lock will be unlocked when the inode reaches the disk.
2995 * The flags indicate how the inode's buffer should be written out.
3002 xfs_inode_log_item_t *iip;
3010 int clcount; /* count of inodes clustered */
3012 enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
3015 XFS_STATS_INC(xs_iflush_count);
3017 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
3018 ASSERT(valusema(&ip->i_flock) <= 0);
3019 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3020 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3026 * If the inode isn't dirty, then just release the inode
3027 * flush lock and do nothing.
3029 if ((ip->i_update_core == 0) &&
3030 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3031 ASSERT((iip != NULL) ?
3032 !(iip->ili_item.li_flags & XFS_LI_IN_AIL) : 1);
3038 * We can't flush the inode until it is unpinned, so
3039 * wait for it. We know noone new can pin it, because
3040 * we are holding the inode lock shared and you need
3041 * to hold it exclusively to pin the inode.
3043 xfs_iunpin_wait(ip);
3046 * This may have been unpinned because the filesystem is shutting
3047 * down forcibly. If that's the case we must not write this inode
3048 * to disk, because the log record didn't make it to disk!
3050 if (XFS_FORCED_SHUTDOWN(mp)) {
3051 ip->i_update_core = 0;
3053 iip->ili_format.ilf_fields = 0;
3055 return XFS_ERROR(EIO);
3059 * Get the buffer containing the on-disk inode.
3061 error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0, 0);
3068 * Decide how buffer will be flushed out. This is done before
3069 * the call to xfs_iflush_int because this field is zeroed by it.
3071 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3073 * Flush out the inode buffer according to the directions
3074 * of the caller. In the cases where the caller has given
3075 * us a choice choose the non-delwri case. This is because
3076 * the inode is in the AIL and we need to get it out soon.
3079 case XFS_IFLUSH_SYNC:
3080 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3083 case XFS_IFLUSH_ASYNC:
3084 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3087 case XFS_IFLUSH_DELWRI:
3097 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3098 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3099 case XFS_IFLUSH_DELWRI:
3102 case XFS_IFLUSH_ASYNC:
3105 case XFS_IFLUSH_SYNC:
3116 * First flush out the inode that xfs_iflush was called with.
3118 error = xfs_iflush_int(ip, bp);
3125 * see if other inodes can be gathered into this write
3128 ip->i_chash->chl_buf = bp;
3130 ch = XFS_CHASH(mp, ip->i_blkno);
3131 s = mutex_spinlock(&ch->ch_lock);
3134 for (iq = ip->i_cnext; iq != ip; iq = iq->i_cnext) {
3136 * Do an un-protected check to see if the inode is dirty and
3137 * is a candidate for flushing. These checks will be repeated
3138 * later after the appropriate locks are acquired.
3141 if ((iq->i_update_core == 0) &&
3143 !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
3144 xfs_ipincount(iq) == 0) {
3149 * Try to get locks. If any are unavailable,
3150 * then this inode cannot be flushed and is skipped.
3153 /* get inode locks (just i_lock) */
3154 if (xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) {
3155 /* get inode flush lock */
3156 if (xfs_iflock_nowait(iq)) {
3157 /* check if pinned */
3158 if (xfs_ipincount(iq) == 0) {
3159 /* arriving here means that
3160 * this inode can be flushed.
3161 * first re-check that it's
3165 if ((iq->i_update_core != 0)||
3167 (iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3169 error = xfs_iflush_int(iq, bp);
3173 goto cluster_corrupt_out;
3182 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3185 mutex_spinunlock(&ch->ch_lock, s);
3188 XFS_STATS_INC(xs_icluster_flushcnt);
3189 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
3193 * If the buffer is pinned then push on the log so we won't
3194 * get stuck waiting in the write for too long.
3196 if (XFS_BUF_ISPINNED(bp)){
3197 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
3200 if (flags & INT_DELWRI) {
3201 xfs_bdwrite(mp, bp);
3202 } else if (flags & INT_ASYNC) {
3203 xfs_bawrite(mp, bp);
3205 error = xfs_bwrite(mp, bp);
3211 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3212 xfs_iflush_abort(ip);
3214 * Unlocks the flush lock
3216 return XFS_ERROR(EFSCORRUPTED);
3218 cluster_corrupt_out:
3219 /* Corruption detected in the clustering loop. Invalidate the
3220 * inode buffer and shut down the filesystem.
3222 mutex_spinunlock(&ch->ch_lock, s);
3225 * Clean up the buffer. If it was B_DELWRI, just release it --
3226 * brelse can handle it with no problems. If not, shut down the
3227 * filesystem before releasing the buffer.
3229 if ((bufwasdelwri= XFS_BUF_ISDELAYWRITE(bp))) {
3233 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3237 * Just like incore_relse: if we have b_iodone functions,
3238 * mark the buffer as an error and call them. Otherwise
3239 * mark it as stale and brelse.
3241 if (XFS_BUF_IODONE_FUNC(bp)) {
3242 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
3246 XFS_BUF_ERROR(bp,EIO);
3254 xfs_iflush_abort(iq);
3256 * Unlocks the flush lock
3258 return XFS_ERROR(EFSCORRUPTED);
3267 xfs_inode_log_item_t *iip;
3270 #ifdef XFS_TRANS_DEBUG
3275 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
3276 ASSERT(valusema(&ip->i_flock) <= 0);
3277 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3278 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3285 * If the inode isn't dirty, then just release the inode
3286 * flush lock and do nothing.
3288 if ((ip->i_update_core == 0) &&
3289 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3294 /* set *dip = inode's place in the buffer */
3295 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset);
3298 * Clear i_update_core before copying out the data.
3299 * This is for coordination with our timestamp updates
3300 * that don't hold the inode lock. They will always
3301 * update the timestamps BEFORE setting i_update_core,
3302 * so if we clear i_update_core after they set it we
3303 * are guaranteed to see their updates to the timestamps.
3304 * I believe that this depends on strongly ordered memory
3305 * semantics, but we have that. We use the SYNCHRONIZE
3306 * macro to make sure that the compiler does not reorder
3307 * the i_update_core access below the data copy below.
3309 ip->i_update_core = 0;
3313 * Make sure to get the latest atime from the Linux inode.
3315 xfs_synchronize_atime(ip);
3317 if (XFS_TEST_ERROR(INT_GET(dip->di_core.di_magic,ARCH_CONVERT) != XFS_DINODE_MAGIC,
3318 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3319 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3320 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3321 ip->i_ino, (int) INT_GET(dip->di_core.di_magic, ARCH_CONVERT), dip);
3324 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3325 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3326 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3327 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3328 ip->i_ino, ip, ip->i_d.di_magic);
3331 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3333 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3334 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3335 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3336 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3337 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3341 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
3343 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3344 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3345 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3346 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3347 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3348 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3353 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3354 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3355 XFS_RANDOM_IFLUSH_5)) {
3356 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3357 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3359 ip->i_d.di_nextents + ip->i_d.di_anextents,
3364 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3365 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3366 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3367 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3368 ip->i_ino, ip->i_d.di_forkoff, ip);
3372 * bump the flush iteration count, used to detect flushes which
3373 * postdate a log record during recovery.
3376 ip->i_d.di_flushiter++;
3379 * Copy the dirty parts of the inode into the on-disk
3380 * inode. We always copy out the core of the inode,
3381 * because if the inode is dirty at all the core must
3384 xfs_xlate_dinode_core((xfs_caddr_t)&(dip->di_core), &(ip->i_d), -1);
3386 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3387 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3388 ip->i_d.di_flushiter = 0;
3391 * If this is really an old format inode and the superblock version
3392 * has not been updated to support only new format inodes, then
3393 * convert back to the old inode format. If the superblock version
3394 * has been updated, then make the conversion permanent.
3396 ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 ||
3397 XFS_SB_VERSION_HASNLINK(&mp->m_sb));
3398 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
3399 if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) {
3403 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3404 INT_SET(dip->di_core.di_onlink, ARCH_CONVERT, ip->i_d.di_nlink);
3407 * The superblock version has already been bumped,
3408 * so just make the conversion to the new inode
3411 ip->i_d.di_version = XFS_DINODE_VERSION_2;
3412 INT_SET(dip->di_core.di_version, ARCH_CONVERT, XFS_DINODE_VERSION_2);
3413 ip->i_d.di_onlink = 0;
3414 dip->di_core.di_onlink = 0;
3415 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3416 memset(&(dip->di_core.di_pad[0]), 0,
3417 sizeof(dip->di_core.di_pad));
3418 ASSERT(ip->i_d.di_projid == 0);
3422 if (xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp) == EFSCORRUPTED) {
3426 if (XFS_IFORK_Q(ip)) {
3428 * The only error from xfs_iflush_fork is on the data fork.
3430 (void) xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3432 xfs_inobp_check(mp, bp);
3435 * We've recorded everything logged in the inode, so we'd
3436 * like to clear the ilf_fields bits so we don't log and
3437 * flush things unnecessarily. However, we can't stop
3438 * logging all this information until the data we've copied
3439 * into the disk buffer is written to disk. If we did we might
3440 * overwrite the copy of the inode in the log with all the
3441 * data after re-logging only part of it, and in the face of
3442 * a crash we wouldn't have all the data we need to recover.
3444 * What we do is move the bits to the ili_last_fields field.
3445 * When logging the inode, these bits are moved back to the
3446 * ilf_fields field. In the xfs_iflush_done() routine we
3447 * clear ili_last_fields, since we know that the information
3448 * those bits represent is permanently on disk. As long as
3449 * the flush completes before the inode is logged again, then
3450 * both ilf_fields and ili_last_fields will be cleared.
3452 * We can play with the ilf_fields bits here, because the inode
3453 * lock must be held exclusively in order to set bits there
3454 * and the flush lock protects the ili_last_fields bits.
3455 * Set ili_logged so the flush done
3456 * routine can tell whether or not to look in the AIL.
3457 * Also, store the current LSN of the inode so that we can tell
3458 * whether the item has moved in the AIL from xfs_iflush_done().
3459 * In order to read the lsn we need the AIL lock, because
3460 * it is a 64 bit value that cannot be read atomically.
3462 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3463 iip->ili_last_fields = iip->ili_format.ilf_fields;
3464 iip->ili_format.ilf_fields = 0;
3465 iip->ili_logged = 1;
3467 ASSERT(sizeof(xfs_lsn_t) == 8); /* don't lock if it shrinks */
3469 iip->ili_flush_lsn = iip->ili_item.li_lsn;
3473 * Attach the function xfs_iflush_done to the inode's
3474 * buffer. This will remove the inode from the AIL
3475 * and unlock the inode's flush lock when the inode is
3476 * completely written to disk.
3478 xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
3479 xfs_iflush_done, (xfs_log_item_t *)iip);
3481 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3482 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3485 * We're flushing an inode which is not in the AIL and has
3486 * not been logged but has i_update_core set. For this
3487 * case we can use a B_DELWRI flush and immediately drop
3488 * the inode flush lock because we can avoid the whole
3489 * AIL state thing. It's OK to drop the flush lock now,
3490 * because we've already locked the buffer and to do anything
3491 * you really need both.
3494 ASSERT(iip->ili_logged == 0);
3495 ASSERT(iip->ili_last_fields == 0);
3496 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3504 return XFS_ERROR(EFSCORRUPTED);
3509 * Flush all inactive inodes in mp.
3519 XFS_MOUNT_ILOCK(mp);
3525 /* Make sure we skip markers inserted by sync */
3526 if (ip->i_mount == NULL) {
3531 vp = XFS_ITOV_NULL(ip);
3533 XFS_MOUNT_IUNLOCK(mp);
3534 xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC);
3538 ASSERT(vn_count(vp) == 0);
3541 } while (ip != mp->m_inodes);
3543 XFS_MOUNT_IUNLOCK(mp);
3547 * xfs_iaccess: check accessibility of inode for mode.
3556 mode_t orgmode = mode;
3557 struct inode *inode = vn_to_inode(XFS_ITOV(ip));
3559 if (mode & S_IWUSR) {
3560 umode_t imode = inode->i_mode;
3562 if (IS_RDONLY(inode) &&
3563 (S_ISREG(imode) || S_ISDIR(imode) || S_ISLNK(imode)))
3564 return XFS_ERROR(EROFS);
3566 if (IS_IMMUTABLE(inode))
3567 return XFS_ERROR(EACCES);
3571 * If there's an Access Control List it's used instead of
3574 if ((error = _ACL_XFS_IACCESS(ip, mode, cr)) != -1)
3575 return error ? XFS_ERROR(error) : 0;
3577 if (current_fsuid(cr) != ip->i_d.di_uid) {
3579 if (!in_group_p((gid_t)ip->i_d.di_gid))
3584 * If the DACs are ok we don't need any capability check.
3586 if ((ip->i_d.di_mode & mode) == mode)
3589 * Read/write DACs are always overridable.
3590 * Executable DACs are overridable if at least one exec bit is set.
3592 if (!(orgmode & S_IXUSR) ||
3593 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
3594 if (capable_cred(cr, CAP_DAC_OVERRIDE))
3597 if ((orgmode == S_IRUSR) ||
3598 (S_ISDIR(inode->i_mode) && (!(orgmode & S_IWUSR)))) {
3599 if (capable_cred(cr, CAP_DAC_READ_SEARCH))
3602 cmn_err(CE_NOTE, "Ick: mode=%o, orgmode=%o", mode, orgmode);
3604 return XFS_ERROR(EACCES);
3606 return XFS_ERROR(EACCES);
3610 * xfs_iroundup: round up argument to next power of two
3619 if ((v & (v - 1)) == 0)
3621 ASSERT((v & 0x80000000) == 0);
3622 if ((v & (v + 1)) == 0)
3624 for (i = 0, m = 1; i < 31; i++, m <<= 1) {
3628 if ((v & (v + 1)) == 0)
3635 #ifdef XFS_ILOCK_TRACE
3636 ktrace_t *xfs_ilock_trace_buf;
3639 xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
3641 ktrace_enter(ip->i_lock_trace,
3643 (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
3644 (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
3645 (void *)ra, /* caller of ilock */
3646 (void *)(unsigned long)current_cpu(),
3647 (void *)(unsigned long)current_pid(),
3648 NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL);
3653 * Return a pointer to the extent record at file index idx.
3657 xfs_ifork_t *ifp, /* inode fork pointer */
3658 xfs_extnum_t idx) /* index of target extent */
3661 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
3662 return ifp->if_u1.if_ext_irec->er_extbuf;
3663 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3664 xfs_ext_irec_t *erp; /* irec pointer */
3665 int erp_idx = 0; /* irec index */
3666 xfs_extnum_t page_idx = idx; /* ext index in target list */
3668 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3669 return &erp->er_extbuf[page_idx];
3670 } else if (ifp->if_bytes) {
3671 return &ifp->if_u1.if_extents[idx];
3678 * Insert new item(s) into the extent records for incore inode
3679 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3683 xfs_ifork_t *ifp, /* inode fork pointer */
3684 xfs_extnum_t idx, /* starting index of new items */
3685 xfs_extnum_t count, /* number of inserted items */
3686 xfs_bmbt_irec_t *new) /* items to insert */
3688 xfs_bmbt_rec_t *ep; /* extent record pointer */
3689 xfs_extnum_t i; /* extent record index */
3691 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3692 xfs_iext_add(ifp, idx, count);
3693 for (i = idx; i < idx + count; i++, new++) {
3694 ep = xfs_iext_get_ext(ifp, i);
3695 xfs_bmbt_set_all(ep, new);
3700 * This is called when the amount of space required for incore file
3701 * extents needs to be increased. The ext_diff parameter stores the
3702 * number of new extents being added and the idx parameter contains
3703 * the extent index where the new extents will be added. If the new
3704 * extents are being appended, then we just need to (re)allocate and
3705 * initialize the space. Otherwise, if the new extents are being
3706 * inserted into the middle of the existing entries, a bit more work
3707 * is required to make room for the new extents to be inserted. The
3708 * caller is responsible for filling in the new extent entries upon
3713 xfs_ifork_t *ifp, /* inode fork pointer */
3714 xfs_extnum_t idx, /* index to begin adding exts */
3715 int ext_diff) /* number of extents to add */
3717 int byte_diff; /* new bytes being added */
3718 int new_size; /* size of extents after adding */
3719 xfs_extnum_t nextents; /* number of extents in file */
3721 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3722 ASSERT((idx >= 0) && (idx <= nextents));
3723 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
3724 new_size = ifp->if_bytes + byte_diff;
3726 * If the new number of extents (nextents + ext_diff)
3727 * fits inside the inode, then continue to use the inline
3730 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
3731 if (idx < nextents) {
3732 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
3733 &ifp->if_u2.if_inline_ext[idx],
3734 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3735 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
3737 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3738 ifp->if_real_bytes = 0;
3739 ifp->if_lastex = nextents + ext_diff;
3742 * Otherwise use a linear (direct) extent list.
3743 * If the extents are currently inside the inode,
3744 * xfs_iext_realloc_direct will switch us from
3745 * inline to direct extent allocation mode.
3747 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
3748 xfs_iext_realloc_direct(ifp, new_size);
3749 if (idx < nextents) {
3750 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
3751 &ifp->if_u1.if_extents[idx],
3752 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3753 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
3756 /* Indirection array */
3758 xfs_ext_irec_t *erp;
3762 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
3763 if (ifp->if_flags & XFS_IFEXTIREC) {
3764 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3766 xfs_iext_irec_init(ifp);
3767 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3768 erp = ifp->if_u1.if_ext_irec;
3770 /* Extents fit in target extent page */
3771 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3772 if (page_idx < erp->er_extcount) {
3773 memmove(&erp->er_extbuf[page_idx + ext_diff],
3774 &erp->er_extbuf[page_idx],
3775 (erp->er_extcount - page_idx) *
3776 sizeof(xfs_bmbt_rec_t));
3777 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3779 erp->er_extcount += ext_diff;
3780 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3782 /* Insert a new extent page */
3784 xfs_iext_add_indirect_multi(ifp,
3785 erp_idx, page_idx, ext_diff);
3788 * If extent(s) are being appended to the last page in
3789 * the indirection array and the new extent(s) don't fit
3790 * in the page, then erp is NULL and erp_idx is set to
3791 * the next index needed in the indirection array.
3794 int count = ext_diff;
3797 erp = xfs_iext_irec_new(ifp, erp_idx);
3798 erp->er_extcount = count;
3799 count -= MIN(count, (int)XFS_LINEAR_EXTS);
3806 ifp->if_bytes = new_size;
3810 * This is called when incore extents are being added to the indirection
3811 * array and the new extents do not fit in the target extent list. The
3812 * erp_idx parameter contains the irec index for the target extent list
3813 * in the indirection array, and the idx parameter contains the extent
3814 * index within the list. The number of extents being added is stored
3815 * in the count parameter.
3817 * |-------| |-------|
3818 * | | | | idx - number of extents before idx
3820 * | | | | count - number of extents being inserted at idx
3821 * |-------| |-------|
3822 * | count | | nex2 | nex2 - number of extents after idx + count
3823 * |-------| |-------|
3826 xfs_iext_add_indirect_multi(
3827 xfs_ifork_t *ifp, /* inode fork pointer */
3828 int erp_idx, /* target extent irec index */
3829 xfs_extnum_t idx, /* index within target list */
3830 int count) /* new extents being added */
3832 int byte_diff; /* new bytes being added */
3833 xfs_ext_irec_t *erp; /* pointer to irec entry */
3834 xfs_extnum_t ext_diff; /* number of extents to add */
3835 xfs_extnum_t ext_cnt; /* new extents still needed */
3836 xfs_extnum_t nex2; /* extents after idx + count */
3837 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3838 int nlists; /* number of irec's (lists) */
3840 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3841 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3842 nex2 = erp->er_extcount - idx;
3843 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3846 * Save second part of target extent list
3847 * (all extents past */
3849 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3850 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_SLEEP);
3851 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3852 erp->er_extcount -= nex2;
3853 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3854 memset(&erp->er_extbuf[idx], 0, byte_diff);
3858 * Add the new extents to the end of the target
3859 * list, then allocate new irec record(s) and
3860 * extent buffer(s) as needed to store the rest
3861 * of the new extents.
3864 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3866 erp->er_extcount += ext_diff;
3867 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3868 ext_cnt -= ext_diff;
3872 erp = xfs_iext_irec_new(ifp, erp_idx);
3873 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3874 erp->er_extcount = ext_diff;
3875 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3876 ext_cnt -= ext_diff;
3879 /* Add nex2 extents back to indirection array */
3881 xfs_extnum_t ext_avail;
3884 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3885 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3888 * If nex2 extents fit in the current page, append
3889 * nex2_ep after the new extents.
3891 if (nex2 <= ext_avail) {
3892 i = erp->er_extcount;
3895 * Otherwise, check if space is available in the
3898 else if ((erp_idx < nlists - 1) &&
3899 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3900 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3903 /* Create a hole for nex2 extents */
3904 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3905 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3908 * Final choice, create a new extent page for
3913 erp = xfs_iext_irec_new(ifp, erp_idx);
3915 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
3916 kmem_free(nex2_ep, byte_diff);
3917 erp->er_extcount += nex2;
3918 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3923 * This is called when the amount of space required for incore file
3924 * extents needs to be decreased. The ext_diff parameter stores the
3925 * number of extents to be removed and the idx parameter contains
3926 * the extent index where the extents will be removed from.
3928 * If the amount of space needed has decreased below the linear
3929 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3930 * extent array. Otherwise, use kmem_realloc() to adjust the
3931 * size to what is needed.
3935 xfs_ifork_t *ifp, /* inode fork pointer */
3936 xfs_extnum_t idx, /* index to begin removing exts */
3937 int ext_diff) /* number of extents to remove */
3939 xfs_extnum_t nextents; /* number of extents in file */
3940 int new_size; /* size of extents after removal */
3942 ASSERT(ext_diff > 0);
3943 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3944 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3946 if (new_size == 0) {
3947 xfs_iext_destroy(ifp);
3948 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3949 xfs_iext_remove_indirect(ifp, idx, ext_diff);
3950 } else if (ifp->if_real_bytes) {
3951 xfs_iext_remove_direct(ifp, idx, ext_diff);
3953 xfs_iext_remove_inline(ifp, idx, ext_diff);
3955 ifp->if_bytes = new_size;
3959 * This removes ext_diff extents from the inline buffer, beginning
3960 * at extent index idx.
3963 xfs_iext_remove_inline(
3964 xfs_ifork_t *ifp, /* inode fork pointer */
3965 xfs_extnum_t idx, /* index to begin removing exts */
3966 int ext_diff) /* number of extents to remove */
3968 int nextents; /* number of extents in file */
3970 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3971 ASSERT(idx < XFS_INLINE_EXTS);
3972 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3973 ASSERT(((nextents - ext_diff) > 0) &&
3974 (nextents - ext_diff) < XFS_INLINE_EXTS);
3976 if (idx + ext_diff < nextents) {
3977 memmove(&ifp->if_u2.if_inline_ext[idx],
3978 &ifp->if_u2.if_inline_ext[idx + ext_diff],
3979 (nextents - (idx + ext_diff)) *
3980 sizeof(xfs_bmbt_rec_t));
3981 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
3982 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3984 memset(&ifp->if_u2.if_inline_ext[idx], 0,
3985 ext_diff * sizeof(xfs_bmbt_rec_t));
3990 * This removes ext_diff extents from a linear (direct) extent list,
3991 * beginning at extent index idx. If the extents are being removed
3992 * from the end of the list (ie. truncate) then we just need to re-
3993 * allocate the list to remove the extra space. Otherwise, if the
3994 * extents are being removed from the middle of the existing extent
3995 * entries, then we first need to move the extent records beginning
3996 * at idx + ext_diff up in the list to overwrite the records being
3997 * removed, then remove the extra space via kmem_realloc.
4000 xfs_iext_remove_direct(
4001 xfs_ifork_t *ifp, /* inode fork pointer */
4002 xfs_extnum_t idx, /* index to begin removing exts */
4003 int ext_diff) /* number of extents to remove */
4005 xfs_extnum_t nextents; /* number of extents in file */
4006 int new_size; /* size of extents after removal */
4008 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4009 new_size = ifp->if_bytes -
4010 (ext_diff * sizeof(xfs_bmbt_rec_t));
4011 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4013 if (new_size == 0) {
4014 xfs_iext_destroy(ifp);
4017 /* Move extents up in the list (if needed) */
4018 if (idx + ext_diff < nextents) {
4019 memmove(&ifp->if_u1.if_extents[idx],
4020 &ifp->if_u1.if_extents[idx + ext_diff],
4021 (nextents - (idx + ext_diff)) *
4022 sizeof(xfs_bmbt_rec_t));
4024 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
4025 0, ext_diff * sizeof(xfs_bmbt_rec_t));
4027 * Reallocate the direct extent list. If the extents
4028 * will fit inside the inode then xfs_iext_realloc_direct
4029 * will switch from direct to inline extent allocation
4032 xfs_iext_realloc_direct(ifp, new_size);
4033 ifp->if_bytes = new_size;
4037 * This is called when incore extents are being removed from the
4038 * indirection array and the extents being removed span multiple extent
4039 * buffers. The idx parameter contains the file extent index where we
4040 * want to begin removing extents, and the count parameter contains
4041 * how many extents need to be removed.
4043 * |-------| |-------|
4044 * | nex1 | | | nex1 - number of extents before idx
4045 * |-------| | count |
4046 * | | | | count - number of extents being removed at idx
4047 * | count | |-------|
4048 * | | | nex2 | nex2 - number of extents after idx + count
4049 * |-------| |-------|
4052 xfs_iext_remove_indirect(
4053 xfs_ifork_t *ifp, /* inode fork pointer */
4054 xfs_extnum_t idx, /* index to begin removing extents */
4055 int count) /* number of extents to remove */
4057 xfs_ext_irec_t *erp; /* indirection array pointer */
4058 int erp_idx = 0; /* indirection array index */
4059 xfs_extnum_t ext_cnt; /* extents left to remove */
4060 xfs_extnum_t ext_diff; /* extents to remove in current list */
4061 xfs_extnum_t nex1; /* number of extents before idx */
4062 xfs_extnum_t nex2; /* extents after idx + count */
4063 int nlists; /* entries in indirection array */
4064 int page_idx = idx; /* index in target extent list */
4066 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4067 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
4068 ASSERT(erp != NULL);
4069 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4073 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
4074 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
4076 * Check for deletion of entire list;
4077 * xfs_iext_irec_remove() updates extent offsets.
4079 if (ext_diff == erp->er_extcount) {
4080 xfs_iext_irec_remove(ifp, erp_idx);
4081 ext_cnt -= ext_diff;
4084 ASSERT(erp_idx < ifp->if_real_bytes /
4086 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4093 /* Move extents up (if needed) */
4095 memmove(&erp->er_extbuf[nex1],
4096 &erp->er_extbuf[nex1 + ext_diff],
4097 nex2 * sizeof(xfs_bmbt_rec_t));
4099 /* Zero out rest of page */
4100 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
4101 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
4102 /* Update remaining counters */
4103 erp->er_extcount -= ext_diff;
4104 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
4105 ext_cnt -= ext_diff;
4110 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
4111 xfs_iext_irec_compact(ifp);
4115 * Create, destroy, or resize a linear (direct) block of extents.
4118 xfs_iext_realloc_direct(
4119 xfs_ifork_t *ifp, /* inode fork pointer */
4120 int new_size) /* new size of extents */
4122 int rnew_size; /* real new size of extents */
4124 rnew_size = new_size;
4126 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
4127 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
4128 (new_size != ifp->if_real_bytes)));
4130 /* Free extent records */
4131 if (new_size == 0) {
4132 xfs_iext_destroy(ifp);
4134 /* Resize direct extent list and zero any new bytes */
4135 else if (ifp->if_real_bytes) {
4136 /* Check if extents will fit inside the inode */
4137 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
4138 xfs_iext_direct_to_inline(ifp, new_size /
4139 (uint)sizeof(xfs_bmbt_rec_t));
4140 ifp->if_bytes = new_size;
4143 if ((new_size & (new_size - 1)) != 0) {
4144 rnew_size = xfs_iroundup(new_size);
4146 if (rnew_size != ifp->if_real_bytes) {
4147 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
4148 kmem_realloc(ifp->if_u1.if_extents,
4153 if (rnew_size > ifp->if_real_bytes) {
4154 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
4155 (uint)sizeof(xfs_bmbt_rec_t)], 0,
4156 rnew_size - ifp->if_real_bytes);
4160 * Switch from the inline extent buffer to a direct
4161 * extent list. Be sure to include the inline extent
4162 * bytes in new_size.
4165 new_size += ifp->if_bytes;
4166 if ((new_size & (new_size - 1)) != 0) {
4167 rnew_size = xfs_iroundup(new_size);
4169 xfs_iext_inline_to_direct(ifp, rnew_size);
4171 ifp->if_real_bytes = rnew_size;
4172 ifp->if_bytes = new_size;
4176 * Switch from linear (direct) extent records to inline buffer.
4179 xfs_iext_direct_to_inline(
4180 xfs_ifork_t *ifp, /* inode fork pointer */
4181 xfs_extnum_t nextents) /* number of extents in file */
4183 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4184 ASSERT(nextents <= XFS_INLINE_EXTS);
4186 * The inline buffer was zeroed when we switched
4187 * from inline to direct extent allocation mode,
4188 * so we don't need to clear it here.
4190 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
4191 nextents * sizeof(xfs_bmbt_rec_t));
4192 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
4193 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
4194 ifp->if_real_bytes = 0;
4198 * Switch from inline buffer to linear (direct) extent records.
4199 * new_size should already be rounded up to the next power of 2
4200 * by the caller (when appropriate), so use new_size as it is.
4201 * However, since new_size may be rounded up, we can't update
4202 * if_bytes here. It is the caller's responsibility to update
4203 * if_bytes upon return.
4206 xfs_iext_inline_to_direct(
4207 xfs_ifork_t *ifp, /* inode fork pointer */
4208 int new_size) /* number of extents in file */
4210 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
4211 kmem_alloc(new_size, KM_SLEEP);
4212 memset(ifp->if_u1.if_extents, 0, new_size);
4213 if (ifp->if_bytes) {
4214 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
4216 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4217 sizeof(xfs_bmbt_rec_t));
4219 ifp->if_real_bytes = new_size;
4223 * Resize an extent indirection array to new_size bytes.
4226 xfs_iext_realloc_indirect(
4227 xfs_ifork_t *ifp, /* inode fork pointer */
4228 int new_size) /* new indirection array size */
4230 int nlists; /* number of irec's (ex lists) */
4231 int size; /* current indirection array size */
4233 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4234 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4235 size = nlists * sizeof(xfs_ext_irec_t);
4236 ASSERT(ifp->if_real_bytes);
4237 ASSERT((new_size >= 0) && (new_size != size));
4238 if (new_size == 0) {
4239 xfs_iext_destroy(ifp);
4241 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
4242 kmem_realloc(ifp->if_u1.if_ext_irec,
4243 new_size, size, KM_SLEEP);
4248 * Switch from indirection array to linear (direct) extent allocations.
4251 xfs_iext_indirect_to_direct(
4252 xfs_ifork_t *ifp) /* inode fork pointer */
4254 xfs_bmbt_rec_t *ep; /* extent record pointer */
4255 xfs_extnum_t nextents; /* number of extents in file */
4256 int size; /* size of file extents */
4258 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4259 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4260 ASSERT(nextents <= XFS_LINEAR_EXTS);
4261 size = nextents * sizeof(xfs_bmbt_rec_t);
4263 xfs_iext_irec_compact_full(ifp);
4264 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
4266 ep = ifp->if_u1.if_ext_irec->er_extbuf;
4267 kmem_free(ifp->if_u1.if_ext_irec, sizeof(xfs_ext_irec_t));
4268 ifp->if_flags &= ~XFS_IFEXTIREC;
4269 ifp->if_u1.if_extents = ep;
4270 ifp->if_bytes = size;
4271 if (nextents < XFS_LINEAR_EXTS) {
4272 xfs_iext_realloc_direct(ifp, size);
4277 * Free incore file extents.
4281 xfs_ifork_t *ifp) /* inode fork pointer */
4283 if (ifp->if_flags & XFS_IFEXTIREC) {
4287 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4288 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
4289 xfs_iext_irec_remove(ifp, erp_idx);
4291 ifp->if_flags &= ~XFS_IFEXTIREC;
4292 } else if (ifp->if_real_bytes) {
4293 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
4294 } else if (ifp->if_bytes) {
4295 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4296 sizeof(xfs_bmbt_rec_t));
4298 ifp->if_u1.if_extents = NULL;
4299 ifp->if_real_bytes = 0;
4304 * Return a pointer to the extent record for file system block bno.
4306 xfs_bmbt_rec_t * /* pointer to found extent record */
4307 xfs_iext_bno_to_ext(
4308 xfs_ifork_t *ifp, /* inode fork pointer */
4309 xfs_fileoff_t bno, /* block number to search for */
4310 xfs_extnum_t *idxp) /* index of target extent */
4312 xfs_bmbt_rec_t *base; /* pointer to first extent */
4313 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
4314 xfs_bmbt_rec_t *ep = NULL; /* pointer to target extent */
4315 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4316 int high; /* upper boundary in search */
4317 xfs_extnum_t idx = 0; /* index of target extent */
4318 int low; /* lower boundary in search */
4319 xfs_extnum_t nextents; /* number of file extents */
4320 xfs_fileoff_t startoff = 0; /* start offset of extent */
4322 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4323 if (nextents == 0) {
4328 if (ifp->if_flags & XFS_IFEXTIREC) {
4329 /* Find target extent list */
4331 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
4332 base = erp->er_extbuf;
4333 high = erp->er_extcount - 1;
4335 base = ifp->if_u1.if_extents;
4336 high = nextents - 1;
4338 /* Binary search extent records */
4339 while (low <= high) {
4340 idx = (low + high) >> 1;
4342 startoff = xfs_bmbt_get_startoff(ep);
4343 blockcount = xfs_bmbt_get_blockcount(ep);
4344 if (bno < startoff) {
4346 } else if (bno >= startoff + blockcount) {
4349 /* Convert back to file-based extent index */
4350 if (ifp->if_flags & XFS_IFEXTIREC) {
4351 idx += erp->er_extoff;
4357 /* Convert back to file-based extent index */
4358 if (ifp->if_flags & XFS_IFEXTIREC) {
4359 idx += erp->er_extoff;
4361 if (bno >= startoff + blockcount) {
4362 if (++idx == nextents) {
4365 ep = xfs_iext_get_ext(ifp, idx);
4373 * Return a pointer to the indirection array entry containing the
4374 * extent record for filesystem block bno. Store the index of the
4375 * target irec in *erp_idxp.
4377 xfs_ext_irec_t * /* pointer to found extent record */
4378 xfs_iext_bno_to_irec(
4379 xfs_ifork_t *ifp, /* inode fork pointer */
4380 xfs_fileoff_t bno, /* block number to search for */
4381 int *erp_idxp) /* irec index of target ext list */
4383 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4384 xfs_ext_irec_t *erp_next; /* next indirection array entry */
4385 int erp_idx; /* indirection array index */
4386 int nlists; /* number of extent irec's (lists) */
4387 int high; /* binary search upper limit */
4388 int low; /* binary search lower limit */
4390 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4391 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4395 while (low <= high) {
4396 erp_idx = (low + high) >> 1;
4397 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4398 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
4399 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
4401 } else if (erp_next && bno >=
4402 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
4408 *erp_idxp = erp_idx;
4413 * Return a pointer to the indirection array entry containing the
4414 * extent record at file extent index *idxp. Store the index of the
4415 * target irec in *erp_idxp and store the page index of the target
4416 * extent record in *idxp.
4419 xfs_iext_idx_to_irec(
4420 xfs_ifork_t *ifp, /* inode fork pointer */
4421 xfs_extnum_t *idxp, /* extent index (file -> page) */
4422 int *erp_idxp, /* pointer to target irec */
4423 int realloc) /* new bytes were just added */
4425 xfs_ext_irec_t *prev; /* pointer to previous irec */
4426 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
4427 int erp_idx; /* indirection array index */
4428 int nlists; /* number of irec's (ex lists) */
4429 int high; /* binary search upper limit */
4430 int low; /* binary search lower limit */
4431 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
4433 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4434 ASSERT(page_idx >= 0 && page_idx <=
4435 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
4436 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4441 /* Binary search extent irec's */
4442 while (low <= high) {
4443 erp_idx = (low + high) >> 1;
4444 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4445 prev = erp_idx > 0 ? erp - 1 : NULL;
4446 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
4447 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
4449 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
4450 (page_idx == erp->er_extoff + erp->er_extcount &&
4453 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
4454 erp->er_extcount == XFS_LINEAR_EXTS) {
4458 erp = erp_idx < nlists ? erp + 1 : NULL;
4461 page_idx -= erp->er_extoff;
4466 *erp_idxp = erp_idx;
4471 * Allocate and initialize an indirection array once the space needed
4472 * for incore extents increases above XFS_IEXT_BUFSZ.
4476 xfs_ifork_t *ifp) /* inode fork pointer */
4478 xfs_ext_irec_t *erp; /* indirection array pointer */
4479 xfs_extnum_t nextents; /* number of extents in file */
4481 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4482 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4483 ASSERT(nextents <= XFS_LINEAR_EXTS);
4485 erp = (xfs_ext_irec_t *)
4486 kmem_alloc(sizeof(xfs_ext_irec_t), KM_SLEEP);
4488 if (nextents == 0) {
4489 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
4490 kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
4491 } else if (!ifp->if_real_bytes) {
4492 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
4493 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
4494 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
4496 erp->er_extbuf = ifp->if_u1.if_extents;
4497 erp->er_extcount = nextents;
4500 ifp->if_flags |= XFS_IFEXTIREC;
4501 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
4502 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
4503 ifp->if_u1.if_ext_irec = erp;
4509 * Allocate and initialize a new entry in the indirection array.
4513 xfs_ifork_t *ifp, /* inode fork pointer */
4514 int erp_idx) /* index for new irec */
4516 xfs_ext_irec_t *erp; /* indirection array pointer */
4517 int i; /* loop counter */
4518 int nlists; /* number of irec's (ex lists) */
4520 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4521 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4523 /* Resize indirection array */
4524 xfs_iext_realloc_indirect(ifp, ++nlists *
4525 sizeof(xfs_ext_irec_t));
4527 * Move records down in the array so the
4528 * new page can use erp_idx.
4530 erp = ifp->if_u1.if_ext_irec;
4531 for (i = nlists - 1; i > erp_idx; i--) {
4532 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
4534 ASSERT(i == erp_idx);
4536 /* Initialize new extent record */
4537 erp = ifp->if_u1.if_ext_irec;
4538 erp[erp_idx].er_extbuf = (xfs_bmbt_rec_t *)
4539 kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
4540 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4541 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
4542 erp[erp_idx].er_extcount = 0;
4543 erp[erp_idx].er_extoff = erp_idx > 0 ?
4544 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
4545 return (&erp[erp_idx]);
4549 * Remove a record from the indirection array.
4552 xfs_iext_irec_remove(
4553 xfs_ifork_t *ifp, /* inode fork pointer */
4554 int erp_idx) /* irec index to remove */
4556 xfs_ext_irec_t *erp; /* indirection array pointer */
4557 int i; /* loop counter */
4558 int nlists; /* number of irec's (ex lists) */
4560 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4561 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4562 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4563 if (erp->er_extbuf) {
4564 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
4566 kmem_free(erp->er_extbuf, XFS_IEXT_BUFSZ);
4568 /* Compact extent records */
4569 erp = ifp->if_u1.if_ext_irec;
4570 for (i = erp_idx; i < nlists - 1; i++) {
4571 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
4574 * Manually free the last extent record from the indirection
4575 * array. A call to xfs_iext_realloc_indirect() with a size
4576 * of zero would result in a call to xfs_iext_destroy() which
4577 * would in turn call this function again, creating a nasty
4581 xfs_iext_realloc_indirect(ifp,
4582 nlists * sizeof(xfs_ext_irec_t));
4584 kmem_free(ifp->if_u1.if_ext_irec,
4585 sizeof(xfs_ext_irec_t));
4587 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4591 * This is called to clean up large amounts of unused memory allocated
4592 * by the indirection array. Before compacting anything though, verify
4593 * that the indirection array is still needed and switch back to the
4594 * linear extent list (or even the inline buffer) if possible. The
4595 * compaction policy is as follows:
4597 * Full Compaction: Extents fit into a single page (or inline buffer)
4598 * Full Compaction: Extents occupy less than 10% of allocated space
4599 * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
4600 * No Compaction: Extents occupy at least 50% of allocated space
4603 xfs_iext_irec_compact(
4604 xfs_ifork_t *ifp) /* inode fork pointer */
4606 xfs_extnum_t nextents; /* number of extents in file */
4607 int nlists; /* number of irec's (ex lists) */
4609 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4610 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4611 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4613 if (nextents == 0) {
4614 xfs_iext_destroy(ifp);
4615 } else if (nextents <= XFS_INLINE_EXTS) {
4616 xfs_iext_indirect_to_direct(ifp);
4617 xfs_iext_direct_to_inline(ifp, nextents);
4618 } else if (nextents <= XFS_LINEAR_EXTS) {
4619 xfs_iext_indirect_to_direct(ifp);
4620 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 3) {
4621 xfs_iext_irec_compact_full(ifp);
4622 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
4623 xfs_iext_irec_compact_pages(ifp);
4628 * Combine extents from neighboring extent pages.
4631 xfs_iext_irec_compact_pages(
4632 xfs_ifork_t *ifp) /* inode fork pointer */
4634 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
4635 int erp_idx = 0; /* indirection array index */
4636 int nlists; /* number of irec's (ex lists) */
4638 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4639 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4640 while (erp_idx < nlists - 1) {
4641 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4643 if (erp_next->er_extcount <=
4644 (XFS_LINEAR_EXTS - erp->er_extcount)) {
4645 memmove(&erp->er_extbuf[erp->er_extcount],
4646 erp_next->er_extbuf, erp_next->er_extcount *
4647 sizeof(xfs_bmbt_rec_t));
4648 erp->er_extcount += erp_next->er_extcount;
4650 * Free page before removing extent record
4651 * so er_extoffs don't get modified in
4652 * xfs_iext_irec_remove.
4654 kmem_free(erp_next->er_extbuf, XFS_IEXT_BUFSZ);
4655 erp_next->er_extbuf = NULL;
4656 xfs_iext_irec_remove(ifp, erp_idx + 1);
4657 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4665 * Fully compact the extent records managed by the indirection array.
4668 xfs_iext_irec_compact_full(
4669 xfs_ifork_t *ifp) /* inode fork pointer */
4671 xfs_bmbt_rec_t *ep, *ep_next; /* extent record pointers */
4672 xfs_ext_irec_t *erp, *erp_next; /* extent irec pointers */
4673 int erp_idx = 0; /* extent irec index */
4674 int ext_avail; /* empty entries in ex list */
4675 int ext_diff; /* number of exts to add */
4676 int nlists; /* number of irec's (ex lists) */
4678 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4679 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4680 erp = ifp->if_u1.if_ext_irec;
4681 ep = &erp->er_extbuf[erp->er_extcount];
4683 ep_next = erp_next->er_extbuf;
4684 while (erp_idx < nlists - 1) {
4685 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
4686 ext_diff = MIN(ext_avail, erp_next->er_extcount);
4687 memcpy(ep, ep_next, ext_diff * sizeof(xfs_bmbt_rec_t));
4688 erp->er_extcount += ext_diff;
4689 erp_next->er_extcount -= ext_diff;
4690 /* Remove next page */
4691 if (erp_next->er_extcount == 0) {
4693 * Free page before removing extent record
4694 * so er_extoffs don't get modified in
4695 * xfs_iext_irec_remove.
4697 kmem_free(erp_next->er_extbuf,
4698 erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
4699 erp_next->er_extbuf = NULL;
4700 xfs_iext_irec_remove(ifp, erp_idx + 1);
4701 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4702 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4703 /* Update next page */
4705 /* Move rest of page up to become next new page */
4706 memmove(erp_next->er_extbuf, ep_next,
4707 erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
4708 ep_next = erp_next->er_extbuf;
4709 memset(&ep_next[erp_next->er_extcount], 0,
4710 (XFS_LINEAR_EXTS - erp_next->er_extcount) *
4711 sizeof(xfs_bmbt_rec_t));
4713 if (erp->er_extcount == XFS_LINEAR_EXTS) {
4715 if (erp_idx < nlists)
4716 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4720 ep = &erp->er_extbuf[erp->er_extcount];
4722 ep_next = erp_next->er_extbuf;
4727 * This is called to update the er_extoff field in the indirection
4728 * array when extents have been added or removed from one of the
4729 * extent lists. erp_idx contains the irec index to begin updating
4730 * at and ext_diff contains the number of extents that were added
4734 xfs_iext_irec_update_extoffs(
4735 xfs_ifork_t *ifp, /* inode fork pointer */
4736 int erp_idx, /* irec index to update */
4737 int ext_diff) /* number of new extents */
4739 int i; /* loop counter */
4740 int nlists; /* number of irec's (ex lists */
4742 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4743 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4744 for (i = erp_idx; i < nlists; i++) {
4745 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;