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);
73 * Make sure that the extents in the given memory buffer
88 for (i = 0; i < nrecs; i++) {
89 ep = xfs_iext_get_ext(ifp, i);
90 rec.l0 = get_unaligned((__uint64_t*)&ep->l0);
91 rec.l1 = get_unaligned((__uint64_t*)&ep->l1);
93 xfs_bmbt_disk_get_all(&rec, &irec);
95 xfs_bmbt_get_all(&rec, &irec);
96 if (fmt == XFS_EXTFMT_NOSTATE)
97 ASSERT(irec.br_state == XFS_EXT_NORM);
101 #define xfs_validate_extents(ifp, nrecs, disk, fmt)
105 * Check that none of the inode's in the buffer have a next
106 * unlinked field of 0.
118 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
120 for (i = 0; i < j; i++) {
121 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
122 i * mp->m_sb.sb_inodesize);
123 if (!dip->di_next_unlinked) {
124 xfs_fs_cmn_err(CE_ALERT, mp,
125 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
127 ASSERT(dip->di_next_unlinked);
134 * This routine is called to map an inode number within a file
135 * system to the buffer containing the on-disk version of the
136 * inode. It returns a pointer to the buffer containing the
137 * on-disk inode in the bpp parameter, and in the dip parameter
138 * it returns a pointer to the on-disk inode within that buffer.
140 * If a non-zero error is returned, then the contents of bpp and
141 * dipp are undefined.
143 * Use xfs_imap() to determine the size and location of the
144 * buffer to read from disk.
162 * Call the space management code to find the location of the
166 error = xfs_imap(mp, tp, ino, &imap, XFS_IMAP_LOOKUP);
169 "xfs_inotobp: xfs_imap() returned an "
170 "error %d on %s. Returning error.", error, mp->m_fsname);
175 * If the inode number maps to a block outside the bounds of the
176 * file system then return NULL rather than calling read_buf
177 * and panicing when we get an error from the driver.
179 if ((imap.im_blkno + imap.im_len) >
180 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
182 "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
183 "of the file system %s. Returning EINVAL.",
184 (unsigned long long)imap.im_blkno,
185 imap.im_len, mp->m_fsname);
186 return XFS_ERROR(EINVAL);
190 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
191 * default to just a read_buf() call.
193 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
194 (int)imap.im_len, XFS_BUF_LOCK, &bp);
198 "xfs_inotobp: xfs_trans_read_buf() returned an "
199 "error %d on %s. Returning error.", error, mp->m_fsname);
202 dip = (xfs_dinode_t *)xfs_buf_offset(bp, 0);
204 INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
205 XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
206 if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP,
207 XFS_RANDOM_ITOBP_INOTOBP))) {
208 XFS_CORRUPTION_ERROR("xfs_inotobp", XFS_ERRLEVEL_LOW, mp, dip);
209 xfs_trans_brelse(tp, bp);
211 "xfs_inotobp: XFS_TEST_ERROR() returned an "
212 "error on %s. Returning EFSCORRUPTED.", mp->m_fsname);
213 return XFS_ERROR(EFSCORRUPTED);
216 xfs_inobp_check(mp, bp);
219 * Set *dipp to point to the on-disk inode in the buffer.
221 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
223 *offset = imap.im_boffset;
229 * This routine is called to map an inode to the buffer containing
230 * the on-disk version of the inode. It returns a pointer to the
231 * buffer containing the on-disk inode in the bpp parameter, and in
232 * the dip parameter it returns a pointer to the on-disk inode within
235 * If a non-zero error is returned, then the contents of bpp and
236 * dipp are undefined.
238 * If the inode is new and has not yet been initialized, use xfs_imap()
239 * to determine the size and location of the buffer to read from disk.
240 * If the inode has already been mapped to its buffer and read in once,
241 * then use the mapping information stored in the inode rather than
242 * calling xfs_imap(). This allows us to avoid the overhead of looking
243 * at the inode btree for small block file systems (see xfs_dilocate()).
244 * We can tell whether the inode has been mapped in before by comparing
245 * its disk block address to 0. Only uninitialized inodes will have
246 * 0 for the disk block address.
266 if (ip->i_blkno == (xfs_daddr_t)0) {
268 * Call the space management code to find the location of the
272 if ((error = xfs_imap(mp, tp, ip->i_ino, &imap,
273 XFS_IMAP_LOOKUP | imap_flags)))
277 * If the inode number maps to a block outside the bounds
278 * of the file system then return NULL rather than calling
279 * read_buf and panicing when we get an error from the
282 if ((imap.im_blkno + imap.im_len) >
283 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
285 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
286 "(imap.im_blkno (0x%llx) "
287 "+ imap.im_len (0x%llx)) > "
288 " XFS_FSB_TO_BB(mp, "
289 "mp->m_sb.sb_dblocks) (0x%llx)",
290 (unsigned long long) imap.im_blkno,
291 (unsigned long long) imap.im_len,
292 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
294 return XFS_ERROR(EINVAL);
298 * Fill in the fields in the inode that will be used to
299 * map the inode to its buffer from now on.
301 ip->i_blkno = imap.im_blkno;
302 ip->i_len = imap.im_len;
303 ip->i_boffset = imap.im_boffset;
306 * We've already mapped the inode once, so just use the
307 * mapping that we saved the first time.
309 imap.im_blkno = ip->i_blkno;
310 imap.im_len = ip->i_len;
311 imap.im_boffset = ip->i_boffset;
313 ASSERT(bno == 0 || bno == imap.im_blkno);
316 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
317 * default to just a read_buf() call.
319 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
320 (int)imap.im_len, XFS_BUF_LOCK, &bp);
324 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
325 "xfs_trans_read_buf() returned error %d, "
326 "imap.im_blkno 0x%llx, imap.im_len 0x%llx",
327 error, (unsigned long long) imap.im_blkno,
328 (unsigned long long) imap.im_len);
334 * Validate the magic number and version of every inode in the buffer
335 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
338 ni = (imap_flags & XFS_IMAP_BULKSTAT) ? 0 :
339 (BBTOB(imap.im_len) >> mp->m_sb.sb_inodelog);
341 ni = (imap_flags & XFS_IMAP_BULKSTAT) ? 0 : 1;
343 for (i = 0; i < ni; i++) {
347 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
348 (i << mp->m_sb.sb_inodelog));
349 di_ok = INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
350 XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
351 if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP,
352 XFS_RANDOM_ITOBP_INOTOBP))) {
354 prdev("bad inode magic/vsn daddr %lld #%d (magic=%x)",
356 (unsigned long long)imap.im_blkno, i,
357 INT_GET(dip->di_core.di_magic, ARCH_CONVERT));
359 XFS_CORRUPTION_ERROR("xfs_itobp", XFS_ERRLEVEL_HIGH,
361 xfs_trans_brelse(tp, bp);
362 return XFS_ERROR(EFSCORRUPTED);
365 #endif /* __KERNEL__ */
367 xfs_inobp_check(mp, bp);
370 * Mark the buffer as an inode buffer now that it looks good
372 XFS_BUF_SET_VTYPE(bp, B_FS_INO);
375 * Set *dipp to point to the on-disk inode in the buffer.
377 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
383 * Move inode type and inode format specific information from the
384 * on-disk inode to the in-core inode. For fifos, devs, and sockets
385 * this means set if_rdev to the proper value. For files, directories,
386 * and symlinks this means to bring in the in-line data or extent
387 * pointers. For a file in B-tree format, only the root is immediately
388 * brought in-core. The rest will be in-lined in if_extents when it
389 * is first referenced (see xfs_iread_extents()).
396 xfs_attr_shortform_t *atp;
400 ip->i_df.if_ext_max =
401 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
405 INT_GET(dip->di_core.di_nextents, ARCH_CONVERT) +
406 INT_GET(dip->di_core.di_anextents, ARCH_CONVERT) >
407 INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT))) {
408 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
409 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
410 (unsigned long long)ip->i_ino,
411 (int)(INT_GET(dip->di_core.di_nextents, ARCH_CONVERT)
412 + INT_GET(dip->di_core.di_anextents, ARCH_CONVERT)),
414 INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT));
415 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
417 return XFS_ERROR(EFSCORRUPTED);
420 if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) {
421 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
422 "corrupt dinode %Lu, forkoff = 0x%x.",
423 (unsigned long long)ip->i_ino,
424 (int)(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT)));
425 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
427 return XFS_ERROR(EFSCORRUPTED);
430 switch (ip->i_d.di_mode & S_IFMT) {
435 if (unlikely(INT_GET(dip->di_core.di_format, ARCH_CONVERT) != XFS_DINODE_FMT_DEV)) {
436 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
438 return XFS_ERROR(EFSCORRUPTED);
441 ip->i_df.if_u2.if_rdev = INT_GET(dip->di_u.di_dev, ARCH_CONVERT);
447 switch (INT_GET(dip->di_core.di_format, ARCH_CONVERT)) {
448 case XFS_DINODE_FMT_LOCAL:
450 * no local regular files yet
452 if (unlikely((INT_GET(dip->di_core.di_mode, ARCH_CONVERT) & S_IFMT) == S_IFREG)) {
453 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
455 "(local format for regular file).",
456 (unsigned long long) ip->i_ino);
457 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
460 return XFS_ERROR(EFSCORRUPTED);
463 di_size = INT_GET(dip->di_core.di_size, ARCH_CONVERT);
464 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
465 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
467 "(bad size %Ld for local inode).",
468 (unsigned long long) ip->i_ino,
469 (long long) di_size);
470 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
473 return XFS_ERROR(EFSCORRUPTED);
477 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
479 case XFS_DINODE_FMT_EXTENTS:
480 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
482 case XFS_DINODE_FMT_BTREE:
483 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
486 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
488 return XFS_ERROR(EFSCORRUPTED);
493 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
494 return XFS_ERROR(EFSCORRUPTED);
499 if (!XFS_DFORK_Q(dip))
501 ASSERT(ip->i_afp == NULL);
502 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
503 ip->i_afp->if_ext_max =
504 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
505 switch (INT_GET(dip->di_core.di_aformat, ARCH_CONVERT)) {
506 case XFS_DINODE_FMT_LOCAL:
507 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
508 size = be16_to_cpu(atp->hdr.totsize);
509 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
511 case XFS_DINODE_FMT_EXTENTS:
512 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
514 case XFS_DINODE_FMT_BTREE:
515 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
518 error = XFS_ERROR(EFSCORRUPTED);
522 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
524 xfs_idestroy_fork(ip, XFS_DATA_FORK);
530 * The file is in-lined in the on-disk inode.
531 * If it fits into if_inline_data, then copy
532 * it there, otherwise allocate a buffer for it
533 * and copy the data there. Either way, set
534 * if_data to point at the data.
535 * If we allocate a buffer for the data, make
536 * sure that its size is a multiple of 4 and
537 * record the real size in i_real_bytes.
550 * If the size is unreasonable, then something
551 * is wrong and we just bail out rather than crash in
552 * kmem_alloc() or memcpy() below.
554 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
555 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
557 "(bad size %d for local fork, size = %d).",
558 (unsigned long long) ip->i_ino, size,
559 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
560 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
562 return XFS_ERROR(EFSCORRUPTED);
564 ifp = XFS_IFORK_PTR(ip, whichfork);
567 ifp->if_u1.if_data = NULL;
568 else if (size <= sizeof(ifp->if_u2.if_inline_data))
569 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
571 real_size = roundup(size, 4);
572 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
574 ifp->if_bytes = size;
575 ifp->if_real_bytes = real_size;
577 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
578 ifp->if_flags &= ~XFS_IFEXTENTS;
579 ifp->if_flags |= XFS_IFINLINE;
584 * The file consists of a set of extents all
585 * of which fit into the on-disk inode.
586 * If there are few enough extents to fit into
587 * the if_inline_ext, then copy them there.
588 * Otherwise allocate a buffer for them and copy
589 * them into it. Either way, set if_extents
590 * to point at the extents.
598 xfs_bmbt_rec_t *ep, *dp;
604 ifp = XFS_IFORK_PTR(ip, whichfork);
605 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
606 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
609 * If the number of extents is unreasonable, then something
610 * is wrong and we just bail out rather than crash in
611 * kmem_alloc() or memcpy() below.
613 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
614 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
615 "corrupt inode %Lu ((a)extents = %d).",
616 (unsigned long long) ip->i_ino, nex);
617 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
619 return XFS_ERROR(EFSCORRUPTED);
622 ifp->if_real_bytes = 0;
624 ifp->if_u1.if_extents = NULL;
625 else if (nex <= XFS_INLINE_EXTS)
626 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
628 xfs_iext_add(ifp, 0, nex);
630 ifp->if_bytes = size;
632 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
633 xfs_validate_extents(ifp, nex, 1, XFS_EXTFMT_INODE(ip));
634 for (i = 0; i < nex; i++, dp++) {
635 ep = xfs_iext_get_ext(ifp, i);
636 ep->l0 = INT_GET(get_unaligned((__uint64_t*)&dp->l0),
638 ep->l1 = INT_GET(get_unaligned((__uint64_t*)&dp->l1),
641 xfs_bmap_trace_exlist("xfs_iformat_extents", ip, nex,
643 if (whichfork != XFS_DATA_FORK ||
644 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
645 if (unlikely(xfs_check_nostate_extents(
647 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
650 return XFS_ERROR(EFSCORRUPTED);
653 ifp->if_flags |= XFS_IFEXTENTS;
658 * The file has too many extents to fit into
659 * the inode, so they are in B-tree format.
660 * Allocate a buffer for the root of the B-tree
661 * and copy the root into it. The i_extents
662 * field will remain NULL until all of the
663 * extents are read in (when they are needed).
671 xfs_bmdr_block_t *dfp;
677 ifp = XFS_IFORK_PTR(ip, whichfork);
678 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
679 size = XFS_BMAP_BROOT_SPACE(dfp);
680 nrecs = XFS_BMAP_BROOT_NUMRECS(dfp);
683 * blow out if -- fork has less extents than can fit in
684 * fork (fork shouldn't be a btree format), root btree
685 * block has more records than can fit into the fork,
686 * or the number of extents is greater than the number of
689 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
690 || XFS_BMDR_SPACE_CALC(nrecs) >
691 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
692 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
693 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
694 "corrupt inode %Lu (btree).",
695 (unsigned long long) ip->i_ino);
696 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
698 return XFS_ERROR(EFSCORRUPTED);
701 ifp->if_broot_bytes = size;
702 ifp->if_broot = kmem_alloc(size, KM_SLEEP);
703 ASSERT(ifp->if_broot != NULL);
705 * Copy and convert from the on-disk structure
706 * to the in-memory structure.
708 xfs_bmdr_to_bmbt(dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
709 ifp->if_broot, size);
710 ifp->if_flags &= ~XFS_IFEXTENTS;
711 ifp->if_flags |= XFS_IFBROOT;
717 * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
720 * buf = on-disk representation
721 * dip = native representation
722 * dir = direction - +ve -> disk to native
723 * -ve -> native to disk
726 xfs_xlate_dinode_core(
728 xfs_dinode_core_t *dip,
731 xfs_dinode_core_t *buf_core = (xfs_dinode_core_t *)buf;
732 xfs_dinode_core_t *mem_core = (xfs_dinode_core_t *)dip;
733 xfs_arch_t arch = ARCH_CONVERT;
737 INT_XLATE(buf_core->di_magic, mem_core->di_magic, dir, arch);
738 INT_XLATE(buf_core->di_mode, mem_core->di_mode, dir, arch);
739 INT_XLATE(buf_core->di_version, mem_core->di_version, dir, arch);
740 INT_XLATE(buf_core->di_format, mem_core->di_format, dir, arch);
741 INT_XLATE(buf_core->di_onlink, mem_core->di_onlink, dir, arch);
742 INT_XLATE(buf_core->di_uid, mem_core->di_uid, dir, arch);
743 INT_XLATE(buf_core->di_gid, mem_core->di_gid, dir, arch);
744 INT_XLATE(buf_core->di_nlink, mem_core->di_nlink, dir, arch);
745 INT_XLATE(buf_core->di_projid, mem_core->di_projid, dir, arch);
748 memcpy(mem_core->di_pad, buf_core->di_pad,
749 sizeof(buf_core->di_pad));
751 memcpy(buf_core->di_pad, mem_core->di_pad,
752 sizeof(buf_core->di_pad));
755 INT_XLATE(buf_core->di_flushiter, mem_core->di_flushiter, dir, arch);
757 INT_XLATE(buf_core->di_atime.t_sec, mem_core->di_atime.t_sec,
759 INT_XLATE(buf_core->di_atime.t_nsec, mem_core->di_atime.t_nsec,
761 INT_XLATE(buf_core->di_mtime.t_sec, mem_core->di_mtime.t_sec,
763 INT_XLATE(buf_core->di_mtime.t_nsec, mem_core->di_mtime.t_nsec,
765 INT_XLATE(buf_core->di_ctime.t_sec, mem_core->di_ctime.t_sec,
767 INT_XLATE(buf_core->di_ctime.t_nsec, mem_core->di_ctime.t_nsec,
769 INT_XLATE(buf_core->di_size, mem_core->di_size, dir, arch);
770 INT_XLATE(buf_core->di_nblocks, mem_core->di_nblocks, dir, arch);
771 INT_XLATE(buf_core->di_extsize, mem_core->di_extsize, dir, arch);
772 INT_XLATE(buf_core->di_nextents, mem_core->di_nextents, dir, arch);
773 INT_XLATE(buf_core->di_anextents, mem_core->di_anextents, dir, arch);
774 INT_XLATE(buf_core->di_forkoff, mem_core->di_forkoff, dir, arch);
775 INT_XLATE(buf_core->di_aformat, mem_core->di_aformat, dir, arch);
776 INT_XLATE(buf_core->di_dmevmask, mem_core->di_dmevmask, dir, arch);
777 INT_XLATE(buf_core->di_dmstate, mem_core->di_dmstate, dir, arch);
778 INT_XLATE(buf_core->di_flags, mem_core->di_flags, dir, arch);
779 INT_XLATE(buf_core->di_gen, mem_core->di_gen, dir, arch);
784 xfs_dinode_core_t *dic,
789 if (di_flags & XFS_DIFLAG_ANY) {
790 if (di_flags & XFS_DIFLAG_REALTIME)
791 flags |= XFS_XFLAG_REALTIME;
792 if (di_flags & XFS_DIFLAG_PREALLOC)
793 flags |= XFS_XFLAG_PREALLOC;
794 if (di_flags & XFS_DIFLAG_IMMUTABLE)
795 flags |= XFS_XFLAG_IMMUTABLE;
796 if (di_flags & XFS_DIFLAG_APPEND)
797 flags |= XFS_XFLAG_APPEND;
798 if (di_flags & XFS_DIFLAG_SYNC)
799 flags |= XFS_XFLAG_SYNC;
800 if (di_flags & XFS_DIFLAG_NOATIME)
801 flags |= XFS_XFLAG_NOATIME;
802 if (di_flags & XFS_DIFLAG_NODUMP)
803 flags |= XFS_XFLAG_NODUMP;
804 if (di_flags & XFS_DIFLAG_RTINHERIT)
805 flags |= XFS_XFLAG_RTINHERIT;
806 if (di_flags & XFS_DIFLAG_PROJINHERIT)
807 flags |= XFS_XFLAG_PROJINHERIT;
808 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
809 flags |= XFS_XFLAG_NOSYMLINKS;
810 if (di_flags & XFS_DIFLAG_EXTSIZE)
811 flags |= XFS_XFLAG_EXTSIZE;
812 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
813 flags |= XFS_XFLAG_EXTSZINHERIT;
823 xfs_dinode_core_t *dic = &ip->i_d;
825 return _xfs_dic2xflags(dic, dic->di_flags) |
826 (XFS_CFORK_Q(dic) ? XFS_XFLAG_HASATTR : 0);
831 xfs_dinode_core_t *dic)
833 return _xfs_dic2xflags(dic, INT_GET(dic->di_flags, ARCH_CONVERT)) |
834 (XFS_CFORK_Q_DISK(dic) ? XFS_XFLAG_HASATTR : 0);
838 * Given a mount structure and an inode number, return a pointer
839 * to a newly allocated in-core inode corresponding to the given
842 * Initialize the inode's attributes and extent pointers if it
843 * already has them (it will not if the inode has no links).
858 ASSERT(xfs_inode_zone != NULL);
860 ip = kmem_zone_zalloc(xfs_inode_zone, KM_SLEEP);
865 * Get pointer's to the on-disk inode and the buffer containing it.
866 * If the inode number refers to a block outside the file system
867 * then xfs_itobp() will return NULL. In this case we should
868 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
869 * know that this is a new incore inode.
871 error = xfs_itobp(mp, tp, ip, &dip, &bp, bno, 0);
873 kmem_zone_free(xfs_inode_zone, ip);
878 * Initialize inode's trace buffers.
879 * Do this before xfs_iformat in case it adds entries.
881 #ifdef XFS_BMAP_TRACE
882 ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_SLEEP);
884 #ifdef XFS_BMBT_TRACE
885 ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_SLEEP);
888 ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_SLEEP);
890 #ifdef XFS_ILOCK_TRACE
891 ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_SLEEP);
893 #ifdef XFS_DIR2_TRACE
894 ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_SLEEP);
898 * If we got something that isn't an inode it means someone
899 * (nfs or dmi) has a stale handle.
901 if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC) {
902 kmem_zone_free(xfs_inode_zone, ip);
903 xfs_trans_brelse(tp, bp);
905 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
906 "dip->di_core.di_magic (0x%x) != "
907 "XFS_DINODE_MAGIC (0x%x)",
908 INT_GET(dip->di_core.di_magic, ARCH_CONVERT),
911 return XFS_ERROR(EINVAL);
915 * If the on-disk inode is already linked to a directory
916 * entry, copy all of the inode into the in-core inode.
917 * xfs_iformat() handles copying in the inode format
918 * specific information.
919 * Otherwise, just get the truly permanent information.
921 if (dip->di_core.di_mode) {
922 xfs_xlate_dinode_core((xfs_caddr_t)&dip->di_core,
924 error = xfs_iformat(ip, dip);
926 kmem_zone_free(xfs_inode_zone, ip);
927 xfs_trans_brelse(tp, bp);
929 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
930 "xfs_iformat() returned error %d",
936 ip->i_d.di_magic = INT_GET(dip->di_core.di_magic, ARCH_CONVERT);
937 ip->i_d.di_version = INT_GET(dip->di_core.di_version, ARCH_CONVERT);
938 ip->i_d.di_gen = INT_GET(dip->di_core.di_gen, ARCH_CONVERT);
939 ip->i_d.di_flushiter = INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT);
941 * Make sure to pull in the mode here as well in
942 * case the inode is released without being used.
943 * This ensures that xfs_inactive() will see that
944 * the inode is already free and not try to mess
945 * with the uninitialized part of it.
949 * Initialize the per-fork minima and maxima for a new
950 * inode here. xfs_iformat will do it for old inodes.
952 ip->i_df.if_ext_max =
953 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
957 INIT_LIST_HEAD(&ip->i_reclaim);
959 bzero(&ip->i_reclaim,sizeof(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 = curthread->td_ucred->cr_uid;
1127 ip->i_d.di_gid = curthread->td_ucred->cr_groups[0];
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 (!groupmember((gid_t)ip->i_d.di_gid, curthread->td_ucred))) {
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 ip->i_d.di_flags |= di_flags;
1232 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1233 ip->i_df.if_flags = XFS_IFEXTENTS;
1234 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1235 ip->i_df.if_u1.if_extents = NULL;
1241 * Attribute fork settings for new inode.
1243 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1244 ip->i_d.di_anextents = 0;
1247 * Log the new values stuffed into the inode.
1249 xfs_trans_log_inode(tp, ip, flags);
1251 /* now that we have an i_mode we can set Linux inode ops (& unlock) */
1252 XVFS_INIT_VNODE(XFS_MTOVFS(tp->t_mountp), vp, XFS_ITOBHV(ip), 1);
1259 * Check to make sure that there are no blocks allocated to the
1260 * file beyond the size of the file. We don't check this for
1261 * files with fixed size extents or real time extents, but we
1262 * at least do it for regular files.
1271 xfs_fileoff_t map_first;
1273 xfs_bmbt_irec_t imaps[2];
1275 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1278 if (ip->i_d.di_flags & (XFS_DIFLAG_REALTIME | XFS_DIFLAG_EXTSIZE))
1282 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1284 * The filesystem could be shutting down, so bmapi may return
1287 if (xfs_bmapi(NULL, ip, map_first,
1289 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1291 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1294 ASSERT(nimaps == 1);
1295 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1300 * Calculate the last possible buffered byte in a file. This must
1301 * include data that was buffered beyond the EOF by the write code.
1302 * This also needs to deal with overflowing the xfs_fsize_t type
1303 * which can happen for sizes near the limit.
1305 * We also need to take into account any blocks beyond the EOF. It
1306 * may be the case that they were buffered by a write which failed.
1307 * In that case the pages will still be in memory, but the inode size
1308 * will never have been updated.
1315 xfs_fsize_t last_byte;
1316 xfs_fileoff_t last_block;
1317 xfs_fileoff_t size_last_block;
1320 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE | MR_ACCESS));
1324 * Only check for blocks beyond the EOF if the extents have
1325 * been read in. This eliminates the need for the inode lock,
1326 * and it also saves us from looking when it really isn't
1329 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1330 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1338 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_d.di_size);
1339 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1341 last_byte = XFS_FSB_TO_B(mp, last_block);
1342 if (last_byte < 0) {
1343 return XFS_MAXIOFFSET(mp);
1345 last_byte += (1 << mp->m_writeio_log);
1346 if (last_byte < 0) {
1347 return XFS_MAXIOFFSET(mp);
1352 #if defined(XFS_RW_TRACE)
1358 xfs_fsize_t new_size,
1359 xfs_off_t toss_start,
1360 xfs_off_t toss_finish)
1362 if (ip->i_rwtrace == NULL) {
1366 ktrace_enter(ip->i_rwtrace,
1369 (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
1370 (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
1371 (void*)((long)flag),
1372 (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
1373 (void*)(unsigned long)(new_size & 0xffffffff),
1374 (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
1375 (void*)(unsigned long)(toss_start & 0xffffffff),
1376 (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
1377 (void*)(unsigned long)(toss_finish & 0xffffffff),
1378 (void*)(unsigned long)current_cpu(),
1379 (void*)(unsigned long)current_pid(),
1385 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1389 * Start the truncation of the file to new_size. The new size
1390 * must be smaller than the current size. This routine will
1391 * clear the buffer and page caches of file data in the removed
1392 * range, and xfs_itruncate_finish() will remove the underlying
1395 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1396 * must NOT have the inode lock held at all. This is because we're
1397 * calling into the buffer/page cache code and we can't hold the
1398 * inode lock when we do so.
1400 * We need to wait for any direct I/Os in flight to complete before we
1401 * proceed with the truncate. This is needed to prevent the extents
1402 * being read or written by the direct I/Os from being removed while the
1403 * I/O is in flight as there is no other method of synchronising
1404 * direct I/O with the truncate operation. Also, because we hold
1405 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1406 * started until the truncate completes and drops the lock. Essentially,
1407 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1408 * between direct I/Os and the truncate operation.
1410 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1411 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1412 * in the case that the caller is locking things out of order and
1413 * may not be able to call xfs_itruncate_finish() with the inode lock
1414 * held without dropping the I/O lock. If the caller must drop the
1415 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1416 * must be called again with all the same restrictions as the initial
1420 xfs_itruncate_start(
1423 xfs_fsize_t new_size)
1425 xfs_fsize_t last_byte;
1426 xfs_off_t toss_start;
1430 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1431 ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
1432 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1433 (flags == XFS_ITRUNC_MAYBE));
1438 vn_iowait(vp); /* wait for the completion of any pending DIOs */
1441 * Call VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES() to get rid of pages and buffers
1442 * overlapping the region being removed. We have to use
1443 * the less efficient VOP_FLUSHINVAL_PAGES() in the case that the
1444 * caller may not be able to finish the truncate without
1445 * dropping the inode's I/O lock. Make sure
1446 * to catch any pages brought in by buffers overlapping
1447 * the EOF by searching out beyond the isize by our
1448 * block size. We round new_size up to a block boundary
1449 * so that we don't toss things on the same block as
1450 * new_size but before it.
1452 * Before calling VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES(), make sure to
1453 * call remapf() over the same region if the file is mapped.
1454 * This frees up mapped file references to the pages in the
1455 * given range and for the VOP_FLUSHINVAL_PAGES() case it ensures
1456 * that we get the latest mapped changes flushed out.
1458 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1459 toss_start = XFS_FSB_TO_B(mp, toss_start);
1460 if (toss_start < 0) {
1462 * The place to start tossing is beyond our maximum
1463 * file size, so there is no way that the data extended
1468 last_byte = xfs_file_last_byte(ip);
1469 xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
1471 if (last_byte > toss_start) {
1472 if (flags & XFS_ITRUNC_DEFINITE) {
1473 XVOP_TOSS_PAGES(vp, toss_start, -1, FI_REMAPF_LOCKED);
1475 XVOP_FLUSHINVAL_PAGES(vp, toss_start, -1, FI_REMAPF_LOCKED);
1480 if (new_size == 0) {
1481 ASSERT(VN_CACHED(vp) == 0);
1487 * Shrink the file to the given new_size. The new
1488 * size must be smaller than the current size.
1489 * This will free up the underlying blocks
1490 * in the removed range after a call to xfs_itruncate_start()
1491 * or xfs_atruncate_start().
1493 * The transaction passed to this routine must have made
1494 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1495 * This routine may commit the given transaction and
1496 * start new ones, so make sure everything involved in
1497 * the transaction is tidy before calling here.
1498 * Some transaction will be returned to the caller to be
1499 * committed. The incoming transaction must already include
1500 * the inode, and both inode locks must be held exclusively.
1501 * The inode must also be "held" within the transaction. On
1502 * return the inode will be "held" within the returned transaction.
1503 * This routine does NOT require any disk space to be reserved
1504 * for it within the transaction.
1506 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1507 * and it indicates the fork which is to be truncated. For the
1508 * attribute fork we only support truncation to size 0.
1510 * We use the sync parameter to indicate whether or not the first
1511 * transaction we perform might have to be synchronous. For the attr fork,
1512 * it needs to be so if the unlink of the inode is not yet known to be
1513 * permanent in the log. This keeps us from freeing and reusing the
1514 * blocks of the attribute fork before the unlink of the inode becomes
1517 * For the data fork, we normally have to run synchronously if we're
1518 * being called out of the inactive path or we're being called
1519 * out of the create path where we're truncating an existing file.
1520 * Either way, the truncate needs to be sync so blocks don't reappear
1521 * in the file with altered data in case of a crash. wsync filesystems
1522 * can run the first case async because anything that shrinks the inode
1523 * has to run sync so by the time we're called here from inactive, the
1524 * inode size is permanently set to 0.
1526 * Calls from the truncate path always need to be sync unless we're
1527 * in a wsync filesystem and the file has already been unlinked.
1529 * The caller is responsible for correctly setting the sync parameter.
1530 * It gets too hard for us to guess here which path we're being called
1531 * out of just based on inode state.
1534 xfs_itruncate_finish(
1537 xfs_fsize_t new_size,
1541 xfs_fsblock_t first_block;
1542 xfs_fileoff_t first_unmap_block;
1543 xfs_fileoff_t last_block;
1544 xfs_filblks_t unmap_len=0;
1549 xfs_bmap_free_t free_list;
1552 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1553 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
1554 ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
1555 ASSERT(*tp != NULL);
1556 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1557 ASSERT(ip->i_transp == *tp);
1558 ASSERT(ip->i_itemp != NULL);
1559 ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
1563 mp = (ntp)->t_mountp;
1564 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1567 * We only support truncating the entire attribute fork.
1569 if (fork == XFS_ATTR_FORK) {
1572 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1573 xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
1575 * The first thing we do is set the size to new_size permanently
1576 * on disk. This way we don't have to worry about anyone ever
1577 * being able to look at the data being freed even in the face
1578 * of a crash. What we're getting around here is the case where
1579 * we free a block, it is allocated to another file, it is written
1580 * to, and then we crash. If the new data gets written to the
1581 * file but the log buffers containing the free and reallocation
1582 * don't, then we'd end up with garbage in the blocks being freed.
1583 * As long as we make the new_size permanent before actually
1584 * freeing any blocks it doesn't matter if they get writtten to.
1586 * The callers must signal into us whether or not the size
1587 * setting here must be synchronous. There are a few cases
1588 * where it doesn't have to be synchronous. Those cases
1589 * occur if the file is unlinked and we know the unlink is
1590 * permanent or if the blocks being truncated are guaranteed
1591 * to be beyond the inode eof (regardless of the link count)
1592 * and the eof value is permanent. Both of these cases occur
1593 * only on wsync-mounted filesystems. In those cases, we're
1594 * guaranteed that no user will ever see the data in the blocks
1595 * that are being truncated so the truncate can run async.
1596 * In the free beyond eof case, the file may wind up with
1597 * more blocks allocated to it than it needs if we crash
1598 * and that won't get fixed until the next time the file
1599 * is re-opened and closed but that's ok as that shouldn't
1600 * be too many blocks.
1602 * However, we can't just make all wsync xactions run async
1603 * because there's one call out of the create path that needs
1604 * to run sync where it's truncating an existing file to size
1605 * 0 whose size is > 0.
1607 * It's probably possible to come up with a test in this
1608 * routine that would correctly distinguish all the above
1609 * cases from the values of the function parameters and the
1610 * inode state but for sanity's sake, I've decided to let the
1611 * layers above just tell us. It's simpler to correctly figure
1612 * out in the layer above exactly under what conditions we
1613 * can run async and I think it's easier for others read and
1614 * follow the logic in case something has to be changed.
1615 * cscope is your friend -- rcc.
1617 * The attribute fork is much simpler.
1619 * For the attribute fork we allow the caller to tell us whether
1620 * the unlink of the inode that led to this call is yet permanent
1621 * in the on disk log. If it is not and we will be freeing extents
1622 * in this inode then we make the first transaction synchronous
1623 * to make sure that the unlink is permanent by the time we free
1626 if (fork == XFS_DATA_FORK) {
1627 if (ip->i_d.di_nextents > 0) {
1628 ip->i_d.di_size = new_size;
1629 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1632 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1633 if (ip->i_d.di_anextents > 0)
1634 xfs_trans_set_sync(ntp);
1636 ASSERT(fork == XFS_DATA_FORK ||
1637 (fork == XFS_ATTR_FORK &&
1638 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1639 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1642 * Since it is possible for space to become allocated beyond
1643 * the end of the file (in a crash where the space is allocated
1644 * but the inode size is not yet updated), simply remove any
1645 * blocks which show up between the new EOF and the maximum
1646 * possible file size. If the first block to be removed is
1647 * beyond the maximum file size (ie it is the same as last_block),
1648 * then there is nothing to do.
1650 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1651 ASSERT(first_unmap_block <= last_block);
1653 if (last_block == first_unmap_block) {
1656 unmap_len = last_block - first_unmap_block + 1;
1660 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1661 * will tell us whether it freed the entire range or
1662 * not. If this is a synchronous mount (wsync),
1663 * then we can tell bunmapi to keep all the
1664 * transactions asynchronous since the unlink
1665 * transaction that made this inode inactive has
1666 * already hit the disk. There's no danger of
1667 * the freed blocks being reused, there being a
1668 * crash, and the reused blocks suddenly reappearing
1669 * in this file with garbage in them once recovery
1672 XFS_BMAP_INIT(&free_list, &first_block);
1673 error = XFS_BUNMAPI(mp, ntp, &ip->i_iocore,
1674 first_unmap_block, unmap_len,
1675 XFS_BMAPI_AFLAG(fork) |
1676 (sync ? 0 : XFS_BMAPI_ASYNC),
1677 XFS_ITRUNC_MAX_EXTENTS,
1678 &first_block, &free_list,
1682 * If the bunmapi call encounters an error,
1683 * return to the caller where the transaction
1684 * can be properly aborted. We just need to
1685 * make sure we're not holding any resources
1686 * that we were not when we came in.
1688 xfs_bmap_cancel(&free_list);
1693 * Duplicate the transaction that has the permanent
1694 * reservation and commit the old transaction.
1696 error = xfs_bmap_finish(tp, &free_list, first_block,
1701 * If the bmap finish call encounters an error,
1702 * return to the caller where the transaction
1703 * can be properly aborted. We just need to
1704 * make sure we're not holding any resources
1705 * that we were not when we came in.
1707 * Aborting from this point might lose some
1708 * blocks in the file system, but oh well.
1710 xfs_bmap_cancel(&free_list);
1713 * If the passed in transaction committed
1714 * in xfs_bmap_finish(), then we want to
1715 * add the inode to this one before returning.
1716 * This keeps things simple for the higher
1717 * level code, because it always knows that
1718 * the inode is locked and held in the
1719 * transaction that returns to it whether
1720 * errors occur or not. We don't mark the
1721 * inode dirty so that this transaction can
1722 * be easily aborted if possible.
1724 xfs_trans_ijoin(ntp, ip,
1725 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1726 xfs_trans_ihold(ntp, ip);
1733 * The first xact was committed,
1734 * so add the inode to the new one.
1735 * Mark it dirty so it will be logged
1736 * and moved forward in the log as
1737 * part of every commit.
1739 xfs_trans_ijoin(ntp, ip,
1740 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1741 xfs_trans_ihold(ntp, ip);
1742 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1744 ntp = xfs_trans_dup(ntp);
1745 (void) xfs_trans_commit(*tp, 0, NULL);
1747 error = xfs_trans_reserve(ntp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1748 XFS_TRANS_PERM_LOG_RES,
1749 XFS_ITRUNCATE_LOG_COUNT);
1751 * Add the inode being truncated to the next chained
1754 xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1755 xfs_trans_ihold(ntp, ip);
1760 * Only update the size in the case of the data fork, but
1761 * always re-log the inode so that our permanent transaction
1762 * can keep on rolling it forward in the log.
1764 if (fork == XFS_DATA_FORK) {
1765 xfs_isize_check(mp, ip, new_size);
1766 ip->i_d.di_size = new_size;
1768 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1769 ASSERT((new_size != 0) ||
1770 (fork == XFS_ATTR_FORK) ||
1771 (ip->i_delayed_blks == 0));
1772 ASSERT((new_size != 0) ||
1773 (fork == XFS_ATTR_FORK) ||
1774 (ip->i_d.di_nextents == 0));
1775 xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
1783 * Do the first part of growing a file: zero any data in the last
1784 * block that is beyond the old EOF. We need to do this before
1785 * the inode is joined to the transaction to modify the i_size.
1786 * That way we can drop the inode lock and call into the buffer
1787 * cache to get the buffer mapping the EOF.
1792 xfs_fsize_t new_size,
1797 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1798 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1799 ASSERT(new_size > ip->i_d.di_size);
1802 * Zero any pages that may have been created by
1803 * xfs_write_file() beyond the end of the file
1804 * and any blocks between the old and new file sizes.
1806 error = xfs_zero_eof(XFS_ITOV(ip), &ip->i_iocore, new_size,
1807 ip->i_d.di_size, new_size);
1814 * This routine is called to extend the size of a file.
1815 * The inode must have both the iolock and the ilock locked
1816 * for update and it must be a part of the current transaction.
1817 * The xfs_igrow_start() function must have been called previously.
1818 * If the change_flag is not zero, the inode change timestamp will
1825 xfs_fsize_t new_size,
1828 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1829 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1830 ASSERT(ip->i_transp == tp);
1831 ASSERT(new_size > ip->i_d.di_size);
1834 * Update the file size. Update the inode change timestamp
1835 * if change_flag set.
1837 ip->i_d.di_size = new_size;
1839 xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
1840 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1846 * This is called when the inode's link count goes to 0.
1847 * We place the on-disk inode on a list in the AGI. It
1848 * will be pulled from this list when the inode is freed.
1860 xfs_agnumber_t agno;
1861 xfs_daddr_t agdaddr;
1868 ASSERT(ip->i_d.di_nlink == 0);
1869 ASSERT(ip->i_d.di_mode != 0);
1870 ASSERT(ip->i_transp == tp);
1874 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1875 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1878 * Get the agi buffer first. It ensures lock ordering
1881 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1882 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1887 * Validate the magic number of the agi block.
1889 agi = XFS_BUF_TO_AGI(agibp);
1891 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1892 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
1893 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK,
1894 XFS_RANDOM_IUNLINK))) {
1895 XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi);
1896 xfs_trans_brelse(tp, agibp);
1897 return XFS_ERROR(EFSCORRUPTED);
1900 * Get the index into the agi hash table for the
1901 * list this inode will go on.
1903 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1905 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1906 ASSERT(agi->agi_unlinked[bucket_index]);
1907 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1909 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
1911 * There is already another inode in the bucket we need
1912 * to add ourselves to. Add us at the front of the list.
1913 * Here we put the head pointer into our next pointer,
1914 * and then we fall through to point the head at us.
1916 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
1920 ASSERT(INT_GET(dip->di_next_unlinked, ARCH_CONVERT) == NULLAGINO);
1921 ASSERT(dip->di_next_unlinked);
1922 /* both on-disk, don't endian flip twice */
1923 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1924 offset = ip->i_boffset +
1925 offsetof(xfs_dinode_t, di_next_unlinked);
1926 xfs_trans_inode_buf(tp, ibp);
1927 xfs_trans_log_buf(tp, ibp, offset,
1928 (offset + sizeof(xfs_agino_t) - 1));
1929 xfs_inobp_check(mp, ibp);
1933 * Point the bucket head pointer at the inode being inserted.
1936 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1937 offset = offsetof(xfs_agi_t, agi_unlinked) +
1938 (sizeof(xfs_agino_t) * bucket_index);
1939 xfs_trans_log_buf(tp, agibp, offset,
1940 (offset + sizeof(xfs_agino_t) - 1));
1945 * Pull the on-disk inode from the AGI unlinked list.
1958 xfs_agnumber_t agno;
1959 xfs_daddr_t agdaddr;
1961 xfs_agino_t next_agino;
1962 xfs_buf_t *last_ibp;
1963 xfs_dinode_t *last_dip = NULL;
1965 int offset, last_offset = 0;
1970 * First pull the on-disk inode from the AGI unlinked list.
1974 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1975 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1978 * Get the agi buffer first. It ensures lock ordering
1981 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1982 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1985 "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
1986 error, mp->m_fsname);
1990 * Validate the magic number of the agi block.
1992 agi = XFS_BUF_TO_AGI(agibp);
1994 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1995 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
1996 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE,
1997 XFS_RANDOM_IUNLINK_REMOVE))) {
1998 XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW,
2000 xfs_trans_brelse(tp, agibp);
2002 "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
2004 return XFS_ERROR(EFSCORRUPTED);
2007 * Get the index into the agi hash table for the
2008 * list this inode will go on.
2010 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2012 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2013 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
2014 ASSERT(agi->agi_unlinked[bucket_index]);
2016 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2018 * We're at the head of the list. Get the inode's
2019 * on-disk buffer to see if there is anyone after us
2020 * on the list. Only modify our next pointer if it
2021 * is not already NULLAGINO. This saves us the overhead
2022 * of dealing with the buffer when there is no need to
2025 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
2028 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2029 error, mp->m_fsname);
2032 next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
2033 ASSERT(next_agino != 0);
2034 if (next_agino != NULLAGINO) {
2035 INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
2036 offset = ip->i_boffset +
2037 offsetof(xfs_dinode_t, di_next_unlinked);
2038 xfs_trans_inode_buf(tp, ibp);
2039 xfs_trans_log_buf(tp, ibp, offset,
2040 (offset + sizeof(xfs_agino_t) - 1));
2041 xfs_inobp_check(mp, ibp);
2043 xfs_trans_brelse(tp, ibp);
2046 * Point the bucket head pointer at the next inode.
2048 ASSERT(next_agino != 0);
2049 ASSERT(next_agino != agino);
2050 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2051 offset = offsetof(xfs_agi_t, agi_unlinked) +
2052 (sizeof(xfs_agino_t) * bucket_index);
2053 xfs_trans_log_buf(tp, agibp, offset,
2054 (offset + sizeof(xfs_agino_t) - 1));
2057 * We need to search the list for the inode being freed.
2059 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2061 while (next_agino != agino) {
2063 * If the last inode wasn't the one pointing to
2064 * us, then release its buffer since we're not
2065 * going to do anything with it.
2067 if (last_ibp != NULL) {
2068 xfs_trans_brelse(tp, last_ibp);
2070 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2071 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
2072 &last_ibp, &last_offset);
2075 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
2076 error, mp->m_fsname);
2079 next_agino = INT_GET(last_dip->di_next_unlinked, ARCH_CONVERT);
2080 ASSERT(next_agino != NULLAGINO);
2081 ASSERT(next_agino != 0);
2084 * Now last_ibp points to the buffer previous to us on
2085 * the unlinked list. Pull us from the list.
2087 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
2090 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2091 error, mp->m_fsname);
2094 next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
2095 ASSERT(next_agino != 0);
2096 ASSERT(next_agino != agino);
2097 if (next_agino != NULLAGINO) {
2098 INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
2099 offset = ip->i_boffset +
2100 offsetof(xfs_dinode_t, di_next_unlinked);
2101 xfs_trans_inode_buf(tp, ibp);
2102 xfs_trans_log_buf(tp, ibp, offset,
2103 (offset + sizeof(xfs_agino_t) - 1));
2104 xfs_inobp_check(mp, ibp);
2106 xfs_trans_brelse(tp, ibp);
2109 * Point the previous inode on the list to the next inode.
2111 INT_SET(last_dip->di_next_unlinked, ARCH_CONVERT, next_agino);
2112 ASSERT(next_agino != 0);
2113 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2114 xfs_trans_inode_buf(tp, last_ibp);
2115 xfs_trans_log_buf(tp, last_ibp, offset,
2116 (offset + sizeof(xfs_agino_t) - 1));
2117 xfs_inobp_check(mp, last_ibp);
2122 static __inline__ int xfs_inode_clean(xfs_inode_t *ip)
2124 return (((ip->i_itemp == NULL) ||
2125 !(ip->i_itemp->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
2126 (ip->i_update_core == 0));
2131 xfs_inode_t *free_ip,
2135 xfs_mount_t *mp = free_ip->i_mount;
2136 int blks_per_cluster;
2139 int i, j, found, pre_flushed;
2143 xfs_inode_t *ip, **ip_found;
2144 xfs_inode_log_item_t *iip;
2145 xfs_log_item_t *lip;
2148 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
2149 blks_per_cluster = 1;
2150 ninodes = mp->m_sb.sb_inopblock;
2151 nbufs = XFS_IALLOC_BLOCKS(mp);
2153 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
2154 mp->m_sb.sb_blocksize;
2155 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
2156 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
2159 ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
2161 for (j = 0; j < nbufs; j++, inum += ninodes) {
2162 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2163 XFS_INO_TO_AGBNO(mp, inum));
2167 * Look for each inode in memory and attempt to lock it,
2168 * we can be racing with flush and tail pushing here.
2169 * any inode we get the locks on, add to an array of
2170 * inode items to process later.
2172 * The get the buffer lock, we could beat a flush
2173 * or tail pushing thread to the lock here, in which
2174 * case they will go looking for the inode buffer
2175 * and fail, we need some other form of interlock
2179 for (i = 0; i < ninodes; i++) {
2180 ih = XFS_IHASH(mp, inum + i);
2181 read_lock(&ih->ih_lock);
2182 for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
2183 if (ip->i_ino == inum + i)
2187 /* Inode not in memory or we found it already,
2190 if (!ip || (ip->i_flags & XFS_ISTALE)) {
2191 read_unlock(&ih->ih_lock);
2195 if (xfs_inode_clean(ip)) {
2196 read_unlock(&ih->ih_lock);
2200 /* If we can get the locks then add it to the
2201 * list, otherwise by the time we get the bp lock
2202 * below it will already be attached to the
2206 /* This inode will already be locked - by us, lets
2210 if (ip == free_ip) {
2211 if (xfs_iflock_nowait(ip)) {
2212 ip->i_flags |= XFS_ISTALE;
2214 if (xfs_inode_clean(ip)) {
2217 ip_found[found++] = ip;
2220 read_unlock(&ih->ih_lock);
2224 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2225 if (xfs_iflock_nowait(ip)) {
2226 ip->i_flags |= XFS_ISTALE;
2228 if (xfs_inode_clean(ip)) {
2230 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2232 ip_found[found++] = ip;
2235 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2239 read_unlock(&ih->ih_lock);
2242 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2243 mp->m_bsize * blks_per_cluster,
2247 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
2249 if (lip->li_type == XFS_LI_INODE) {
2250 iip = (xfs_inode_log_item_t *)lip;
2251 ASSERT(iip->ili_logged == 1);
2252 lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
2254 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2256 iip->ili_inode->i_flags |= XFS_ISTALE;
2259 lip = lip->li_bio_list;
2262 for (i = 0; i < found; i++) {
2267 ip->i_update_core = 0;
2269 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2273 iip->ili_last_fields = iip->ili_format.ilf_fields;
2274 iip->ili_format.ilf_fields = 0;
2275 iip->ili_logged = 1;
2277 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2280 xfs_buf_attach_iodone(bp,
2281 (void(*)(xfs_buf_t*,xfs_log_item_t*))
2282 xfs_istale_done, (xfs_log_item_t *)iip);
2283 if (ip != free_ip) {
2284 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2288 if (found || pre_flushed)
2289 xfs_trans_stale_inode_buf(tp, bp);
2290 xfs_trans_binval(tp, bp);
2293 kmem_free(ip_found, ninodes * sizeof(xfs_inode_t *));
2297 * This is called to return an inode to the inode free list.
2298 * The inode should already be truncated to 0 length and have
2299 * no pages associated with it. This routine also assumes that
2300 * the inode is already a part of the transaction.
2302 * The on-disk copy of the inode will have been added to the list
2303 * of unlinked inodes in the AGI. We need to remove the inode from
2304 * that list atomically with respect to freeing it here.
2310 xfs_bmap_free_t *flist)
2314 xfs_ino_t first_ino;
2316 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2317 ASSERT(ip->i_transp == tp);
2318 ASSERT(ip->i_d.di_nlink == 0);
2319 ASSERT(ip->i_d.di_nextents == 0);
2320 ASSERT(ip->i_d.di_anextents == 0);
2321 ASSERT((ip->i_d.di_size == 0) ||
2322 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2323 ASSERT(ip->i_d.di_nblocks == 0);
2326 * Pull the on-disk inode from the AGI unlinked list.
2328 error = xfs_iunlink_remove(tp, ip);
2333 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2337 ip->i_d.di_mode = 0; /* mark incore inode as free */
2338 ip->i_d.di_flags = 0;
2339 ip->i_d.di_dmevmask = 0;
2340 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2341 ip->i_df.if_ext_max =
2342 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2343 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2344 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2346 * Bump the generation count so no one will be confused
2347 * by reincarnations of this inode.
2350 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2353 xfs_ifree_cluster(ip, tp, first_ino);
2360 * Reallocate the space for if_broot based on the number of records
2361 * being added or deleted as indicated in rec_diff. Move the records
2362 * and pointers in if_broot to fit the new size. When shrinking this
2363 * will eliminate holes between the records and pointers created by
2364 * the caller. When growing this will create holes to be filled in
2367 * The caller must not request to add more records than would fit in
2368 * the on-disk inode root. If the if_broot is currently NULL, then
2369 * if we adding records one will be allocated. The caller must also
2370 * not request that the number of records go below zero, although
2371 * it can go to zero.
2373 * ip -- the inode whose if_broot area is changing
2374 * ext_diff -- the change in the number of records, positive or negative,
2375 * requested for the if_broot array.
2385 xfs_bmbt_block_t *new_broot;
2392 * Handle the degenerate case quietly.
2394 if (rec_diff == 0) {
2398 ifp = XFS_IFORK_PTR(ip, whichfork);
2401 * If there wasn't any memory allocated before, just
2402 * allocate it now and get out.
2404 if (ifp->if_broot_bytes == 0) {
2405 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
2406 ifp->if_broot = (xfs_bmbt_block_t*)kmem_alloc(new_size,
2408 ifp->if_broot_bytes = (int)new_size;
2413 * If there is already an existing if_broot, then we need
2414 * to realloc() it and shift the pointers to their new
2415 * location. The records don't change location because
2416 * they are kept butted up against the btree block header.
2418 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2419 new_max = cur_max + rec_diff;
2420 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2421 ifp->if_broot = (xfs_bmbt_block_t *)
2422 kmem_realloc(ifp->if_broot,
2424 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2426 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2427 ifp->if_broot_bytes);
2428 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2430 ifp->if_broot_bytes = (int)new_size;
2431 ASSERT(ifp->if_broot_bytes <=
2432 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2433 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2438 * rec_diff is less than 0. In this case, we are shrinking the
2439 * if_broot buffer. It must already exist. If we go to zero
2440 * records, just get rid of the root and clear the status bit.
2442 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
2443 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2444 new_max = cur_max + rec_diff;
2445 ASSERT(new_max >= 0);
2447 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2451 new_broot = (xfs_bmbt_block_t *)kmem_alloc(new_size, KM_SLEEP);
2453 * First copy over the btree block header.
2455 memcpy(new_broot, ifp->if_broot, sizeof(xfs_bmbt_block_t));
2458 ifp->if_flags &= ~XFS_IFBROOT;
2462 * Only copy the records and pointers if there are any.
2466 * First copy the records.
2468 op = (char *)XFS_BMAP_BROOT_REC_ADDR(ifp->if_broot, 1,
2469 ifp->if_broot_bytes);
2470 np = (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot, 1,
2472 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2475 * Then copy the pointers.
2477 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2478 ifp->if_broot_bytes);
2479 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot, 1,
2481 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2483 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2484 ifp->if_broot = new_broot;
2485 ifp->if_broot_bytes = (int)new_size;
2486 ASSERT(ifp->if_broot_bytes <=
2487 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2493 * This is called when the amount of space needed for if_data
2494 * is increased or decreased. The change in size is indicated by
2495 * the number of bytes that need to be added or deleted in the
2496 * byte_diff parameter.
2498 * If the amount of space needed has decreased below the size of the
2499 * inline buffer, then switch to using the inline buffer. Otherwise,
2500 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2501 * to what is needed.
2503 * ip -- the inode whose if_data area is changing
2504 * byte_diff -- the change in the number of bytes, positive or negative,
2505 * requested for the if_data array.
2517 if (byte_diff == 0) {
2521 ifp = XFS_IFORK_PTR(ip, whichfork);
2522 new_size = (int)ifp->if_bytes + byte_diff;
2523 ASSERT(new_size >= 0);
2525 if (new_size == 0) {
2526 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2527 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2529 ifp->if_u1.if_data = NULL;
2531 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2533 * If the valid extents/data can fit in if_inline_ext/data,
2534 * copy them from the malloc'd vector and free it.
2536 if (ifp->if_u1.if_data == NULL) {
2537 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2538 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2539 ASSERT(ifp->if_real_bytes != 0);
2540 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2542 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2543 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2548 * Stuck with malloc/realloc.
2549 * For inline data, the underlying buffer must be
2550 * a multiple of 4 bytes in size so that it can be
2551 * logged and stay on word boundaries. We enforce
2554 real_size = roundup(new_size, 4);
2555 if (ifp->if_u1.if_data == NULL) {
2556 ASSERT(ifp->if_real_bytes == 0);
2557 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2558 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2560 * Only do the realloc if the underlying size
2561 * is really changing.
2563 if (ifp->if_real_bytes != real_size) {
2564 ifp->if_u1.if_data =
2565 kmem_realloc(ifp->if_u1.if_data,
2571 ASSERT(ifp->if_real_bytes == 0);
2572 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2573 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2577 ifp->if_real_bytes = real_size;
2578 ifp->if_bytes = new_size;
2579 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2586 * Map inode to disk block and offset.
2588 * mp -- the mount point structure for the current file system
2589 * tp -- the current transaction
2590 * ino -- the inode number of the inode to be located
2591 * imap -- this structure is filled in with the information necessary
2592 * to retrieve the given inode from disk
2593 * flags -- flags to pass to xfs_dilocate indicating whether or not
2594 * lookups in the inode btree were OK or not
2604 xfs_fsblock_t fsbno;
2609 fsbno = imap->im_blkno ?
2610 XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK;
2611 error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags);
2615 imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno);
2616 imap->im_len = XFS_FSB_TO_BB(mp, len);
2617 imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno);
2618 imap->im_ioffset = (ushort)off;
2619 imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog);
2630 ifp = XFS_IFORK_PTR(ip, whichfork);
2631 if (ifp->if_broot != NULL) {
2632 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2633 ifp->if_broot = NULL;
2637 * If the format is local, then we can't have an extents
2638 * array so just look for an inline data array. If we're
2639 * not local then we may or may not have an extents list,
2640 * so check and free it up if we do.
2642 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2643 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2644 (ifp->if_u1.if_data != NULL)) {
2645 ASSERT(ifp->if_real_bytes != 0);
2646 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2647 ifp->if_u1.if_data = NULL;
2648 ifp->if_real_bytes = 0;
2650 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2651 ((ifp->if_flags & XFS_IFEXTIREC) ||
2652 ((ifp->if_u1.if_extents != NULL) &&
2653 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
2654 ASSERT(ifp->if_real_bytes != 0);
2655 xfs_iext_destroy(ifp);
2657 ASSERT(ifp->if_u1.if_extents == NULL ||
2658 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2659 ASSERT(ifp->if_real_bytes == 0);
2660 if (whichfork == XFS_ATTR_FORK) {
2661 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2667 * This is called free all the memory associated with an inode.
2668 * It must free the inode itself and any buffers allocated for
2669 * if_extents/if_data and if_broot. It must also free the lock
2670 * associated with the inode.
2677 switch (ip->i_d.di_mode & S_IFMT) {
2681 xfs_idestroy_fork(ip, XFS_DATA_FORK);
2685 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
2686 mrfree(&ip->i_lock);
2687 mrfree(&ip->i_iolock);
2688 freesema(&ip->i_flock);
2689 #ifdef XFS_BMAP_TRACE
2690 ktrace_free(ip->i_xtrace);
2692 #ifdef XFS_BMBT_TRACE
2693 ktrace_free(ip->i_btrace);
2696 ktrace_free(ip->i_rwtrace);
2698 #ifdef XFS_ILOCK_TRACE
2699 ktrace_free(ip->i_lock_trace);
2701 #ifdef XFS_DIR2_TRACE
2702 ktrace_free(ip->i_dir_trace);
2705 /* XXXdpd should be able to assert this but shutdown
2706 * is leaving the AIL behind. */
2707 ASSERT(((ip->i_itemp->ili_item.li_flags & XFS_LI_IN_AIL) == 0) ||
2708 XFS_FORCED_SHUTDOWN(ip->i_mount));
2709 xfs_inode_item_destroy(ip);
2711 kmem_zone_free(xfs_inode_zone, ip);
2716 * Increment the pin count of the given buffer.
2717 * This value is protected by ipinlock spinlock in the mount structure.
2723 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2725 atomic_inc(&ip->i_pincount);
2729 * Decrement the pin count of the given inode, and wake up
2730 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2731 * inode must have been previously pinned with a call to xfs_ipin().
2737 ASSERT(atomic_read(&ip->i_pincount) > 0);
2739 if (atomic_dec_and_test(&ip->i_pincount)) {
2741 * If the inode is currently being reclaimed, the
2742 * linux inode _and_ the xfs vnode may have been
2743 * freed so we cannot reference either of them safely.
2744 * Hence we should not try to do anything to them
2745 * if the xfs inode is currently in the reclaim
2748 * However, we still need to issue the unpin wakeup
2749 * call as the inode reclaim may be blocked waiting for
2750 * the inode to become unpinned.
2752 if (!(ip->i_flags & (XFS_IRECLAIM|XFS_IRECLAIMABLE))) {
2754 * Should I mark FreeBSD vnode as dirty here?
2756 printf("xfs_iunpin: REC RECABLE ip %p\n",ip);
2758 xfs_vnode_t *vp = XFS_ITOV_NULL(ip);
2760 /* make sync come back and flush this inode */
2762 struct inode *inode = vn_to_inode(vp);
2764 if (!(inode->i_state & I_NEW))
2765 mark_inode_dirty_sync(inode);
2769 wakeup(&ip->i_ipin_wait);
2774 * This is called to wait for the given inode to be unpinned.
2775 * It will sleep until this happens. The caller must have the
2776 * inode locked in at least shared mode so that the buffer cannot
2777 * be subsequently pinned once someone is waiting for it to be
2784 xfs_inode_log_item_t *iip;
2787 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE | MR_ACCESS));
2789 if (atomic_read(&ip->i_pincount) == 0) {
2794 if (iip && iip->ili_last_lsn) {
2795 lsn = iip->ili_last_lsn;
2801 * Give the log a push so we don't wait here too long.
2803 xfs_log_force(ip->i_mount, lsn, XFS_LOG_FORCE);
2806 * XXXKAN: xfs_iunpin is not locking inode
2809 while(atomic_read(&ip->i_pincount) != 0)
2810 tsleep(&ip->i_ipin_wait, PRIBIO, "iunpin", 0);
2815 * xfs_iextents_copy()
2817 * This is called to copy the REAL extents (as opposed to the delayed
2818 * allocation extents) from the inode into the given buffer. It
2819 * returns the number of bytes copied into the buffer.
2821 * If there are no delayed allocation extents, then we can just
2822 * memcpy() the extents into the buffer. Otherwise, we need to
2823 * examine each extent in turn and skip those which are delayed.
2828 xfs_bmbt_rec_t *buffer,
2832 xfs_bmbt_rec_t *dest_ep;
2834 #ifdef XFS_BMAP_TRACE
2835 static char fname[] = "xfs_iextents_copy";
2840 xfs_fsblock_t start_block;
2842 ifp = XFS_IFORK_PTR(ip, whichfork);
2843 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
2844 ASSERT(ifp->if_bytes > 0);
2846 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2847 xfs_bmap_trace_exlist(fname, ip, nrecs, whichfork);
2851 * There are some delayed allocation extents in the
2852 * inode, so copy the extents one at a time and skip
2853 * the delayed ones. There must be at least one
2854 * non-delayed extent.
2858 for (i = 0; i < nrecs; i++) {
2859 ep = xfs_iext_get_ext(ifp, i);
2860 start_block = xfs_bmbt_get_startblock(ep);
2861 if (ISNULLSTARTBLOCK(start_block)) {
2863 * It's a delayed allocation extent, so skip it.
2868 /* Translate to on disk format */
2869 put_unaligned(INT_GET(ep->l0, ARCH_CONVERT),
2870 (__uint64_t*)&dest_ep->l0);
2871 put_unaligned(INT_GET(ep->l1, ARCH_CONVERT),
2872 (__uint64_t*)&dest_ep->l1);
2876 ASSERT(copied != 0);
2877 xfs_validate_extents(ifp, copied, 1, XFS_EXTFMT_INODE(ip));
2879 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2883 * Each of the following cases stores data into the same region
2884 * of the on-disk inode, so only one of them can be valid at
2885 * any given time. While it is possible to have conflicting formats
2886 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2887 * in EXTENTS format, this can only happen when the fork has
2888 * changed formats after being modified but before being flushed.
2889 * In these cases, the format always takes precedence, because the
2890 * format indicates the current state of the fork.
2897 xfs_inode_log_item_t *iip,
2904 #ifdef XFS_TRANS_DEBUG
2907 static const short brootflag[2] =
2908 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2909 static const short dataflag[2] =
2910 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2911 static const short extflag[2] =
2912 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2916 ifp = XFS_IFORK_PTR(ip, whichfork);
2918 * This can happen if we gave up in iformat in an error path,
2919 * for the attribute fork.
2922 ASSERT(whichfork == XFS_ATTR_FORK);
2925 cp = XFS_DFORK_PTR(dip, whichfork);
2927 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2928 case XFS_DINODE_FMT_LOCAL:
2929 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2930 (ifp->if_bytes > 0)) {
2931 ASSERT(ifp->if_u1.if_data != NULL);
2932 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2933 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2935 if (whichfork == XFS_DATA_FORK) {
2936 if (unlikely(XFS_DIR_SHORTFORM_VALIDATE_ONDISK(mp, dip))) {
2937 XFS_ERROR_REPORT("xfs_iflush_fork",
2938 XFS_ERRLEVEL_LOW, mp);
2939 return XFS_ERROR(EFSCORRUPTED);
2944 case XFS_DINODE_FMT_EXTENTS:
2945 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2946 !(iip->ili_format.ilf_fields & extflag[whichfork]));
2947 ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
2948 (ifp->if_bytes == 0));
2949 ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
2950 (ifp->if_bytes > 0));
2951 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
2952 (ifp->if_bytes > 0)) {
2953 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2954 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2959 case XFS_DINODE_FMT_BTREE:
2960 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
2961 (ifp->if_broot_bytes > 0)) {
2962 ASSERT(ifp->if_broot != NULL);
2963 ASSERT(ifp->if_broot_bytes <=
2964 (XFS_IFORK_SIZE(ip, whichfork) +
2965 XFS_BROOT_SIZE_ADJ));
2966 xfs_bmbt_to_bmdr(ifp->if_broot, ifp->if_broot_bytes,
2967 (xfs_bmdr_block_t *)cp,
2968 XFS_DFORK_SIZE(dip, mp, whichfork));
2972 case XFS_DINODE_FMT_DEV:
2973 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
2974 ASSERT(whichfork == XFS_DATA_FORK);
2975 INT_SET(dip->di_u.di_dev, ARCH_CONVERT, ip->i_df.if_u2.if_rdev);
2979 case XFS_DINODE_FMT_UUID:
2980 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
2981 ASSERT(whichfork == XFS_DATA_FORK);
2982 memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid,
2996 * xfs_iflush() will write a modified inode's changes out to the
2997 * inode's on disk home. The caller must have the inode lock held
2998 * in at least shared mode and the inode flush semaphore must be
2999 * held as well. The inode lock will still be held upon return from
3000 * the call and the caller is free to unlock it.
3001 * The inode flush lock will be unlocked when the inode reaches the disk.
3002 * The flags indicate how the inode's buffer should be written out.
3009 xfs_inode_log_item_t *iip;
3017 int clcount; /* count of inodes clustered */
3019 enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
3022 XFS_STATS_INC(xs_iflush_count);
3025 printf("xfs_iflush: ip %p i_ino %lld\n",ip,ip->i_ino);
3026 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
3027 ASSERT(valusema(&ip->i_flock) <= 0);
3028 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3029 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3035 * If the inode isn't dirty, then just release the inode
3036 * flush lock and do nothing.
3038 if ((ip->i_update_core == 0) &&
3039 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3040 ASSERT((iip != NULL) ?
3041 !(iip->ili_item.li_flags & XFS_LI_IN_AIL) : 1);
3047 * We can't flush the inode until it is unpinned, so
3048 * wait for it. We know noone new can pin it, because
3049 * we are holding the inode lock shared and you need
3050 * to hold it exclusively to pin the inode.
3052 xfs_iunpin_wait(ip);
3055 * This may have been unpinned because the filesystem is shutting
3056 * down forcibly. If that's the case we must not write this inode
3057 * to disk, because the log record didn't make it to disk!
3059 if (XFS_FORCED_SHUTDOWN(mp)) {
3060 ip->i_update_core = 0;
3062 iip->ili_format.ilf_fields = 0;
3064 return XFS_ERROR(EIO);
3068 * Get the buffer containing the on-disk inode.
3070 error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0, 0);
3077 * Decide how buffer will be flushed out. This is done before
3078 * the call to xfs_iflush_int because this field is zeroed by it.
3080 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3082 * Flush out the inode buffer according to the directions
3083 * of the caller. In the cases where the caller has given
3084 * us a choice choose the non-delwri case. This is because
3085 * the inode is in the AIL and we need to get it out soon.
3088 case XFS_IFLUSH_SYNC:
3089 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3092 case XFS_IFLUSH_ASYNC:
3093 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3096 case XFS_IFLUSH_DELWRI:
3106 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3107 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3108 case XFS_IFLUSH_DELWRI:
3111 case XFS_IFLUSH_ASYNC:
3114 case XFS_IFLUSH_SYNC:
3125 * First flush out the inode that xfs_iflush was called with.
3127 error = xfs_iflush_int(ip, bp);
3134 * see if other inodes can be gathered into this write
3137 ip->i_chash->chl_buf = bp;
3139 ch = XFS_CHASH(mp, ip->i_blkno);
3140 s = mutex_spinlock(&ch->ch_lock);
3143 for (iq = ip->i_cnext; iq != ip; iq = iq->i_cnext) {
3145 * Do an un-protected check to see if the inode is dirty and
3146 * is a candidate for flushing. These checks will be repeated
3147 * later after the appropriate locks are acquired.
3150 if ((iq->i_update_core == 0) &&
3152 !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
3153 xfs_ipincount(iq) == 0) {
3158 * Try to get locks. If any are unavailable,
3159 * then this inode cannot be flushed and is skipped.
3162 /* get inode locks (just i_lock) */
3163 if (xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) {
3164 /* get inode flush lock */
3165 if (xfs_iflock_nowait(iq)) {
3166 /* check if pinned */
3167 if (xfs_ipincount(iq) == 0) {
3168 /* arriving here means that
3169 * this inode can be flushed.
3170 * first re-check that it's
3174 if ((iq->i_update_core != 0)||
3176 (iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3178 error = xfs_iflush_int(iq, bp);
3182 goto cluster_corrupt_out;
3191 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3194 mutex_spinunlock(&ch->ch_lock, s);
3197 XFS_STATS_INC(xs_icluster_flushcnt);
3198 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
3202 * If the buffer is pinned then push on the log so we won't
3203 * get stuck waiting in the write for too long.
3205 if (XFS_BUF_ISPINNED(bp)){
3206 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
3209 if (flags & INT_DELWRI) {
3210 xfs_bdwrite(mp, bp);
3211 } else if (flags & INT_ASYNC) {
3212 xfs_bawrite(mp, bp);
3214 error = xfs_bwrite(mp, bp);
3220 xfs_force_shutdown(mp, XFS_CORRUPT_INCORE);
3221 xfs_iflush_abort(ip);
3223 * Unlocks the flush lock
3225 return XFS_ERROR(EFSCORRUPTED);
3227 cluster_corrupt_out:
3228 /* Corruption detected in the clustering loop. Invalidate the
3229 * inode buffer and shut down the filesystem.
3231 mutex_spinunlock(&ch->ch_lock, s);
3234 * Clean up the buffer. If it was B_DELWRI, just release it --
3235 * brelse can handle it with no problems. If not, shut down the
3236 * filesystem before releasing the buffer.
3238 if ((bufwasdelwri= XFS_BUF_ISDELAYWRITE(bp))) {
3242 xfs_force_shutdown(mp, XFS_CORRUPT_INCORE);
3246 * Just like incore_relse: if we have b_iodone functions,
3247 * mark the buffer as an error and call them. Otherwise
3248 * mark it as stale and brelse.
3250 if (XFS_BUF_IODONE_FUNC(bp)) {
3251 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
3255 XFS_BUF_ERROR(bp,EIO);
3263 xfs_iflush_abort(iq);
3265 * Unlocks the flush lock
3267 return XFS_ERROR(EFSCORRUPTED);
3276 xfs_inode_log_item_t *iip;
3279 #ifdef XFS_TRANS_DEBUG
3284 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
3285 ASSERT(valusema(&ip->i_flock) <= 0);
3286 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3287 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3294 * If the inode isn't dirty, then just release the inode
3295 * flush lock and do nothing.
3297 if ((ip->i_update_core == 0) &&
3298 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3303 /* set *dip = inode's place in the buffer */
3304 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset);
3307 * Clear i_update_core before copying out the data.
3308 * This is for coordination with our timestamp updates
3309 * that don't hold the inode lock. They will always
3310 * update the timestamps BEFORE setting i_update_core,
3311 * so if we clear i_update_core after they set it we
3312 * are guaranteed to see their updates to the timestamps.
3313 * I believe that this depends on strongly ordered memory
3314 * semantics, but we have that. We use the SYNCHRONIZE
3315 * macro to make sure that the compiler does not reorder
3316 * the i_update_core access below the data copy below.
3318 ip->i_update_core = 0;
3322 * Make sure to get the latest atime from the Linux inode.
3324 xfs_synchronize_atime(ip);
3326 if (XFS_TEST_ERROR(INT_GET(dip->di_core.di_magic,ARCH_CONVERT) != XFS_DINODE_MAGIC,
3327 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3328 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3329 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3330 ip->i_ino, (int) INT_GET(dip->di_core.di_magic, ARCH_CONVERT), dip);
3333 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3334 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3335 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3336 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3337 ip->i_ino, ip, ip->i_d.di_magic);
3340 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3342 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3343 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3344 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3345 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3346 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3350 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
3352 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3353 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3354 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3355 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3356 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3357 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3362 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3363 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3364 XFS_RANDOM_IFLUSH_5)) {
3365 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3366 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3368 ip->i_d.di_nextents + ip->i_d.di_anextents,
3373 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3374 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3375 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3376 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3377 ip->i_ino, ip->i_d.di_forkoff, ip);
3381 * bump the flush iteration count, used to detect flushes which
3382 * postdate a log record during recovery.
3385 ip->i_d.di_flushiter++;
3388 * Copy the dirty parts of the inode into the on-disk
3389 * inode. We always copy out the core of the inode,
3390 * because if the inode is dirty at all the core must
3393 xfs_xlate_dinode_core((xfs_caddr_t)&(dip->di_core), &(ip->i_d), -1);
3395 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3396 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3397 ip->i_d.di_flushiter = 0;
3400 * If this is really an old format inode and the superblock version
3401 * has not been updated to support only new format inodes, then
3402 * convert back to the old inode format. If the superblock version
3403 * has been updated, then make the conversion permanent.
3405 ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 ||
3406 XFS_SB_VERSION_HASNLINK(&mp->m_sb));
3407 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
3408 if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) {
3412 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3413 INT_SET(dip->di_core.di_onlink, ARCH_CONVERT, ip->i_d.di_nlink);
3416 * The superblock version has already been bumped,
3417 * so just make the conversion to the new inode
3420 ip->i_d.di_version = XFS_DINODE_VERSION_2;
3421 INT_SET(dip->di_core.di_version, ARCH_CONVERT, XFS_DINODE_VERSION_2);
3422 ip->i_d.di_onlink = 0;
3423 dip->di_core.di_onlink = 0;
3424 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3425 memset(&(dip->di_core.di_pad[0]), 0,
3426 sizeof(dip->di_core.di_pad));
3427 ASSERT(ip->i_d.di_projid == 0);
3431 if (xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp) == EFSCORRUPTED) {
3435 if (XFS_IFORK_Q(ip)) {
3437 * The only error from xfs_iflush_fork is on the data fork.
3439 (void) xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3441 xfs_inobp_check(mp, bp);
3444 * We've recorded everything logged in the inode, so we'd
3445 * like to clear the ilf_fields bits so we don't log and
3446 * flush things unnecessarily. However, we can't stop
3447 * logging all this information until the data we've copied
3448 * into the disk buffer is written to disk. If we did we might
3449 * overwrite the copy of the inode in the log with all the
3450 * data after re-logging only part of it, and in the face of
3451 * a crash we wouldn't have all the data we need to recover.
3453 * What we do is move the bits to the ili_last_fields field.
3454 * When logging the inode, these bits are moved back to the
3455 * ilf_fields field. In the xfs_iflush_done() routine we
3456 * clear ili_last_fields, since we know that the information
3457 * those bits represent is permanently on disk. As long as
3458 * the flush completes before the inode is logged again, then
3459 * both ilf_fields and ili_last_fields will be cleared.
3461 * We can play with the ilf_fields bits here, because the inode
3462 * lock must be held exclusively in order to set bits there
3463 * and the flush lock protects the ili_last_fields bits.
3464 * Set ili_logged so the flush done
3465 * routine can tell whether or not to look in the AIL.
3466 * Also, store the current LSN of the inode so that we can tell
3467 * whether the item has moved in the AIL from xfs_iflush_done().
3468 * In order to read the lsn we need the AIL lock, because
3469 * it is a 64 bit value that cannot be read atomically.
3471 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3472 iip->ili_last_fields = iip->ili_format.ilf_fields;
3473 iip->ili_format.ilf_fields = 0;
3474 iip->ili_logged = 1;
3476 ASSERT(sizeof(xfs_lsn_t) == 8); /* don't lock if it shrinks */
3478 iip->ili_flush_lsn = iip->ili_item.li_lsn;
3482 * Attach the function xfs_iflush_done to the inode's
3483 * buffer. This will remove the inode from the AIL
3484 * and unlock the inode's flush lock when the inode is
3485 * completely written to disk.
3487 xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
3488 xfs_iflush_done, (xfs_log_item_t *)iip);
3490 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3491 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3494 * We're flushing an inode which is not in the AIL and has
3495 * not been logged but has i_update_core set. For this
3496 * case we can use a B_DELWRI flush and immediately drop
3497 * the inode flush lock because we can avoid the whole
3498 * AIL state thing. It's OK to drop the flush lock now,
3499 * because we've already locked the buffer and to do anything
3500 * you really need both.
3503 ASSERT(iip->ili_logged == 0);
3504 ASSERT(iip->ili_last_fields == 0);
3505 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3513 return XFS_ERROR(EFSCORRUPTED);
3518 * Flush all inactive inodes in mp.
3532 XFS_MOUNT_ILOCK(mp);
3538 /* Make sure we skip markers inserted by sync */
3539 if (ip->i_mount == NULL) {
3545 * It's up to our caller to purge the root
3546 * and quota vnodes later.
3548 vp = XFS_ITOV_NULL(ip);
3551 XFS_MOUNT_IUNLOCK(mp);
3552 xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC);
3557 if (vn_count(vp) != 0) {
3558 if (vn_count(vp) == 1 &&
3559 (ip == mp->m_rootip ||
3561 (ip->i_ino == mp->m_sb.sb_uquotino ||
3562 ip->i_ino == mp->m_sb.sb_gquotino)))) {
3567 * Ignore busy inodes but continue flushing
3574 * Sample vp mapping while holding mp locked on MP
3575 * systems, so we don't purge a reclaimed or
3576 * nonexistent vnode. We break from the loop
3577 * since we know that we modify
3578 * it by pulling ourselves from it in xfs_reclaim()
3579 * called via vn_purge() below. Set ip to the next
3580 * entry in the list anyway so we'll know below
3581 * whether we reached the end or not.
3584 XFS_MOUNT_IUNLOCK(mp);
3588 } while (ip != mp->m_inodes);
3590 * We need to distinguish between when we exit the loop
3591 * after a purge and when we simply hit the end of the
3592 * list. We can't use the (ip == mp->m_inodes) test,
3593 * because when we purge an inode at the start of the list
3594 * the next inode on the list becomes mp->m_inodes. That
3595 * would cause such a test to bail out early. The purged
3596 * variable tells us how we got out of the loop.
3602 XFS_MOUNT_IUNLOCK(mp);
3606 * xfs_iaccess: check accessibility of inode for mode.
3607 * This function is quite linuxy now
3608 * probably should be move to a os specfic location
3622 /* FreeBSD local change here */
3623 imode = (ip->i_d.di_mode & MODEMASK) | VTTOIF(vp->v_vnode->v_type);
3625 * Verify that the MAC policy allows the requested access.
3627 if ((error = _MAC_XFS_IACCESS(ip, accmode, cr)))
3628 return XFS_ERROR(error);
3630 if (accmode & VWRITE) {
3631 xfs_mount_t *mp = ip->i_mount;
3633 if ((XVFSTOMNT(XFS_MTOVFS(mp))->mnt_flag & MNT_RDONLY) &&
3634 (S_ISREG(imode) || S_ISDIR(imode) || S_ISLNK(imode)))
3635 return XFS_ERROR(EROFS);
3638 if (IS_IMMUTABLE(inode))
3639 return XFS_ERROR(EACCES);
3644 * If there's an Access Control List it's used instead of
3647 if ((error = _ACL_XFS_IACCESS(ip, accmode, cr)) != -1)
3648 return error ? XFS_ERROR(error) : 0;
3651 /* FreeBSD local change here */
3652 error = vaccess(vp->v_vnode->v_type, imode, ip->i_d.di_uid, ip->i_d.di_gid,
3659 * xfs_iroundup: round up argument to next power of two
3668 if ((v & (v - 1)) == 0)
3670 ASSERT((v & 0x80000000) == 0);
3671 if ((v & (v + 1)) == 0)
3673 for (i = 0, m = 1; i < 31; i++, m <<= 1) {
3677 if ((v & (v + 1)) == 0)
3684 #ifdef XFS_ILOCK_TRACE
3685 ktrace_t *xfs_ilock_trace_buf;
3688 xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
3690 ktrace_enter(ip->i_lock_trace,
3692 (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
3693 (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
3694 (void *)ra, /* caller of ilock */
3695 (void *)(unsigned long)current_cpu(),
3696 (void *)(unsigned long)current_pid(),
3697 NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL);
3702 * Return a pointer to the extent record at file index idx.
3706 xfs_ifork_t *ifp, /* inode fork pointer */
3707 xfs_extnum_t idx) /* index of target extent */
3710 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
3711 return ifp->if_u1.if_ext_irec->er_extbuf;
3712 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3713 xfs_ext_irec_t *erp; /* irec pointer */
3714 int erp_idx = 0; /* irec index */
3715 xfs_extnum_t page_idx = idx; /* ext index in target list */
3717 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3718 return &erp->er_extbuf[page_idx];
3719 } else if (ifp->if_bytes) {
3720 return &ifp->if_u1.if_extents[idx];
3727 * Insert new item(s) into the extent records for incore inode
3728 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3732 xfs_ifork_t *ifp, /* inode fork pointer */
3733 xfs_extnum_t idx, /* starting index of new items */
3734 xfs_extnum_t count, /* number of inserted items */
3735 xfs_bmbt_irec_t *new) /* items to insert */
3737 xfs_bmbt_rec_t *ep; /* extent record pointer */
3738 xfs_extnum_t i; /* extent record index */
3740 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3741 xfs_iext_add(ifp, idx, count);
3742 for (i = idx; i < idx + count; i++, new++) {
3743 ep = xfs_iext_get_ext(ifp, i);
3744 xfs_bmbt_set_all(ep, new);
3749 * This is called when the amount of space required for incore file
3750 * extents needs to be increased. The ext_diff parameter stores the
3751 * number of new extents being added and the idx parameter contains
3752 * the extent index where the new extents will be added. If the new
3753 * extents are being appended, then we just need to (re)allocate and
3754 * initialize the space. Otherwise, if the new extents are being
3755 * inserted into the middle of the existing entries, a bit more work
3756 * is required to make room for the new extents to be inserted. The
3757 * caller is responsible for filling in the new extent entries upon
3762 xfs_ifork_t *ifp, /* inode fork pointer */
3763 xfs_extnum_t idx, /* index to begin adding exts */
3764 int ext_diff) /* number of extents to add */
3766 int byte_diff; /* new bytes being added */
3767 int new_size; /* size of extents after adding */
3768 xfs_extnum_t nextents; /* number of extents in file */
3770 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3771 ASSERT((idx >= 0) && (idx <= nextents));
3772 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
3773 new_size = ifp->if_bytes + byte_diff;
3775 * If the new number of extents (nextents + ext_diff)
3776 * fits inside the inode, then continue to use the inline
3779 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
3780 if (idx < nextents) {
3781 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
3782 &ifp->if_u2.if_inline_ext[idx],
3783 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3784 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
3786 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3787 ifp->if_real_bytes = 0;
3788 ifp->if_lastex = nextents + ext_diff;
3791 * Otherwise use a linear (direct) extent list.
3792 * If the extents are currently inside the inode,
3793 * xfs_iext_realloc_direct will switch us from
3794 * inline to direct extent allocation mode.
3796 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
3797 xfs_iext_realloc_direct(ifp, new_size);
3798 if (idx < nextents) {
3799 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
3800 &ifp->if_u1.if_extents[idx],
3801 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3802 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
3805 /* Indirection array */
3807 xfs_ext_irec_t *erp;
3811 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
3812 if (ifp->if_flags & XFS_IFEXTIREC) {
3813 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3815 xfs_iext_irec_init(ifp);
3816 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3817 erp = ifp->if_u1.if_ext_irec;
3819 /* Extents fit in target extent page */
3820 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3821 if (page_idx < erp->er_extcount) {
3822 memmove(&erp->er_extbuf[page_idx + ext_diff],
3823 &erp->er_extbuf[page_idx],
3824 (erp->er_extcount - page_idx) *
3825 sizeof(xfs_bmbt_rec_t));
3826 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3828 erp->er_extcount += ext_diff;
3829 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3831 /* Insert a new extent page */
3833 xfs_iext_add_indirect_multi(ifp,
3834 erp_idx, page_idx, ext_diff);
3837 * If extent(s) are being appended to the last page in
3838 * the indirection array and the new extent(s) don't fit
3839 * in the page, then erp is NULL and erp_idx is set to
3840 * the next index needed in the indirection array.
3843 int count = ext_diff;
3846 erp = xfs_iext_irec_new(ifp, erp_idx);
3847 erp->er_extcount = count;
3848 count -= MIN(count, (int)XFS_LINEAR_EXTS);
3855 ifp->if_bytes = new_size;
3859 * This is called when incore extents are being added to the indirection
3860 * array and the new extents do not fit in the target extent list. The
3861 * erp_idx parameter contains the irec index for the target extent list
3862 * in the indirection array, and the idx parameter contains the extent
3863 * index within the list. The number of extents being added is stored
3864 * in the count parameter.
3866 * |-------| |-------|
3867 * | | | | idx - number of extents before idx
3869 * | | | | count - number of extents being inserted at idx
3870 * |-------| |-------|
3871 * | count | | nex2 | nex2 - number of extents after idx + count
3872 * |-------| |-------|
3875 xfs_iext_add_indirect_multi(
3876 xfs_ifork_t *ifp, /* inode fork pointer */
3877 int erp_idx, /* target extent irec index */
3878 xfs_extnum_t idx, /* index within target list */
3879 int count) /* new extents being added */
3881 int byte_diff; /* new bytes being added */
3882 xfs_ext_irec_t *erp; /* pointer to irec entry */
3883 xfs_extnum_t ext_diff; /* number of extents to add */
3884 xfs_extnum_t ext_cnt; /* new extents still needed */
3885 xfs_extnum_t nex2; /* extents after idx + count */
3886 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3887 int nlists; /* number of irec's (lists) */
3889 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3890 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3891 nex2 = erp->er_extcount - idx;
3892 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3895 * Save second part of target extent list
3896 * (all extents past */
3898 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3899 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_SLEEP);
3900 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3901 erp->er_extcount -= nex2;
3902 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3903 memset(&erp->er_extbuf[idx], 0, byte_diff);
3907 * Add the new extents to the end of the target
3908 * list, then allocate new irec record(s) and
3909 * extent buffer(s) as needed to store the rest
3910 * of the new extents.
3913 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3915 erp->er_extcount += ext_diff;
3916 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3917 ext_cnt -= ext_diff;
3921 erp = xfs_iext_irec_new(ifp, erp_idx);
3922 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3923 erp->er_extcount = ext_diff;
3924 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3925 ext_cnt -= ext_diff;
3928 /* Add nex2 extents back to indirection array */
3930 xfs_extnum_t ext_avail;
3933 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3934 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3937 * If nex2 extents fit in the current page, append
3938 * nex2_ep after the new extents.
3940 if (nex2 <= ext_avail) {
3941 i = erp->er_extcount;
3944 * Otherwise, check if space is available in the
3947 else if ((erp_idx < nlists - 1) &&
3948 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3949 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3952 /* Create a hole for nex2 extents */
3953 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3954 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3957 * Final choice, create a new extent page for
3962 erp = xfs_iext_irec_new(ifp, erp_idx);
3964 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
3965 kmem_free(nex2_ep, byte_diff);
3966 erp->er_extcount += nex2;
3967 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3972 * This is called when the amount of space required for incore file
3973 * extents needs to be decreased. The ext_diff parameter stores the
3974 * number of extents to be removed and the idx parameter contains
3975 * the extent index where the extents will be removed from.
3977 * If the amount of space needed has decreased below the linear
3978 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3979 * extent array. Otherwise, use kmem_realloc() to adjust the
3980 * size to what is needed.
3984 xfs_ifork_t *ifp, /* inode fork pointer */
3985 xfs_extnum_t idx, /* index to begin removing exts */
3986 int ext_diff) /* number of extents to remove */
3988 xfs_extnum_t nextents; /* number of extents in file */
3989 int new_size; /* size of extents after removal */
3991 ASSERT(ext_diff > 0);
3992 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3993 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3995 if (new_size == 0) {
3996 xfs_iext_destroy(ifp);
3997 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3998 xfs_iext_remove_indirect(ifp, idx, ext_diff);
3999 } else if (ifp->if_real_bytes) {
4000 xfs_iext_remove_direct(ifp, idx, ext_diff);
4002 xfs_iext_remove_inline(ifp, idx, ext_diff);
4004 ifp->if_bytes = new_size;
4008 * This removes ext_diff extents from the inline buffer, beginning
4009 * at extent index idx.
4012 xfs_iext_remove_inline(
4013 xfs_ifork_t *ifp, /* inode fork pointer */
4014 xfs_extnum_t idx, /* index to begin removing exts */
4015 int ext_diff) /* number of extents to remove */
4017 int nextents; /* number of extents in file */
4019 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4020 ASSERT(idx < XFS_INLINE_EXTS);
4021 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4022 ASSERT(((nextents - ext_diff) > 0) &&
4023 (nextents - ext_diff) < XFS_INLINE_EXTS);
4025 if (idx + ext_diff < nextents) {
4026 memmove(&ifp->if_u2.if_inline_ext[idx],
4027 &ifp->if_u2.if_inline_ext[idx + ext_diff],
4028 (nextents - (idx + ext_diff)) *
4029 sizeof(xfs_bmbt_rec_t));
4030 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
4031 0, ext_diff * sizeof(xfs_bmbt_rec_t));
4033 memset(&ifp->if_u2.if_inline_ext[idx], 0,
4034 ext_diff * sizeof(xfs_bmbt_rec_t));
4039 * This removes ext_diff extents from a linear (direct) extent list,
4040 * beginning at extent index idx. If the extents are being removed
4041 * from the end of the list (ie. truncate) then we just need to re-
4042 * allocate the list to remove the extra space. Otherwise, if the
4043 * extents are being removed from the middle of the existing extent
4044 * entries, then we first need to move the extent records beginning
4045 * at idx + ext_diff up in the list to overwrite the records being
4046 * removed, then remove the extra space via kmem_realloc.
4049 xfs_iext_remove_direct(
4050 xfs_ifork_t *ifp, /* inode fork pointer */
4051 xfs_extnum_t idx, /* index to begin removing exts */
4052 int ext_diff) /* number of extents to remove */
4054 xfs_extnum_t nextents; /* number of extents in file */
4055 int new_size; /* size of extents after removal */
4057 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4058 new_size = ifp->if_bytes -
4059 (ext_diff * sizeof(xfs_bmbt_rec_t));
4060 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4062 if (new_size == 0) {
4063 xfs_iext_destroy(ifp);
4066 /* Move extents up in the list (if needed) */
4067 if (idx + ext_diff < nextents) {
4068 memmove(&ifp->if_u1.if_extents[idx],
4069 &ifp->if_u1.if_extents[idx + ext_diff],
4070 (nextents - (idx + ext_diff)) *
4071 sizeof(xfs_bmbt_rec_t));
4073 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
4074 0, ext_diff * sizeof(xfs_bmbt_rec_t));
4076 * Reallocate the direct extent list. If the extents
4077 * will fit inside the inode then xfs_iext_realloc_direct
4078 * will switch from direct to inline extent allocation
4081 xfs_iext_realloc_direct(ifp, new_size);
4082 ifp->if_bytes = new_size;
4086 * This is called when incore extents are being removed from the
4087 * indirection array and the extents being removed span multiple extent
4088 * buffers. The idx parameter contains the file extent index where we
4089 * want to begin removing extents, and the count parameter contains
4090 * how many extents need to be removed.
4092 * |-------| |-------|
4093 * | nex1 | | | nex1 - number of extents before idx
4094 * |-------| | count |
4095 * | | | | count - number of extents being removed at idx
4096 * | count | |-------|
4097 * | | | nex2 | nex2 - number of extents after idx + count
4098 * |-------| |-------|
4101 xfs_iext_remove_indirect(
4102 xfs_ifork_t *ifp, /* inode fork pointer */
4103 xfs_extnum_t idx, /* index to begin removing extents */
4104 int count) /* number of extents to remove */
4106 xfs_ext_irec_t *erp; /* indirection array pointer */
4107 int erp_idx = 0; /* indirection array index */
4108 xfs_extnum_t ext_cnt; /* extents left to remove */
4109 xfs_extnum_t ext_diff; /* extents to remove in current list */
4110 xfs_extnum_t nex1; /* number of extents before idx */
4111 xfs_extnum_t nex2; /* extents after idx + count */
4112 int nlists; /* entries in indirection array */
4113 int page_idx = idx; /* index in target extent list */
4115 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4116 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
4117 ASSERT(erp != NULL);
4118 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4122 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
4123 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
4125 * Check for deletion of entire list;
4126 * xfs_iext_irec_remove() updates extent offsets.
4128 if (ext_diff == erp->er_extcount) {
4129 xfs_iext_irec_remove(ifp, erp_idx);
4130 ext_cnt -= ext_diff;
4133 ASSERT(erp_idx < ifp->if_real_bytes /
4135 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4142 /* Move extents up (if needed) */
4144 memmove(&erp->er_extbuf[nex1],
4145 &erp->er_extbuf[nex1 + ext_diff],
4146 nex2 * sizeof(xfs_bmbt_rec_t));
4148 /* Zero out rest of page */
4149 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
4150 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
4151 /* Update remaining counters */
4152 erp->er_extcount -= ext_diff;
4153 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
4154 ext_cnt -= ext_diff;
4159 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
4160 xfs_iext_irec_compact(ifp);
4164 * Create, destroy, or resize a linear (direct) block of extents.
4167 xfs_iext_realloc_direct(
4168 xfs_ifork_t *ifp, /* inode fork pointer */
4169 int new_size) /* new size of extents */
4171 int rnew_size; /* real new size of extents */
4173 rnew_size = new_size;
4175 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
4176 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
4177 (new_size != ifp->if_real_bytes)));
4179 /* Free extent records */
4180 if (new_size == 0) {
4181 xfs_iext_destroy(ifp);
4183 /* Resize direct extent list and zero any new bytes */
4184 else if (ifp->if_real_bytes) {
4185 /* Check if extents will fit inside the inode */
4186 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
4187 xfs_iext_direct_to_inline(ifp, new_size /
4188 (uint)sizeof(xfs_bmbt_rec_t));
4189 ifp->if_bytes = new_size;
4192 if ((new_size & (new_size - 1)) != 0) {
4193 rnew_size = xfs_iroundup(new_size);
4195 if (rnew_size != ifp->if_real_bytes) {
4196 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
4197 kmem_realloc(ifp->if_u1.if_extents,
4202 if (rnew_size > ifp->if_real_bytes) {
4203 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
4204 (uint)sizeof(xfs_bmbt_rec_t)], 0,
4205 rnew_size - ifp->if_real_bytes);
4209 * Switch from the inline extent buffer to a direct
4210 * extent list. Be sure to include the inline extent
4211 * bytes in new_size.
4214 new_size += ifp->if_bytes;
4215 if ((new_size & (new_size - 1)) != 0) {
4216 rnew_size = xfs_iroundup(new_size);
4218 xfs_iext_inline_to_direct(ifp, rnew_size);
4220 ifp->if_real_bytes = rnew_size;
4221 ifp->if_bytes = new_size;
4225 * Switch from linear (direct) extent records to inline buffer.
4228 xfs_iext_direct_to_inline(
4229 xfs_ifork_t *ifp, /* inode fork pointer */
4230 xfs_extnum_t nextents) /* number of extents in file */
4232 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4233 ASSERT(nextents <= XFS_INLINE_EXTS);
4235 * The inline buffer was zeroed when we switched
4236 * from inline to direct extent allocation mode,
4237 * so we don't need to clear it here.
4239 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
4240 nextents * sizeof(xfs_bmbt_rec_t));
4241 kmem_free(ifp->if_u1.if_extents, KM_SLEEP);
4242 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
4243 ifp->if_real_bytes = 0;
4247 * Switch from inline buffer to linear (direct) extent records.
4248 * new_size should already be rounded up to the next power of 2
4249 * by the caller (when appropriate), so use new_size as it is.
4250 * However, since new_size may be rounded up, we can't update
4251 * if_bytes here. It is the caller's responsibility to update
4252 * if_bytes upon return.
4255 xfs_iext_inline_to_direct(
4256 xfs_ifork_t *ifp, /* inode fork pointer */
4257 int new_size) /* number of extents in file */
4259 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
4260 kmem_alloc(new_size, KM_SLEEP);
4261 memset(ifp->if_u1.if_extents, 0, new_size);
4262 if (ifp->if_bytes) {
4263 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
4265 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4266 sizeof(xfs_bmbt_rec_t));
4268 ifp->if_real_bytes = new_size;
4272 * Resize an extent indirection array to new_size bytes.
4275 xfs_iext_realloc_indirect(
4276 xfs_ifork_t *ifp, /* inode fork pointer */
4277 int new_size) /* new indirection array size */
4279 int nlists; /* number of irec's (ex lists) */
4280 int size; /* current indirection array size */
4282 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4283 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4284 size = nlists * sizeof(xfs_ext_irec_t);
4285 ASSERT(ifp->if_real_bytes);
4286 ASSERT((new_size >= 0) && (new_size != size));
4287 if (new_size == 0) {
4288 xfs_iext_destroy(ifp);
4290 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
4291 kmem_realloc(ifp->if_u1.if_ext_irec,
4292 new_size, size, KM_SLEEP);
4297 * Switch from indirection array to linear (direct) extent allocations.
4300 xfs_iext_indirect_to_direct(
4301 xfs_ifork_t *ifp) /* inode fork pointer */
4303 xfs_bmbt_rec_t *ep; /* extent record pointer */
4304 xfs_extnum_t nextents; /* number of extents in file */
4305 int size; /* size of file extents */
4307 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4308 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4309 ASSERT(nextents <= XFS_LINEAR_EXTS);
4310 size = nextents * sizeof(xfs_bmbt_rec_t);
4312 xfs_iext_irec_compact_full(ifp);
4313 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
4315 ep = ifp->if_u1.if_ext_irec->er_extbuf;
4316 kmem_free(ifp->if_u1.if_ext_irec, sizeof(xfs_ext_irec_t));
4317 ifp->if_flags &= ~XFS_IFEXTIREC;
4318 ifp->if_u1.if_extents = ep;
4319 ifp->if_bytes = size;
4320 if (nextents < XFS_LINEAR_EXTS) {
4321 xfs_iext_realloc_direct(ifp, size);
4326 * Free incore file extents.
4330 xfs_ifork_t *ifp) /* inode fork pointer */
4332 if (ifp->if_flags & XFS_IFEXTIREC) {
4336 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4337 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
4338 xfs_iext_irec_remove(ifp, erp_idx);
4340 ifp->if_flags &= ~XFS_IFEXTIREC;
4341 } else if (ifp->if_real_bytes) {
4342 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
4343 } else if (ifp->if_bytes) {
4344 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4345 sizeof(xfs_bmbt_rec_t));
4347 ifp->if_u1.if_extents = NULL;
4348 ifp->if_real_bytes = 0;
4353 * Return a pointer to the extent record for file system block bno.
4355 xfs_bmbt_rec_t * /* pointer to found extent record */
4356 xfs_iext_bno_to_ext(
4357 xfs_ifork_t *ifp, /* inode fork pointer */
4358 xfs_fileoff_t bno, /* block number to search for */
4359 xfs_extnum_t *idxp) /* index of target extent */
4361 xfs_bmbt_rec_t *base; /* pointer to first extent */
4362 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
4363 xfs_bmbt_rec_t *ep = NULL; /* pointer to target extent */
4364 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4365 int high; /* upper boundary in search */
4366 xfs_extnum_t idx = 0; /* index of target extent */
4367 int low; /* lower boundary in search */
4368 xfs_extnum_t nextents; /* number of file extents */
4369 xfs_fileoff_t startoff = 0; /* start offset of extent */
4371 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4372 if (nextents == 0) {
4377 if (ifp->if_flags & XFS_IFEXTIREC) {
4378 /* Find target extent list */
4380 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
4381 base = erp->er_extbuf;
4382 high = erp->er_extcount - 1;
4384 base = ifp->if_u1.if_extents;
4385 high = nextents - 1;
4387 /* Binary search extent records */
4388 while (low <= high) {
4389 idx = (low + high) >> 1;
4391 startoff = xfs_bmbt_get_startoff(ep);
4392 blockcount = xfs_bmbt_get_blockcount(ep);
4393 if (bno < startoff) {
4395 } else if (bno >= startoff + blockcount) {
4398 /* Convert back to file-based extent index */
4399 if (ifp->if_flags & XFS_IFEXTIREC) {
4400 idx += erp->er_extoff;
4406 /* Convert back to file-based extent index */
4407 if (ifp->if_flags & XFS_IFEXTIREC) {
4408 idx += erp->er_extoff;
4410 if (bno >= startoff + blockcount) {
4411 if (++idx == nextents) {
4414 ep = xfs_iext_get_ext(ifp, idx);
4422 * Return a pointer to the indirection array entry containing the
4423 * extent record for filesystem block bno. Store the index of the
4424 * target irec in *erp_idxp.
4426 xfs_ext_irec_t * /* pointer to found extent record */
4427 xfs_iext_bno_to_irec(
4428 xfs_ifork_t *ifp, /* inode fork pointer */
4429 xfs_fileoff_t bno, /* block number to search for */
4430 int *erp_idxp) /* irec index of target ext list */
4432 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4433 xfs_ext_irec_t *erp_next; /* next indirection array entry */
4434 int erp_idx; /* indirection array index */
4435 int nlists; /* number of extent irec's (lists) */
4436 int high; /* binary search upper limit */
4437 int low; /* binary search lower limit */
4439 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4440 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4444 while (low <= high) {
4445 erp_idx = (low + high) >> 1;
4446 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4447 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
4448 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
4450 } else if (erp_next && bno >=
4451 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
4457 *erp_idxp = erp_idx;
4462 * Return a pointer to the indirection array entry containing the
4463 * extent record at file extent index *idxp. Store the index of the
4464 * target irec in *erp_idxp and store the page index of the target
4465 * extent record in *idxp.
4468 xfs_iext_idx_to_irec(
4469 xfs_ifork_t *ifp, /* inode fork pointer */
4470 xfs_extnum_t *idxp, /* extent index (file -> page) */
4471 int *erp_idxp, /* pointer to target irec */
4472 int realloc) /* new bytes were just added */
4474 xfs_ext_irec_t *prev; /* pointer to previous irec */
4475 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
4476 int erp_idx; /* indirection array index */
4477 int nlists; /* number of irec's (ex lists) */
4478 int high; /* binary search upper limit */
4479 int low; /* binary search lower limit */
4480 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
4482 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4483 ASSERT(page_idx >= 0 && page_idx <=
4484 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
4485 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4490 /* Binary search extent irec's */
4491 while (low <= high) {
4492 erp_idx = (low + high) >> 1;
4493 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4494 prev = erp_idx > 0 ? erp - 1 : NULL;
4495 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
4496 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
4498 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
4499 (page_idx == erp->er_extoff + erp->er_extcount &&
4502 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
4503 erp->er_extcount == XFS_LINEAR_EXTS) {
4507 erp = erp_idx < nlists ? erp + 1 : NULL;
4510 page_idx -= erp->er_extoff;
4515 *erp_idxp = erp_idx;
4520 * Allocate and initialize an indirection array once the space needed
4521 * for incore extents increases above XFS_IEXT_BUFSZ.
4525 xfs_ifork_t *ifp) /* inode fork pointer */
4527 xfs_ext_irec_t *erp; /* indirection array pointer */
4528 xfs_extnum_t nextents; /* number of extents in file */
4530 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4531 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4532 ASSERT(nextents <= XFS_LINEAR_EXTS);
4534 erp = (xfs_ext_irec_t *)
4535 kmem_alloc(sizeof(xfs_ext_irec_t), KM_SLEEP);
4537 if (nextents == 0) {
4538 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
4539 kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
4540 } else if (!ifp->if_real_bytes) {
4541 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
4542 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
4543 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
4545 erp->er_extbuf = ifp->if_u1.if_extents;
4546 erp->er_extcount = nextents;
4549 ifp->if_flags |= XFS_IFEXTIREC;
4550 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
4551 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
4552 ifp->if_u1.if_ext_irec = erp;
4558 * Allocate and initialize a new entry in the indirection array.
4562 xfs_ifork_t *ifp, /* inode fork pointer */
4563 int erp_idx) /* index for new irec */
4565 xfs_ext_irec_t *erp; /* indirection array pointer */
4566 int i; /* loop counter */
4567 int nlists; /* number of irec's (ex lists) */
4569 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4570 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4572 /* Resize indirection array */
4573 xfs_iext_realloc_indirect(ifp, ++nlists *
4574 sizeof(xfs_ext_irec_t));
4576 * Move records down in the array so the
4577 * new page can use erp_idx.
4579 erp = ifp->if_u1.if_ext_irec;
4580 for (i = nlists - 1; i > erp_idx; i--) {
4581 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
4583 ASSERT(i == erp_idx);
4585 /* Initialize new extent record */
4586 erp = ifp->if_u1.if_ext_irec;
4587 erp[erp_idx].er_extbuf = (xfs_bmbt_rec_t *)
4588 kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
4589 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4590 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
4591 erp[erp_idx].er_extcount = 0;
4592 erp[erp_idx].er_extoff = erp_idx > 0 ?
4593 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
4594 return (&erp[erp_idx]);
4598 * Remove a record from the indirection array.
4601 xfs_iext_irec_remove(
4602 xfs_ifork_t *ifp, /* inode fork pointer */
4603 int erp_idx) /* irec index to remove */
4605 xfs_ext_irec_t *erp; /* indirection array pointer */
4606 int i; /* loop counter */
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 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4612 if (erp->er_extbuf) {
4613 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
4615 kmem_free(erp->er_extbuf, XFS_IEXT_BUFSZ);
4617 /* Compact extent records */
4618 erp = ifp->if_u1.if_ext_irec;
4619 for (i = erp_idx; i < nlists - 1; i++) {
4620 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
4623 * Manually free the last extent record from the indirection
4624 * array. A call to xfs_iext_realloc_indirect() with a size
4625 * of zero would result in a call to xfs_iext_destroy() which
4626 * would in turn call this function again, creating a nasty
4630 xfs_iext_realloc_indirect(ifp,
4631 nlists * sizeof(xfs_ext_irec_t));
4633 kmem_free(ifp->if_u1.if_ext_irec,
4634 sizeof(xfs_ext_irec_t));
4636 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4640 * This is called to clean up large amounts of unused memory allocated
4641 * by the indirection array. Before compacting anything though, verify
4642 * that the indirection array is still needed and switch back to the
4643 * linear extent list (or even the inline buffer) if possible. The
4644 * compaction policy is as follows:
4646 * Full Compaction: Extents fit into a single page (or inline buffer)
4647 * Full Compaction: Extents occupy less than 10% of allocated space
4648 * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
4649 * No Compaction: Extents occupy at least 50% of allocated space
4652 xfs_iext_irec_compact(
4653 xfs_ifork_t *ifp) /* inode fork pointer */
4655 xfs_extnum_t nextents; /* number of extents in file */
4656 int nlists; /* number of irec's (ex lists) */
4658 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4659 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4660 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4662 if (nextents == 0) {
4663 xfs_iext_destroy(ifp);
4664 } else if (nextents <= XFS_INLINE_EXTS) {
4665 xfs_iext_indirect_to_direct(ifp);
4666 xfs_iext_direct_to_inline(ifp, nextents);
4667 } else if (nextents <= XFS_LINEAR_EXTS) {
4668 xfs_iext_indirect_to_direct(ifp);
4669 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 3) {
4670 xfs_iext_irec_compact_full(ifp);
4671 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
4672 xfs_iext_irec_compact_pages(ifp);
4677 * Combine extents from neighboring extent pages.
4680 xfs_iext_irec_compact_pages(
4681 xfs_ifork_t *ifp) /* inode fork pointer */
4683 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
4684 int erp_idx = 0; /* indirection array index */
4685 int nlists; /* number of irec's (ex lists) */
4687 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4688 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4689 while (erp_idx < nlists - 1) {
4690 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4692 if (erp_next->er_extcount <=
4693 (XFS_LINEAR_EXTS - erp->er_extcount)) {
4694 memmove(&erp->er_extbuf[erp->er_extcount],
4695 erp_next->er_extbuf, erp_next->er_extcount *
4696 sizeof(xfs_bmbt_rec_t));
4697 erp->er_extcount += erp_next->er_extcount;
4699 * Free page before removing extent record
4700 * so er_extoffs don't get modified in
4701 * xfs_iext_irec_remove.
4703 kmem_free(erp_next->er_extbuf, XFS_IEXT_BUFSZ);
4704 erp_next->er_extbuf = NULL;
4705 xfs_iext_irec_remove(ifp, erp_idx + 1);
4706 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4714 * Fully compact the extent records managed by the indirection array.
4717 xfs_iext_irec_compact_full(
4718 xfs_ifork_t *ifp) /* inode fork pointer */
4720 xfs_bmbt_rec_t *ep, *ep_next; /* extent record pointers */
4721 xfs_ext_irec_t *erp, *erp_next; /* extent irec pointers */
4722 int erp_idx = 0; /* extent irec index */
4723 int ext_avail; /* empty entries in ex list */
4724 int ext_diff; /* number of exts to add */
4725 int nlists; /* number of irec's (ex lists) */
4727 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4728 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4729 erp = ifp->if_u1.if_ext_irec;
4730 ep = &erp->er_extbuf[erp->er_extcount];
4732 ep_next = erp_next->er_extbuf;
4733 while (erp_idx < nlists - 1) {
4734 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
4735 ext_diff = MIN(ext_avail, erp_next->er_extcount);
4736 memcpy(ep, ep_next, ext_diff * sizeof(xfs_bmbt_rec_t));
4737 erp->er_extcount += ext_diff;
4738 erp_next->er_extcount -= ext_diff;
4739 /* Remove next page */
4740 if (erp_next->er_extcount == 0) {
4742 * Free page before removing extent record
4743 * so er_extoffs don't get modified in
4744 * xfs_iext_irec_remove.
4746 kmem_free(erp_next->er_extbuf,
4747 erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
4748 erp_next->er_extbuf = NULL;
4749 xfs_iext_irec_remove(ifp, erp_idx + 1);
4750 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4751 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4752 /* Update next page */
4754 /* Move rest of page up to become next new page */
4755 memmove(erp_next->er_extbuf, ep_next,
4756 erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
4757 ep_next = erp_next->er_extbuf;
4758 memset(&ep_next[erp_next->er_extcount], 0,
4759 (XFS_LINEAR_EXTS - erp_next->er_extcount) *
4760 sizeof(xfs_bmbt_rec_t));
4762 if (erp->er_extcount == XFS_LINEAR_EXTS) {
4764 if (erp_idx < nlists)
4765 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4769 ep = &erp->er_extbuf[erp->er_extcount];
4771 ep_next = erp_next->er_extbuf;
4776 * This is called to update the er_extoff field in the indirection
4777 * array when extents have been added or removed from one of the
4778 * extent lists. erp_idx contains the irec index to begin updating
4779 * at and ext_diff contains the number of extents that were added
4783 xfs_iext_irec_update_extoffs(
4784 xfs_ifork_t *ifp, /* inode fork pointer */
4785 int erp_idx, /* irec index to update */
4786 int ext_diff) /* number of new extents */
4788 int i; /* loop counter */
4789 int nlists; /* number of irec's (ex lists */
4791 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4792 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4793 for (i = erp_idx; i < nlists; i++) {
4794 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;