2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
25 #include "xfs_trans.h"
26 #include "xfs_trans_priv.h"
31 #include "xfs_dmapi.h"
32 #include "xfs_mount.h"
33 #include "xfs_bmap_btree.h"
34 #include "xfs_alloc_btree.h"
35 #include "xfs_ialloc_btree.h"
36 #include "xfs_dir_sf.h"
37 #include "xfs_dir2_sf.h"
38 #include "xfs_attr_sf.h"
39 #include "xfs_dinode.h"
40 #include "xfs_inode.h"
41 #include "xfs_buf_item.h"
42 #include "xfs_inode_item.h"
43 #include "xfs_btree.h"
44 #include "xfs_alloc.h"
45 #include "xfs_ialloc.h"
48 #include "xfs_error.h"
49 #include "xfs_utils.h"
50 #include "xfs_dir2_trace.h"
51 #include "xfs_quota.h"
56 kmem_zone_t *xfs_ifork_zone;
57 kmem_zone_t *xfs_inode_zone;
58 kmem_zone_t *xfs_chashlist_zone;
61 * Used in xfs_itruncate(). This is the maximum number of extents
62 * freed from a file in a single transaction.
64 #define XFS_ITRUNC_MAX_EXTENTS 2
66 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
67 STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
68 STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
69 STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
74 * Make sure that the extents in the given memory buffer
89 for (i = 0; i < nrecs; i++) {
90 ep = xfs_iext_get_ext(ifp, i);
91 rec.l0 = get_unaligned((__uint64_t*)&ep->l0);
92 rec.l1 = get_unaligned((__uint64_t*)&ep->l1);
94 xfs_bmbt_disk_get_all(&rec, &irec);
96 xfs_bmbt_get_all(&rec, &irec);
97 if (fmt == XFS_EXTFMT_NOSTATE)
98 ASSERT(irec.br_state == XFS_EXT_NORM);
102 #define xfs_validate_extents(ifp, nrecs, disk, fmt)
106 * Check that none of the inode's in the buffer have a next
107 * unlinked field of 0.
119 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
121 for (i = 0; i < j; i++) {
122 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
123 i * mp->m_sb.sb_inodesize);
124 if (!dip->di_next_unlinked) {
125 xfs_fs_cmn_err(CE_ALERT, mp,
126 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
128 ASSERT(dip->di_next_unlinked);
135 * This routine is called to map an inode number within a file
136 * system to the buffer containing the on-disk version of the
137 * inode. It returns a pointer to the buffer containing the
138 * on-disk inode in the bpp parameter, and in the dip parameter
139 * it returns a pointer to the on-disk inode within that buffer.
141 * If a non-zero error is returned, then the contents of bpp and
142 * dipp are undefined.
144 * Use xfs_imap() to determine the size and location of the
145 * buffer to read from disk.
163 * Call the space management code to find the location of the
167 error = xfs_imap(mp, tp, ino, &imap, XFS_IMAP_LOOKUP);
170 "xfs_inotobp: xfs_imap() returned an "
171 "error %d on %s. Returning error.", error, mp->m_fsname);
176 * If the inode number maps to a block outside the bounds of the
177 * file system then return NULL rather than calling read_buf
178 * and panicing when we get an error from the driver.
180 if ((imap.im_blkno + imap.im_len) >
181 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
183 "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
184 "of the file system %s. Returning EINVAL.",
185 (unsigned long long)imap.im_blkno,
186 imap.im_len, mp->m_fsname);
187 return XFS_ERROR(EINVAL);
191 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
192 * default to just a read_buf() call.
194 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
195 (int)imap.im_len, XFS_BUF_LOCK, &bp);
199 "xfs_inotobp: xfs_trans_read_buf() returned an "
200 "error %d on %s. Returning error.", error, mp->m_fsname);
203 dip = (xfs_dinode_t *)xfs_buf_offset(bp, 0);
205 INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
206 XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
207 if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP,
208 XFS_RANDOM_ITOBP_INOTOBP))) {
209 XFS_CORRUPTION_ERROR("xfs_inotobp", XFS_ERRLEVEL_LOW, mp, dip);
210 xfs_trans_brelse(tp, bp);
212 "xfs_inotobp: XFS_TEST_ERROR() returned an "
213 "error on %s. Returning EFSCORRUPTED.", mp->m_fsname);
214 return XFS_ERROR(EFSCORRUPTED);
217 xfs_inobp_check(mp, bp);
220 * Set *dipp to point to the on-disk inode in the buffer.
222 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
224 *offset = imap.im_boffset;
230 * This routine is called to map an inode to the buffer containing
231 * the on-disk version of the inode. It returns a pointer to the
232 * buffer containing the on-disk inode in the bpp parameter, and in
233 * the dip parameter it returns a pointer to the on-disk inode within
236 * If a non-zero error is returned, then the contents of bpp and
237 * dipp are undefined.
239 * If the inode is new and has not yet been initialized, use xfs_imap()
240 * to determine the size and location of the buffer to read from disk.
241 * If the inode has already been mapped to its buffer and read in once,
242 * then use the mapping information stored in the inode rather than
243 * calling xfs_imap(). This allows us to avoid the overhead of looking
244 * at the inode btree for small block file systems (see xfs_dilocate()).
245 * We can tell whether the inode has been mapped in before by comparing
246 * its disk block address to 0. Only uninitialized inodes will have
247 * 0 for the disk block address.
267 if (ip->i_blkno == (xfs_daddr_t)0) {
269 * Call the space management code to find the location of the
273 if ((error = xfs_imap(mp, tp, ip->i_ino, &imap,
274 XFS_IMAP_LOOKUP | imap_flags)))
278 * If the inode number maps to a block outside the bounds
279 * of the file system then return NULL rather than calling
280 * read_buf and panicing when we get an error from the
283 if ((imap.im_blkno + imap.im_len) >
284 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
286 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
287 "(imap.im_blkno (0x%llx) "
288 "+ imap.im_len (0x%llx)) > "
289 " XFS_FSB_TO_BB(mp, "
290 "mp->m_sb.sb_dblocks) (0x%llx)",
291 (unsigned long long) imap.im_blkno,
292 (unsigned long long) imap.im_len,
293 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
295 return XFS_ERROR(EINVAL);
299 * Fill in the fields in the inode that will be used to
300 * map the inode to its buffer from now on.
302 ip->i_blkno = imap.im_blkno;
303 ip->i_len = imap.im_len;
304 ip->i_boffset = imap.im_boffset;
307 * We've already mapped the inode once, so just use the
308 * mapping that we saved the first time.
310 imap.im_blkno = ip->i_blkno;
311 imap.im_len = ip->i_len;
312 imap.im_boffset = ip->i_boffset;
314 ASSERT(bno == 0 || bno == imap.im_blkno);
317 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
318 * default to just a read_buf() call.
320 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
321 (int)imap.im_len, XFS_BUF_LOCK, &bp);
325 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
326 "xfs_trans_read_buf() returned error %d, "
327 "imap.im_blkno 0x%llx, imap.im_len 0x%llx",
328 error, (unsigned long long) imap.im_blkno,
329 (unsigned long long) imap.im_len);
335 * Validate the magic number and version of every inode in the buffer
336 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
339 ni = (imap_flags & XFS_IMAP_BULKSTAT) ? 0 :
340 (BBTOB(imap.im_len) >> mp->m_sb.sb_inodelog);
342 ni = (imap_flags & XFS_IMAP_BULKSTAT) ? 0 : 1;
344 for (i = 0; i < ni; i++) {
348 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
349 (i << mp->m_sb.sb_inodelog));
350 di_ok = INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
351 XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
352 if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP,
353 XFS_RANDOM_ITOBP_INOTOBP))) {
355 prdev("bad inode magic/vsn daddr %lld #%d (magic=%x)",
357 (unsigned long long)imap.im_blkno, i,
358 INT_GET(dip->di_core.di_magic, ARCH_CONVERT));
360 XFS_CORRUPTION_ERROR("xfs_itobp", XFS_ERRLEVEL_HIGH,
362 xfs_trans_brelse(tp, bp);
363 return XFS_ERROR(EFSCORRUPTED);
366 #endif /* __KERNEL__ */
368 xfs_inobp_check(mp, bp);
371 * Mark the buffer as an inode buffer now that it looks good
373 XFS_BUF_SET_VTYPE(bp, B_FS_INO);
376 * Set *dipp to point to the on-disk inode in the buffer.
378 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
384 * Move inode type and inode format specific information from the
385 * on-disk inode to the in-core inode. For fifos, devs, and sockets
386 * this means set if_rdev to the proper value. For files, directories,
387 * and symlinks this means to bring in the in-line data or extent
388 * pointers. For a file in B-tree format, only the root is immediately
389 * brought in-core. The rest will be in-lined in if_extents when it
390 * is first referenced (see xfs_iread_extents()).
397 xfs_attr_shortform_t *atp;
401 ip->i_df.if_ext_max =
402 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
406 INT_GET(dip->di_core.di_nextents, ARCH_CONVERT) +
407 INT_GET(dip->di_core.di_anextents, ARCH_CONVERT) >
408 INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT))) {
409 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
410 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
411 (unsigned long long)ip->i_ino,
412 (int)(INT_GET(dip->di_core.di_nextents, ARCH_CONVERT)
413 + INT_GET(dip->di_core.di_anextents, ARCH_CONVERT)),
415 INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT));
416 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
418 return XFS_ERROR(EFSCORRUPTED);
421 if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) {
422 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
423 "corrupt dinode %Lu, forkoff = 0x%x.",
424 (unsigned long long)ip->i_ino,
425 (int)(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT)));
426 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
428 return XFS_ERROR(EFSCORRUPTED);
431 switch (ip->i_d.di_mode & S_IFMT) {
436 if (unlikely(INT_GET(dip->di_core.di_format, ARCH_CONVERT) != XFS_DINODE_FMT_DEV)) {
437 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
439 return XFS_ERROR(EFSCORRUPTED);
442 ip->i_df.if_u2.if_rdev = INT_GET(dip->di_u.di_dev, ARCH_CONVERT);
448 switch (INT_GET(dip->di_core.di_format, ARCH_CONVERT)) {
449 case XFS_DINODE_FMT_LOCAL:
451 * no local regular files yet
453 if (unlikely((INT_GET(dip->di_core.di_mode, ARCH_CONVERT) & S_IFMT) == S_IFREG)) {
454 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
456 "(local format for regular file).",
457 (unsigned long long) ip->i_ino);
458 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
461 return XFS_ERROR(EFSCORRUPTED);
464 di_size = INT_GET(dip->di_core.di_size, ARCH_CONVERT);
465 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
466 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
468 "(bad size %Ld for local inode).",
469 (unsigned long long) ip->i_ino,
470 (long long) di_size);
471 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
474 return XFS_ERROR(EFSCORRUPTED);
478 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
480 case XFS_DINODE_FMT_EXTENTS:
481 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
483 case XFS_DINODE_FMT_BTREE:
484 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
487 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
489 return XFS_ERROR(EFSCORRUPTED);
494 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
495 return XFS_ERROR(EFSCORRUPTED);
500 if (!XFS_DFORK_Q(dip))
502 ASSERT(ip->i_afp == NULL);
503 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
504 ip->i_afp->if_ext_max =
505 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
506 switch (INT_GET(dip->di_core.di_aformat, ARCH_CONVERT)) {
507 case XFS_DINODE_FMT_LOCAL:
508 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
509 size = be16_to_cpu(atp->hdr.totsize);
510 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
512 case XFS_DINODE_FMT_EXTENTS:
513 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
515 case XFS_DINODE_FMT_BTREE:
516 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
519 error = XFS_ERROR(EFSCORRUPTED);
523 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
525 xfs_idestroy_fork(ip, XFS_DATA_FORK);
531 * The file is in-lined in the on-disk inode.
532 * If it fits into if_inline_data, then copy
533 * it there, otherwise allocate a buffer for it
534 * and copy the data there. Either way, set
535 * if_data to point at the data.
536 * If we allocate a buffer for the data, make
537 * sure that its size is a multiple of 4 and
538 * record the real size in i_real_bytes.
551 * If the size is unreasonable, then something
552 * is wrong and we just bail out rather than crash in
553 * kmem_alloc() or memcpy() below.
555 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
556 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
558 "(bad size %d for local fork, size = %d).",
559 (unsigned long long) ip->i_ino, size,
560 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
561 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
563 return XFS_ERROR(EFSCORRUPTED);
565 ifp = XFS_IFORK_PTR(ip, whichfork);
568 ifp->if_u1.if_data = NULL;
569 else if (size <= sizeof(ifp->if_u2.if_inline_data))
570 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
572 real_size = roundup(size, 4);
573 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
575 ifp->if_bytes = size;
576 ifp->if_real_bytes = real_size;
578 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
579 ifp->if_flags &= ~XFS_IFEXTENTS;
580 ifp->if_flags |= XFS_IFINLINE;
585 * The file consists of a set of extents all
586 * of which fit into the on-disk inode.
587 * If there are few enough extents to fit into
588 * the if_inline_ext, then copy them there.
589 * Otherwise allocate a buffer for them and copy
590 * them into it. Either way, set if_extents
591 * to point at the extents.
599 xfs_bmbt_rec_t *ep, *dp;
605 ifp = XFS_IFORK_PTR(ip, whichfork);
606 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
607 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
610 * If the number of extents is unreasonable, then something
611 * is wrong and we just bail out rather than crash in
612 * kmem_alloc() or memcpy() below.
614 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
615 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
616 "corrupt inode %Lu ((a)extents = %d).",
617 (unsigned long long) ip->i_ino, nex);
618 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
620 return XFS_ERROR(EFSCORRUPTED);
623 ifp->if_real_bytes = 0;
625 ifp->if_u1.if_extents = NULL;
626 else if (nex <= XFS_INLINE_EXTS)
627 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
629 xfs_iext_add(ifp, 0, nex);
631 ifp->if_bytes = size;
633 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
634 xfs_validate_extents(ifp, nex, 1, XFS_EXTFMT_INODE(ip));
635 for (i = 0; i < nex; i++, dp++) {
636 ep = xfs_iext_get_ext(ifp, i);
637 ep->l0 = INT_GET(get_unaligned((__uint64_t*)&dp->l0),
639 ep->l1 = INT_GET(get_unaligned((__uint64_t*)&dp->l1),
642 xfs_bmap_trace_exlist("xfs_iformat_extents", ip, nex,
644 if (whichfork != XFS_DATA_FORK ||
645 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
646 if (unlikely(xfs_check_nostate_extents(
648 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
651 return XFS_ERROR(EFSCORRUPTED);
654 ifp->if_flags |= XFS_IFEXTENTS;
659 * The file has too many extents to fit into
660 * the inode, so they are in B-tree format.
661 * Allocate a buffer for the root of the B-tree
662 * and copy the root into it. The i_extents
663 * field will remain NULL until all of the
664 * extents are read in (when they are needed).
672 xfs_bmdr_block_t *dfp;
678 ifp = XFS_IFORK_PTR(ip, whichfork);
679 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
680 size = XFS_BMAP_BROOT_SPACE(dfp);
681 nrecs = XFS_BMAP_BROOT_NUMRECS(dfp);
684 * blow out if -- fork has less extents than can fit in
685 * fork (fork shouldn't be a btree format), root btree
686 * block has more records than can fit into the fork,
687 * or the number of extents is greater than the number of
690 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
691 || XFS_BMDR_SPACE_CALC(nrecs) >
692 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
693 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
694 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
695 "corrupt inode %Lu (btree).",
696 (unsigned long long) ip->i_ino);
697 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
699 return XFS_ERROR(EFSCORRUPTED);
702 ifp->if_broot_bytes = size;
703 ifp->if_broot = kmem_alloc(size, KM_SLEEP);
704 ASSERT(ifp->if_broot != NULL);
706 * Copy and convert from the on-disk structure
707 * to the in-memory structure.
709 xfs_bmdr_to_bmbt(dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
710 ifp->if_broot, size);
711 ifp->if_flags &= ~XFS_IFEXTENTS;
712 ifp->if_flags |= XFS_IFBROOT;
718 * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
721 * buf = on-disk representation
722 * dip = native representation
723 * dir = direction - +ve -> disk to native
724 * -ve -> native to disk
727 xfs_xlate_dinode_core(
729 xfs_dinode_core_t *dip,
732 xfs_dinode_core_t *buf_core = (xfs_dinode_core_t *)buf;
733 xfs_dinode_core_t *mem_core = (xfs_dinode_core_t *)dip;
734 xfs_arch_t arch = ARCH_CONVERT;
738 INT_XLATE(buf_core->di_magic, mem_core->di_magic, dir, arch);
739 INT_XLATE(buf_core->di_mode, mem_core->di_mode, dir, arch);
740 INT_XLATE(buf_core->di_version, mem_core->di_version, dir, arch);
741 INT_XLATE(buf_core->di_format, mem_core->di_format, dir, arch);
742 INT_XLATE(buf_core->di_onlink, mem_core->di_onlink, dir, arch);
743 INT_XLATE(buf_core->di_uid, mem_core->di_uid, dir, arch);
744 INT_XLATE(buf_core->di_gid, mem_core->di_gid, dir, arch);
745 INT_XLATE(buf_core->di_nlink, mem_core->di_nlink, dir, arch);
746 INT_XLATE(buf_core->di_projid, mem_core->di_projid, dir, arch);
749 memcpy(mem_core->di_pad, buf_core->di_pad,
750 sizeof(buf_core->di_pad));
752 memcpy(buf_core->di_pad, mem_core->di_pad,
753 sizeof(buf_core->di_pad));
756 INT_XLATE(buf_core->di_flushiter, mem_core->di_flushiter, dir, arch);
758 INT_XLATE(buf_core->di_atime.t_sec, mem_core->di_atime.t_sec,
760 INT_XLATE(buf_core->di_atime.t_nsec, mem_core->di_atime.t_nsec,
762 INT_XLATE(buf_core->di_mtime.t_sec, mem_core->di_mtime.t_sec,
764 INT_XLATE(buf_core->di_mtime.t_nsec, mem_core->di_mtime.t_nsec,
766 INT_XLATE(buf_core->di_ctime.t_sec, mem_core->di_ctime.t_sec,
768 INT_XLATE(buf_core->di_ctime.t_nsec, mem_core->di_ctime.t_nsec,
770 INT_XLATE(buf_core->di_size, mem_core->di_size, dir, arch);
771 INT_XLATE(buf_core->di_nblocks, mem_core->di_nblocks, dir, arch);
772 INT_XLATE(buf_core->di_extsize, mem_core->di_extsize, dir, arch);
773 INT_XLATE(buf_core->di_nextents, mem_core->di_nextents, dir, arch);
774 INT_XLATE(buf_core->di_anextents, mem_core->di_anextents, dir, arch);
775 INT_XLATE(buf_core->di_forkoff, mem_core->di_forkoff, dir, arch);
776 INT_XLATE(buf_core->di_aformat, mem_core->di_aformat, dir, arch);
777 INT_XLATE(buf_core->di_dmevmask, mem_core->di_dmevmask, dir, arch);
778 INT_XLATE(buf_core->di_dmstate, mem_core->di_dmstate, dir, arch);
779 INT_XLATE(buf_core->di_flags, mem_core->di_flags, dir, arch);
780 INT_XLATE(buf_core->di_gen, mem_core->di_gen, dir, arch);
785 xfs_dinode_core_t *dic,
790 if (di_flags & XFS_DIFLAG_ANY) {
791 if (di_flags & XFS_DIFLAG_REALTIME)
792 flags |= XFS_XFLAG_REALTIME;
793 if (di_flags & XFS_DIFLAG_PREALLOC)
794 flags |= XFS_XFLAG_PREALLOC;
795 if (di_flags & XFS_DIFLAG_IMMUTABLE)
796 flags |= XFS_XFLAG_IMMUTABLE;
797 if (di_flags & XFS_DIFLAG_APPEND)
798 flags |= XFS_XFLAG_APPEND;
799 if (di_flags & XFS_DIFLAG_SYNC)
800 flags |= XFS_XFLAG_SYNC;
801 if (di_flags & XFS_DIFLAG_NOATIME)
802 flags |= XFS_XFLAG_NOATIME;
803 if (di_flags & XFS_DIFLAG_NODUMP)
804 flags |= XFS_XFLAG_NODUMP;
805 if (di_flags & XFS_DIFLAG_RTINHERIT)
806 flags |= XFS_XFLAG_RTINHERIT;
807 if (di_flags & XFS_DIFLAG_PROJINHERIT)
808 flags |= XFS_XFLAG_PROJINHERIT;
809 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
810 flags |= XFS_XFLAG_NOSYMLINKS;
811 if (di_flags & XFS_DIFLAG_EXTSIZE)
812 flags |= XFS_XFLAG_EXTSIZE;
813 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
814 flags |= XFS_XFLAG_EXTSZINHERIT;
824 xfs_dinode_core_t *dic = &ip->i_d;
826 return _xfs_dic2xflags(dic, dic->di_flags) |
827 (XFS_CFORK_Q(dic) ? XFS_XFLAG_HASATTR : 0);
832 xfs_dinode_core_t *dic)
834 return _xfs_dic2xflags(dic, INT_GET(dic->di_flags, ARCH_CONVERT)) |
835 (XFS_CFORK_Q_DISK(dic) ? XFS_XFLAG_HASATTR : 0);
839 * Given a mount structure and an inode number, return a pointer
840 * to a newly allocated in-core inode corresponding to the given
843 * Initialize the inode's attributes and extent pointers if it
844 * already has them (it will not if the inode has no links).
859 ASSERT(xfs_inode_zone != NULL);
861 ip = kmem_zone_zalloc(xfs_inode_zone, KM_SLEEP);
866 * Get pointer's to the on-disk inode and the buffer containing it.
867 * If the inode number refers to a block outside the file system
868 * then xfs_itobp() will return NULL. In this case we should
869 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
870 * know that this is a new incore inode.
872 error = xfs_itobp(mp, tp, ip, &dip, &bp, bno, 0);
874 kmem_zone_free(xfs_inode_zone, ip);
879 * Initialize inode's trace buffers.
880 * Do this before xfs_iformat in case it adds entries.
882 #ifdef XFS_BMAP_TRACE
883 ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_SLEEP);
885 #ifdef XFS_BMBT_TRACE
886 ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_SLEEP);
889 ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_SLEEP);
891 #ifdef XFS_ILOCK_TRACE
892 ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_SLEEP);
894 #ifdef XFS_DIR2_TRACE
895 ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_SLEEP);
899 * If we got something that isn't an inode it means someone
900 * (nfs or dmi) has a stale handle.
902 if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC) {
903 kmem_zone_free(xfs_inode_zone, ip);
904 xfs_trans_brelse(tp, bp);
906 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
907 "dip->di_core.di_magic (0x%x) != "
908 "XFS_DINODE_MAGIC (0x%x)",
909 INT_GET(dip->di_core.di_magic, ARCH_CONVERT),
912 return XFS_ERROR(EINVAL);
916 * If the on-disk inode is already linked to a directory
917 * entry, copy all of the inode into the in-core inode.
918 * xfs_iformat() handles copying in the inode format
919 * specific information.
920 * Otherwise, just get the truly permanent information.
922 if (dip->di_core.di_mode) {
923 xfs_xlate_dinode_core((xfs_caddr_t)&dip->di_core,
925 error = xfs_iformat(ip, dip);
927 kmem_zone_free(xfs_inode_zone, ip);
928 xfs_trans_brelse(tp, bp);
930 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
931 "xfs_iformat() returned error %d",
937 ip->i_d.di_magic = INT_GET(dip->di_core.di_magic, ARCH_CONVERT);
938 ip->i_d.di_version = INT_GET(dip->di_core.di_version, ARCH_CONVERT);
939 ip->i_d.di_gen = INT_GET(dip->di_core.di_gen, ARCH_CONVERT);
940 ip->i_d.di_flushiter = INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT);
942 * Make sure to pull in the mode here as well in
943 * case the inode is released without being used.
944 * This ensures that xfs_inactive() will see that
945 * the inode is already free and not try to mess
946 * with the uninitialized part of it.
950 * Initialize the per-fork minima and maxima for a new
951 * inode here. xfs_iformat will do it for old inodes.
953 ip->i_df.if_ext_max =
954 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
957 INIT_LIST_HEAD(&ip->i_reclaim);
960 * The inode format changed when we moved the link count and
961 * made it 32 bits long. If this is an old format inode,
962 * convert it in memory to look like a new one. If it gets
963 * flushed to disk we will convert back before flushing or
964 * logging it. We zero out the new projid field and the old link
965 * count field. We'll handle clearing the pad field (the remains
966 * of the old uuid field) when we actually convert the inode to
967 * the new format. We don't change the version number so that we
968 * can distinguish this from a real new format inode.
970 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
971 ip->i_d.di_nlink = ip->i_d.di_onlink;
972 ip->i_d.di_onlink = 0;
973 ip->i_d.di_projid = 0;
976 ip->i_delayed_blks = 0;
979 * Mark the buffer containing the inode as something to keep
980 * around for a while. This helps to keep recently accessed
981 * meta-data in-core longer.
983 XFS_BUF_SET_REF(bp, XFS_INO_REF);
986 * Use xfs_trans_brelse() to release the buffer containing the
987 * on-disk inode, because it was acquired with xfs_trans_read_buf()
988 * in xfs_itobp() above. If tp is NULL, this is just a normal
989 * brelse(). If we're within a transaction, then xfs_trans_brelse()
990 * will only release the buffer if it is not dirty within the
991 * transaction. It will be OK to release the buffer in this case,
992 * because inodes on disk are never destroyed and we will be
993 * locking the new in-core inode before putting it in the hash
994 * table where other processes can find it. Thus we don't have
995 * to worry about the inode being changed just because we released
998 xfs_trans_brelse(tp, bp);
1004 * Read in extents from a btree-format inode.
1005 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1015 xfs_extnum_t nextents;
1018 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
1019 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
1021 return XFS_ERROR(EFSCORRUPTED);
1023 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
1024 size = nextents * sizeof(xfs_bmbt_rec_t);
1025 ifp = XFS_IFORK_PTR(ip, whichfork);
1028 * We know that the size is valid (it's checked in iformat_btree)
1030 ifp->if_lastex = NULLEXTNUM;
1031 ifp->if_bytes = ifp->if_real_bytes = 0;
1032 ifp->if_flags |= XFS_IFEXTENTS;
1033 xfs_iext_add(ifp, 0, nextents);
1034 error = xfs_bmap_read_extents(tp, ip, whichfork);
1036 xfs_iext_destroy(ifp);
1037 ifp->if_flags &= ~XFS_IFEXTENTS;
1040 xfs_validate_extents(ifp, nextents, 0, XFS_EXTFMT_INODE(ip));
1045 * Allocate an inode on disk and return a copy of its in-core version.
1046 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1047 * appropriately within the inode. The uid and gid for the inode are
1048 * set according to the contents of the given cred structure.
1050 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1051 * has a free inode available, call xfs_iget()
1052 * to obtain the in-core version of the allocated inode. Finally,
1053 * fill in the inode and log its initial contents. In this case,
1054 * ialloc_context would be set to NULL and call_again set to false.
1056 * If xfs_dialloc() does not have an available inode,
1057 * it will replenish its supply by doing an allocation. Since we can
1058 * only do one allocation within a transaction without deadlocks, we
1059 * must commit the current transaction before returning the inode itself.
1060 * In this case, therefore, we will set call_again to true and return.
1061 * The caller should then commit the current transaction, start a new
1062 * transaction, and call xfs_ialloc() again to actually get the inode.
1064 * To ensure that some other process does not grab the inode that
1065 * was allocated during the first call to xfs_ialloc(), this routine
1066 * also returns the [locked] bp pointing to the head of the freelist
1067 * as ialloc_context. The caller should hold this buffer across
1068 * the commit and pass it back into this routine on the second call.
1080 xfs_buf_t **ialloc_context,
1081 boolean_t *call_again,
1091 * Call the space management code to pick
1092 * the on-disk inode to be allocated.
1094 error = xfs_dialloc(tp, pip->i_ino, mode, okalloc,
1095 ialloc_context, call_again, &ino);
1099 if (*call_again || ino == NULLFSINO) {
1103 ASSERT(*ialloc_context == NULL);
1106 * Get the in-core inode with the lock held exclusively.
1107 * This is because we're setting fields here we need
1108 * to prevent others from looking at until we're done.
1110 error = xfs_trans_iget(tp->t_mountp, tp, ino,
1111 IGET_CREATE, XFS_ILOCK_EXCL, &ip);
1118 ip->i_d.di_mode = (__uint16_t)mode;
1119 ip->i_d.di_onlink = 0;
1120 ip->i_d.di_nlink = nlink;
1121 ASSERT(ip->i_d.di_nlink == nlink);
1122 ip->i_d.di_uid = current_fsuid(cr);
1123 ip->i_d.di_gid = current_fsgid(cr);
1124 ip->i_d.di_projid = prid;
1125 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1128 * If the superblock version is up to where we support new format
1129 * inodes and this is currently an old format inode, then change
1130 * the inode version number now. This way we only do the conversion
1131 * here rather than here and in the flush/logging code.
1133 if (XFS_SB_VERSION_HASNLINK(&tp->t_mountp->m_sb) &&
1134 ip->i_d.di_version == XFS_DINODE_VERSION_1) {
1135 ip->i_d.di_version = XFS_DINODE_VERSION_2;
1137 * We've already zeroed the old link count, the projid field,
1138 * and the pad field.
1143 * Project ids won't be stored on disk if we are using a version 1 inode.
1145 if ( (prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1))
1146 xfs_bump_ino_vers2(tp, ip);
1148 if (XFS_INHERIT_GID(pip, vp->v_vfsp)) {
1149 ip->i_d.di_gid = pip->i_d.di_gid;
1150 if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1151 ip->i_d.di_mode |= S_ISGID;
1156 * If the group ID of the new file does not match the effective group
1157 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1158 * (and only if the irix_sgid_inherit compatibility variable is set).
1160 if ((irix_sgid_inherit) &&
1161 (ip->i_d.di_mode & S_ISGID) &&
1162 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1163 ip->i_d.di_mode &= ~S_ISGID;
1166 ip->i_d.di_size = 0;
1167 ip->i_d.di_nextents = 0;
1168 ASSERT(ip->i_d.di_nblocks == 0);
1169 xfs_ichgtime(ip, XFS_ICHGTIME_CHG|XFS_ICHGTIME_ACC|XFS_ICHGTIME_MOD);
1171 * di_gen will have been taken care of in xfs_iread.
1173 ip->i_d.di_extsize = 0;
1174 ip->i_d.di_dmevmask = 0;
1175 ip->i_d.di_dmstate = 0;
1176 ip->i_d.di_flags = 0;
1177 flags = XFS_ILOG_CORE;
1178 switch (mode & S_IFMT) {
1183 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1184 ip->i_df.if_u2.if_rdev = rdev;
1185 ip->i_df.if_flags = 0;
1186 flags |= XFS_ILOG_DEV;
1190 if (unlikely(pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
1193 if ((mode & S_IFMT) == S_IFDIR) {
1194 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1195 di_flags |= XFS_DIFLAG_RTINHERIT;
1196 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1197 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1198 ip->i_d.di_extsize = pip->i_d.di_extsize;
1200 } else if ((mode & S_IFMT) == S_IFREG) {
1201 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) {
1202 di_flags |= XFS_DIFLAG_REALTIME;
1203 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
1205 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1206 di_flags |= XFS_DIFLAG_EXTSIZE;
1207 ip->i_d.di_extsize = pip->i_d.di_extsize;
1210 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1211 xfs_inherit_noatime)
1212 di_flags |= XFS_DIFLAG_NOATIME;
1213 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1215 di_flags |= XFS_DIFLAG_NODUMP;
1216 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1218 di_flags |= XFS_DIFLAG_SYNC;
1219 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1220 xfs_inherit_nosymlinks)
1221 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1222 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1223 di_flags |= XFS_DIFLAG_PROJINHERIT;
1224 ip->i_d.di_flags |= di_flags;
1228 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1229 ip->i_df.if_flags = XFS_IFEXTENTS;
1230 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1231 ip->i_df.if_u1.if_extents = NULL;
1237 * Attribute fork settings for new inode.
1239 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1240 ip->i_d.di_anextents = 0;
1243 * Log the new values stuffed into the inode.
1245 xfs_trans_log_inode(tp, ip, flags);
1247 /* now that we have an i_mode we can set Linux inode ops (& unlock) */
1248 VFS_INIT_VNODE(XFS_MTOVFS(tp->t_mountp), vp, XFS_ITOBHV(ip), 1);
1255 * Check to make sure that there are no blocks allocated to the
1256 * file beyond the size of the file. We don't check this for
1257 * files with fixed size extents or real time extents, but we
1258 * at least do it for regular files.
1267 xfs_fileoff_t map_first;
1269 xfs_bmbt_irec_t imaps[2];
1271 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1274 if (ip->i_d.di_flags & (XFS_DIFLAG_REALTIME | XFS_DIFLAG_EXTSIZE))
1278 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1280 * The filesystem could be shutting down, so bmapi may return
1283 if (xfs_bmapi(NULL, ip, map_first,
1285 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1287 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1290 ASSERT(nimaps == 1);
1291 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1296 * Calculate the last possible buffered byte in a file. This must
1297 * include data that was buffered beyond the EOF by the write code.
1298 * This also needs to deal with overflowing the xfs_fsize_t type
1299 * which can happen for sizes near the limit.
1301 * We also need to take into account any blocks beyond the EOF. It
1302 * may be the case that they were buffered by a write which failed.
1303 * In that case the pages will still be in memory, but the inode size
1304 * will never have been updated.
1311 xfs_fsize_t last_byte;
1312 xfs_fileoff_t last_block;
1313 xfs_fileoff_t size_last_block;
1316 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE | MR_ACCESS));
1320 * Only check for blocks beyond the EOF if the extents have
1321 * been read in. This eliminates the need for the inode lock,
1322 * and it also saves us from looking when it really isn't
1325 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1326 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1334 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_d.di_size);
1335 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1337 last_byte = XFS_FSB_TO_B(mp, last_block);
1338 if (last_byte < 0) {
1339 return XFS_MAXIOFFSET(mp);
1341 last_byte += (1 << mp->m_writeio_log);
1342 if (last_byte < 0) {
1343 return XFS_MAXIOFFSET(mp);
1348 #if defined(XFS_RW_TRACE)
1354 xfs_fsize_t new_size,
1355 xfs_off_t toss_start,
1356 xfs_off_t toss_finish)
1358 if (ip->i_rwtrace == NULL) {
1362 ktrace_enter(ip->i_rwtrace,
1365 (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
1366 (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
1367 (void*)((long)flag),
1368 (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
1369 (void*)(unsigned long)(new_size & 0xffffffff),
1370 (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
1371 (void*)(unsigned long)(toss_start & 0xffffffff),
1372 (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
1373 (void*)(unsigned long)(toss_finish & 0xffffffff),
1374 (void*)(unsigned long)current_cpu(),
1375 (void*)(unsigned long)current_pid(),
1381 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1385 * Start the truncation of the file to new_size. The new size
1386 * must be smaller than the current size. This routine will
1387 * clear the buffer and page caches of file data in the removed
1388 * range, and xfs_itruncate_finish() will remove the underlying
1391 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1392 * must NOT have the inode lock held at all. This is because we're
1393 * calling into the buffer/page cache code and we can't hold the
1394 * inode lock when we do so.
1396 * We need to wait for any direct I/Os in flight to complete before we
1397 * proceed with the truncate. This is needed to prevent the extents
1398 * being read or written by the direct I/Os from being removed while the
1399 * I/O is in flight as there is no other method of synchronising
1400 * direct I/O with the truncate operation. Also, because we hold
1401 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1402 * started until the truncate completes and drops the lock. Essentially,
1403 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1404 * between direct I/Os and the truncate operation.
1406 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1407 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1408 * in the case that the caller is locking things out of order and
1409 * may not be able to call xfs_itruncate_finish() with the inode lock
1410 * held without dropping the I/O lock. If the caller must drop the
1411 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1412 * must be called again with all the same restrictions as the initial
1416 xfs_itruncate_start(
1419 xfs_fsize_t new_size)
1421 xfs_fsize_t last_byte;
1422 xfs_off_t toss_start;
1426 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1427 ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
1428 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1429 (flags == XFS_ITRUNC_MAYBE));
1434 vn_iowait(vp); /* wait for the completion of any pending DIOs */
1437 * Call VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES() to get rid of pages and buffers
1438 * overlapping the region being removed. We have to use
1439 * the less efficient VOP_FLUSHINVAL_PAGES() in the case that the
1440 * caller may not be able to finish the truncate without
1441 * dropping the inode's I/O lock. Make sure
1442 * to catch any pages brought in by buffers overlapping
1443 * the EOF by searching out beyond the isize by our
1444 * block size. We round new_size up to a block boundary
1445 * so that we don't toss things on the same block as
1446 * new_size but before it.
1448 * Before calling VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES(), make sure to
1449 * call remapf() over the same region if the file is mapped.
1450 * This frees up mapped file references to the pages in the
1451 * given range and for the VOP_FLUSHINVAL_PAGES() case it ensures
1452 * that we get the latest mapped changes flushed out.
1454 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1455 toss_start = XFS_FSB_TO_B(mp, toss_start);
1456 if (toss_start < 0) {
1458 * The place to start tossing is beyond our maximum
1459 * file size, so there is no way that the data extended
1464 last_byte = xfs_file_last_byte(ip);
1465 xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
1467 if (last_byte > toss_start) {
1468 if (flags & XFS_ITRUNC_DEFINITE) {
1469 VOP_TOSS_PAGES(vp, toss_start, -1, FI_REMAPF_LOCKED);
1471 VOP_FLUSHINVAL_PAGES(vp, toss_start, -1, FI_REMAPF_LOCKED);
1476 if (new_size == 0) {
1477 ASSERT(VN_CACHED(vp) == 0);
1483 * Shrink the file to the given new_size. The new
1484 * size must be smaller than the current size.
1485 * This will free up the underlying blocks
1486 * in the removed range after a call to xfs_itruncate_start()
1487 * or xfs_atruncate_start().
1489 * The transaction passed to this routine must have made
1490 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1491 * This routine may commit the given transaction and
1492 * start new ones, so make sure everything involved in
1493 * the transaction is tidy before calling here.
1494 * Some transaction will be returned to the caller to be
1495 * committed. The incoming transaction must already include
1496 * the inode, and both inode locks must be held exclusively.
1497 * The inode must also be "held" within the transaction. On
1498 * return the inode will be "held" within the returned transaction.
1499 * This routine does NOT require any disk space to be reserved
1500 * for it within the transaction.
1502 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1503 * and it indicates the fork which is to be truncated. For the
1504 * attribute fork we only support truncation to size 0.
1506 * We use the sync parameter to indicate whether or not the first
1507 * transaction we perform might have to be synchronous. For the attr fork,
1508 * it needs to be so if the unlink of the inode is not yet known to be
1509 * permanent in the log. This keeps us from freeing and reusing the
1510 * blocks of the attribute fork before the unlink of the inode becomes
1513 * For the data fork, we normally have to run synchronously if we're
1514 * being called out of the inactive path or we're being called
1515 * out of the create path where we're truncating an existing file.
1516 * Either way, the truncate needs to be sync so blocks don't reappear
1517 * in the file with altered data in case of a crash. wsync filesystems
1518 * can run the first case async because anything that shrinks the inode
1519 * has to run sync so by the time we're called here from inactive, the
1520 * inode size is permanently set to 0.
1522 * Calls from the truncate path always need to be sync unless we're
1523 * in a wsync filesystem and the file has already been unlinked.
1525 * The caller is responsible for correctly setting the sync parameter.
1526 * It gets too hard for us to guess here which path we're being called
1527 * out of just based on inode state.
1530 xfs_itruncate_finish(
1533 xfs_fsize_t new_size,
1537 xfs_fsblock_t first_block;
1538 xfs_fileoff_t first_unmap_block;
1539 xfs_fileoff_t last_block;
1540 xfs_filblks_t unmap_len=0;
1545 xfs_bmap_free_t free_list;
1548 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1549 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
1550 ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
1551 ASSERT(*tp != NULL);
1552 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1553 ASSERT(ip->i_transp == *tp);
1554 ASSERT(ip->i_itemp != NULL);
1555 ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
1559 mp = (ntp)->t_mountp;
1560 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1563 * We only support truncating the entire attribute fork.
1565 if (fork == XFS_ATTR_FORK) {
1568 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1569 xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
1571 * The first thing we do is set the size to new_size permanently
1572 * on disk. This way we don't have to worry about anyone ever
1573 * being able to look at the data being freed even in the face
1574 * of a crash. What we're getting around here is the case where
1575 * we free a block, it is allocated to another file, it is written
1576 * to, and then we crash. If the new data gets written to the
1577 * file but the log buffers containing the free and reallocation
1578 * don't, then we'd end up with garbage in the blocks being freed.
1579 * As long as we make the new_size permanent before actually
1580 * freeing any blocks it doesn't matter if they get writtten to.
1582 * The callers must signal into us whether or not the size
1583 * setting here must be synchronous. There are a few cases
1584 * where it doesn't have to be synchronous. Those cases
1585 * occur if the file is unlinked and we know the unlink is
1586 * permanent or if the blocks being truncated are guaranteed
1587 * to be beyond the inode eof (regardless of the link count)
1588 * and the eof value is permanent. Both of these cases occur
1589 * only on wsync-mounted filesystems. In those cases, we're
1590 * guaranteed that no user will ever see the data in the blocks
1591 * that are being truncated so the truncate can run async.
1592 * In the free beyond eof case, the file may wind up with
1593 * more blocks allocated to it than it needs if we crash
1594 * and that won't get fixed until the next time the file
1595 * is re-opened and closed but that's ok as that shouldn't
1596 * be too many blocks.
1598 * However, we can't just make all wsync xactions run async
1599 * because there's one call out of the create path that needs
1600 * to run sync where it's truncating an existing file to size
1601 * 0 whose size is > 0.
1603 * It's probably possible to come up with a test in this
1604 * routine that would correctly distinguish all the above
1605 * cases from the values of the function parameters and the
1606 * inode state but for sanity's sake, I've decided to let the
1607 * layers above just tell us. It's simpler to correctly figure
1608 * out in the layer above exactly under what conditions we
1609 * can run async and I think it's easier for others read and
1610 * follow the logic in case something has to be changed.
1611 * cscope is your friend -- rcc.
1613 * The attribute fork is much simpler.
1615 * For the attribute fork we allow the caller to tell us whether
1616 * the unlink of the inode that led to this call is yet permanent
1617 * in the on disk log. If it is not and we will be freeing extents
1618 * in this inode then we make the first transaction synchronous
1619 * to make sure that the unlink is permanent by the time we free
1622 if (fork == XFS_DATA_FORK) {
1623 if (ip->i_d.di_nextents > 0) {
1624 ip->i_d.di_size = new_size;
1625 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1628 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1629 if (ip->i_d.di_anextents > 0)
1630 xfs_trans_set_sync(ntp);
1632 ASSERT(fork == XFS_DATA_FORK ||
1633 (fork == XFS_ATTR_FORK &&
1634 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1635 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1638 * Since it is possible for space to become allocated beyond
1639 * the end of the file (in a crash where the space is allocated
1640 * but the inode size is not yet updated), simply remove any
1641 * blocks which show up between the new EOF and the maximum
1642 * possible file size. If the first block to be removed is
1643 * beyond the maximum file size (ie it is the same as last_block),
1644 * then there is nothing to do.
1646 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1647 ASSERT(first_unmap_block <= last_block);
1649 if (last_block == first_unmap_block) {
1652 unmap_len = last_block - first_unmap_block + 1;
1656 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1657 * will tell us whether it freed the entire range or
1658 * not. If this is a synchronous mount (wsync),
1659 * then we can tell bunmapi to keep all the
1660 * transactions asynchronous since the unlink
1661 * transaction that made this inode inactive has
1662 * already hit the disk. There's no danger of
1663 * the freed blocks being reused, there being a
1664 * crash, and the reused blocks suddenly reappearing
1665 * in this file with garbage in them once recovery
1668 XFS_BMAP_INIT(&free_list, &first_block);
1669 error = XFS_BUNMAPI(mp, ntp, &ip->i_iocore,
1670 first_unmap_block, unmap_len,
1671 XFS_BMAPI_AFLAG(fork) |
1672 (sync ? 0 : XFS_BMAPI_ASYNC),
1673 XFS_ITRUNC_MAX_EXTENTS,
1674 &first_block, &free_list,
1678 * If the bunmapi call encounters an error,
1679 * return to the caller where the transaction
1680 * can be properly aborted. We just need to
1681 * make sure we're not holding any resources
1682 * that we were not when we came in.
1684 xfs_bmap_cancel(&free_list);
1689 * Duplicate the transaction that has the permanent
1690 * reservation and commit the old transaction.
1692 error = xfs_bmap_finish(tp, &free_list, first_block,
1697 * If the bmap finish call encounters an error,
1698 * return to the caller where the transaction
1699 * can be properly aborted. We just need to
1700 * make sure we're not holding any resources
1701 * that we were not when we came in.
1703 * Aborting from this point might lose some
1704 * blocks in the file system, but oh well.
1706 xfs_bmap_cancel(&free_list);
1709 * If the passed in transaction committed
1710 * in xfs_bmap_finish(), then we want to
1711 * add the inode to this one before returning.
1712 * This keeps things simple for the higher
1713 * level code, because it always knows that
1714 * the inode is locked and held in the
1715 * transaction that returns to it whether
1716 * errors occur or not. We don't mark the
1717 * inode dirty so that this transaction can
1718 * be easily aborted if possible.
1720 xfs_trans_ijoin(ntp, ip,
1721 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1722 xfs_trans_ihold(ntp, ip);
1729 * The first xact was committed,
1730 * so add the inode to the new one.
1731 * Mark it dirty so it will be logged
1732 * and moved forward in the log as
1733 * part of every commit.
1735 xfs_trans_ijoin(ntp, ip,
1736 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1737 xfs_trans_ihold(ntp, ip);
1738 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1740 ntp = xfs_trans_dup(ntp);
1741 (void) xfs_trans_commit(*tp, 0, NULL);
1743 error = xfs_trans_reserve(ntp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1744 XFS_TRANS_PERM_LOG_RES,
1745 XFS_ITRUNCATE_LOG_COUNT);
1747 * Add the inode being truncated to the next chained
1750 xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1751 xfs_trans_ihold(ntp, ip);
1756 * Only update the size in the case of the data fork, but
1757 * always re-log the inode so that our permanent transaction
1758 * can keep on rolling it forward in the log.
1760 if (fork == XFS_DATA_FORK) {
1761 xfs_isize_check(mp, ip, new_size);
1762 ip->i_d.di_size = new_size;
1764 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1765 ASSERT((new_size != 0) ||
1766 (fork == XFS_ATTR_FORK) ||
1767 (ip->i_delayed_blks == 0));
1768 ASSERT((new_size != 0) ||
1769 (fork == XFS_ATTR_FORK) ||
1770 (ip->i_d.di_nextents == 0));
1771 xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
1779 * Do the first part of growing a file: zero any data in the last
1780 * block that is beyond the old EOF. We need to do this before
1781 * the inode is joined to the transaction to modify the i_size.
1782 * That way we can drop the inode lock and call into the buffer
1783 * cache to get the buffer mapping the EOF.
1788 xfs_fsize_t new_size,
1793 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1794 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1795 ASSERT(new_size > ip->i_d.di_size);
1798 * Zero any pages that may have been created by
1799 * xfs_write_file() beyond the end of the file
1800 * and any blocks between the old and new file sizes.
1802 error = xfs_zero_eof(XFS_ITOV(ip), &ip->i_iocore, new_size,
1803 ip->i_d.di_size, new_size);
1810 * This routine is called to extend the size of a file.
1811 * The inode must have both the iolock and the ilock locked
1812 * for update and it must be a part of the current transaction.
1813 * The xfs_igrow_start() function must have been called previously.
1814 * If the change_flag is not zero, the inode change timestamp will
1821 xfs_fsize_t new_size,
1824 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1825 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1826 ASSERT(ip->i_transp == tp);
1827 ASSERT(new_size > ip->i_d.di_size);
1830 * Update the file size. Update the inode change timestamp
1831 * if change_flag set.
1833 ip->i_d.di_size = new_size;
1835 xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
1836 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1842 * This is called when the inode's link count goes to 0.
1843 * We place the on-disk inode on a list in the AGI. It
1844 * will be pulled from this list when the inode is freed.
1856 xfs_agnumber_t agno;
1857 xfs_daddr_t agdaddr;
1864 ASSERT(ip->i_d.di_nlink == 0);
1865 ASSERT(ip->i_d.di_mode != 0);
1866 ASSERT(ip->i_transp == tp);
1870 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1871 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1874 * Get the agi buffer first. It ensures lock ordering
1877 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1878 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1883 * Validate the magic number of the agi block.
1885 agi = XFS_BUF_TO_AGI(agibp);
1887 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1888 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
1889 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK,
1890 XFS_RANDOM_IUNLINK))) {
1891 XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi);
1892 xfs_trans_brelse(tp, agibp);
1893 return XFS_ERROR(EFSCORRUPTED);
1896 * Get the index into the agi hash table for the
1897 * list this inode will go on.
1899 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1901 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1902 ASSERT(agi->agi_unlinked[bucket_index]);
1903 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1905 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
1907 * There is already another inode in the bucket we need
1908 * to add ourselves to. Add us at the front of the list.
1909 * Here we put the head pointer into our next pointer,
1910 * and then we fall through to point the head at us.
1912 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
1916 ASSERT(INT_GET(dip->di_next_unlinked, ARCH_CONVERT) == NULLAGINO);
1917 ASSERT(dip->di_next_unlinked);
1918 /* both on-disk, don't endian flip twice */
1919 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1920 offset = ip->i_boffset +
1921 offsetof(xfs_dinode_t, di_next_unlinked);
1922 xfs_trans_inode_buf(tp, ibp);
1923 xfs_trans_log_buf(tp, ibp, offset,
1924 (offset + sizeof(xfs_agino_t) - 1));
1925 xfs_inobp_check(mp, ibp);
1929 * Point the bucket head pointer at the inode being inserted.
1932 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1933 offset = offsetof(xfs_agi_t, agi_unlinked) +
1934 (sizeof(xfs_agino_t) * bucket_index);
1935 xfs_trans_log_buf(tp, agibp, offset,
1936 (offset + sizeof(xfs_agino_t) - 1));
1941 * Pull the on-disk inode from the AGI unlinked list.
1954 xfs_agnumber_t agno;
1955 xfs_daddr_t agdaddr;
1957 xfs_agino_t next_agino;
1958 xfs_buf_t *last_ibp;
1959 xfs_dinode_t *last_dip;
1961 int offset, last_offset;
1966 * First pull the on-disk inode from the AGI unlinked list.
1970 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1971 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1974 * Get the agi buffer first. It ensures lock ordering
1977 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1978 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1981 "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
1982 error, mp->m_fsname);
1986 * Validate the magic number of the agi block.
1988 agi = XFS_BUF_TO_AGI(agibp);
1990 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1991 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
1992 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE,
1993 XFS_RANDOM_IUNLINK_REMOVE))) {
1994 XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW,
1996 xfs_trans_brelse(tp, agibp);
1998 "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
2000 return XFS_ERROR(EFSCORRUPTED);
2003 * Get the index into the agi hash table for the
2004 * list this inode will go on.
2006 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2008 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2009 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
2010 ASSERT(agi->agi_unlinked[bucket_index]);
2012 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2014 * We're at the head of the list. Get the inode's
2015 * on-disk buffer to see if there is anyone after us
2016 * on the list. Only modify our next pointer if it
2017 * is not already NULLAGINO. This saves us the overhead
2018 * of dealing with the buffer when there is no need to
2021 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
2024 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2025 error, mp->m_fsname);
2028 next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
2029 ASSERT(next_agino != 0);
2030 if (next_agino != NULLAGINO) {
2031 INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
2032 offset = ip->i_boffset +
2033 offsetof(xfs_dinode_t, di_next_unlinked);
2034 xfs_trans_inode_buf(tp, ibp);
2035 xfs_trans_log_buf(tp, ibp, offset,
2036 (offset + sizeof(xfs_agino_t) - 1));
2037 xfs_inobp_check(mp, ibp);
2039 xfs_trans_brelse(tp, ibp);
2042 * Point the bucket head pointer at the next inode.
2044 ASSERT(next_agino != 0);
2045 ASSERT(next_agino != agino);
2046 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2047 offset = offsetof(xfs_agi_t, agi_unlinked) +
2048 (sizeof(xfs_agino_t) * bucket_index);
2049 xfs_trans_log_buf(tp, agibp, offset,
2050 (offset + sizeof(xfs_agino_t) - 1));
2053 * We need to search the list for the inode being freed.
2055 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2057 while (next_agino != agino) {
2059 * If the last inode wasn't the one pointing to
2060 * us, then release its buffer since we're not
2061 * going to do anything with it.
2063 if (last_ibp != NULL) {
2064 xfs_trans_brelse(tp, last_ibp);
2066 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2067 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
2068 &last_ibp, &last_offset);
2071 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
2072 error, mp->m_fsname);
2075 next_agino = INT_GET(last_dip->di_next_unlinked, ARCH_CONVERT);
2076 ASSERT(next_agino != NULLAGINO);
2077 ASSERT(next_agino != 0);
2080 * Now last_ibp points to the buffer previous to us on
2081 * the unlinked list. Pull us from the list.
2083 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
2086 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2087 error, mp->m_fsname);
2090 next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
2091 ASSERT(next_agino != 0);
2092 ASSERT(next_agino != agino);
2093 if (next_agino != NULLAGINO) {
2094 INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
2095 offset = ip->i_boffset +
2096 offsetof(xfs_dinode_t, di_next_unlinked);
2097 xfs_trans_inode_buf(tp, ibp);
2098 xfs_trans_log_buf(tp, ibp, offset,
2099 (offset + sizeof(xfs_agino_t) - 1));
2100 xfs_inobp_check(mp, ibp);
2102 xfs_trans_brelse(tp, ibp);
2105 * Point the previous inode on the list to the next inode.
2107 INT_SET(last_dip->di_next_unlinked, ARCH_CONVERT, next_agino);
2108 ASSERT(next_agino != 0);
2109 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2110 xfs_trans_inode_buf(tp, last_ibp);
2111 xfs_trans_log_buf(tp, last_ibp, offset,
2112 (offset + sizeof(xfs_agino_t) - 1));
2113 xfs_inobp_check(mp, last_ibp);
2118 static __inline__ int xfs_inode_clean(xfs_inode_t *ip)
2120 return (((ip->i_itemp == NULL) ||
2121 !(ip->i_itemp->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
2122 (ip->i_update_core == 0));
2127 xfs_inode_t *free_ip,
2131 xfs_mount_t *mp = free_ip->i_mount;
2132 int blks_per_cluster;
2135 int i, j, found, pre_flushed;
2139 xfs_inode_t *ip, **ip_found;
2140 xfs_inode_log_item_t *iip;
2141 xfs_log_item_t *lip;
2144 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
2145 blks_per_cluster = 1;
2146 ninodes = mp->m_sb.sb_inopblock;
2147 nbufs = XFS_IALLOC_BLOCKS(mp);
2149 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
2150 mp->m_sb.sb_blocksize;
2151 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
2152 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
2155 ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
2157 for (j = 0; j < nbufs; j++, inum += ninodes) {
2158 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2159 XFS_INO_TO_AGBNO(mp, inum));
2163 * Look for each inode in memory and attempt to lock it,
2164 * we can be racing with flush and tail pushing here.
2165 * any inode we get the locks on, add to an array of
2166 * inode items to process later.
2168 * The get the buffer lock, we could beat a flush
2169 * or tail pushing thread to the lock here, in which
2170 * case they will go looking for the inode buffer
2171 * and fail, we need some other form of interlock
2175 for (i = 0; i < ninodes; i++) {
2176 ih = XFS_IHASH(mp, inum + i);
2177 read_lock(&ih->ih_lock);
2178 for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
2179 if (ip->i_ino == inum + i)
2183 /* Inode not in memory or we found it already,
2186 if (!ip || (ip->i_flags & XFS_ISTALE)) {
2187 read_unlock(&ih->ih_lock);
2191 if (xfs_inode_clean(ip)) {
2192 read_unlock(&ih->ih_lock);
2196 /* If we can get the locks then add it to the
2197 * list, otherwise by the time we get the bp lock
2198 * below it will already be attached to the
2202 /* This inode will already be locked - by us, lets
2206 if (ip == free_ip) {
2207 if (xfs_iflock_nowait(ip)) {
2208 ip->i_flags |= XFS_ISTALE;
2210 if (xfs_inode_clean(ip)) {
2213 ip_found[found++] = ip;
2216 read_unlock(&ih->ih_lock);
2220 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2221 if (xfs_iflock_nowait(ip)) {
2222 ip->i_flags |= XFS_ISTALE;
2224 if (xfs_inode_clean(ip)) {
2226 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2228 ip_found[found++] = ip;
2231 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2235 read_unlock(&ih->ih_lock);
2238 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2239 mp->m_bsize * blks_per_cluster,
2243 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
2245 if (lip->li_type == XFS_LI_INODE) {
2246 iip = (xfs_inode_log_item_t *)lip;
2247 ASSERT(iip->ili_logged == 1);
2248 lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
2250 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2252 iip->ili_inode->i_flags |= XFS_ISTALE;
2255 lip = lip->li_bio_list;
2258 for (i = 0; i < found; i++) {
2263 ip->i_update_core = 0;
2265 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2269 iip->ili_last_fields = iip->ili_format.ilf_fields;
2270 iip->ili_format.ilf_fields = 0;
2271 iip->ili_logged = 1;
2273 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2276 xfs_buf_attach_iodone(bp,
2277 (void(*)(xfs_buf_t*,xfs_log_item_t*))
2278 xfs_istale_done, (xfs_log_item_t *)iip);
2279 if (ip != free_ip) {
2280 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2284 if (found || pre_flushed)
2285 xfs_trans_stale_inode_buf(tp, bp);
2286 xfs_trans_binval(tp, bp);
2289 kmem_free(ip_found, ninodes * sizeof(xfs_inode_t *));
2293 * This is called to return an inode to the inode free list.
2294 * The inode should already be truncated to 0 length and have
2295 * no pages associated with it. This routine also assumes that
2296 * the inode is already a part of the transaction.
2298 * The on-disk copy of the inode will have been added to the list
2299 * of unlinked inodes in the AGI. We need to remove the inode from
2300 * that list atomically with respect to freeing it here.
2306 xfs_bmap_free_t *flist)
2310 xfs_ino_t first_ino;
2312 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2313 ASSERT(ip->i_transp == tp);
2314 ASSERT(ip->i_d.di_nlink == 0);
2315 ASSERT(ip->i_d.di_nextents == 0);
2316 ASSERT(ip->i_d.di_anextents == 0);
2317 ASSERT((ip->i_d.di_size == 0) ||
2318 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2319 ASSERT(ip->i_d.di_nblocks == 0);
2322 * Pull the on-disk inode from the AGI unlinked list.
2324 error = xfs_iunlink_remove(tp, ip);
2329 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2333 ip->i_d.di_mode = 0; /* mark incore inode as free */
2334 ip->i_d.di_flags = 0;
2335 ip->i_d.di_dmevmask = 0;
2336 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2337 ip->i_df.if_ext_max =
2338 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2339 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2340 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2342 * Bump the generation count so no one will be confused
2343 * by reincarnations of this inode.
2346 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2349 xfs_ifree_cluster(ip, tp, first_ino);
2356 * Reallocate the space for if_broot based on the number of records
2357 * being added or deleted as indicated in rec_diff. Move the records
2358 * and pointers in if_broot to fit the new size. When shrinking this
2359 * will eliminate holes between the records and pointers created by
2360 * the caller. When growing this will create holes to be filled in
2363 * The caller must not request to add more records than would fit in
2364 * the on-disk inode root. If the if_broot is currently NULL, then
2365 * if we adding records one will be allocated. The caller must also
2366 * not request that the number of records go below zero, although
2367 * it can go to zero.
2369 * ip -- the inode whose if_broot area is changing
2370 * ext_diff -- the change in the number of records, positive or negative,
2371 * requested for the if_broot array.
2381 xfs_bmbt_block_t *new_broot;
2388 * Handle the degenerate case quietly.
2390 if (rec_diff == 0) {
2394 ifp = XFS_IFORK_PTR(ip, whichfork);
2397 * If there wasn't any memory allocated before, just
2398 * allocate it now and get out.
2400 if (ifp->if_broot_bytes == 0) {
2401 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
2402 ifp->if_broot = (xfs_bmbt_block_t*)kmem_alloc(new_size,
2404 ifp->if_broot_bytes = (int)new_size;
2409 * If there is already an existing if_broot, then we need
2410 * to realloc() it and shift the pointers to their new
2411 * location. The records don't change location because
2412 * they are kept butted up against the btree block header.
2414 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2415 new_max = cur_max + rec_diff;
2416 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2417 ifp->if_broot = (xfs_bmbt_block_t *)
2418 kmem_realloc(ifp->if_broot,
2420 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2422 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2423 ifp->if_broot_bytes);
2424 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2426 ifp->if_broot_bytes = (int)new_size;
2427 ASSERT(ifp->if_broot_bytes <=
2428 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2429 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2434 * rec_diff is less than 0. In this case, we are shrinking the
2435 * if_broot buffer. It must already exist. If we go to zero
2436 * records, just get rid of the root and clear the status bit.
2438 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
2439 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2440 new_max = cur_max + rec_diff;
2441 ASSERT(new_max >= 0);
2443 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2447 new_broot = (xfs_bmbt_block_t *)kmem_alloc(new_size, KM_SLEEP);
2449 * First copy over the btree block header.
2451 memcpy(new_broot, ifp->if_broot, sizeof(xfs_bmbt_block_t));
2454 ifp->if_flags &= ~XFS_IFBROOT;
2458 * Only copy the records and pointers if there are any.
2462 * First copy the records.
2464 op = (char *)XFS_BMAP_BROOT_REC_ADDR(ifp->if_broot, 1,
2465 ifp->if_broot_bytes);
2466 np = (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot, 1,
2468 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2471 * Then copy the pointers.
2473 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2474 ifp->if_broot_bytes);
2475 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot, 1,
2477 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2479 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2480 ifp->if_broot = new_broot;
2481 ifp->if_broot_bytes = (int)new_size;
2482 ASSERT(ifp->if_broot_bytes <=
2483 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2489 * This is called when the amount of space needed for if_data
2490 * is increased or decreased. The change in size is indicated by
2491 * the number of bytes that need to be added or deleted in the
2492 * byte_diff parameter.
2494 * If the amount of space needed has decreased below the size of the
2495 * inline buffer, then switch to using the inline buffer. Otherwise,
2496 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2497 * to what is needed.
2499 * ip -- the inode whose if_data area is changing
2500 * byte_diff -- the change in the number of bytes, positive or negative,
2501 * requested for the if_data array.
2513 if (byte_diff == 0) {
2517 ifp = XFS_IFORK_PTR(ip, whichfork);
2518 new_size = (int)ifp->if_bytes + byte_diff;
2519 ASSERT(new_size >= 0);
2521 if (new_size == 0) {
2522 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2523 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2525 ifp->if_u1.if_data = NULL;
2527 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2529 * If the valid extents/data can fit in if_inline_ext/data,
2530 * copy them from the malloc'd vector and free it.
2532 if (ifp->if_u1.if_data == NULL) {
2533 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2534 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2535 ASSERT(ifp->if_real_bytes != 0);
2536 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2538 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2539 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2544 * Stuck with malloc/realloc.
2545 * For inline data, the underlying buffer must be
2546 * a multiple of 4 bytes in size so that it can be
2547 * logged and stay on word boundaries. We enforce
2550 real_size = roundup(new_size, 4);
2551 if (ifp->if_u1.if_data == NULL) {
2552 ASSERT(ifp->if_real_bytes == 0);
2553 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2554 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2556 * Only do the realloc if the underlying size
2557 * is really changing.
2559 if (ifp->if_real_bytes != real_size) {
2560 ifp->if_u1.if_data =
2561 kmem_realloc(ifp->if_u1.if_data,
2567 ASSERT(ifp->if_real_bytes == 0);
2568 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2569 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2573 ifp->if_real_bytes = real_size;
2574 ifp->if_bytes = new_size;
2575 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2582 * Map inode to disk block and offset.
2584 * mp -- the mount point structure for the current file system
2585 * tp -- the current transaction
2586 * ino -- the inode number of the inode to be located
2587 * imap -- this structure is filled in with the information necessary
2588 * to retrieve the given inode from disk
2589 * flags -- flags to pass to xfs_dilocate indicating whether or not
2590 * lookups in the inode btree were OK or not
2600 xfs_fsblock_t fsbno;
2605 fsbno = imap->im_blkno ?
2606 XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK;
2607 error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags);
2611 imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno);
2612 imap->im_len = XFS_FSB_TO_BB(mp, len);
2613 imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno);
2614 imap->im_ioffset = (ushort)off;
2615 imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog);
2626 ifp = XFS_IFORK_PTR(ip, whichfork);
2627 if (ifp->if_broot != NULL) {
2628 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2629 ifp->if_broot = NULL;
2633 * If the format is local, then we can't have an extents
2634 * array so just look for an inline data array. If we're
2635 * not local then we may or may not have an extents list,
2636 * so check and free it up if we do.
2638 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2639 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2640 (ifp->if_u1.if_data != NULL)) {
2641 ASSERT(ifp->if_real_bytes != 0);
2642 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2643 ifp->if_u1.if_data = NULL;
2644 ifp->if_real_bytes = 0;
2646 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2647 ((ifp->if_flags & XFS_IFEXTIREC) ||
2648 ((ifp->if_u1.if_extents != NULL) &&
2649 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
2650 ASSERT(ifp->if_real_bytes != 0);
2651 xfs_iext_destroy(ifp);
2653 ASSERT(ifp->if_u1.if_extents == NULL ||
2654 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2655 ASSERT(ifp->if_real_bytes == 0);
2656 if (whichfork == XFS_ATTR_FORK) {
2657 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2663 * This is called free all the memory associated with an inode.
2664 * It must free the inode itself and any buffers allocated for
2665 * if_extents/if_data and if_broot. It must also free the lock
2666 * associated with the inode.
2673 switch (ip->i_d.di_mode & S_IFMT) {
2677 xfs_idestroy_fork(ip, XFS_DATA_FORK);
2681 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
2682 mrfree(&ip->i_lock);
2683 mrfree(&ip->i_iolock);
2684 freesema(&ip->i_flock);
2685 #ifdef XFS_BMAP_TRACE
2686 ktrace_free(ip->i_xtrace);
2688 #ifdef XFS_BMBT_TRACE
2689 ktrace_free(ip->i_btrace);
2692 ktrace_free(ip->i_rwtrace);
2694 #ifdef XFS_ILOCK_TRACE
2695 ktrace_free(ip->i_lock_trace);
2697 #ifdef XFS_DIR2_TRACE
2698 ktrace_free(ip->i_dir_trace);
2701 /* XXXdpd should be able to assert this but shutdown
2702 * is leaving the AIL behind. */
2703 ASSERT(((ip->i_itemp->ili_item.li_flags & XFS_LI_IN_AIL) == 0) ||
2704 XFS_FORCED_SHUTDOWN(ip->i_mount));
2705 xfs_inode_item_destroy(ip);
2707 kmem_zone_free(xfs_inode_zone, ip);
2712 * Increment the pin count of the given buffer.
2713 * This value is protected by ipinlock spinlock in the mount structure.
2719 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2721 atomic_inc(&ip->i_pincount);
2725 * Decrement the pin count of the given inode, and wake up
2726 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2727 * inode must have been previously pinned with a call to xfs_ipin().
2733 ASSERT(atomic_read(&ip->i_pincount) > 0);
2735 if (atomic_dec_and_test(&ip->i_pincount)) {
2737 * If the inode is currently being reclaimed, the
2738 * linux inode _and_ the xfs vnode may have been
2739 * freed so we cannot reference either of them safely.
2740 * Hence we should not try to do anything to them
2741 * if the xfs inode is currently in the reclaim
2744 * However, we still need to issue the unpin wakeup
2745 * call as the inode reclaim may be blocked waiting for
2746 * the inode to become unpinned.
2748 if (!(ip->i_flags & (XFS_IRECLAIM|XFS_IRECLAIMABLE))) {
2749 vnode_t *vp = XFS_ITOV_NULL(ip);
2751 /* make sync come back and flush this inode */
2753 struct inode *inode = vn_to_inode(vp);
2755 if (!(inode->i_state & I_NEW))
2756 mark_inode_dirty_sync(inode);
2759 wake_up(&ip->i_ipin_wait);
2764 * This is called to wait for the given inode to be unpinned.
2765 * It will sleep until this happens. The caller must have the
2766 * inode locked in at least shared mode so that the buffer cannot
2767 * be subsequently pinned once someone is waiting for it to be
2774 xfs_inode_log_item_t *iip;
2777 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE | MR_ACCESS));
2779 if (atomic_read(&ip->i_pincount) == 0) {
2784 if (iip && iip->ili_last_lsn) {
2785 lsn = iip->ili_last_lsn;
2791 * Give the log a push so we don't wait here too long.
2793 xfs_log_force(ip->i_mount, lsn, XFS_LOG_FORCE);
2795 wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
2800 * xfs_iextents_copy()
2802 * This is called to copy the REAL extents (as opposed to the delayed
2803 * allocation extents) from the inode into the given buffer. It
2804 * returns the number of bytes copied into the buffer.
2806 * If there are no delayed allocation extents, then we can just
2807 * memcpy() the extents into the buffer. Otherwise, we need to
2808 * examine each extent in turn and skip those which are delayed.
2813 xfs_bmbt_rec_t *buffer,
2817 xfs_bmbt_rec_t *dest_ep;
2819 #ifdef XFS_BMAP_TRACE
2820 static char fname[] = "xfs_iextents_copy";
2825 xfs_fsblock_t start_block;
2827 ifp = XFS_IFORK_PTR(ip, whichfork);
2828 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
2829 ASSERT(ifp->if_bytes > 0);
2831 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2832 xfs_bmap_trace_exlist(fname, ip, nrecs, whichfork);
2836 * There are some delayed allocation extents in the
2837 * inode, so copy the extents one at a time and skip
2838 * the delayed ones. There must be at least one
2839 * non-delayed extent.
2843 for (i = 0; i < nrecs; i++) {
2844 ep = xfs_iext_get_ext(ifp, i);
2845 start_block = xfs_bmbt_get_startblock(ep);
2846 if (ISNULLSTARTBLOCK(start_block)) {
2848 * It's a delayed allocation extent, so skip it.
2853 /* Translate to on disk format */
2854 put_unaligned(INT_GET(ep->l0, ARCH_CONVERT),
2855 (__uint64_t*)&dest_ep->l0);
2856 put_unaligned(INT_GET(ep->l1, ARCH_CONVERT),
2857 (__uint64_t*)&dest_ep->l1);
2861 ASSERT(copied != 0);
2862 xfs_validate_extents(ifp, copied, 1, XFS_EXTFMT_INODE(ip));
2864 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2868 * Each of the following cases stores data into the same region
2869 * of the on-disk inode, so only one of them can be valid at
2870 * any given time. While it is possible to have conflicting formats
2871 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2872 * in EXTENTS format, this can only happen when the fork has
2873 * changed formats after being modified but before being flushed.
2874 * In these cases, the format always takes precedence, because the
2875 * format indicates the current state of the fork.
2882 xfs_inode_log_item_t *iip,
2889 #ifdef XFS_TRANS_DEBUG
2892 static const short brootflag[2] =
2893 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2894 static const short dataflag[2] =
2895 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2896 static const short extflag[2] =
2897 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2901 ifp = XFS_IFORK_PTR(ip, whichfork);
2903 * This can happen if we gave up in iformat in an error path,
2904 * for the attribute fork.
2907 ASSERT(whichfork == XFS_ATTR_FORK);
2910 cp = XFS_DFORK_PTR(dip, whichfork);
2912 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2913 case XFS_DINODE_FMT_LOCAL:
2914 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2915 (ifp->if_bytes > 0)) {
2916 ASSERT(ifp->if_u1.if_data != NULL);
2917 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2918 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2920 if (whichfork == XFS_DATA_FORK) {
2921 if (unlikely(XFS_DIR_SHORTFORM_VALIDATE_ONDISK(mp, dip))) {
2922 XFS_ERROR_REPORT("xfs_iflush_fork",
2923 XFS_ERRLEVEL_LOW, mp);
2924 return XFS_ERROR(EFSCORRUPTED);
2929 case XFS_DINODE_FMT_EXTENTS:
2930 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2931 !(iip->ili_format.ilf_fields & extflag[whichfork]));
2932 ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
2933 (ifp->if_bytes == 0));
2934 ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
2935 (ifp->if_bytes > 0));
2936 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
2937 (ifp->if_bytes > 0)) {
2938 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2939 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2944 case XFS_DINODE_FMT_BTREE:
2945 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
2946 (ifp->if_broot_bytes > 0)) {
2947 ASSERT(ifp->if_broot != NULL);
2948 ASSERT(ifp->if_broot_bytes <=
2949 (XFS_IFORK_SIZE(ip, whichfork) +
2950 XFS_BROOT_SIZE_ADJ));
2951 xfs_bmbt_to_bmdr(ifp->if_broot, ifp->if_broot_bytes,
2952 (xfs_bmdr_block_t *)cp,
2953 XFS_DFORK_SIZE(dip, mp, whichfork));
2957 case XFS_DINODE_FMT_DEV:
2958 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
2959 ASSERT(whichfork == XFS_DATA_FORK);
2960 INT_SET(dip->di_u.di_dev, ARCH_CONVERT, ip->i_df.if_u2.if_rdev);
2964 case XFS_DINODE_FMT_UUID:
2965 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
2966 ASSERT(whichfork == XFS_DATA_FORK);
2967 memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid,
2981 * xfs_iflush() will write a modified inode's changes out to the
2982 * inode's on disk home. The caller must have the inode lock held
2983 * in at least shared mode and the inode flush semaphore must be
2984 * held as well. The inode lock will still be held upon return from
2985 * the call and the caller is free to unlock it.
2986 * The inode flush lock will be unlocked when the inode reaches the disk.
2987 * The flags indicate how the inode's buffer should be written out.
2994 xfs_inode_log_item_t *iip;
3002 int clcount; /* count of inodes clustered */
3004 enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
3007 XFS_STATS_INC(xs_iflush_count);
3009 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
3010 ASSERT(valusema(&ip->i_flock) <= 0);
3011 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3012 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3018 * If the inode isn't dirty, then just release the inode
3019 * flush lock and do nothing.
3021 if ((ip->i_update_core == 0) &&
3022 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3023 ASSERT((iip != NULL) ?
3024 !(iip->ili_item.li_flags & XFS_LI_IN_AIL) : 1);
3030 * We can't flush the inode until it is unpinned, so
3031 * wait for it. We know noone new can pin it, because
3032 * we are holding the inode lock shared and you need
3033 * to hold it exclusively to pin the inode.
3035 xfs_iunpin_wait(ip);
3038 * This may have been unpinned because the filesystem is shutting
3039 * down forcibly. If that's the case we must not write this inode
3040 * to disk, because the log record didn't make it to disk!
3042 if (XFS_FORCED_SHUTDOWN(mp)) {
3043 ip->i_update_core = 0;
3045 iip->ili_format.ilf_fields = 0;
3047 return XFS_ERROR(EIO);
3051 * Get the buffer containing the on-disk inode.
3053 error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0, 0);
3060 * Decide how buffer will be flushed out. This is done before
3061 * the call to xfs_iflush_int because this field is zeroed by it.
3063 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3065 * Flush out the inode buffer according to the directions
3066 * of the caller. In the cases where the caller has given
3067 * us a choice choose the non-delwri case. This is because
3068 * the inode is in the AIL and we need to get it out soon.
3071 case XFS_IFLUSH_SYNC:
3072 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3075 case XFS_IFLUSH_ASYNC:
3076 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3079 case XFS_IFLUSH_DELWRI:
3089 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3090 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3091 case XFS_IFLUSH_DELWRI:
3094 case XFS_IFLUSH_ASYNC:
3097 case XFS_IFLUSH_SYNC:
3108 * First flush out the inode that xfs_iflush was called with.
3110 error = xfs_iflush_int(ip, bp);
3117 * see if other inodes can be gathered into this write
3120 ip->i_chash->chl_buf = bp;
3122 ch = XFS_CHASH(mp, ip->i_blkno);
3123 s = mutex_spinlock(&ch->ch_lock);
3126 for (iq = ip->i_cnext; iq != ip; iq = iq->i_cnext) {
3128 * Do an un-protected check to see if the inode is dirty and
3129 * is a candidate for flushing. These checks will be repeated
3130 * later after the appropriate locks are acquired.
3133 if ((iq->i_update_core == 0) &&
3135 !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
3136 xfs_ipincount(iq) == 0) {
3141 * Try to get locks. If any are unavailable,
3142 * then this inode cannot be flushed and is skipped.
3145 /* get inode locks (just i_lock) */
3146 if (xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) {
3147 /* get inode flush lock */
3148 if (xfs_iflock_nowait(iq)) {
3149 /* check if pinned */
3150 if (xfs_ipincount(iq) == 0) {
3151 /* arriving here means that
3152 * this inode can be flushed.
3153 * first re-check that it's
3157 if ((iq->i_update_core != 0)||
3159 (iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3161 error = xfs_iflush_int(iq, bp);
3165 goto cluster_corrupt_out;
3174 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3177 mutex_spinunlock(&ch->ch_lock, s);
3180 XFS_STATS_INC(xs_icluster_flushcnt);
3181 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
3185 * If the buffer is pinned then push on the log so we won't
3186 * get stuck waiting in the write for too long.
3188 if (XFS_BUF_ISPINNED(bp)){
3189 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
3192 if (flags & INT_DELWRI) {
3193 xfs_bdwrite(mp, bp);
3194 } else if (flags & INT_ASYNC) {
3195 xfs_bawrite(mp, bp);
3197 error = xfs_bwrite(mp, bp);
3203 xfs_force_shutdown(mp, XFS_CORRUPT_INCORE);
3204 xfs_iflush_abort(ip);
3206 * Unlocks the flush lock
3208 return XFS_ERROR(EFSCORRUPTED);
3210 cluster_corrupt_out:
3211 /* Corruption detected in the clustering loop. Invalidate the
3212 * inode buffer and shut down the filesystem.
3214 mutex_spinunlock(&ch->ch_lock, s);
3217 * Clean up the buffer. If it was B_DELWRI, just release it --
3218 * brelse can handle it with no problems. If not, shut down the
3219 * filesystem before releasing the buffer.
3221 if ((bufwasdelwri= XFS_BUF_ISDELAYWRITE(bp))) {
3225 xfs_force_shutdown(mp, XFS_CORRUPT_INCORE);
3229 * Just like incore_relse: if we have b_iodone functions,
3230 * mark the buffer as an error and call them. Otherwise
3231 * mark it as stale and brelse.
3233 if (XFS_BUF_IODONE_FUNC(bp)) {
3234 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
3238 XFS_BUF_ERROR(bp,EIO);
3246 xfs_iflush_abort(iq);
3248 * Unlocks the flush lock
3250 return XFS_ERROR(EFSCORRUPTED);
3259 xfs_inode_log_item_t *iip;
3262 #ifdef XFS_TRANS_DEBUG
3267 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
3268 ASSERT(valusema(&ip->i_flock) <= 0);
3269 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3270 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3277 * If the inode isn't dirty, then just release the inode
3278 * flush lock and do nothing.
3280 if ((ip->i_update_core == 0) &&
3281 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3286 /* set *dip = inode's place in the buffer */
3287 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset);
3290 * Clear i_update_core before copying out the data.
3291 * This is for coordination with our timestamp updates
3292 * that don't hold the inode lock. They will always
3293 * update the timestamps BEFORE setting i_update_core,
3294 * so if we clear i_update_core after they set it we
3295 * are guaranteed to see their updates to the timestamps.
3296 * I believe that this depends on strongly ordered memory
3297 * semantics, but we have that. We use the SYNCHRONIZE
3298 * macro to make sure that the compiler does not reorder
3299 * the i_update_core access below the data copy below.
3301 ip->i_update_core = 0;
3305 * Make sure to get the latest atime from the Linux inode.
3307 xfs_synchronize_atime(ip);
3309 if (XFS_TEST_ERROR(INT_GET(dip->di_core.di_magic,ARCH_CONVERT) != XFS_DINODE_MAGIC,
3310 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3311 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3312 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3313 ip->i_ino, (int) INT_GET(dip->di_core.di_magic, ARCH_CONVERT), dip);
3316 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3317 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3318 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3319 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3320 ip->i_ino, ip, ip->i_d.di_magic);
3323 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3325 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3326 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3327 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3328 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3329 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3333 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
3335 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3336 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3337 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3338 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3339 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3340 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3345 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3346 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3347 XFS_RANDOM_IFLUSH_5)) {
3348 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3349 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3351 ip->i_d.di_nextents + ip->i_d.di_anextents,
3356 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3357 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3358 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3359 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3360 ip->i_ino, ip->i_d.di_forkoff, ip);
3364 * bump the flush iteration count, used to detect flushes which
3365 * postdate a log record during recovery.
3368 ip->i_d.di_flushiter++;
3371 * Copy the dirty parts of the inode into the on-disk
3372 * inode. We always copy out the core of the inode,
3373 * because if the inode is dirty at all the core must
3376 xfs_xlate_dinode_core((xfs_caddr_t)&(dip->di_core), &(ip->i_d), -1);
3378 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3379 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3380 ip->i_d.di_flushiter = 0;
3383 * If this is really an old format inode and the superblock version
3384 * has not been updated to support only new format inodes, then
3385 * convert back to the old inode format. If the superblock version
3386 * has been updated, then make the conversion permanent.
3388 ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 ||
3389 XFS_SB_VERSION_HASNLINK(&mp->m_sb));
3390 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
3391 if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) {
3395 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3396 INT_SET(dip->di_core.di_onlink, ARCH_CONVERT, ip->i_d.di_nlink);
3399 * The superblock version has already been bumped,
3400 * so just make the conversion to the new inode
3403 ip->i_d.di_version = XFS_DINODE_VERSION_2;
3404 INT_SET(dip->di_core.di_version, ARCH_CONVERT, XFS_DINODE_VERSION_2);
3405 ip->i_d.di_onlink = 0;
3406 dip->di_core.di_onlink = 0;
3407 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3408 memset(&(dip->di_core.di_pad[0]), 0,
3409 sizeof(dip->di_core.di_pad));
3410 ASSERT(ip->i_d.di_projid == 0);
3414 if (xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp) == EFSCORRUPTED) {
3418 if (XFS_IFORK_Q(ip)) {
3420 * The only error from xfs_iflush_fork is on the data fork.
3422 (void) xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3424 xfs_inobp_check(mp, bp);
3427 * We've recorded everything logged in the inode, so we'd
3428 * like to clear the ilf_fields bits so we don't log and
3429 * flush things unnecessarily. However, we can't stop
3430 * logging all this information until the data we've copied
3431 * into the disk buffer is written to disk. If we did we might
3432 * overwrite the copy of the inode in the log with all the
3433 * data after re-logging only part of it, and in the face of
3434 * a crash we wouldn't have all the data we need to recover.
3436 * What we do is move the bits to the ili_last_fields field.
3437 * When logging the inode, these bits are moved back to the
3438 * ilf_fields field. In the xfs_iflush_done() routine we
3439 * clear ili_last_fields, since we know that the information
3440 * those bits represent is permanently on disk. As long as
3441 * the flush completes before the inode is logged again, then
3442 * both ilf_fields and ili_last_fields will be cleared.
3444 * We can play with the ilf_fields bits here, because the inode
3445 * lock must be held exclusively in order to set bits there
3446 * and the flush lock protects the ili_last_fields bits.
3447 * Set ili_logged so the flush done
3448 * routine can tell whether or not to look in the AIL.
3449 * Also, store the current LSN of the inode so that we can tell
3450 * whether the item has moved in the AIL from xfs_iflush_done().
3451 * In order to read the lsn we need the AIL lock, because
3452 * it is a 64 bit value that cannot be read atomically.
3454 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3455 iip->ili_last_fields = iip->ili_format.ilf_fields;
3456 iip->ili_format.ilf_fields = 0;
3457 iip->ili_logged = 1;
3459 ASSERT(sizeof(xfs_lsn_t) == 8); /* don't lock if it shrinks */
3461 iip->ili_flush_lsn = iip->ili_item.li_lsn;
3465 * Attach the function xfs_iflush_done to the inode's
3466 * buffer. This will remove the inode from the AIL
3467 * and unlock the inode's flush lock when the inode is
3468 * completely written to disk.
3470 xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
3471 xfs_iflush_done, (xfs_log_item_t *)iip);
3473 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3474 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3477 * We're flushing an inode which is not in the AIL and has
3478 * not been logged but has i_update_core set. For this
3479 * case we can use a B_DELWRI flush and immediately drop
3480 * the inode flush lock because we can avoid the whole
3481 * AIL state thing. It's OK to drop the flush lock now,
3482 * because we've already locked the buffer and to do anything
3483 * you really need both.
3486 ASSERT(iip->ili_logged == 0);
3487 ASSERT(iip->ili_last_fields == 0);
3488 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3496 return XFS_ERROR(EFSCORRUPTED);
3501 * Flush all inactive inodes in mp.
3511 XFS_MOUNT_ILOCK(mp);
3517 /* Make sure we skip markers inserted by sync */
3518 if (ip->i_mount == NULL) {
3523 vp = XFS_ITOV_NULL(ip);
3525 XFS_MOUNT_IUNLOCK(mp);
3526 xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC);
3530 ASSERT(vn_count(vp) == 0);
3533 } while (ip != mp->m_inodes);
3535 XFS_MOUNT_IUNLOCK(mp);
3539 * xfs_iaccess: check accessibility of inode for mode.
3548 mode_t orgmode = mode;
3549 struct inode *inode = vn_to_inode(XFS_ITOV(ip));
3551 if (mode & S_IWUSR) {
3552 umode_t imode = inode->i_mode;
3554 if (IS_RDONLY(inode) &&
3555 (S_ISREG(imode) || S_ISDIR(imode) || S_ISLNK(imode)))
3556 return XFS_ERROR(EROFS);
3558 if (IS_IMMUTABLE(inode))
3559 return XFS_ERROR(EACCES);
3563 * If there's an Access Control List it's used instead of
3566 if ((error = _ACL_XFS_IACCESS(ip, mode, cr)) != -1)
3567 return error ? XFS_ERROR(error) : 0;
3569 if (current_fsuid(cr) != ip->i_d.di_uid) {
3571 if (!in_group_p((gid_t)ip->i_d.di_gid))
3576 * If the DACs are ok we don't need any capability check.
3578 if ((ip->i_d.di_mode & mode) == mode)
3581 * Read/write DACs are always overridable.
3582 * Executable DACs are overridable if at least one exec bit is set.
3584 if (!(orgmode & S_IXUSR) ||
3585 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
3586 if (capable_cred(cr, CAP_DAC_OVERRIDE))
3589 if ((orgmode == S_IRUSR) ||
3590 (S_ISDIR(inode->i_mode) && (!(orgmode & S_IWUSR)))) {
3591 if (capable_cred(cr, CAP_DAC_READ_SEARCH))
3594 cmn_err(CE_NOTE, "Ick: mode=%o, orgmode=%o", mode, orgmode);
3596 return XFS_ERROR(EACCES);
3598 return XFS_ERROR(EACCES);
3602 * xfs_iroundup: round up argument to next power of two
3611 if ((v & (v - 1)) == 0)
3613 ASSERT((v & 0x80000000) == 0);
3614 if ((v & (v + 1)) == 0)
3616 for (i = 0, m = 1; i < 31; i++, m <<= 1) {
3620 if ((v & (v + 1)) == 0)
3627 #ifdef XFS_ILOCK_TRACE
3628 ktrace_t *xfs_ilock_trace_buf;
3631 xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
3633 ktrace_enter(ip->i_lock_trace,
3635 (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
3636 (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
3637 (void *)ra, /* caller of ilock */
3638 (void *)(unsigned long)current_cpu(),
3639 (void *)(unsigned long)current_pid(),
3640 NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL);
3645 * Return a pointer to the extent record at file index idx.
3649 xfs_ifork_t *ifp, /* inode fork pointer */
3650 xfs_extnum_t idx) /* index of target extent */
3653 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
3654 return ifp->if_u1.if_ext_irec->er_extbuf;
3655 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3656 xfs_ext_irec_t *erp; /* irec pointer */
3657 int erp_idx = 0; /* irec index */
3658 xfs_extnum_t page_idx = idx; /* ext index in target list */
3660 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3661 return &erp->er_extbuf[page_idx];
3662 } else if (ifp->if_bytes) {
3663 return &ifp->if_u1.if_extents[idx];
3670 * Insert new item(s) into the extent records for incore inode
3671 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3675 xfs_ifork_t *ifp, /* inode fork pointer */
3676 xfs_extnum_t idx, /* starting index of new items */
3677 xfs_extnum_t count, /* number of inserted items */
3678 xfs_bmbt_irec_t *new) /* items to insert */
3680 xfs_bmbt_rec_t *ep; /* extent record pointer */
3681 xfs_extnum_t i; /* extent record index */
3683 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3684 xfs_iext_add(ifp, idx, count);
3685 for (i = idx; i < idx + count; i++, new++) {
3686 ep = xfs_iext_get_ext(ifp, i);
3687 xfs_bmbt_set_all(ep, new);
3692 * This is called when the amount of space required for incore file
3693 * extents needs to be increased. The ext_diff parameter stores the
3694 * number of new extents being added and the idx parameter contains
3695 * the extent index where the new extents will be added. If the new
3696 * extents are being appended, then we just need to (re)allocate and
3697 * initialize the space. Otherwise, if the new extents are being
3698 * inserted into the middle of the existing entries, a bit more work
3699 * is required to make room for the new extents to be inserted. The
3700 * caller is responsible for filling in the new extent entries upon
3705 xfs_ifork_t *ifp, /* inode fork pointer */
3706 xfs_extnum_t idx, /* index to begin adding exts */
3707 int ext_diff) /* number of extents to add */
3709 int byte_diff; /* new bytes being added */
3710 int new_size; /* size of extents after adding */
3711 xfs_extnum_t nextents; /* number of extents in file */
3713 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3714 ASSERT((idx >= 0) && (idx <= nextents));
3715 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
3716 new_size = ifp->if_bytes + byte_diff;
3718 * If the new number of extents (nextents + ext_diff)
3719 * fits inside the inode, then continue to use the inline
3722 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
3723 if (idx < nextents) {
3724 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
3725 &ifp->if_u2.if_inline_ext[idx],
3726 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3727 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
3729 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3730 ifp->if_real_bytes = 0;
3731 ifp->if_lastex = nextents + ext_diff;
3734 * Otherwise use a linear (direct) extent list.
3735 * If the extents are currently inside the inode,
3736 * xfs_iext_realloc_direct will switch us from
3737 * inline to direct extent allocation mode.
3739 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
3740 xfs_iext_realloc_direct(ifp, new_size);
3741 if (idx < nextents) {
3742 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
3743 &ifp->if_u1.if_extents[idx],
3744 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3745 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
3748 /* Indirection array */
3750 xfs_ext_irec_t *erp;
3754 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
3755 if (ifp->if_flags & XFS_IFEXTIREC) {
3756 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3758 xfs_iext_irec_init(ifp);
3759 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3760 erp = ifp->if_u1.if_ext_irec;
3762 /* Extents fit in target extent page */
3763 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3764 if (page_idx < erp->er_extcount) {
3765 memmove(&erp->er_extbuf[page_idx + ext_diff],
3766 &erp->er_extbuf[page_idx],
3767 (erp->er_extcount - page_idx) *
3768 sizeof(xfs_bmbt_rec_t));
3769 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3771 erp->er_extcount += ext_diff;
3772 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3774 /* Insert a new extent page */
3776 xfs_iext_add_indirect_multi(ifp,
3777 erp_idx, page_idx, ext_diff);
3780 * If extent(s) are being appended to the last page in
3781 * the indirection array and the new extent(s) don't fit
3782 * in the page, then erp is NULL and erp_idx is set to
3783 * the next index needed in the indirection array.
3786 int count = ext_diff;
3789 erp = xfs_iext_irec_new(ifp, erp_idx);
3790 erp->er_extcount = count;
3791 count -= MIN(count, (int)XFS_LINEAR_EXTS);
3798 ifp->if_bytes = new_size;
3802 * This is called when incore extents are being added to the indirection
3803 * array and the new extents do not fit in the target extent list. The
3804 * erp_idx parameter contains the irec index for the target extent list
3805 * in the indirection array, and the idx parameter contains the extent
3806 * index within the list. The number of extents being added is stored
3807 * in the count parameter.
3809 * |-------| |-------|
3810 * | | | | idx - number of extents before idx
3812 * | | | | count - number of extents being inserted at idx
3813 * |-------| |-------|
3814 * | count | | nex2 | nex2 - number of extents after idx + count
3815 * |-------| |-------|
3818 xfs_iext_add_indirect_multi(
3819 xfs_ifork_t *ifp, /* inode fork pointer */
3820 int erp_idx, /* target extent irec index */
3821 xfs_extnum_t idx, /* index within target list */
3822 int count) /* new extents being added */
3824 int byte_diff; /* new bytes being added */
3825 xfs_ext_irec_t *erp; /* pointer to irec entry */
3826 xfs_extnum_t ext_diff; /* number of extents to add */
3827 xfs_extnum_t ext_cnt; /* new extents still needed */
3828 xfs_extnum_t nex2; /* extents after idx + count */
3829 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3830 int nlists; /* number of irec's (lists) */
3832 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3833 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3834 nex2 = erp->er_extcount - idx;
3835 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3838 * Save second part of target extent list
3839 * (all extents past */
3841 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3842 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_SLEEP);
3843 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3844 erp->er_extcount -= nex2;
3845 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3846 memset(&erp->er_extbuf[idx], 0, byte_diff);
3850 * Add the new extents to the end of the target
3851 * list, then allocate new irec record(s) and
3852 * extent buffer(s) as needed to store the rest
3853 * of the new extents.
3856 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3858 erp->er_extcount += ext_diff;
3859 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3860 ext_cnt -= ext_diff;
3864 erp = xfs_iext_irec_new(ifp, erp_idx);
3865 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3866 erp->er_extcount = ext_diff;
3867 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3868 ext_cnt -= ext_diff;
3871 /* Add nex2 extents back to indirection array */
3873 xfs_extnum_t ext_avail;
3876 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3877 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3880 * If nex2 extents fit in the current page, append
3881 * nex2_ep after the new extents.
3883 if (nex2 <= ext_avail) {
3884 i = erp->er_extcount;
3887 * Otherwise, check if space is available in the
3890 else if ((erp_idx < nlists - 1) &&
3891 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3892 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3895 /* Create a hole for nex2 extents */
3896 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3897 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3900 * Final choice, create a new extent page for
3905 erp = xfs_iext_irec_new(ifp, erp_idx);
3907 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
3908 kmem_free(nex2_ep, byte_diff);
3909 erp->er_extcount += nex2;
3910 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3915 * This is called when the amount of space required for incore file
3916 * extents needs to be decreased. The ext_diff parameter stores the
3917 * number of extents to be removed and the idx parameter contains
3918 * the extent index where the extents will be removed from.
3920 * If the amount of space needed has decreased below the linear
3921 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3922 * extent array. Otherwise, use kmem_realloc() to adjust the
3923 * size to what is needed.
3927 xfs_ifork_t *ifp, /* inode fork pointer */
3928 xfs_extnum_t idx, /* index to begin removing exts */
3929 int ext_diff) /* number of extents to remove */
3931 xfs_extnum_t nextents; /* number of extents in file */
3932 int new_size; /* size of extents after removal */
3934 ASSERT(ext_diff > 0);
3935 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3936 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3938 if (new_size == 0) {
3939 xfs_iext_destroy(ifp);
3940 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3941 xfs_iext_remove_indirect(ifp, idx, ext_diff);
3942 } else if (ifp->if_real_bytes) {
3943 xfs_iext_remove_direct(ifp, idx, ext_diff);
3945 xfs_iext_remove_inline(ifp, idx, ext_diff);
3947 ifp->if_bytes = new_size;
3951 * This removes ext_diff extents from the inline buffer, beginning
3952 * at extent index idx.
3955 xfs_iext_remove_inline(
3956 xfs_ifork_t *ifp, /* inode fork pointer */
3957 xfs_extnum_t idx, /* index to begin removing exts */
3958 int ext_diff) /* number of extents to remove */
3960 int nextents; /* number of extents in file */
3962 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3963 ASSERT(idx < XFS_INLINE_EXTS);
3964 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3965 ASSERT(((nextents - ext_diff) > 0) &&
3966 (nextents - ext_diff) < XFS_INLINE_EXTS);
3968 if (idx + ext_diff < nextents) {
3969 memmove(&ifp->if_u2.if_inline_ext[idx],
3970 &ifp->if_u2.if_inline_ext[idx + ext_diff],
3971 (nextents - (idx + ext_diff)) *
3972 sizeof(xfs_bmbt_rec_t));
3973 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
3974 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3976 memset(&ifp->if_u2.if_inline_ext[idx], 0,
3977 ext_diff * sizeof(xfs_bmbt_rec_t));
3982 * This removes ext_diff extents from a linear (direct) extent list,
3983 * beginning at extent index idx. If the extents are being removed
3984 * from the end of the list (ie. truncate) then we just need to re-
3985 * allocate the list to remove the extra space. Otherwise, if the
3986 * extents are being removed from the middle of the existing extent
3987 * entries, then we first need to move the extent records beginning
3988 * at idx + ext_diff up in the list to overwrite the records being
3989 * removed, then remove the extra space via kmem_realloc.
3992 xfs_iext_remove_direct(
3993 xfs_ifork_t *ifp, /* inode fork pointer */
3994 xfs_extnum_t idx, /* index to begin removing exts */
3995 int ext_diff) /* number of extents to remove */
3997 xfs_extnum_t nextents; /* number of extents in file */
3998 int new_size; /* size of extents after removal */
4000 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4001 new_size = ifp->if_bytes -
4002 (ext_diff * sizeof(xfs_bmbt_rec_t));
4003 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4005 if (new_size == 0) {
4006 xfs_iext_destroy(ifp);
4009 /* Move extents up in the list (if needed) */
4010 if (idx + ext_diff < nextents) {
4011 memmove(&ifp->if_u1.if_extents[idx],
4012 &ifp->if_u1.if_extents[idx + ext_diff],
4013 (nextents - (idx + ext_diff)) *
4014 sizeof(xfs_bmbt_rec_t));
4016 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
4017 0, ext_diff * sizeof(xfs_bmbt_rec_t));
4019 * Reallocate the direct extent list. If the extents
4020 * will fit inside the inode then xfs_iext_realloc_direct
4021 * will switch from direct to inline extent allocation
4024 xfs_iext_realloc_direct(ifp, new_size);
4025 ifp->if_bytes = new_size;
4029 * This is called when incore extents are being removed from the
4030 * indirection array and the extents being removed span multiple extent
4031 * buffers. The idx parameter contains the file extent index where we
4032 * want to begin removing extents, and the count parameter contains
4033 * how many extents need to be removed.
4035 * |-------| |-------|
4036 * | nex1 | | | nex1 - number of extents before idx
4037 * |-------| | count |
4038 * | | | | count - number of extents being removed at idx
4039 * | count | |-------|
4040 * | | | nex2 | nex2 - number of extents after idx + count
4041 * |-------| |-------|
4044 xfs_iext_remove_indirect(
4045 xfs_ifork_t *ifp, /* inode fork pointer */
4046 xfs_extnum_t idx, /* index to begin removing extents */
4047 int count) /* number of extents to remove */
4049 xfs_ext_irec_t *erp; /* indirection array pointer */
4050 int erp_idx = 0; /* indirection array index */
4051 xfs_extnum_t ext_cnt; /* extents left to remove */
4052 xfs_extnum_t ext_diff; /* extents to remove in current list */
4053 xfs_extnum_t nex1; /* number of extents before idx */
4054 xfs_extnum_t nex2; /* extents after idx + count */
4055 int nlists; /* entries in indirection array */
4056 int page_idx = idx; /* index in target extent list */
4058 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4059 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
4060 ASSERT(erp != NULL);
4061 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4065 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
4066 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
4068 * Check for deletion of entire list;
4069 * xfs_iext_irec_remove() updates extent offsets.
4071 if (ext_diff == erp->er_extcount) {
4072 xfs_iext_irec_remove(ifp, erp_idx);
4073 ext_cnt -= ext_diff;
4076 ASSERT(erp_idx < ifp->if_real_bytes /
4078 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4085 /* Move extents up (if needed) */
4087 memmove(&erp->er_extbuf[nex1],
4088 &erp->er_extbuf[nex1 + ext_diff],
4089 nex2 * sizeof(xfs_bmbt_rec_t));
4091 /* Zero out rest of page */
4092 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
4093 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
4094 /* Update remaining counters */
4095 erp->er_extcount -= ext_diff;
4096 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
4097 ext_cnt -= ext_diff;
4102 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
4103 xfs_iext_irec_compact(ifp);
4107 * Create, destroy, or resize a linear (direct) block of extents.
4110 xfs_iext_realloc_direct(
4111 xfs_ifork_t *ifp, /* inode fork pointer */
4112 int new_size) /* new size of extents */
4114 int rnew_size; /* real new size of extents */
4116 rnew_size = new_size;
4118 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
4119 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
4120 (new_size != ifp->if_real_bytes)));
4122 /* Free extent records */
4123 if (new_size == 0) {
4124 xfs_iext_destroy(ifp);
4126 /* Resize direct extent list and zero any new bytes */
4127 else if (ifp->if_real_bytes) {
4128 /* Check if extents will fit inside the inode */
4129 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
4130 xfs_iext_direct_to_inline(ifp, new_size /
4131 (uint)sizeof(xfs_bmbt_rec_t));
4132 ifp->if_bytes = new_size;
4135 if ((new_size & (new_size - 1)) != 0) {
4136 rnew_size = xfs_iroundup(new_size);
4138 if (rnew_size != ifp->if_real_bytes) {
4139 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
4140 kmem_realloc(ifp->if_u1.if_extents,
4145 if (rnew_size > ifp->if_real_bytes) {
4146 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
4147 (uint)sizeof(xfs_bmbt_rec_t)], 0,
4148 rnew_size - ifp->if_real_bytes);
4152 * Switch from the inline extent buffer to a direct
4153 * extent list. Be sure to include the inline extent
4154 * bytes in new_size.
4157 new_size += ifp->if_bytes;
4158 if ((new_size & (new_size - 1)) != 0) {
4159 rnew_size = xfs_iroundup(new_size);
4161 xfs_iext_inline_to_direct(ifp, rnew_size);
4163 ifp->if_real_bytes = rnew_size;
4164 ifp->if_bytes = new_size;
4168 * Switch from linear (direct) extent records to inline buffer.
4171 xfs_iext_direct_to_inline(
4172 xfs_ifork_t *ifp, /* inode fork pointer */
4173 xfs_extnum_t nextents) /* number of extents in file */
4175 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4176 ASSERT(nextents <= XFS_INLINE_EXTS);
4178 * The inline buffer was zeroed when we switched
4179 * from inline to direct extent allocation mode,
4180 * so we don't need to clear it here.
4182 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
4183 nextents * sizeof(xfs_bmbt_rec_t));
4184 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
4185 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
4186 ifp->if_real_bytes = 0;
4190 * Switch from inline buffer to linear (direct) extent records.
4191 * new_size should already be rounded up to the next power of 2
4192 * by the caller (when appropriate), so use new_size as it is.
4193 * However, since new_size may be rounded up, we can't update
4194 * if_bytes here. It is the caller's responsibility to update
4195 * if_bytes upon return.
4198 xfs_iext_inline_to_direct(
4199 xfs_ifork_t *ifp, /* inode fork pointer */
4200 int new_size) /* number of extents in file */
4202 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
4203 kmem_alloc(new_size, KM_SLEEP);
4204 memset(ifp->if_u1.if_extents, 0, new_size);
4205 if (ifp->if_bytes) {
4206 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
4208 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4209 sizeof(xfs_bmbt_rec_t));
4211 ifp->if_real_bytes = new_size;
4215 * Resize an extent indirection array to new_size bytes.
4218 xfs_iext_realloc_indirect(
4219 xfs_ifork_t *ifp, /* inode fork pointer */
4220 int new_size) /* new indirection array size */
4222 int nlists; /* number of irec's (ex lists) */
4223 int size; /* current indirection array size */
4225 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4226 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4227 size = nlists * sizeof(xfs_ext_irec_t);
4228 ASSERT(ifp->if_real_bytes);
4229 ASSERT((new_size >= 0) && (new_size != size));
4230 if (new_size == 0) {
4231 xfs_iext_destroy(ifp);
4233 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
4234 kmem_realloc(ifp->if_u1.if_ext_irec,
4235 new_size, size, KM_SLEEP);
4240 * Switch from indirection array to linear (direct) extent allocations.
4243 xfs_iext_indirect_to_direct(
4244 xfs_ifork_t *ifp) /* inode fork pointer */
4246 xfs_bmbt_rec_t *ep; /* extent record pointer */
4247 xfs_extnum_t nextents; /* number of extents in file */
4248 int size; /* size of file extents */
4250 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4251 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4252 ASSERT(nextents <= XFS_LINEAR_EXTS);
4253 size = nextents * sizeof(xfs_bmbt_rec_t);
4255 xfs_iext_irec_compact_full(ifp);
4256 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
4258 ep = ifp->if_u1.if_ext_irec->er_extbuf;
4259 kmem_free(ifp->if_u1.if_ext_irec, sizeof(xfs_ext_irec_t));
4260 ifp->if_flags &= ~XFS_IFEXTIREC;
4261 ifp->if_u1.if_extents = ep;
4262 ifp->if_bytes = size;
4263 if (nextents < XFS_LINEAR_EXTS) {
4264 xfs_iext_realloc_direct(ifp, size);
4269 * Free incore file extents.
4273 xfs_ifork_t *ifp) /* inode fork pointer */
4275 if (ifp->if_flags & XFS_IFEXTIREC) {
4279 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4280 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
4281 xfs_iext_irec_remove(ifp, erp_idx);
4283 ifp->if_flags &= ~XFS_IFEXTIREC;
4284 } else if (ifp->if_real_bytes) {
4285 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
4286 } else if (ifp->if_bytes) {
4287 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4288 sizeof(xfs_bmbt_rec_t));
4290 ifp->if_u1.if_extents = NULL;
4291 ifp->if_real_bytes = 0;
4296 * Return a pointer to the extent record for file system block bno.
4298 xfs_bmbt_rec_t * /* pointer to found extent record */
4299 xfs_iext_bno_to_ext(
4300 xfs_ifork_t *ifp, /* inode fork pointer */
4301 xfs_fileoff_t bno, /* block number to search for */
4302 xfs_extnum_t *idxp) /* index of target extent */
4304 xfs_bmbt_rec_t *base; /* pointer to first extent */
4305 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
4306 xfs_bmbt_rec_t *ep = NULL; /* pointer to target extent */
4307 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4308 int high; /* upper boundary in search */
4309 xfs_extnum_t idx = 0; /* index of target extent */
4310 int low; /* lower boundary in search */
4311 xfs_extnum_t nextents; /* number of file extents */
4312 xfs_fileoff_t startoff = 0; /* start offset of extent */
4314 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4315 if (nextents == 0) {
4320 if (ifp->if_flags & XFS_IFEXTIREC) {
4321 /* Find target extent list */
4323 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
4324 base = erp->er_extbuf;
4325 high = erp->er_extcount - 1;
4327 base = ifp->if_u1.if_extents;
4328 high = nextents - 1;
4330 /* Binary search extent records */
4331 while (low <= high) {
4332 idx = (low + high) >> 1;
4334 startoff = xfs_bmbt_get_startoff(ep);
4335 blockcount = xfs_bmbt_get_blockcount(ep);
4336 if (bno < startoff) {
4338 } else if (bno >= startoff + blockcount) {
4341 /* Convert back to file-based extent index */
4342 if (ifp->if_flags & XFS_IFEXTIREC) {
4343 idx += erp->er_extoff;
4349 /* Convert back to file-based extent index */
4350 if (ifp->if_flags & XFS_IFEXTIREC) {
4351 idx += erp->er_extoff;
4353 if (bno >= startoff + blockcount) {
4354 if (++idx == nextents) {
4357 ep = xfs_iext_get_ext(ifp, idx);
4365 * Return a pointer to the indirection array entry containing the
4366 * extent record for filesystem block bno. Store the index of the
4367 * target irec in *erp_idxp.
4369 xfs_ext_irec_t * /* pointer to found extent record */
4370 xfs_iext_bno_to_irec(
4371 xfs_ifork_t *ifp, /* inode fork pointer */
4372 xfs_fileoff_t bno, /* block number to search for */
4373 int *erp_idxp) /* irec index of target ext list */
4375 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4376 xfs_ext_irec_t *erp_next; /* next indirection array entry */
4377 int erp_idx; /* indirection array index */
4378 int nlists; /* number of extent irec's (lists) */
4379 int high; /* binary search upper limit */
4380 int low; /* binary search lower limit */
4382 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4383 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4387 while (low <= high) {
4388 erp_idx = (low + high) >> 1;
4389 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4390 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
4391 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
4393 } else if (erp_next && bno >=
4394 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
4400 *erp_idxp = erp_idx;
4405 * Return a pointer to the indirection array entry containing the
4406 * extent record at file extent index *idxp. Store the index of the
4407 * target irec in *erp_idxp and store the page index of the target
4408 * extent record in *idxp.
4411 xfs_iext_idx_to_irec(
4412 xfs_ifork_t *ifp, /* inode fork pointer */
4413 xfs_extnum_t *idxp, /* extent index (file -> page) */
4414 int *erp_idxp, /* pointer to target irec */
4415 int realloc) /* new bytes were just added */
4417 xfs_ext_irec_t *prev; /* pointer to previous irec */
4418 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
4419 int erp_idx; /* indirection array index */
4420 int nlists; /* number of irec's (ex lists) */
4421 int high; /* binary search upper limit */
4422 int low; /* binary search lower limit */
4423 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
4425 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4426 ASSERT(page_idx >= 0 && page_idx <=
4427 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
4428 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4433 /* Binary search extent irec's */
4434 while (low <= high) {
4435 erp_idx = (low + high) >> 1;
4436 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4437 prev = erp_idx > 0 ? erp - 1 : NULL;
4438 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
4439 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
4441 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
4442 (page_idx == erp->er_extoff + erp->er_extcount &&
4445 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
4446 erp->er_extcount == XFS_LINEAR_EXTS) {
4450 erp = erp_idx < nlists ? erp + 1 : NULL;
4453 page_idx -= erp->er_extoff;
4458 *erp_idxp = erp_idx;
4463 * Allocate and initialize an indirection array once the space needed
4464 * for incore extents increases above XFS_IEXT_BUFSZ.
4468 xfs_ifork_t *ifp) /* inode fork pointer */
4470 xfs_ext_irec_t *erp; /* indirection array pointer */
4471 xfs_extnum_t nextents; /* number of extents in file */
4473 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4474 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4475 ASSERT(nextents <= XFS_LINEAR_EXTS);
4477 erp = (xfs_ext_irec_t *)
4478 kmem_alloc(sizeof(xfs_ext_irec_t), KM_SLEEP);
4480 if (nextents == 0) {
4481 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
4482 kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
4483 } else if (!ifp->if_real_bytes) {
4484 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
4485 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
4486 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
4488 erp->er_extbuf = ifp->if_u1.if_extents;
4489 erp->er_extcount = nextents;
4492 ifp->if_flags |= XFS_IFEXTIREC;
4493 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
4494 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
4495 ifp->if_u1.if_ext_irec = erp;
4501 * Allocate and initialize a new entry in the indirection array.
4505 xfs_ifork_t *ifp, /* inode fork pointer */
4506 int erp_idx) /* index for new irec */
4508 xfs_ext_irec_t *erp; /* indirection array pointer */
4509 int i; /* loop counter */
4510 int nlists; /* number of irec's (ex lists) */
4512 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4513 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4515 /* Resize indirection array */
4516 xfs_iext_realloc_indirect(ifp, ++nlists *
4517 sizeof(xfs_ext_irec_t));
4519 * Move records down in the array so the
4520 * new page can use erp_idx.
4522 erp = ifp->if_u1.if_ext_irec;
4523 for (i = nlists - 1; i > erp_idx; i--) {
4524 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
4526 ASSERT(i == erp_idx);
4528 /* Initialize new extent record */
4529 erp = ifp->if_u1.if_ext_irec;
4530 erp[erp_idx].er_extbuf = (xfs_bmbt_rec_t *)
4531 kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
4532 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4533 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
4534 erp[erp_idx].er_extcount = 0;
4535 erp[erp_idx].er_extoff = erp_idx > 0 ?
4536 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
4537 return (&erp[erp_idx]);
4541 * Remove a record from the indirection array.
4544 xfs_iext_irec_remove(
4545 xfs_ifork_t *ifp, /* inode fork pointer */
4546 int erp_idx) /* irec index to remove */
4548 xfs_ext_irec_t *erp; /* indirection array pointer */
4549 int i; /* loop counter */
4550 int nlists; /* number of irec's (ex lists) */
4552 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4553 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4554 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4555 if (erp->er_extbuf) {
4556 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
4558 kmem_free(erp->er_extbuf, XFS_IEXT_BUFSZ);
4560 /* Compact extent records */
4561 erp = ifp->if_u1.if_ext_irec;
4562 for (i = erp_idx; i < nlists - 1; i++) {
4563 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
4566 * Manually free the last extent record from the indirection
4567 * array. A call to xfs_iext_realloc_indirect() with a size
4568 * of zero would result in a call to xfs_iext_destroy() which
4569 * would in turn call this function again, creating a nasty
4573 xfs_iext_realloc_indirect(ifp,
4574 nlists * sizeof(xfs_ext_irec_t));
4576 kmem_free(ifp->if_u1.if_ext_irec,
4577 sizeof(xfs_ext_irec_t));
4579 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4583 * This is called to clean up large amounts of unused memory allocated
4584 * by the indirection array. Before compacting anything though, verify
4585 * that the indirection array is still needed and switch back to the
4586 * linear extent list (or even the inline buffer) if possible. The
4587 * compaction policy is as follows:
4589 * Full Compaction: Extents fit into a single page (or inline buffer)
4590 * Full Compaction: Extents occupy less than 10% of allocated space
4591 * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
4592 * No Compaction: Extents occupy at least 50% of allocated space
4595 xfs_iext_irec_compact(
4596 xfs_ifork_t *ifp) /* inode fork pointer */
4598 xfs_extnum_t nextents; /* number of extents in file */
4599 int nlists; /* number of irec's (ex lists) */
4601 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4602 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4603 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4605 if (nextents == 0) {
4606 xfs_iext_destroy(ifp);
4607 } else if (nextents <= XFS_INLINE_EXTS) {
4608 xfs_iext_indirect_to_direct(ifp);
4609 xfs_iext_direct_to_inline(ifp, nextents);
4610 } else if (nextents <= XFS_LINEAR_EXTS) {
4611 xfs_iext_indirect_to_direct(ifp);
4612 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 3) {
4613 xfs_iext_irec_compact_full(ifp);
4614 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
4615 xfs_iext_irec_compact_pages(ifp);
4620 * Combine extents from neighboring extent pages.
4623 xfs_iext_irec_compact_pages(
4624 xfs_ifork_t *ifp) /* inode fork pointer */
4626 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
4627 int erp_idx = 0; /* indirection array index */
4628 int nlists; /* number of irec's (ex lists) */
4630 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4631 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4632 while (erp_idx < nlists - 1) {
4633 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4635 if (erp_next->er_extcount <=
4636 (XFS_LINEAR_EXTS - erp->er_extcount)) {
4637 memmove(&erp->er_extbuf[erp->er_extcount],
4638 erp_next->er_extbuf, erp_next->er_extcount *
4639 sizeof(xfs_bmbt_rec_t));
4640 erp->er_extcount += erp_next->er_extcount;
4642 * Free page before removing extent record
4643 * so er_extoffs don't get modified in
4644 * xfs_iext_irec_remove.
4646 kmem_free(erp_next->er_extbuf, XFS_IEXT_BUFSZ);
4647 erp_next->er_extbuf = NULL;
4648 xfs_iext_irec_remove(ifp, erp_idx + 1);
4649 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4657 * Fully compact the extent records managed by the indirection array.
4660 xfs_iext_irec_compact_full(
4661 xfs_ifork_t *ifp) /* inode fork pointer */
4663 xfs_bmbt_rec_t *ep, *ep_next; /* extent record pointers */
4664 xfs_ext_irec_t *erp, *erp_next; /* extent irec pointers */
4665 int erp_idx = 0; /* extent irec index */
4666 int ext_avail; /* empty entries in ex list */
4667 int ext_diff; /* number of exts to add */
4668 int nlists; /* number of irec's (ex lists) */
4670 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4671 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4672 erp = ifp->if_u1.if_ext_irec;
4673 ep = &erp->er_extbuf[erp->er_extcount];
4675 ep_next = erp_next->er_extbuf;
4676 while (erp_idx < nlists - 1) {
4677 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
4678 ext_diff = MIN(ext_avail, erp_next->er_extcount);
4679 memcpy(ep, ep_next, ext_diff * sizeof(xfs_bmbt_rec_t));
4680 erp->er_extcount += ext_diff;
4681 erp_next->er_extcount -= ext_diff;
4682 /* Remove next page */
4683 if (erp_next->er_extcount == 0) {
4685 * Free page before removing extent record
4686 * so er_extoffs don't get modified in
4687 * xfs_iext_irec_remove.
4689 kmem_free(erp_next->er_extbuf,
4690 erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
4691 erp_next->er_extbuf = NULL;
4692 xfs_iext_irec_remove(ifp, erp_idx + 1);
4693 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4694 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4695 /* Update next page */
4697 /* Move rest of page up to become next new page */
4698 memmove(erp_next->er_extbuf, ep_next,
4699 erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
4700 ep_next = erp_next->er_extbuf;
4701 memset(&ep_next[erp_next->er_extcount], 0,
4702 (XFS_LINEAR_EXTS - erp_next->er_extcount) *
4703 sizeof(xfs_bmbt_rec_t));
4705 if (erp->er_extcount == XFS_LINEAR_EXTS) {
4707 if (erp_idx < nlists)
4708 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4712 ep = &erp->er_extbuf[erp->er_extcount];
4714 ep_next = erp_next->er_extbuf;
4719 * This is called to update the er_extoff field in the indirection
4720 * array when extents have been added or removed from one of the
4721 * extent lists. erp_idx contains the irec index to begin updating
4722 * at and ext_diff contains the number of extents that were added
4726 xfs_iext_irec_update_extoffs(
4727 xfs_ifork_t *ifp, /* inode fork pointer */
4728 int erp_idx, /* irec index to update */
4729 int ext_diff) /* number of new extents */
4731 int i; /* loop counter */
4732 int nlists; /* number of irec's (ex lists */
4734 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4735 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4736 for (i = erp_idx; i < nlists; i++) {
4737 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;