2 * Copyright (c) 2000-2003,2005 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"
55 #include <linux/vserver/xid.h>
57 kmem_zone_t *xfs_ifork_zone;
58 kmem_zone_t *xfs_inode_zone;
59 kmem_zone_t *xfs_chashlist_zone;
62 * Used in xfs_itruncate(). This is the maximum number of extents
63 * freed from a file in a single transaction.
65 #define XFS_ITRUNC_MAX_EXTENTS 2
67 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
68 STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
69 STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
70 STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
75 * Make sure that the extents in the given memory buffer
90 for (i = 0; i < nrecs; i++) {
91 ep = xfs_iext_get_ext(ifp, i);
92 rec.l0 = get_unaligned((__uint64_t*)&ep->l0);
93 rec.l1 = get_unaligned((__uint64_t*)&ep->l1);
95 xfs_bmbt_disk_get_all(&rec, &irec);
97 xfs_bmbt_get_all(&rec, &irec);
98 if (fmt == XFS_EXTFMT_NOSTATE)
99 ASSERT(irec.br_state == XFS_EXT_NORM);
103 #define xfs_validate_extents(ifp, nrecs, disk, fmt)
107 * Check that none of the inode's in the buffer have a next
108 * unlinked field of 0.
120 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
122 for (i = 0; i < j; i++) {
123 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
124 i * mp->m_sb.sb_inodesize);
125 if (!dip->di_next_unlinked) {
126 xfs_fs_cmn_err(CE_ALERT, mp,
127 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
129 ASSERT(dip->di_next_unlinked);
136 * This routine is called to map an inode number within a file
137 * system to the buffer containing the on-disk version of the
138 * inode. It returns a pointer to the buffer containing the
139 * on-disk inode in the bpp parameter, and in the dip parameter
140 * it returns a pointer to the on-disk inode within that buffer.
142 * If a non-zero error is returned, then the contents of bpp and
143 * dipp are undefined.
145 * Use xfs_imap() to determine the size and location of the
146 * buffer to read from disk.
164 * Call the space management code to find the location of the
168 error = xfs_imap(mp, tp, ino, &imap, XFS_IMAP_LOOKUP);
171 "xfs_inotobp: xfs_imap() returned an "
172 "error %d on %s. Returning error.", error, mp->m_fsname);
177 * If the inode number maps to a block outside the bounds of the
178 * file system then return NULL rather than calling read_buf
179 * and panicing when we get an error from the driver.
181 if ((imap.im_blkno + imap.im_len) >
182 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
184 "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
185 "of the file system %s. Returning EINVAL.",
186 (unsigned long long)imap.im_blkno,
187 imap.im_len, mp->m_fsname);
188 return XFS_ERROR(EINVAL);
192 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
193 * default to just a read_buf() call.
195 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
196 (int)imap.im_len, XFS_BUF_LOCK, &bp);
200 "xfs_inotobp: xfs_trans_read_buf() returned an "
201 "error %d on %s. Returning error.", error, mp->m_fsname);
204 dip = (xfs_dinode_t *)xfs_buf_offset(bp, 0);
206 INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
207 XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
208 if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP,
209 XFS_RANDOM_ITOBP_INOTOBP))) {
210 XFS_CORRUPTION_ERROR("xfs_inotobp", XFS_ERRLEVEL_LOW, mp, dip);
211 xfs_trans_brelse(tp, bp);
213 "xfs_inotobp: XFS_TEST_ERROR() returned an "
214 "error on %s. Returning EFSCORRUPTED.", mp->m_fsname);
215 return XFS_ERROR(EFSCORRUPTED);
218 xfs_inobp_check(mp, bp);
221 * Set *dipp to point to the on-disk inode in the buffer.
223 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
225 *offset = imap.im_boffset;
231 * This routine is called to map an inode to the buffer containing
232 * the on-disk version of the inode. It returns a pointer to the
233 * buffer containing the on-disk inode in the bpp parameter, and in
234 * the dip parameter it returns a pointer to the on-disk inode within
237 * If a non-zero error is returned, then the contents of bpp and
238 * dipp are undefined.
240 * If the inode is new and has not yet been initialized, use xfs_imap()
241 * to determine the size and location of the buffer to read from disk.
242 * If the inode has already been mapped to its buffer and read in once,
243 * then use the mapping information stored in the inode rather than
244 * calling xfs_imap(). This allows us to avoid the overhead of looking
245 * at the inode btree for small block file systems (see xfs_dilocate()).
246 * We can tell whether the inode has been mapped in before by comparing
247 * its disk block address to 0. Only uninitialized inodes will have
248 * 0 for the disk block address.
268 if (ip->i_blkno == (xfs_daddr_t)0) {
270 * Call the space management code to find the location of the
274 if ((error = xfs_imap(mp, tp, ip->i_ino, &imap,
275 XFS_IMAP_LOOKUP | imap_flags)))
279 * If the inode number maps to a block outside the bounds
280 * of the file system then return NULL rather than calling
281 * read_buf and panicing when we get an error from the
284 if ((imap.im_blkno + imap.im_len) >
285 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
287 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
288 "(imap.im_blkno (0x%llx) "
289 "+ imap.im_len (0x%llx)) > "
290 " XFS_FSB_TO_BB(mp, "
291 "mp->m_sb.sb_dblocks) (0x%llx)",
292 (unsigned long long) imap.im_blkno,
293 (unsigned long long) imap.im_len,
294 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
296 return XFS_ERROR(EINVAL);
300 * Fill in the fields in the inode that will be used to
301 * map the inode to its buffer from now on.
303 ip->i_blkno = imap.im_blkno;
304 ip->i_len = imap.im_len;
305 ip->i_boffset = imap.im_boffset;
308 * We've already mapped the inode once, so just use the
309 * mapping that we saved the first time.
311 imap.im_blkno = ip->i_blkno;
312 imap.im_len = ip->i_len;
313 imap.im_boffset = ip->i_boffset;
315 ASSERT(bno == 0 || bno == imap.im_blkno);
318 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
319 * default to just a read_buf() call.
321 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
322 (int)imap.im_len, XFS_BUF_LOCK, &bp);
326 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
327 "xfs_trans_read_buf() returned error %d, "
328 "imap.im_blkno 0x%llx, imap.im_len 0x%llx",
329 error, (unsigned long long) imap.im_blkno,
330 (unsigned long long) imap.im_len);
336 * Validate the magic number and version of every inode in the buffer
337 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
340 ni = (imap_flags & XFS_IMAP_BULKSTAT) ? 0 :
341 (BBTOB(imap.im_len) >> mp->m_sb.sb_inodelog);
343 ni = (imap_flags & XFS_IMAP_BULKSTAT) ? 0 : 1;
345 for (i = 0; i < ni; i++) {
349 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
350 (i << mp->m_sb.sb_inodelog));
351 di_ok = INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
352 XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
353 if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP,
354 XFS_RANDOM_ITOBP_INOTOBP))) {
356 prdev("bad inode magic/vsn daddr %lld #%d (magic=%x)",
358 (unsigned long long)imap.im_blkno, i,
359 INT_GET(dip->di_core.di_magic, ARCH_CONVERT));
361 XFS_CORRUPTION_ERROR("xfs_itobp", XFS_ERRLEVEL_HIGH,
363 xfs_trans_brelse(tp, bp);
364 return XFS_ERROR(EFSCORRUPTED);
367 #endif /* __KERNEL__ */
369 xfs_inobp_check(mp, bp);
372 * Mark the buffer as an inode buffer now that it looks good
374 XFS_BUF_SET_VTYPE(bp, B_FS_INO);
377 * Set *dipp to point to the on-disk inode in the buffer.
379 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
385 * Move inode type and inode format specific information from the
386 * on-disk inode to the in-core inode. For fifos, devs, and sockets
387 * this means set if_rdev to the proper value. For files, directories,
388 * and symlinks this means to bring in the in-line data or extent
389 * pointers. For a file in B-tree format, only the root is immediately
390 * brought in-core. The rest will be in-lined in if_extents when it
391 * is first referenced (see xfs_iread_extents()).
398 xfs_attr_shortform_t *atp;
402 ip->i_df.if_ext_max =
403 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
407 INT_GET(dip->di_core.di_nextents, ARCH_CONVERT) +
408 INT_GET(dip->di_core.di_anextents, ARCH_CONVERT) >
409 INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT))) {
410 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
411 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
412 (unsigned long long)ip->i_ino,
413 (int)(INT_GET(dip->di_core.di_nextents, ARCH_CONVERT)
414 + INT_GET(dip->di_core.di_anextents, ARCH_CONVERT)),
416 INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT));
417 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
419 return XFS_ERROR(EFSCORRUPTED);
422 if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) {
423 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
424 "corrupt dinode %Lu, forkoff = 0x%x.",
425 (unsigned long long)ip->i_ino,
426 (int)(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT)));
427 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
429 return XFS_ERROR(EFSCORRUPTED);
432 switch (ip->i_d.di_mode & S_IFMT) {
437 if (unlikely(INT_GET(dip->di_core.di_format, ARCH_CONVERT) != XFS_DINODE_FMT_DEV)) {
438 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
440 return XFS_ERROR(EFSCORRUPTED);
443 ip->i_df.if_u2.if_rdev = INT_GET(dip->di_u.di_dev, ARCH_CONVERT);
449 switch (INT_GET(dip->di_core.di_format, ARCH_CONVERT)) {
450 case XFS_DINODE_FMT_LOCAL:
452 * no local regular files yet
454 if (unlikely((INT_GET(dip->di_core.di_mode, ARCH_CONVERT) & S_IFMT) == S_IFREG)) {
455 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
457 "(local format for regular file).",
458 (unsigned long long) ip->i_ino);
459 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
462 return XFS_ERROR(EFSCORRUPTED);
465 di_size = INT_GET(dip->di_core.di_size, ARCH_CONVERT);
466 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
467 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
469 "(bad size %Ld for local inode).",
470 (unsigned long long) ip->i_ino,
471 (long long) di_size);
472 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
475 return XFS_ERROR(EFSCORRUPTED);
479 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
481 case XFS_DINODE_FMT_EXTENTS:
482 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
484 case XFS_DINODE_FMT_BTREE:
485 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
488 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
490 return XFS_ERROR(EFSCORRUPTED);
495 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
496 return XFS_ERROR(EFSCORRUPTED);
501 if (!XFS_DFORK_Q(dip))
503 ASSERT(ip->i_afp == NULL);
504 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
505 ip->i_afp->if_ext_max =
506 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
507 switch (INT_GET(dip->di_core.di_aformat, ARCH_CONVERT)) {
508 case XFS_DINODE_FMT_LOCAL:
509 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
510 size = be16_to_cpu(atp->hdr.totsize);
511 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
513 case XFS_DINODE_FMT_EXTENTS:
514 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
516 case XFS_DINODE_FMT_BTREE:
517 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
520 error = XFS_ERROR(EFSCORRUPTED);
524 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
526 xfs_idestroy_fork(ip, XFS_DATA_FORK);
532 * The file is in-lined in the on-disk inode.
533 * If it fits into if_inline_data, then copy
534 * it there, otherwise allocate a buffer for it
535 * and copy the data there. Either way, set
536 * if_data to point at the data.
537 * If we allocate a buffer for the data, make
538 * sure that its size is a multiple of 4 and
539 * record the real size in i_real_bytes.
552 * If the size is unreasonable, then something
553 * is wrong and we just bail out rather than crash in
554 * kmem_alloc() or memcpy() below.
556 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
557 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
559 "(bad size %d for local fork, size = %d).",
560 (unsigned long long) ip->i_ino, size,
561 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
562 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
564 return XFS_ERROR(EFSCORRUPTED);
566 ifp = XFS_IFORK_PTR(ip, whichfork);
569 ifp->if_u1.if_data = NULL;
570 else if (size <= sizeof(ifp->if_u2.if_inline_data))
571 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
573 real_size = roundup(size, 4);
574 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
576 ifp->if_bytes = size;
577 ifp->if_real_bytes = real_size;
579 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
580 ifp->if_flags &= ~XFS_IFEXTENTS;
581 ifp->if_flags |= XFS_IFINLINE;
586 * The file consists of a set of extents all
587 * of which fit into the on-disk inode.
588 * If there are few enough extents to fit into
589 * the if_inline_ext, then copy them there.
590 * Otherwise allocate a buffer for them and copy
591 * them into it. Either way, set if_extents
592 * to point at the extents.
600 xfs_bmbt_rec_t *ep, *dp;
606 ifp = XFS_IFORK_PTR(ip, whichfork);
607 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
608 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
611 * If the number of extents is unreasonable, then something
612 * is wrong and we just bail out rather than crash in
613 * kmem_alloc() or memcpy() below.
615 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
616 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
617 "corrupt inode %Lu ((a)extents = %d).",
618 (unsigned long long) ip->i_ino, nex);
619 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
621 return XFS_ERROR(EFSCORRUPTED);
624 ifp->if_real_bytes = 0;
626 ifp->if_u1.if_extents = NULL;
627 else if (nex <= XFS_INLINE_EXTS)
628 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
630 xfs_iext_add(ifp, 0, nex);
632 ifp->if_bytes = size;
634 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
635 xfs_validate_extents(ifp, nex, 1, XFS_EXTFMT_INODE(ip));
636 for (i = 0; i < nex; i++, dp++) {
637 ep = xfs_iext_get_ext(ifp, i);
638 ep->l0 = INT_GET(get_unaligned((__uint64_t*)&dp->l0),
640 ep->l1 = INT_GET(get_unaligned((__uint64_t*)&dp->l1),
643 xfs_bmap_trace_exlist("xfs_iformat_extents", ip, nex,
645 if (whichfork != XFS_DATA_FORK ||
646 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
647 if (unlikely(xfs_check_nostate_extents(
649 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
652 return XFS_ERROR(EFSCORRUPTED);
655 ifp->if_flags |= XFS_IFEXTENTS;
660 * The file has too many extents to fit into
661 * the inode, so they are in B-tree format.
662 * Allocate a buffer for the root of the B-tree
663 * and copy the root into it. The i_extents
664 * field will remain NULL until all of the
665 * extents are read in (when they are needed).
673 xfs_bmdr_block_t *dfp;
679 ifp = XFS_IFORK_PTR(ip, whichfork);
680 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
681 size = XFS_BMAP_BROOT_SPACE(dfp);
682 nrecs = XFS_BMAP_BROOT_NUMRECS(dfp);
685 * blow out if -- fork has less extents than can fit in
686 * fork (fork shouldn't be a btree format), root btree
687 * block has more records than can fit into the fork,
688 * or the number of extents is greater than the number of
691 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
692 || XFS_BMDR_SPACE_CALC(nrecs) >
693 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
694 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
695 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
696 "corrupt inode %Lu (btree).",
697 (unsigned long long) ip->i_ino);
698 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
700 return XFS_ERROR(EFSCORRUPTED);
703 ifp->if_broot_bytes = size;
704 ifp->if_broot = kmem_alloc(size, KM_SLEEP);
705 ASSERT(ifp->if_broot != NULL);
707 * Copy and convert from the on-disk structure
708 * to the in-memory structure.
710 xfs_bmdr_to_bmbt(dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
711 ifp->if_broot, size);
712 ifp->if_flags &= ~XFS_IFEXTENTS;
713 ifp->if_flags |= XFS_IFBROOT;
719 * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
722 * buf = on-disk representation
723 * dip = native representation
724 * dir = direction - +ve -> disk to native
725 * -ve -> native to disk
728 xfs_xlate_dinode_core(
730 xfs_dinode_core_t *dip,
733 xfs_dinode_core_t *buf_core = (xfs_dinode_core_t *)buf;
734 xfs_dinode_core_t *mem_core = (xfs_dinode_core_t *)dip;
735 xfs_arch_t arch = ARCH_CONVERT;
736 uint32_t uid = 0, gid = 0;
742 xid = mem_core->di_xid;
743 /* FIXME: supposed to use superblock flag */
744 uid = XIDINO_UID(1, mem_core->di_uid, xid);
745 gid = XIDINO_GID(1, mem_core->di_gid, xid);
746 xid = XIDINO_XID(1, xid);
749 INT_XLATE(buf_core->di_magic, mem_core->di_magic, dir, arch);
750 INT_XLATE(buf_core->di_mode, mem_core->di_mode, dir, arch);
751 INT_XLATE(buf_core->di_version, mem_core->di_version, dir, arch);
752 INT_XLATE(buf_core->di_format, mem_core->di_format, dir, arch);
753 INT_XLATE(buf_core->di_onlink, mem_core->di_onlink, dir, arch);
754 INT_XLATE(buf_core->di_uid, uid, dir, arch);
755 INT_XLATE(buf_core->di_gid, gid, dir, arch);
756 INT_XLATE(buf_core->di_xid, xid, dir, arch);
757 INT_XLATE(buf_core->di_nlink, mem_core->di_nlink, dir, arch);
758 INT_XLATE(buf_core->di_projid, mem_core->di_projid, dir, arch);
761 /* FIXME: supposed to use superblock flag */
762 mem_core->di_uid = INOXID_UID(1, uid, gid);
763 mem_core->di_gid = INOXID_GID(1, uid, gid);
764 mem_core->di_xid = INOXID_XID(1, uid, gid, xid);
765 memcpy(mem_core->di_pad, buf_core->di_pad,
766 sizeof(buf_core->di_pad));
768 memcpy(buf_core->di_pad, mem_core->di_pad,
769 sizeof(buf_core->di_pad));
772 INT_XLATE(buf_core->di_flushiter, mem_core->di_flushiter, dir, arch);
774 INT_XLATE(buf_core->di_atime.t_sec, mem_core->di_atime.t_sec,
776 INT_XLATE(buf_core->di_atime.t_nsec, mem_core->di_atime.t_nsec,
778 INT_XLATE(buf_core->di_mtime.t_sec, mem_core->di_mtime.t_sec,
780 INT_XLATE(buf_core->di_mtime.t_nsec, mem_core->di_mtime.t_nsec,
782 INT_XLATE(buf_core->di_ctime.t_sec, mem_core->di_ctime.t_sec,
784 INT_XLATE(buf_core->di_ctime.t_nsec, mem_core->di_ctime.t_nsec,
786 INT_XLATE(buf_core->di_size, mem_core->di_size, dir, arch);
787 INT_XLATE(buf_core->di_nblocks, mem_core->di_nblocks, dir, arch);
788 INT_XLATE(buf_core->di_extsize, mem_core->di_extsize, dir, arch);
789 INT_XLATE(buf_core->di_nextents, mem_core->di_nextents, dir, arch);
790 INT_XLATE(buf_core->di_anextents, mem_core->di_anextents, dir, arch);
791 INT_XLATE(buf_core->di_forkoff, mem_core->di_forkoff, dir, arch);
792 INT_XLATE(buf_core->di_aformat, mem_core->di_aformat, dir, arch);
793 INT_XLATE(buf_core->di_dmevmask, mem_core->di_dmevmask, dir, arch);
794 INT_XLATE(buf_core->di_dmstate, mem_core->di_dmstate, dir, arch);
795 INT_XLATE(buf_core->di_flags, mem_core->di_flags, dir, arch);
796 INT_XLATE(buf_core->di_gen, mem_core->di_gen, dir, arch);
801 xfs_dinode_core_t *dic,
806 if (di_flags & XFS_DIFLAG_ANY) {
807 if (di_flags & XFS_DIFLAG_REALTIME)
808 flags |= XFS_XFLAG_REALTIME;
809 if (di_flags & XFS_DIFLAG_PREALLOC)
810 flags |= XFS_XFLAG_PREALLOC;
811 if (di_flags & XFS_DIFLAG_IMMUTABLE)
812 flags |= XFS_XFLAG_IMMUTABLE;
813 if (di_flags & XFS_DIFLAG_IUNLINK)
814 flags |= XFS_XFLAG_IUNLINK;
815 if (di_flags & XFS_DIFLAG_BARRIER)
816 flags |= XFS_XFLAG_BARRIER;
817 if (di_flags & XFS_DIFLAG_APPEND)
818 flags |= XFS_XFLAG_APPEND;
819 if (di_flags & XFS_DIFLAG_SYNC)
820 flags |= XFS_XFLAG_SYNC;
821 if (di_flags & XFS_DIFLAG_NOATIME)
822 flags |= XFS_XFLAG_NOATIME;
823 if (di_flags & XFS_DIFLAG_NODUMP)
824 flags |= XFS_XFLAG_NODUMP;
825 if (di_flags & XFS_DIFLAG_RTINHERIT)
826 flags |= XFS_XFLAG_RTINHERIT;
827 if (di_flags & XFS_DIFLAG_PROJINHERIT)
828 flags |= XFS_XFLAG_PROJINHERIT;
829 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
830 flags |= XFS_XFLAG_NOSYMLINKS;
831 if (di_flags & XFS_DIFLAG_EXTSIZE)
832 flags |= XFS_XFLAG_EXTSIZE;
833 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
834 flags |= XFS_XFLAG_EXTSZINHERIT;
844 xfs_dinode_core_t *dic = &ip->i_d;
846 return _xfs_dic2xflags(dic, dic->di_flags) |
847 (XFS_CFORK_Q(dic) ? XFS_XFLAG_HASATTR : 0);
852 xfs_dinode_core_t *dic)
854 return _xfs_dic2xflags(dic, INT_GET(dic->di_flags, ARCH_CONVERT)) |
855 (XFS_CFORK_Q_DISK(dic) ? XFS_XFLAG_HASATTR : 0);
859 * Given a mount structure and an inode number, return a pointer
860 * to a newly allocated in-core inode corresponding to the given
863 * Initialize the inode's attributes and extent pointers if it
864 * already has them (it will not if the inode has no links).
879 ASSERT(xfs_inode_zone != NULL);
881 ip = kmem_zone_zalloc(xfs_inode_zone, KM_SLEEP);
886 * Get pointer's to the on-disk inode and the buffer containing it.
887 * If the inode number refers to a block outside the file system
888 * then xfs_itobp() will return NULL. In this case we should
889 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
890 * know that this is a new incore inode.
892 error = xfs_itobp(mp, tp, ip, &dip, &bp, bno, 0);
894 kmem_zone_free(xfs_inode_zone, ip);
899 * Initialize inode's trace buffers.
900 * Do this before xfs_iformat in case it adds entries.
902 #ifdef XFS_BMAP_TRACE
903 ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_SLEEP);
905 #ifdef XFS_BMBT_TRACE
906 ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_SLEEP);
909 ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_SLEEP);
911 #ifdef XFS_ILOCK_TRACE
912 ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_SLEEP);
914 #ifdef XFS_DIR2_TRACE
915 ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_SLEEP);
919 * If we got something that isn't an inode it means someone
920 * (nfs or dmi) has a stale handle.
922 if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC) {
923 kmem_zone_free(xfs_inode_zone, ip);
924 xfs_trans_brelse(tp, bp);
926 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
927 "dip->di_core.di_magic (0x%x) != "
928 "XFS_DINODE_MAGIC (0x%x)",
929 INT_GET(dip->di_core.di_magic, ARCH_CONVERT),
932 return XFS_ERROR(EINVAL);
936 * If the on-disk inode is already linked to a directory
937 * entry, copy all of the inode into the in-core inode.
938 * xfs_iformat() handles copying in the inode format
939 * specific information.
940 * Otherwise, just get the truly permanent information.
942 if (dip->di_core.di_mode) {
943 xfs_xlate_dinode_core((xfs_caddr_t)&dip->di_core,
945 error = xfs_iformat(ip, dip);
947 kmem_zone_free(xfs_inode_zone, ip);
948 xfs_trans_brelse(tp, bp);
950 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
951 "xfs_iformat() returned error %d",
957 ip->i_d.di_magic = INT_GET(dip->di_core.di_magic, ARCH_CONVERT);
958 ip->i_d.di_version = INT_GET(dip->di_core.di_version, ARCH_CONVERT);
959 ip->i_d.di_gen = INT_GET(dip->di_core.di_gen, ARCH_CONVERT);
960 ip->i_d.di_flushiter = INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT);
962 * Make sure to pull in the mode here as well in
963 * case the inode is released without being used.
964 * This ensures that xfs_inactive() will see that
965 * the inode is already free and not try to mess
966 * with the uninitialized part of it.
970 * Initialize the per-fork minima and maxima for a new
971 * inode here. xfs_iformat will do it for old inodes.
973 ip->i_df.if_ext_max =
974 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
977 INIT_LIST_HEAD(&ip->i_reclaim);
980 * The inode format changed when we moved the link count and
981 * made it 32 bits long. If this is an old format inode,
982 * convert it in memory to look like a new one. If it gets
983 * flushed to disk we will convert back before flushing or
984 * logging it. We zero out the new projid field and the old link
985 * count field. We'll handle clearing the pad field (the remains
986 * of the old uuid field) when we actually convert the inode to
987 * the new format. We don't change the version number so that we
988 * can distinguish this from a real new format inode.
990 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
991 ip->i_d.di_nlink = ip->i_d.di_onlink;
992 ip->i_d.di_onlink = 0;
993 ip->i_d.di_projid = 0;
996 ip->i_delayed_blks = 0;
999 * Mark the buffer containing the inode as something to keep
1000 * around for a while. This helps to keep recently accessed
1001 * meta-data in-core longer.
1003 XFS_BUF_SET_REF(bp, XFS_INO_REF);
1006 * Use xfs_trans_brelse() to release the buffer containing the
1007 * on-disk inode, because it was acquired with xfs_trans_read_buf()
1008 * in xfs_itobp() above. If tp is NULL, this is just a normal
1009 * brelse(). If we're within a transaction, then xfs_trans_brelse()
1010 * will only release the buffer if it is not dirty within the
1011 * transaction. It will be OK to release the buffer in this case,
1012 * because inodes on disk are never destroyed and we will be
1013 * locking the new in-core inode before putting it in the hash
1014 * table where other processes can find it. Thus we don't have
1015 * to worry about the inode being changed just because we released
1018 xfs_trans_brelse(tp, bp);
1024 * Read in extents from a btree-format inode.
1025 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1035 xfs_extnum_t nextents;
1038 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
1039 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
1041 return XFS_ERROR(EFSCORRUPTED);
1043 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
1044 size = nextents * sizeof(xfs_bmbt_rec_t);
1045 ifp = XFS_IFORK_PTR(ip, whichfork);
1048 * We know that the size is valid (it's checked in iformat_btree)
1050 ifp->if_lastex = NULLEXTNUM;
1051 ifp->if_bytes = ifp->if_real_bytes = 0;
1052 ifp->if_flags |= XFS_IFEXTENTS;
1053 xfs_iext_add(ifp, 0, nextents);
1054 error = xfs_bmap_read_extents(tp, ip, whichfork);
1056 xfs_iext_destroy(ifp);
1057 ifp->if_flags &= ~XFS_IFEXTENTS;
1060 xfs_validate_extents(ifp, nextents, 0, XFS_EXTFMT_INODE(ip));
1065 * Allocate an inode on disk and return a copy of its in-core version.
1066 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1067 * appropriately within the inode. The uid and gid for the inode are
1068 * set according to the contents of the given cred structure.
1070 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1071 * has a free inode available, call xfs_iget()
1072 * to obtain the in-core version of the allocated inode. Finally,
1073 * fill in the inode and log its initial contents. In this case,
1074 * ialloc_context would be set to NULL and call_again set to false.
1076 * If xfs_dialloc() does not have an available inode,
1077 * it will replenish its supply by doing an allocation. Since we can
1078 * only do one allocation within a transaction without deadlocks, we
1079 * must commit the current transaction before returning the inode itself.
1080 * In this case, therefore, we will set call_again to true and return.
1081 * The caller should then commit the current transaction, start a new
1082 * transaction, and call xfs_ialloc() again to actually get the inode.
1084 * To ensure that some other process does not grab the inode that
1085 * was allocated during the first call to xfs_ialloc(), this routine
1086 * also returns the [locked] bp pointing to the head of the freelist
1087 * as ialloc_context. The caller should hold this buffer across
1088 * the commit and pass it back into this routine on the second call.
1100 xfs_buf_t **ialloc_context,
1101 boolean_t *call_again,
1111 * Call the space management code to pick
1112 * the on-disk inode to be allocated.
1114 error = xfs_dialloc(tp, pip->i_ino, mode, okalloc,
1115 ialloc_context, call_again, &ino);
1119 if (*call_again || ino == NULLFSINO) {
1123 ASSERT(*ialloc_context == NULL);
1126 * Get the in-core inode with the lock held exclusively.
1127 * This is because we're setting fields here we need
1128 * to prevent others from looking at until we're done.
1130 error = xfs_trans_iget(tp->t_mountp, tp, ino,
1131 IGET_CREATE, XFS_ILOCK_EXCL, &ip);
1138 ip->i_d.di_mode = (__uint16_t)mode;
1139 ip->i_d.di_onlink = 0;
1140 ip->i_d.di_nlink = nlink;
1141 ASSERT(ip->i_d.di_nlink == nlink);
1142 ip->i_d.di_uid = current_fsuid(cr);
1143 ip->i_d.di_gid = current_fsgid(cr);
1144 ip->i_d.di_xid = current_fsxid(cr, vp);
1145 ip->i_d.di_projid = prid;
1146 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1149 * If the superblock version is up to where we support new format
1150 * inodes and this is currently an old format inode, then change
1151 * the inode version number now. This way we only do the conversion
1152 * here rather than here and in the flush/logging code.
1154 if (XFS_SB_VERSION_HASNLINK(&tp->t_mountp->m_sb) &&
1155 ip->i_d.di_version == XFS_DINODE_VERSION_1) {
1156 ip->i_d.di_version = XFS_DINODE_VERSION_2;
1158 * We've already zeroed the old link count, the projid field,
1159 * and the pad field.
1164 * Project ids won't be stored on disk if we are using a version 1 inode.
1166 if ( (prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1))
1167 xfs_bump_ino_vers2(tp, ip);
1169 if (XFS_INHERIT_GID(pip, vp->v_vfsp)) {
1170 ip->i_d.di_gid = pip->i_d.di_gid;
1171 if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1172 ip->i_d.di_mode |= S_ISGID;
1177 * If the group ID of the new file does not match the effective group
1178 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1179 * (and only if the irix_sgid_inherit compatibility variable is set).
1181 if ((irix_sgid_inherit) &&
1182 (ip->i_d.di_mode & S_ISGID) &&
1183 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1184 ip->i_d.di_mode &= ~S_ISGID;
1187 ip->i_d.di_size = 0;
1188 ip->i_d.di_nextents = 0;
1189 ASSERT(ip->i_d.di_nblocks == 0);
1190 xfs_ichgtime(ip, XFS_ICHGTIME_CHG|XFS_ICHGTIME_ACC|XFS_ICHGTIME_MOD);
1192 * di_gen will have been taken care of in xfs_iread.
1194 ip->i_d.di_extsize = 0;
1195 ip->i_d.di_dmevmask = 0;
1196 ip->i_d.di_dmstate = 0;
1197 ip->i_d.di_flags = 0;
1198 flags = XFS_ILOG_CORE;
1199 switch (mode & S_IFMT) {
1204 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1205 ip->i_df.if_u2.if_rdev = rdev;
1206 ip->i_df.if_flags = 0;
1207 flags |= XFS_ILOG_DEV;
1211 if (unlikely(pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
1214 if ((mode & S_IFMT) == S_IFDIR) {
1215 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1216 di_flags |= XFS_DIFLAG_RTINHERIT;
1217 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1218 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1219 ip->i_d.di_extsize = pip->i_d.di_extsize;
1221 } else if ((mode & S_IFMT) == S_IFREG) {
1222 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) {
1223 di_flags |= XFS_DIFLAG_REALTIME;
1224 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
1226 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1227 di_flags |= XFS_DIFLAG_EXTSIZE;
1228 ip->i_d.di_extsize = pip->i_d.di_extsize;
1231 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1232 xfs_inherit_noatime)
1233 di_flags |= XFS_DIFLAG_NOATIME;
1234 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1236 di_flags |= XFS_DIFLAG_NODUMP;
1237 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1239 di_flags |= XFS_DIFLAG_SYNC;
1240 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1241 xfs_inherit_nosymlinks)
1242 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1243 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1244 di_flags |= XFS_DIFLAG_PROJINHERIT;
1245 ip->i_d.di_flags |= di_flags;
1249 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1250 ip->i_df.if_flags = XFS_IFEXTENTS;
1251 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1252 ip->i_df.if_u1.if_extents = NULL;
1258 * Attribute fork settings for new inode.
1260 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1261 ip->i_d.di_anextents = 0;
1264 * Log the new values stuffed into the inode.
1266 xfs_trans_log_inode(tp, ip, flags);
1268 /* now that we have an i_mode we can set Linux inode ops (& unlock) */
1269 VFS_INIT_VNODE(XFS_MTOVFS(tp->t_mountp), vp, XFS_ITOBHV(ip), 1);
1276 * Check to make sure that there are no blocks allocated to the
1277 * file beyond the size of the file. We don't check this for
1278 * files with fixed size extents or real time extents, but we
1279 * at least do it for regular files.
1288 xfs_fileoff_t map_first;
1290 xfs_bmbt_irec_t imaps[2];
1292 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1295 if (ip->i_d.di_flags & (XFS_DIFLAG_REALTIME | XFS_DIFLAG_EXTSIZE))
1299 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1301 * The filesystem could be shutting down, so bmapi may return
1304 if (xfs_bmapi(NULL, ip, map_first,
1306 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1308 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1311 ASSERT(nimaps == 1);
1312 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1317 * Calculate the last possible buffered byte in a file. This must
1318 * include data that was buffered beyond the EOF by the write code.
1319 * This also needs to deal with overflowing the xfs_fsize_t type
1320 * which can happen for sizes near the limit.
1322 * We also need to take into account any blocks beyond the EOF. It
1323 * may be the case that they were buffered by a write which failed.
1324 * In that case the pages will still be in memory, but the inode size
1325 * will never have been updated.
1332 xfs_fsize_t last_byte;
1333 xfs_fileoff_t last_block;
1334 xfs_fileoff_t size_last_block;
1337 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE | MR_ACCESS));
1341 * Only check for blocks beyond the EOF if the extents have
1342 * been read in. This eliminates the need for the inode lock,
1343 * and it also saves us from looking when it really isn't
1346 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1347 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1355 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_d.di_size);
1356 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1358 last_byte = XFS_FSB_TO_B(mp, last_block);
1359 if (last_byte < 0) {
1360 return XFS_MAXIOFFSET(mp);
1362 last_byte += (1 << mp->m_writeio_log);
1363 if (last_byte < 0) {
1364 return XFS_MAXIOFFSET(mp);
1369 #if defined(XFS_RW_TRACE)
1375 xfs_fsize_t new_size,
1376 xfs_off_t toss_start,
1377 xfs_off_t toss_finish)
1379 if (ip->i_rwtrace == NULL) {
1383 ktrace_enter(ip->i_rwtrace,
1386 (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
1387 (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
1388 (void*)((long)flag),
1389 (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
1390 (void*)(unsigned long)(new_size & 0xffffffff),
1391 (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
1392 (void*)(unsigned long)(toss_start & 0xffffffff),
1393 (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
1394 (void*)(unsigned long)(toss_finish & 0xffffffff),
1395 (void*)(unsigned long)current_cpu(),
1396 (void*)(unsigned long)current_pid(),
1402 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1406 * Start the truncation of the file to new_size. The new size
1407 * must be smaller than the current size. This routine will
1408 * clear the buffer and page caches of file data in the removed
1409 * range, and xfs_itruncate_finish() will remove the underlying
1412 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1413 * must NOT have the inode lock held at all. This is because we're
1414 * calling into the buffer/page cache code and we can't hold the
1415 * inode lock when we do so.
1417 * We need to wait for any direct I/Os in flight to complete before we
1418 * proceed with the truncate. This is needed to prevent the extents
1419 * being read or written by the direct I/Os from being removed while the
1420 * I/O is in flight as there is no other method of synchronising
1421 * direct I/O with the truncate operation. Also, because we hold
1422 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1423 * started until the truncate completes and drops the lock. Essentially,
1424 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1425 * between direct I/Os and the truncate operation.
1427 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1428 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1429 * in the case that the caller is locking things out of order and
1430 * may not be able to call xfs_itruncate_finish() with the inode lock
1431 * held without dropping the I/O lock. If the caller must drop the
1432 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1433 * must be called again with all the same restrictions as the initial
1437 xfs_itruncate_start(
1440 xfs_fsize_t new_size)
1442 xfs_fsize_t last_byte;
1443 xfs_off_t toss_start;
1447 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1448 ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
1449 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1450 (flags == XFS_ITRUNC_MAYBE));
1455 vn_iowait(vp); /* wait for the completion of any pending DIOs */
1458 * Call VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES() to get rid of pages and buffers
1459 * overlapping the region being removed. We have to use
1460 * the less efficient VOP_FLUSHINVAL_PAGES() in the case that the
1461 * caller may not be able to finish the truncate without
1462 * dropping the inode's I/O lock. Make sure
1463 * to catch any pages brought in by buffers overlapping
1464 * the EOF by searching out beyond the isize by our
1465 * block size. We round new_size up to a block boundary
1466 * so that we don't toss things on the same block as
1467 * new_size but before it.
1469 * Before calling VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES(), make sure to
1470 * call remapf() over the same region if the file is mapped.
1471 * This frees up mapped file references to the pages in the
1472 * given range and for the VOP_FLUSHINVAL_PAGES() case it ensures
1473 * that we get the latest mapped changes flushed out.
1475 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1476 toss_start = XFS_FSB_TO_B(mp, toss_start);
1477 if (toss_start < 0) {
1479 * The place to start tossing is beyond our maximum
1480 * file size, so there is no way that the data extended
1485 last_byte = xfs_file_last_byte(ip);
1486 xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
1488 if (last_byte > toss_start) {
1489 if (flags & XFS_ITRUNC_DEFINITE) {
1490 VOP_TOSS_PAGES(vp, toss_start, -1, FI_REMAPF_LOCKED);
1492 VOP_FLUSHINVAL_PAGES(vp, toss_start, -1, FI_REMAPF_LOCKED);
1497 if (new_size == 0) {
1498 ASSERT(VN_CACHED(vp) == 0);
1504 * Shrink the file to the given new_size. The new
1505 * size must be smaller than the current size.
1506 * This will free up the underlying blocks
1507 * in the removed range after a call to xfs_itruncate_start()
1508 * or xfs_atruncate_start().
1510 * The transaction passed to this routine must have made
1511 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1512 * This routine may commit the given transaction and
1513 * start new ones, so make sure everything involved in
1514 * the transaction is tidy before calling here.
1515 * Some transaction will be returned to the caller to be
1516 * committed. The incoming transaction must already include
1517 * the inode, and both inode locks must be held exclusively.
1518 * The inode must also be "held" within the transaction. On
1519 * return the inode will be "held" within the returned transaction.
1520 * This routine does NOT require any disk space to be reserved
1521 * for it within the transaction.
1523 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1524 * and it indicates the fork which is to be truncated. For the
1525 * attribute fork we only support truncation to size 0.
1527 * We use the sync parameter to indicate whether or not the first
1528 * transaction we perform might have to be synchronous. For the attr fork,
1529 * it needs to be so if the unlink of the inode is not yet known to be
1530 * permanent in the log. This keeps us from freeing and reusing the
1531 * blocks of the attribute fork before the unlink of the inode becomes
1534 * For the data fork, we normally have to run synchronously if we're
1535 * being called out of the inactive path or we're being called
1536 * out of the create path where we're truncating an existing file.
1537 * Either way, the truncate needs to be sync so blocks don't reappear
1538 * in the file with altered data in case of a crash. wsync filesystems
1539 * can run the first case async because anything that shrinks the inode
1540 * has to run sync so by the time we're called here from inactive, the
1541 * inode size is permanently set to 0.
1543 * Calls from the truncate path always need to be sync unless we're
1544 * in a wsync filesystem and the file has already been unlinked.
1546 * The caller is responsible for correctly setting the sync parameter.
1547 * It gets too hard for us to guess here which path we're being called
1548 * out of just based on inode state.
1551 xfs_itruncate_finish(
1554 xfs_fsize_t new_size,
1558 xfs_fsblock_t first_block;
1559 xfs_fileoff_t first_unmap_block;
1560 xfs_fileoff_t last_block;
1561 xfs_filblks_t unmap_len=0;
1566 xfs_bmap_free_t free_list;
1569 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1570 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
1571 ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
1572 ASSERT(*tp != NULL);
1573 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1574 ASSERT(ip->i_transp == *tp);
1575 ASSERT(ip->i_itemp != NULL);
1576 ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
1580 mp = (ntp)->t_mountp;
1581 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1584 * We only support truncating the entire attribute fork.
1586 if (fork == XFS_ATTR_FORK) {
1589 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1590 xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
1592 * The first thing we do is set the size to new_size permanently
1593 * on disk. This way we don't have to worry about anyone ever
1594 * being able to look at the data being freed even in the face
1595 * of a crash. What we're getting around here is the case where
1596 * we free a block, it is allocated to another file, it is written
1597 * to, and then we crash. If the new data gets written to the
1598 * file but the log buffers containing the free and reallocation
1599 * don't, then we'd end up with garbage in the blocks being freed.
1600 * As long as we make the new_size permanent before actually
1601 * freeing any blocks it doesn't matter if they get writtten to.
1603 * The callers must signal into us whether or not the size
1604 * setting here must be synchronous. There are a few cases
1605 * where it doesn't have to be synchronous. Those cases
1606 * occur if the file is unlinked and we know the unlink is
1607 * permanent or if the blocks being truncated are guaranteed
1608 * to be beyond the inode eof (regardless of the link count)
1609 * and the eof value is permanent. Both of these cases occur
1610 * only on wsync-mounted filesystems. In those cases, we're
1611 * guaranteed that no user will ever see the data in the blocks
1612 * that are being truncated so the truncate can run async.
1613 * In the free beyond eof case, the file may wind up with
1614 * more blocks allocated to it than it needs if we crash
1615 * and that won't get fixed until the next time the file
1616 * is re-opened and closed but that's ok as that shouldn't
1617 * be too many blocks.
1619 * However, we can't just make all wsync xactions run async
1620 * because there's one call out of the create path that needs
1621 * to run sync where it's truncating an existing file to size
1622 * 0 whose size is > 0.
1624 * It's probably possible to come up with a test in this
1625 * routine that would correctly distinguish all the above
1626 * cases from the values of the function parameters and the
1627 * inode state but for sanity's sake, I've decided to let the
1628 * layers above just tell us. It's simpler to correctly figure
1629 * out in the layer above exactly under what conditions we
1630 * can run async and I think it's easier for others read and
1631 * follow the logic in case something has to be changed.
1632 * cscope is your friend -- rcc.
1634 * The attribute fork is much simpler.
1636 * For the attribute fork we allow the caller to tell us whether
1637 * the unlink of the inode that led to this call is yet permanent
1638 * in the on disk log. If it is not and we will be freeing extents
1639 * in this inode then we make the first transaction synchronous
1640 * to make sure that the unlink is permanent by the time we free
1643 if (fork == XFS_DATA_FORK) {
1644 if (ip->i_d.di_nextents > 0) {
1645 ip->i_d.di_size = new_size;
1646 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1649 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1650 if (ip->i_d.di_anextents > 0)
1651 xfs_trans_set_sync(ntp);
1653 ASSERT(fork == XFS_DATA_FORK ||
1654 (fork == XFS_ATTR_FORK &&
1655 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1656 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1659 * Since it is possible for space to become allocated beyond
1660 * the end of the file (in a crash where the space is allocated
1661 * but the inode size is not yet updated), simply remove any
1662 * blocks which show up between the new EOF and the maximum
1663 * possible file size. If the first block to be removed is
1664 * beyond the maximum file size (ie it is the same as last_block),
1665 * then there is nothing to do.
1667 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1668 ASSERT(first_unmap_block <= last_block);
1670 if (last_block == first_unmap_block) {
1673 unmap_len = last_block - first_unmap_block + 1;
1677 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1678 * will tell us whether it freed the entire range or
1679 * not. If this is a synchronous mount (wsync),
1680 * then we can tell bunmapi to keep all the
1681 * transactions asynchronous since the unlink
1682 * transaction that made this inode inactive has
1683 * already hit the disk. There's no danger of
1684 * the freed blocks being reused, there being a
1685 * crash, and the reused blocks suddenly reappearing
1686 * in this file with garbage in them once recovery
1689 XFS_BMAP_INIT(&free_list, &first_block);
1690 error = xfs_bunmapi(ntp, ip, first_unmap_block,
1692 XFS_BMAPI_AFLAG(fork) |
1693 (sync ? 0 : XFS_BMAPI_ASYNC),
1694 XFS_ITRUNC_MAX_EXTENTS,
1695 &first_block, &free_list, &done);
1698 * If the bunmapi call encounters an error,
1699 * return to the caller where the transaction
1700 * can be properly aborted. We just need to
1701 * make sure we're not holding any resources
1702 * that we were not when we came in.
1704 xfs_bmap_cancel(&free_list);
1709 * Duplicate the transaction that has the permanent
1710 * reservation and commit the old transaction.
1712 error = xfs_bmap_finish(tp, &free_list, first_block,
1717 * If the bmap finish call encounters an error,
1718 * return to the caller where the transaction
1719 * can be properly aborted. We just need to
1720 * make sure we're not holding any resources
1721 * that we were not when we came in.
1723 * Aborting from this point might lose some
1724 * blocks in the file system, but oh well.
1726 xfs_bmap_cancel(&free_list);
1729 * If the passed in transaction committed
1730 * in xfs_bmap_finish(), then we want to
1731 * add the inode to this one before returning.
1732 * This keeps things simple for the higher
1733 * level code, because it always knows that
1734 * the inode is locked and held in the
1735 * transaction that returns to it whether
1736 * errors occur or not. We don't mark the
1737 * inode dirty so that this transaction can
1738 * be easily aborted if possible.
1740 xfs_trans_ijoin(ntp, ip,
1741 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1742 xfs_trans_ihold(ntp, ip);
1749 * The first xact was committed,
1750 * so add the inode to the new one.
1751 * Mark it dirty so it will be logged
1752 * and moved forward in the log as
1753 * part of every commit.
1755 xfs_trans_ijoin(ntp, ip,
1756 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1757 xfs_trans_ihold(ntp, ip);
1758 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1760 ntp = xfs_trans_dup(ntp);
1761 (void) xfs_trans_commit(*tp, 0, NULL);
1763 error = xfs_trans_reserve(ntp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1764 XFS_TRANS_PERM_LOG_RES,
1765 XFS_ITRUNCATE_LOG_COUNT);
1767 * Add the inode being truncated to the next chained
1770 xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1771 xfs_trans_ihold(ntp, ip);
1776 * Only update the size in the case of the data fork, but
1777 * always re-log the inode so that our permanent transaction
1778 * can keep on rolling it forward in the log.
1780 if (fork == XFS_DATA_FORK) {
1781 xfs_isize_check(mp, ip, new_size);
1782 ip->i_d.di_size = new_size;
1784 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1785 ASSERT((new_size != 0) ||
1786 (fork == XFS_ATTR_FORK) ||
1787 (ip->i_delayed_blks == 0));
1788 ASSERT((new_size != 0) ||
1789 (fork == XFS_ATTR_FORK) ||
1790 (ip->i_d.di_nextents == 0));
1791 xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
1799 * Do the first part of growing a file: zero any data in the last
1800 * block that is beyond the old EOF. We need to do this before
1801 * the inode is joined to the transaction to modify the i_size.
1802 * That way we can drop the inode lock and call into the buffer
1803 * cache to get the buffer mapping the EOF.
1808 xfs_fsize_t new_size,
1813 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1814 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1815 ASSERT(new_size > ip->i_d.di_size);
1818 * Zero any pages that may have been created by
1819 * xfs_write_file() beyond the end of the file
1820 * and any blocks between the old and new file sizes.
1822 error = xfs_zero_eof(XFS_ITOV(ip), &ip->i_iocore, new_size,
1823 ip->i_d.di_size, new_size);
1830 * This routine is called to extend the size of a file.
1831 * The inode must have both the iolock and the ilock locked
1832 * for update and it must be a part of the current transaction.
1833 * The xfs_igrow_start() function must have been called previously.
1834 * If the change_flag is not zero, the inode change timestamp will
1841 xfs_fsize_t new_size,
1844 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1845 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1846 ASSERT(ip->i_transp == tp);
1847 ASSERT(new_size > ip->i_d.di_size);
1850 * Update the file size. Update the inode change timestamp
1851 * if change_flag set.
1853 ip->i_d.di_size = new_size;
1855 xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
1856 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1862 * This is called when the inode's link count goes to 0.
1863 * We place the on-disk inode on a list in the AGI. It
1864 * will be pulled from this list when the inode is freed.
1876 xfs_agnumber_t agno;
1877 xfs_daddr_t agdaddr;
1884 ASSERT(ip->i_d.di_nlink == 0);
1885 ASSERT(ip->i_d.di_mode != 0);
1886 ASSERT(ip->i_transp == tp);
1890 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1891 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1894 * Get the agi buffer first. It ensures lock ordering
1897 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1898 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1903 * Validate the magic number of the agi block.
1905 agi = XFS_BUF_TO_AGI(agibp);
1907 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1908 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
1909 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK,
1910 XFS_RANDOM_IUNLINK))) {
1911 XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi);
1912 xfs_trans_brelse(tp, agibp);
1913 return XFS_ERROR(EFSCORRUPTED);
1916 * Get the index into the agi hash table for the
1917 * list this inode will go on.
1919 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1921 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1922 ASSERT(agi->agi_unlinked[bucket_index]);
1923 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1925 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
1927 * There is already another inode in the bucket we need
1928 * to add ourselves to. Add us at the front of the list.
1929 * Here we put the head pointer into our next pointer,
1930 * and then we fall through to point the head at us.
1932 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
1936 ASSERT(INT_GET(dip->di_next_unlinked, ARCH_CONVERT) == NULLAGINO);
1937 ASSERT(dip->di_next_unlinked);
1938 /* both on-disk, don't endian flip twice */
1939 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1940 offset = ip->i_boffset +
1941 offsetof(xfs_dinode_t, di_next_unlinked);
1942 xfs_trans_inode_buf(tp, ibp);
1943 xfs_trans_log_buf(tp, ibp, offset,
1944 (offset + sizeof(xfs_agino_t) - 1));
1945 xfs_inobp_check(mp, ibp);
1949 * Point the bucket head pointer at the inode being inserted.
1952 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1953 offset = offsetof(xfs_agi_t, agi_unlinked) +
1954 (sizeof(xfs_agino_t) * bucket_index);
1955 xfs_trans_log_buf(tp, agibp, offset,
1956 (offset + sizeof(xfs_agino_t) - 1));
1961 * Pull the on-disk inode from the AGI unlinked list.
1974 xfs_agnumber_t agno;
1975 xfs_daddr_t agdaddr;
1977 xfs_agino_t next_agino;
1978 xfs_buf_t *last_ibp;
1979 xfs_dinode_t *last_dip;
1981 int offset, last_offset;
1986 * First pull the on-disk inode from the AGI unlinked list.
1990 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1991 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1994 * Get the agi buffer first. It ensures lock ordering
1997 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1998 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
2001 "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
2002 error, mp->m_fsname);
2006 * Validate the magic number of the agi block.
2008 agi = XFS_BUF_TO_AGI(agibp);
2010 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
2011 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
2012 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE,
2013 XFS_RANDOM_IUNLINK_REMOVE))) {
2014 XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW,
2016 xfs_trans_brelse(tp, agibp);
2018 "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
2020 return XFS_ERROR(EFSCORRUPTED);
2023 * Get the index into the agi hash table for the
2024 * list this inode will go on.
2026 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2028 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2029 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
2030 ASSERT(agi->agi_unlinked[bucket_index]);
2032 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2034 * We're at the head of the list. Get the inode's
2035 * on-disk buffer to see if there is anyone after us
2036 * on the list. Only modify our next pointer if it
2037 * is not already NULLAGINO. This saves us the overhead
2038 * of dealing with the buffer when there is no need to
2041 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
2044 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2045 error, mp->m_fsname);
2048 next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
2049 ASSERT(next_agino != 0);
2050 if (next_agino != NULLAGINO) {
2051 INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
2052 offset = ip->i_boffset +
2053 offsetof(xfs_dinode_t, di_next_unlinked);
2054 xfs_trans_inode_buf(tp, ibp);
2055 xfs_trans_log_buf(tp, ibp, offset,
2056 (offset + sizeof(xfs_agino_t) - 1));
2057 xfs_inobp_check(mp, ibp);
2059 xfs_trans_brelse(tp, ibp);
2062 * Point the bucket head pointer at the next inode.
2064 ASSERT(next_agino != 0);
2065 ASSERT(next_agino != agino);
2066 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2067 offset = offsetof(xfs_agi_t, agi_unlinked) +
2068 (sizeof(xfs_agino_t) * bucket_index);
2069 xfs_trans_log_buf(tp, agibp, offset,
2070 (offset + sizeof(xfs_agino_t) - 1));
2073 * We need to search the list for the inode being freed.
2075 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2077 while (next_agino != agino) {
2079 * If the last inode wasn't the one pointing to
2080 * us, then release its buffer since we're not
2081 * going to do anything with it.
2083 if (last_ibp != NULL) {
2084 xfs_trans_brelse(tp, last_ibp);
2086 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2087 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
2088 &last_ibp, &last_offset);
2091 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
2092 error, mp->m_fsname);
2095 next_agino = INT_GET(last_dip->di_next_unlinked, ARCH_CONVERT);
2096 ASSERT(next_agino != NULLAGINO);
2097 ASSERT(next_agino != 0);
2100 * Now last_ibp points to the buffer previous to us on
2101 * the unlinked list. Pull us from the list.
2103 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
2106 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2107 error, mp->m_fsname);
2110 next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
2111 ASSERT(next_agino != 0);
2112 ASSERT(next_agino != agino);
2113 if (next_agino != NULLAGINO) {
2114 INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
2115 offset = ip->i_boffset +
2116 offsetof(xfs_dinode_t, di_next_unlinked);
2117 xfs_trans_inode_buf(tp, ibp);
2118 xfs_trans_log_buf(tp, ibp, offset,
2119 (offset + sizeof(xfs_agino_t) - 1));
2120 xfs_inobp_check(mp, ibp);
2122 xfs_trans_brelse(tp, ibp);
2125 * Point the previous inode on the list to the next inode.
2127 INT_SET(last_dip->di_next_unlinked, ARCH_CONVERT, next_agino);
2128 ASSERT(next_agino != 0);
2129 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2130 xfs_trans_inode_buf(tp, last_ibp);
2131 xfs_trans_log_buf(tp, last_ibp, offset,
2132 (offset + sizeof(xfs_agino_t) - 1));
2133 xfs_inobp_check(mp, last_ibp);
2138 static __inline__ int xfs_inode_clean(xfs_inode_t *ip)
2140 return (((ip->i_itemp == NULL) ||
2141 !(ip->i_itemp->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
2142 (ip->i_update_core == 0));
2147 xfs_inode_t *free_ip,
2151 xfs_mount_t *mp = free_ip->i_mount;
2152 int blks_per_cluster;
2155 int i, j, found, pre_flushed;
2159 xfs_inode_t *ip, **ip_found;
2160 xfs_inode_log_item_t *iip;
2161 xfs_log_item_t *lip;
2164 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
2165 blks_per_cluster = 1;
2166 ninodes = mp->m_sb.sb_inopblock;
2167 nbufs = XFS_IALLOC_BLOCKS(mp);
2169 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
2170 mp->m_sb.sb_blocksize;
2171 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
2172 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
2175 ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
2177 for (j = 0; j < nbufs; j++, inum += ninodes) {
2178 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2179 XFS_INO_TO_AGBNO(mp, inum));
2183 * Look for each inode in memory and attempt to lock it,
2184 * we can be racing with flush and tail pushing here.
2185 * any inode we get the locks on, add to an array of
2186 * inode items to process later.
2188 * The get the buffer lock, we could beat a flush
2189 * or tail pushing thread to the lock here, in which
2190 * case they will go looking for the inode buffer
2191 * and fail, we need some other form of interlock
2195 for (i = 0; i < ninodes; i++) {
2196 ih = XFS_IHASH(mp, inum + i);
2197 read_lock(&ih->ih_lock);
2198 for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
2199 if (ip->i_ino == inum + i)
2203 /* Inode not in memory or we found it already,
2206 if (!ip || (ip->i_flags & XFS_ISTALE)) {
2207 read_unlock(&ih->ih_lock);
2211 if (xfs_inode_clean(ip)) {
2212 read_unlock(&ih->ih_lock);
2216 /* If we can get the locks then add it to the
2217 * list, otherwise by the time we get the bp lock
2218 * below it will already be attached to the
2222 /* This inode will already be locked - by us, lets
2226 if (ip == free_ip) {
2227 if (xfs_iflock_nowait(ip)) {
2228 ip->i_flags |= XFS_ISTALE;
2230 if (xfs_inode_clean(ip)) {
2233 ip_found[found++] = ip;
2236 read_unlock(&ih->ih_lock);
2240 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2241 if (xfs_iflock_nowait(ip)) {
2242 ip->i_flags |= XFS_ISTALE;
2244 if (xfs_inode_clean(ip)) {
2246 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2248 ip_found[found++] = ip;
2251 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2255 read_unlock(&ih->ih_lock);
2258 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2259 mp->m_bsize * blks_per_cluster,
2263 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
2265 if (lip->li_type == XFS_LI_INODE) {
2266 iip = (xfs_inode_log_item_t *)lip;
2267 ASSERT(iip->ili_logged == 1);
2268 lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
2270 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2272 iip->ili_inode->i_flags |= XFS_ISTALE;
2275 lip = lip->li_bio_list;
2278 for (i = 0; i < found; i++) {
2283 ip->i_update_core = 0;
2285 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2289 iip->ili_last_fields = iip->ili_format.ilf_fields;
2290 iip->ili_format.ilf_fields = 0;
2291 iip->ili_logged = 1;
2293 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2296 xfs_buf_attach_iodone(bp,
2297 (void(*)(xfs_buf_t*,xfs_log_item_t*))
2298 xfs_istale_done, (xfs_log_item_t *)iip);
2299 if (ip != free_ip) {
2300 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2304 if (found || pre_flushed)
2305 xfs_trans_stale_inode_buf(tp, bp);
2306 xfs_trans_binval(tp, bp);
2309 kmem_free(ip_found, ninodes * sizeof(xfs_inode_t *));
2313 * This is called to return an inode to the inode free list.
2314 * The inode should already be truncated to 0 length and have
2315 * no pages associated with it. This routine also assumes that
2316 * the inode is already a part of the transaction.
2318 * The on-disk copy of the inode will have been added to the list
2319 * of unlinked inodes in the AGI. We need to remove the inode from
2320 * that list atomically with respect to freeing it here.
2326 xfs_bmap_free_t *flist)
2330 xfs_ino_t first_ino;
2332 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2333 ASSERT(ip->i_transp == tp);
2334 ASSERT(ip->i_d.di_nlink == 0);
2335 ASSERT(ip->i_d.di_nextents == 0);
2336 ASSERT(ip->i_d.di_anextents == 0);
2337 ASSERT((ip->i_d.di_size == 0) ||
2338 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2339 ASSERT(ip->i_d.di_nblocks == 0);
2342 * Pull the on-disk inode from the AGI unlinked list.
2344 error = xfs_iunlink_remove(tp, ip);
2349 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2353 ip->i_d.di_mode = 0; /* mark incore inode as free */
2354 ip->i_d.di_flags = 0;
2355 ip->i_d.di_dmevmask = 0;
2356 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2357 ip->i_df.if_ext_max =
2358 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2359 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2360 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2362 * Bump the generation count so no one will be confused
2363 * by reincarnations of this inode.
2366 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2369 xfs_ifree_cluster(ip, tp, first_ino);
2376 * Reallocate the space for if_broot based on the number of records
2377 * being added or deleted as indicated in rec_diff. Move the records
2378 * and pointers in if_broot to fit the new size. When shrinking this
2379 * will eliminate holes between the records and pointers created by
2380 * the caller. When growing this will create holes to be filled in
2383 * The caller must not request to add more records than would fit in
2384 * the on-disk inode root. If the if_broot is currently NULL, then
2385 * if we adding records one will be allocated. The caller must also
2386 * not request that the number of records go below zero, although
2387 * it can go to zero.
2389 * ip -- the inode whose if_broot area is changing
2390 * ext_diff -- the change in the number of records, positive or negative,
2391 * requested for the if_broot array.
2401 xfs_bmbt_block_t *new_broot;
2408 * Handle the degenerate case quietly.
2410 if (rec_diff == 0) {
2414 ifp = XFS_IFORK_PTR(ip, whichfork);
2417 * If there wasn't any memory allocated before, just
2418 * allocate it now and get out.
2420 if (ifp->if_broot_bytes == 0) {
2421 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
2422 ifp->if_broot = (xfs_bmbt_block_t*)kmem_alloc(new_size,
2424 ifp->if_broot_bytes = (int)new_size;
2429 * If there is already an existing if_broot, then we need
2430 * to realloc() it and shift the pointers to their new
2431 * location. The records don't change location because
2432 * they are kept butted up against the btree block header.
2434 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2435 new_max = cur_max + rec_diff;
2436 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2437 ifp->if_broot = (xfs_bmbt_block_t *)
2438 kmem_realloc(ifp->if_broot,
2440 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2442 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2443 ifp->if_broot_bytes);
2444 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2446 ifp->if_broot_bytes = (int)new_size;
2447 ASSERT(ifp->if_broot_bytes <=
2448 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2449 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2454 * rec_diff is less than 0. In this case, we are shrinking the
2455 * if_broot buffer. It must already exist. If we go to zero
2456 * records, just get rid of the root and clear the status bit.
2458 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
2459 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2460 new_max = cur_max + rec_diff;
2461 ASSERT(new_max >= 0);
2463 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2467 new_broot = (xfs_bmbt_block_t *)kmem_alloc(new_size, KM_SLEEP);
2469 * First copy over the btree block header.
2471 memcpy(new_broot, ifp->if_broot, sizeof(xfs_bmbt_block_t));
2474 ifp->if_flags &= ~XFS_IFBROOT;
2478 * Only copy the records and pointers if there are any.
2482 * First copy the records.
2484 op = (char *)XFS_BMAP_BROOT_REC_ADDR(ifp->if_broot, 1,
2485 ifp->if_broot_bytes);
2486 np = (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot, 1,
2488 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2491 * Then copy the pointers.
2493 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2494 ifp->if_broot_bytes);
2495 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot, 1,
2497 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2499 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2500 ifp->if_broot = new_broot;
2501 ifp->if_broot_bytes = (int)new_size;
2502 ASSERT(ifp->if_broot_bytes <=
2503 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2509 * This is called when the amount of space needed for if_data
2510 * is increased or decreased. The change in size is indicated by
2511 * the number of bytes that need to be added or deleted in the
2512 * byte_diff parameter.
2514 * If the amount of space needed has decreased below the size of the
2515 * inline buffer, then switch to using the inline buffer. Otherwise,
2516 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2517 * to what is needed.
2519 * ip -- the inode whose if_data area is changing
2520 * byte_diff -- the change in the number of bytes, positive or negative,
2521 * requested for the if_data array.
2533 if (byte_diff == 0) {
2537 ifp = XFS_IFORK_PTR(ip, whichfork);
2538 new_size = (int)ifp->if_bytes + byte_diff;
2539 ASSERT(new_size >= 0);
2541 if (new_size == 0) {
2542 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2543 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2545 ifp->if_u1.if_data = NULL;
2547 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2549 * If the valid extents/data can fit in if_inline_ext/data,
2550 * copy them from the malloc'd vector and free it.
2552 if (ifp->if_u1.if_data == NULL) {
2553 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2554 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2555 ASSERT(ifp->if_real_bytes != 0);
2556 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2558 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2559 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2564 * Stuck with malloc/realloc.
2565 * For inline data, the underlying buffer must be
2566 * a multiple of 4 bytes in size so that it can be
2567 * logged and stay on word boundaries. We enforce
2570 real_size = roundup(new_size, 4);
2571 if (ifp->if_u1.if_data == NULL) {
2572 ASSERT(ifp->if_real_bytes == 0);
2573 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2574 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2576 * Only do the realloc if the underlying size
2577 * is really changing.
2579 if (ifp->if_real_bytes != real_size) {
2580 ifp->if_u1.if_data =
2581 kmem_realloc(ifp->if_u1.if_data,
2587 ASSERT(ifp->if_real_bytes == 0);
2588 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2589 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2593 ifp->if_real_bytes = real_size;
2594 ifp->if_bytes = new_size;
2595 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2602 * Map inode to disk block and offset.
2604 * mp -- the mount point structure for the current file system
2605 * tp -- the current transaction
2606 * ino -- the inode number of the inode to be located
2607 * imap -- this structure is filled in with the information necessary
2608 * to retrieve the given inode from disk
2609 * flags -- flags to pass to xfs_dilocate indicating whether or not
2610 * lookups in the inode btree were OK or not
2620 xfs_fsblock_t fsbno;
2625 fsbno = imap->im_blkno ?
2626 XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK;
2627 error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags);
2631 imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno);
2632 imap->im_len = XFS_FSB_TO_BB(mp, len);
2633 imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno);
2634 imap->im_ioffset = (ushort)off;
2635 imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog);
2646 ifp = XFS_IFORK_PTR(ip, whichfork);
2647 if (ifp->if_broot != NULL) {
2648 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2649 ifp->if_broot = NULL;
2653 * If the format is local, then we can't have an extents
2654 * array so just look for an inline data array. If we're
2655 * not local then we may or may not have an extents list,
2656 * so check and free it up if we do.
2658 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2659 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2660 (ifp->if_u1.if_data != NULL)) {
2661 ASSERT(ifp->if_real_bytes != 0);
2662 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2663 ifp->if_u1.if_data = NULL;
2664 ifp->if_real_bytes = 0;
2666 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2667 ((ifp->if_flags & XFS_IFEXTIREC) ||
2668 ((ifp->if_u1.if_extents != NULL) &&
2669 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
2670 ASSERT(ifp->if_real_bytes != 0);
2671 xfs_iext_destroy(ifp);
2673 ASSERT(ifp->if_u1.if_extents == NULL ||
2674 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2675 ASSERT(ifp->if_real_bytes == 0);
2676 if (whichfork == XFS_ATTR_FORK) {
2677 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2683 * This is called free all the memory associated with an inode.
2684 * It must free the inode itself and any buffers allocated for
2685 * if_extents/if_data and if_broot. It must also free the lock
2686 * associated with the inode.
2693 switch (ip->i_d.di_mode & S_IFMT) {
2697 xfs_idestroy_fork(ip, XFS_DATA_FORK);
2701 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
2702 mrfree(&ip->i_lock);
2703 mrfree(&ip->i_iolock);
2704 freesema(&ip->i_flock);
2705 #ifdef XFS_BMAP_TRACE
2706 ktrace_free(ip->i_xtrace);
2708 #ifdef XFS_BMBT_TRACE
2709 ktrace_free(ip->i_btrace);
2712 ktrace_free(ip->i_rwtrace);
2714 #ifdef XFS_ILOCK_TRACE
2715 ktrace_free(ip->i_lock_trace);
2717 #ifdef XFS_DIR2_TRACE
2718 ktrace_free(ip->i_dir_trace);
2721 /* XXXdpd should be able to assert this but shutdown
2722 * is leaving the AIL behind. */
2723 ASSERT(((ip->i_itemp->ili_item.li_flags & XFS_LI_IN_AIL) == 0) ||
2724 XFS_FORCED_SHUTDOWN(ip->i_mount));
2725 xfs_inode_item_destroy(ip);
2727 kmem_zone_free(xfs_inode_zone, ip);
2732 * Increment the pin count of the given buffer.
2733 * This value is protected by ipinlock spinlock in the mount structure.
2739 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2741 atomic_inc(&ip->i_pincount);
2745 * Decrement the pin count of the given inode, and wake up
2746 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2747 * inode must have been previously pinned with a call to xfs_ipin().
2753 ASSERT(atomic_read(&ip->i_pincount) > 0);
2755 if (atomic_dec_and_test(&ip->i_pincount)) {
2757 * If the inode is currently being reclaimed, the
2758 * linux inode _and_ the xfs vnode may have been
2759 * freed so we cannot reference either of them safely.
2760 * Hence we should not try to do anything to them
2761 * if the xfs inode is currently in the reclaim
2764 * However, we still need to issue the unpin wakeup
2765 * call as the inode reclaim may be blocked waiting for
2766 * the inode to become unpinned.
2768 if (!(ip->i_flags & (XFS_IRECLAIM|XFS_IRECLAIMABLE))) {
2769 vnode_t *vp = XFS_ITOV_NULL(ip);
2771 /* make sync come back and flush this inode */
2773 struct inode *inode = vn_to_inode(vp);
2775 if (!(inode->i_state & I_NEW))
2776 mark_inode_dirty_sync(inode);
2779 wake_up(&ip->i_ipin_wait);
2784 * This is called to wait for the given inode to be unpinned.
2785 * It will sleep until this happens. The caller must have the
2786 * inode locked in at least shared mode so that the buffer cannot
2787 * be subsequently pinned once someone is waiting for it to be
2794 xfs_inode_log_item_t *iip;
2797 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE | MR_ACCESS));
2799 if (atomic_read(&ip->i_pincount) == 0) {
2804 if (iip && iip->ili_last_lsn) {
2805 lsn = iip->ili_last_lsn;
2811 * Give the log a push so we don't wait here too long.
2813 xfs_log_force(ip->i_mount, lsn, XFS_LOG_FORCE);
2815 wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
2820 * xfs_iextents_copy()
2822 * This is called to copy the REAL extents (as opposed to the delayed
2823 * allocation extents) from the inode into the given buffer. It
2824 * returns the number of bytes copied into the buffer.
2826 * If there are no delayed allocation extents, then we can just
2827 * memcpy() the extents into the buffer. Otherwise, we need to
2828 * examine each extent in turn and skip those which are delayed.
2833 xfs_bmbt_rec_t *buffer,
2837 xfs_bmbt_rec_t *dest_ep;
2839 #ifdef XFS_BMAP_TRACE
2840 static char fname[] = "xfs_iextents_copy";
2845 xfs_fsblock_t start_block;
2847 ifp = XFS_IFORK_PTR(ip, whichfork);
2848 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
2849 ASSERT(ifp->if_bytes > 0);
2851 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2852 xfs_bmap_trace_exlist(fname, ip, nrecs, whichfork);
2856 * There are some delayed allocation extents in the
2857 * inode, so copy the extents one at a time and skip
2858 * the delayed ones. There must be at least one
2859 * non-delayed extent.
2863 for (i = 0; i < nrecs; i++) {
2864 ep = xfs_iext_get_ext(ifp, i);
2865 start_block = xfs_bmbt_get_startblock(ep);
2866 if (ISNULLSTARTBLOCK(start_block)) {
2868 * It's a delayed allocation extent, so skip it.
2873 /* Translate to on disk format */
2874 put_unaligned(INT_GET(ep->l0, ARCH_CONVERT),
2875 (__uint64_t*)&dest_ep->l0);
2876 put_unaligned(INT_GET(ep->l1, ARCH_CONVERT),
2877 (__uint64_t*)&dest_ep->l1);
2881 ASSERT(copied != 0);
2882 xfs_validate_extents(ifp, copied, 1, XFS_EXTFMT_INODE(ip));
2884 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2888 * Each of the following cases stores data into the same region
2889 * of the on-disk inode, so only one of them can be valid at
2890 * any given time. While it is possible to have conflicting formats
2891 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2892 * in EXTENTS format, this can only happen when the fork has
2893 * changed formats after being modified but before being flushed.
2894 * In these cases, the format always takes precedence, because the
2895 * format indicates the current state of the fork.
2902 xfs_inode_log_item_t *iip,
2909 #ifdef XFS_TRANS_DEBUG
2912 static const short brootflag[2] =
2913 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2914 static const short dataflag[2] =
2915 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2916 static const short extflag[2] =
2917 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2921 ifp = XFS_IFORK_PTR(ip, whichfork);
2923 * This can happen if we gave up in iformat in an error path,
2924 * for the attribute fork.
2927 ASSERT(whichfork == XFS_ATTR_FORK);
2930 cp = XFS_DFORK_PTR(dip, whichfork);
2932 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2933 case XFS_DINODE_FMT_LOCAL:
2934 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2935 (ifp->if_bytes > 0)) {
2936 ASSERT(ifp->if_u1.if_data != NULL);
2937 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2938 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2940 if (whichfork == XFS_DATA_FORK) {
2941 if (unlikely(XFS_DIR_SHORTFORM_VALIDATE_ONDISK(mp, dip))) {
2942 XFS_ERROR_REPORT("xfs_iflush_fork",
2943 XFS_ERRLEVEL_LOW, mp);
2944 return XFS_ERROR(EFSCORRUPTED);
2949 case XFS_DINODE_FMT_EXTENTS:
2950 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2951 !(iip->ili_format.ilf_fields & extflag[whichfork]));
2952 ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
2953 (ifp->if_bytes == 0));
2954 ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
2955 (ifp->if_bytes > 0));
2956 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
2957 (ifp->if_bytes > 0)) {
2958 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2959 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2964 case XFS_DINODE_FMT_BTREE:
2965 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
2966 (ifp->if_broot_bytes > 0)) {
2967 ASSERT(ifp->if_broot != NULL);
2968 ASSERT(ifp->if_broot_bytes <=
2969 (XFS_IFORK_SIZE(ip, whichfork) +
2970 XFS_BROOT_SIZE_ADJ));
2971 xfs_bmbt_to_bmdr(ifp->if_broot, ifp->if_broot_bytes,
2972 (xfs_bmdr_block_t *)cp,
2973 XFS_DFORK_SIZE(dip, mp, whichfork));
2977 case XFS_DINODE_FMT_DEV:
2978 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
2979 ASSERT(whichfork == XFS_DATA_FORK);
2980 INT_SET(dip->di_u.di_dev, ARCH_CONVERT, ip->i_df.if_u2.if_rdev);
2984 case XFS_DINODE_FMT_UUID:
2985 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
2986 ASSERT(whichfork == XFS_DATA_FORK);
2987 memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid,
3001 * xfs_iflush() will write a modified inode's changes out to the
3002 * inode's on disk home. The caller must have the inode lock held
3003 * in at least shared mode and the inode flush semaphore must be
3004 * held as well. The inode lock will still be held upon return from
3005 * the call and the caller is free to unlock it.
3006 * The inode flush lock will be unlocked when the inode reaches the disk.
3007 * The flags indicate how the inode's buffer should be written out.
3014 xfs_inode_log_item_t *iip;
3022 int clcount; /* count of inodes clustered */
3024 enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
3027 XFS_STATS_INC(xs_iflush_count);
3029 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
3030 ASSERT(valusema(&ip->i_flock) <= 0);
3031 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3032 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3038 * If the inode isn't dirty, then just release the inode
3039 * flush lock and do nothing.
3041 if ((ip->i_update_core == 0) &&
3042 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3043 ASSERT((iip != NULL) ?
3044 !(iip->ili_item.li_flags & XFS_LI_IN_AIL) : 1);
3050 * We can't flush the inode until it is unpinned, so
3051 * wait for it. We know noone new can pin it, because
3052 * we are holding the inode lock shared and you need
3053 * to hold it exclusively to pin the inode.
3055 xfs_iunpin_wait(ip);
3058 * This may have been unpinned because the filesystem is shutting
3059 * down forcibly. If that's the case we must not write this inode
3060 * to disk, because the log record didn't make it to disk!
3062 if (XFS_FORCED_SHUTDOWN(mp)) {
3063 ip->i_update_core = 0;
3065 iip->ili_format.ilf_fields = 0;
3067 return XFS_ERROR(EIO);
3071 * Get the buffer containing the on-disk inode.
3073 error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0, 0);
3080 * Decide how buffer will be flushed out. This is done before
3081 * the call to xfs_iflush_int because this field is zeroed by it.
3083 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3085 * Flush out the inode buffer according to the directions
3086 * of the caller. In the cases where the caller has given
3087 * us a choice choose the non-delwri case. This is because
3088 * the inode is in the AIL and we need to get it out soon.
3091 case XFS_IFLUSH_SYNC:
3092 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3095 case XFS_IFLUSH_ASYNC:
3096 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3099 case XFS_IFLUSH_DELWRI:
3109 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3110 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3111 case XFS_IFLUSH_DELWRI:
3114 case XFS_IFLUSH_ASYNC:
3117 case XFS_IFLUSH_SYNC:
3128 * First flush out the inode that xfs_iflush was called with.
3130 error = xfs_iflush_int(ip, bp);
3137 * see if other inodes can be gathered into this write
3140 ip->i_chash->chl_buf = bp;
3142 ch = XFS_CHASH(mp, ip->i_blkno);
3143 s = mutex_spinlock(&ch->ch_lock);
3146 for (iq = ip->i_cnext; iq != ip; iq = iq->i_cnext) {
3148 * Do an un-protected check to see if the inode is dirty and
3149 * is a candidate for flushing. These checks will be repeated
3150 * later after the appropriate locks are acquired.
3153 if ((iq->i_update_core == 0) &&
3155 !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
3156 xfs_ipincount(iq) == 0) {
3161 * Try to get locks. If any are unavailable,
3162 * then this inode cannot be flushed and is skipped.
3165 /* get inode locks (just i_lock) */
3166 if (xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) {
3167 /* get inode flush lock */
3168 if (xfs_iflock_nowait(iq)) {
3169 /* check if pinned */
3170 if (xfs_ipincount(iq) == 0) {
3171 /* arriving here means that
3172 * this inode can be flushed.
3173 * first re-check that it's
3177 if ((iq->i_update_core != 0)||
3179 (iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3181 error = xfs_iflush_int(iq, bp);
3185 goto cluster_corrupt_out;
3194 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3197 mutex_spinunlock(&ch->ch_lock, s);
3200 XFS_STATS_INC(xs_icluster_flushcnt);
3201 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
3205 * If the buffer is pinned then push on the log so we won't
3206 * get stuck waiting in the write for too long.
3208 if (XFS_BUF_ISPINNED(bp)){
3209 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
3212 if (flags & INT_DELWRI) {
3213 xfs_bdwrite(mp, bp);
3214 } else if (flags & INT_ASYNC) {
3215 xfs_bawrite(mp, bp);
3217 error = xfs_bwrite(mp, bp);
3223 xfs_force_shutdown(mp, XFS_CORRUPT_INCORE);
3224 xfs_iflush_abort(ip);
3226 * Unlocks the flush lock
3228 return XFS_ERROR(EFSCORRUPTED);
3230 cluster_corrupt_out:
3231 /* Corruption detected in the clustering loop. Invalidate the
3232 * inode buffer and shut down the filesystem.
3234 mutex_spinunlock(&ch->ch_lock, s);
3237 * Clean up the buffer. If it was B_DELWRI, just release it --
3238 * brelse can handle it with no problems. If not, shut down the
3239 * filesystem before releasing the buffer.
3241 if ((bufwasdelwri= XFS_BUF_ISDELAYWRITE(bp))) {
3245 xfs_force_shutdown(mp, XFS_CORRUPT_INCORE);
3249 * Just like incore_relse: if we have b_iodone functions,
3250 * mark the buffer as an error and call them. Otherwise
3251 * mark it as stale and brelse.
3253 if (XFS_BUF_IODONE_FUNC(bp)) {
3254 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
3258 XFS_BUF_ERROR(bp,EIO);
3266 xfs_iflush_abort(iq);
3268 * Unlocks the flush lock
3270 return XFS_ERROR(EFSCORRUPTED);
3279 xfs_inode_log_item_t *iip;
3282 #ifdef XFS_TRANS_DEBUG
3287 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
3288 ASSERT(valusema(&ip->i_flock) <= 0);
3289 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3290 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3297 * If the inode isn't dirty, then just release the inode
3298 * flush lock and do nothing.
3300 if ((ip->i_update_core == 0) &&
3301 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3306 /* set *dip = inode's place in the buffer */
3307 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset);
3310 * Clear i_update_core before copying out the data.
3311 * This is for coordination with our timestamp updates
3312 * that don't hold the inode lock. They will always
3313 * update the timestamps BEFORE setting i_update_core,
3314 * so if we clear i_update_core after they set it we
3315 * are guaranteed to see their updates to the timestamps.
3316 * I believe that this depends on strongly ordered memory
3317 * semantics, but we have that. We use the SYNCHRONIZE
3318 * macro to make sure that the compiler does not reorder
3319 * the i_update_core access below the data copy below.
3321 ip->i_update_core = 0;
3325 * Make sure to get the latest atime from the Linux inode.
3327 xfs_synchronize_atime(ip);
3329 if (XFS_TEST_ERROR(INT_GET(dip->di_core.di_magic,ARCH_CONVERT) != XFS_DINODE_MAGIC,
3330 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3331 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3332 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3333 ip->i_ino, (int) INT_GET(dip->di_core.di_magic, ARCH_CONVERT), dip);
3336 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3337 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3338 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3339 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3340 ip->i_ino, ip, ip->i_d.di_magic);
3343 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3345 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3346 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3347 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3348 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3349 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3353 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
3355 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3356 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3357 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3358 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3359 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3360 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3365 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3366 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3367 XFS_RANDOM_IFLUSH_5)) {
3368 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3369 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3371 ip->i_d.di_nextents + ip->i_d.di_anextents,
3376 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3377 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3378 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3379 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3380 ip->i_ino, ip->i_d.di_forkoff, ip);
3384 * bump the flush iteration count, used to detect flushes which
3385 * postdate a log record during recovery.
3388 ip->i_d.di_flushiter++;
3391 * Copy the dirty parts of the inode into the on-disk
3392 * inode. We always copy out the core of the inode,
3393 * because if the inode is dirty at all the core must
3396 xfs_xlate_dinode_core((xfs_caddr_t)&(dip->di_core), &(ip->i_d), -1);
3398 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3399 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3400 ip->i_d.di_flushiter = 0;
3403 * If this is really an old format inode and the superblock version
3404 * has not been updated to support only new format inodes, then
3405 * convert back to the old inode format. If the superblock version
3406 * has been updated, then make the conversion permanent.
3408 ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 ||
3409 XFS_SB_VERSION_HASNLINK(&mp->m_sb));
3410 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
3411 if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) {
3415 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3416 INT_SET(dip->di_core.di_onlink, ARCH_CONVERT, ip->i_d.di_nlink);
3419 * The superblock version has already been bumped,
3420 * so just make the conversion to the new inode
3423 ip->i_d.di_version = XFS_DINODE_VERSION_2;
3424 INT_SET(dip->di_core.di_version, ARCH_CONVERT, XFS_DINODE_VERSION_2);
3425 ip->i_d.di_onlink = 0;
3426 dip->di_core.di_onlink = 0;
3427 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3428 memset(&(dip->di_core.di_pad[0]), 0,
3429 sizeof(dip->di_core.di_pad));
3430 ASSERT(ip->i_d.di_projid == 0);
3434 if (xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp) == EFSCORRUPTED) {
3438 if (XFS_IFORK_Q(ip)) {
3440 * The only error from xfs_iflush_fork is on the data fork.
3442 (void) xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3444 xfs_inobp_check(mp, bp);
3447 * We've recorded everything logged in the inode, so we'd
3448 * like to clear the ilf_fields bits so we don't log and
3449 * flush things unnecessarily. However, we can't stop
3450 * logging all this information until the data we've copied
3451 * into the disk buffer is written to disk. If we did we might
3452 * overwrite the copy of the inode in the log with all the
3453 * data after re-logging only part of it, and in the face of
3454 * a crash we wouldn't have all the data we need to recover.
3456 * What we do is move the bits to the ili_last_fields field.
3457 * When logging the inode, these bits are moved back to the
3458 * ilf_fields field. In the xfs_iflush_done() routine we
3459 * clear ili_last_fields, since we know that the information
3460 * those bits represent is permanently on disk. As long as
3461 * the flush completes before the inode is logged again, then
3462 * both ilf_fields and ili_last_fields will be cleared.
3464 * We can play with the ilf_fields bits here, because the inode
3465 * lock must be held exclusively in order to set bits there
3466 * and the flush lock protects the ili_last_fields bits.
3467 * Set ili_logged so the flush done
3468 * routine can tell whether or not to look in the AIL.
3469 * Also, store the current LSN of the inode so that we can tell
3470 * whether the item has moved in the AIL from xfs_iflush_done().
3471 * In order to read the lsn we need the AIL lock, because
3472 * it is a 64 bit value that cannot be read atomically.
3474 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3475 iip->ili_last_fields = iip->ili_format.ilf_fields;
3476 iip->ili_format.ilf_fields = 0;
3477 iip->ili_logged = 1;
3479 ASSERT(sizeof(xfs_lsn_t) == 8); /* don't lock if it shrinks */
3481 iip->ili_flush_lsn = iip->ili_item.li_lsn;
3485 * Attach the function xfs_iflush_done to the inode's
3486 * buffer. This will remove the inode from the AIL
3487 * and unlock the inode's flush lock when the inode is
3488 * completely written to disk.
3490 xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
3491 xfs_iflush_done, (xfs_log_item_t *)iip);
3493 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3494 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3497 * We're flushing an inode which is not in the AIL and has
3498 * not been logged but has i_update_core set. For this
3499 * case we can use a B_DELWRI flush and immediately drop
3500 * the inode flush lock because we can avoid the whole
3501 * AIL state thing. It's OK to drop the flush lock now,
3502 * because we've already locked the buffer and to do anything
3503 * you really need both.
3506 ASSERT(iip->ili_logged == 0);
3507 ASSERT(iip->ili_last_fields == 0);
3508 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3516 return XFS_ERROR(EFSCORRUPTED);
3521 * Flush all inactive inodes in mp.
3531 XFS_MOUNT_ILOCK(mp);
3537 /* Make sure we skip markers inserted by sync */
3538 if (ip->i_mount == NULL) {
3543 vp = XFS_ITOV_NULL(ip);
3545 XFS_MOUNT_IUNLOCK(mp);
3546 xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC);
3550 ASSERT(vn_count(vp) == 0);
3553 } while (ip != mp->m_inodes);
3555 XFS_MOUNT_IUNLOCK(mp);
3559 * xfs_iaccess: check accessibility of inode for mode.
3568 mode_t orgmode = mode;
3569 struct inode *inode = vn_to_inode(XFS_ITOV(ip));
3571 if (mode & S_IWUSR) {
3572 umode_t imode = inode->i_mode;
3574 if (IS_RDONLY(inode) &&
3575 (S_ISREG(imode) || S_ISDIR(imode) || S_ISLNK(imode)))
3576 return XFS_ERROR(EROFS);
3578 if (IS_IMMUTABLE(inode))
3579 return XFS_ERROR(EACCES);
3583 * If there's an Access Control List it's used instead of
3586 if ((error = _ACL_XFS_IACCESS(ip, mode, cr)) != -1)
3587 return error ? XFS_ERROR(error) : 0;
3589 if (current_fsuid(cr) != ip->i_d.di_uid) {
3591 if (!in_group_p((gid_t)ip->i_d.di_gid))
3596 * If the DACs are ok we don't need any capability check.
3598 if ((ip->i_d.di_mode & mode) == mode)
3601 * Read/write DACs are always overridable.
3602 * Executable DACs are overridable if at least one exec bit is set.
3604 if (!(orgmode & S_IXUSR) ||
3605 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
3606 if (capable_cred(cr, CAP_DAC_OVERRIDE))
3609 if ((orgmode == S_IRUSR) ||
3610 (S_ISDIR(inode->i_mode) && (!(orgmode & S_IWUSR)))) {
3611 if (capable_cred(cr, CAP_DAC_READ_SEARCH))
3614 cmn_err(CE_NOTE, "Ick: mode=%o, orgmode=%o", mode, orgmode);
3616 return XFS_ERROR(EACCES);
3618 return XFS_ERROR(EACCES);
3622 * xfs_iroundup: round up argument to next power of two
3631 if ((v & (v - 1)) == 0)
3633 ASSERT((v & 0x80000000) == 0);
3634 if ((v & (v + 1)) == 0)
3636 for (i = 0, m = 1; i < 31; i++, m <<= 1) {
3640 if ((v & (v + 1)) == 0)
3647 #ifdef XFS_ILOCK_TRACE
3648 ktrace_t *xfs_ilock_trace_buf;
3651 xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
3653 ktrace_enter(ip->i_lock_trace,
3655 (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
3656 (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
3657 (void *)ra, /* caller of ilock */
3658 (void *)(unsigned long)current_cpu(),
3659 (void *)(unsigned long)current_pid(),
3660 NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL);
3665 * Return a pointer to the extent record at file index idx.
3669 xfs_ifork_t *ifp, /* inode fork pointer */
3670 xfs_extnum_t idx) /* index of target extent */
3673 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
3674 return ifp->if_u1.if_ext_irec->er_extbuf;
3675 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3676 xfs_ext_irec_t *erp; /* irec pointer */
3677 int erp_idx = 0; /* irec index */
3678 xfs_extnum_t page_idx = idx; /* ext index in target list */
3680 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3681 return &erp->er_extbuf[page_idx];
3682 } else if (ifp->if_bytes) {
3683 return &ifp->if_u1.if_extents[idx];
3690 * Insert new item(s) into the extent records for incore inode
3691 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3695 xfs_ifork_t *ifp, /* inode fork pointer */
3696 xfs_extnum_t idx, /* starting index of new items */
3697 xfs_extnum_t count, /* number of inserted items */
3698 xfs_bmbt_irec_t *new) /* items to insert */
3700 xfs_bmbt_rec_t *ep; /* extent record pointer */
3701 xfs_extnum_t i; /* extent record index */
3703 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3704 xfs_iext_add(ifp, idx, count);
3705 for (i = idx; i < idx + count; i++, new++) {
3706 ep = xfs_iext_get_ext(ifp, i);
3707 xfs_bmbt_set_all(ep, new);
3712 * This is called when the amount of space required for incore file
3713 * extents needs to be increased. The ext_diff parameter stores the
3714 * number of new extents being added and the idx parameter contains
3715 * the extent index where the new extents will be added. If the new
3716 * extents are being appended, then we just need to (re)allocate and
3717 * initialize the space. Otherwise, if the new extents are being
3718 * inserted into the middle of the existing entries, a bit more work
3719 * is required to make room for the new extents to be inserted. The
3720 * caller is responsible for filling in the new extent entries upon
3725 xfs_ifork_t *ifp, /* inode fork pointer */
3726 xfs_extnum_t idx, /* index to begin adding exts */
3727 int ext_diff) /* number of extents to add */
3729 int byte_diff; /* new bytes being added */
3730 int new_size; /* size of extents after adding */
3731 xfs_extnum_t nextents; /* number of extents in file */
3733 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3734 ASSERT((idx >= 0) && (idx <= nextents));
3735 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
3736 new_size = ifp->if_bytes + byte_diff;
3738 * If the new number of extents (nextents + ext_diff)
3739 * fits inside the inode, then continue to use the inline
3742 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
3743 if (idx < nextents) {
3744 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
3745 &ifp->if_u2.if_inline_ext[idx],
3746 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3747 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
3749 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3750 ifp->if_real_bytes = 0;
3751 ifp->if_lastex = nextents + ext_diff;
3754 * Otherwise use a linear (direct) extent list.
3755 * If the extents are currently inside the inode,
3756 * xfs_iext_realloc_direct will switch us from
3757 * inline to direct extent allocation mode.
3759 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
3760 xfs_iext_realloc_direct(ifp, new_size);
3761 if (idx < nextents) {
3762 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
3763 &ifp->if_u1.if_extents[idx],
3764 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3765 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
3768 /* Indirection array */
3770 xfs_ext_irec_t *erp;
3774 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
3775 if (ifp->if_flags & XFS_IFEXTIREC) {
3776 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3778 xfs_iext_irec_init(ifp);
3779 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3780 erp = ifp->if_u1.if_ext_irec;
3782 /* Extents fit in target extent page */
3783 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3784 if (page_idx < erp->er_extcount) {
3785 memmove(&erp->er_extbuf[page_idx + ext_diff],
3786 &erp->er_extbuf[page_idx],
3787 (erp->er_extcount - page_idx) *
3788 sizeof(xfs_bmbt_rec_t));
3789 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3791 erp->er_extcount += ext_diff;
3792 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3794 /* Insert a new extent page */
3796 xfs_iext_add_indirect_multi(ifp,
3797 erp_idx, page_idx, ext_diff);
3800 * If extent(s) are being appended to the last page in
3801 * the indirection array and the new extent(s) don't fit
3802 * in the page, then erp is NULL and erp_idx is set to
3803 * the next index needed in the indirection array.
3806 int count = ext_diff;
3809 erp = xfs_iext_irec_new(ifp, erp_idx);
3810 erp->er_extcount = count;
3811 count -= MIN(count, (int)XFS_LINEAR_EXTS);
3818 ifp->if_bytes = new_size;
3822 * This is called when incore extents are being added to the indirection
3823 * array and the new extents do not fit in the target extent list. The
3824 * erp_idx parameter contains the irec index for the target extent list
3825 * in the indirection array, and the idx parameter contains the extent
3826 * index within the list. The number of extents being added is stored
3827 * in the count parameter.
3829 * |-------| |-------|
3830 * | | | | idx - number of extents before idx
3832 * | | | | count - number of extents being inserted at idx
3833 * |-------| |-------|
3834 * | count | | nex2 | nex2 - number of extents after idx + count
3835 * |-------| |-------|
3838 xfs_iext_add_indirect_multi(
3839 xfs_ifork_t *ifp, /* inode fork pointer */
3840 int erp_idx, /* target extent irec index */
3841 xfs_extnum_t idx, /* index within target list */
3842 int count) /* new extents being added */
3844 int byte_diff; /* new bytes being added */
3845 xfs_ext_irec_t *erp; /* pointer to irec entry */
3846 xfs_extnum_t ext_diff; /* number of extents to add */
3847 xfs_extnum_t ext_cnt; /* new extents still needed */
3848 xfs_extnum_t nex2; /* extents after idx + count */
3849 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3850 int nlists; /* number of irec's (lists) */
3852 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3853 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3854 nex2 = erp->er_extcount - idx;
3855 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3858 * Save second part of target extent list
3859 * (all extents past */
3861 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3862 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_SLEEP);
3863 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3864 erp->er_extcount -= nex2;
3865 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3866 memset(&erp->er_extbuf[idx], 0, byte_diff);
3870 * Add the new extents to the end of the target
3871 * list, then allocate new irec record(s) and
3872 * extent buffer(s) as needed to store the rest
3873 * of the new extents.
3876 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3878 erp->er_extcount += ext_diff;
3879 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3880 ext_cnt -= ext_diff;
3884 erp = xfs_iext_irec_new(ifp, erp_idx);
3885 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3886 erp->er_extcount = ext_diff;
3887 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3888 ext_cnt -= ext_diff;
3891 /* Add nex2 extents back to indirection array */
3893 xfs_extnum_t ext_avail;
3896 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3897 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3900 * If nex2 extents fit in the current page, append
3901 * nex2_ep after the new extents.
3903 if (nex2 <= ext_avail) {
3904 i = erp->er_extcount;
3907 * Otherwise, check if space is available in the
3910 else if ((erp_idx < nlists - 1) &&
3911 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3912 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3915 /* Create a hole for nex2 extents */
3916 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3917 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3920 * Final choice, create a new extent page for
3925 erp = xfs_iext_irec_new(ifp, erp_idx);
3927 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
3928 kmem_free(nex2_ep, byte_diff);
3929 erp->er_extcount += nex2;
3930 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3935 * This is called when the amount of space required for incore file
3936 * extents needs to be decreased. The ext_diff parameter stores the
3937 * number of extents to be removed and the idx parameter contains
3938 * the extent index where the extents will be removed from.
3940 * If the amount of space needed has decreased below the linear
3941 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3942 * extent array. Otherwise, use kmem_realloc() to adjust the
3943 * size to what is needed.
3947 xfs_ifork_t *ifp, /* inode fork pointer */
3948 xfs_extnum_t idx, /* index to begin removing exts */
3949 int ext_diff) /* number of extents to remove */
3951 xfs_extnum_t nextents; /* number of extents in file */
3952 int new_size; /* size of extents after removal */
3954 ASSERT(ext_diff > 0);
3955 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3956 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3958 if (new_size == 0) {
3959 xfs_iext_destroy(ifp);
3960 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3961 xfs_iext_remove_indirect(ifp, idx, ext_diff);
3962 } else if (ifp->if_real_bytes) {
3963 xfs_iext_remove_direct(ifp, idx, ext_diff);
3965 xfs_iext_remove_inline(ifp, idx, ext_diff);
3967 ifp->if_bytes = new_size;
3971 * This removes ext_diff extents from the inline buffer, beginning
3972 * at extent index idx.
3975 xfs_iext_remove_inline(
3976 xfs_ifork_t *ifp, /* inode fork pointer */
3977 xfs_extnum_t idx, /* index to begin removing exts */
3978 int ext_diff) /* number of extents to remove */
3980 int nextents; /* number of extents in file */
3982 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3983 ASSERT(idx < XFS_INLINE_EXTS);
3984 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3985 ASSERT(((nextents - ext_diff) > 0) &&
3986 (nextents - ext_diff) < XFS_INLINE_EXTS);
3988 if (idx + ext_diff < nextents) {
3989 memmove(&ifp->if_u2.if_inline_ext[idx],
3990 &ifp->if_u2.if_inline_ext[idx + ext_diff],
3991 (nextents - (idx + ext_diff)) *
3992 sizeof(xfs_bmbt_rec_t));
3993 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
3994 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3996 memset(&ifp->if_u2.if_inline_ext[idx], 0,
3997 ext_diff * sizeof(xfs_bmbt_rec_t));
4002 * This removes ext_diff extents from a linear (direct) extent list,
4003 * beginning at extent index idx. If the extents are being removed
4004 * from the end of the list (ie. truncate) then we just need to re-
4005 * allocate the list to remove the extra space. Otherwise, if the
4006 * extents are being removed from the middle of the existing extent
4007 * entries, then we first need to move the extent records beginning
4008 * at idx + ext_diff up in the list to overwrite the records being
4009 * removed, then remove the extra space via kmem_realloc.
4012 xfs_iext_remove_direct(
4013 xfs_ifork_t *ifp, /* inode fork pointer */
4014 xfs_extnum_t idx, /* index to begin removing exts */
4015 int ext_diff) /* number of extents to remove */
4017 xfs_extnum_t nextents; /* number of extents in file */
4018 int new_size; /* size of extents after removal */
4020 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4021 new_size = ifp->if_bytes -
4022 (ext_diff * sizeof(xfs_bmbt_rec_t));
4023 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4025 if (new_size == 0) {
4026 xfs_iext_destroy(ifp);
4029 /* Move extents up in the list (if needed) */
4030 if (idx + ext_diff < nextents) {
4031 memmove(&ifp->if_u1.if_extents[idx],
4032 &ifp->if_u1.if_extents[idx + ext_diff],
4033 (nextents - (idx + ext_diff)) *
4034 sizeof(xfs_bmbt_rec_t));
4036 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
4037 0, ext_diff * sizeof(xfs_bmbt_rec_t));
4039 * Reallocate the direct extent list. If the extents
4040 * will fit inside the inode then xfs_iext_realloc_direct
4041 * will switch from direct to inline extent allocation
4044 xfs_iext_realloc_direct(ifp, new_size);
4045 ifp->if_bytes = new_size;
4049 * This is called when incore extents are being removed from the
4050 * indirection array and the extents being removed span multiple extent
4051 * buffers. The idx parameter contains the file extent index where we
4052 * want to begin removing extents, and the count parameter contains
4053 * how many extents need to be removed.
4055 * |-------| |-------|
4056 * | nex1 | | | nex1 - number of extents before idx
4057 * |-------| | count |
4058 * | | | | count - number of extents being removed at idx
4059 * | count | |-------|
4060 * | | | nex2 | nex2 - number of extents after idx + count
4061 * |-------| |-------|
4064 xfs_iext_remove_indirect(
4065 xfs_ifork_t *ifp, /* inode fork pointer */
4066 xfs_extnum_t idx, /* index to begin removing extents */
4067 int count) /* number of extents to remove */
4069 xfs_ext_irec_t *erp; /* indirection array pointer */
4070 int erp_idx = 0; /* indirection array index */
4071 xfs_extnum_t ext_cnt; /* extents left to remove */
4072 xfs_extnum_t ext_diff; /* extents to remove in current list */
4073 xfs_extnum_t nex1; /* number of extents before idx */
4074 xfs_extnum_t nex2; /* extents after idx + count */
4075 int nlists; /* entries in indirection array */
4076 int page_idx = idx; /* index in target extent list */
4078 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4079 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
4080 ASSERT(erp != NULL);
4081 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4085 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
4086 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
4088 * Check for deletion of entire list;
4089 * xfs_iext_irec_remove() updates extent offsets.
4091 if (ext_diff == erp->er_extcount) {
4092 xfs_iext_irec_remove(ifp, erp_idx);
4093 ext_cnt -= ext_diff;
4096 ASSERT(erp_idx < ifp->if_real_bytes /
4098 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4105 /* Move extents up (if needed) */
4107 memmove(&erp->er_extbuf[nex1],
4108 &erp->er_extbuf[nex1 + ext_diff],
4109 nex2 * sizeof(xfs_bmbt_rec_t));
4111 /* Zero out rest of page */
4112 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
4113 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
4114 /* Update remaining counters */
4115 erp->er_extcount -= ext_diff;
4116 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
4117 ext_cnt -= ext_diff;
4122 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
4123 xfs_iext_irec_compact(ifp);
4127 * Create, destroy, or resize a linear (direct) block of extents.
4130 xfs_iext_realloc_direct(
4131 xfs_ifork_t *ifp, /* inode fork pointer */
4132 int new_size) /* new size of extents */
4134 int rnew_size; /* real new size of extents */
4136 rnew_size = new_size;
4138 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
4139 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
4140 (new_size != ifp->if_real_bytes)));
4142 /* Free extent records */
4143 if (new_size == 0) {
4144 xfs_iext_destroy(ifp);
4146 /* Resize direct extent list and zero any new bytes */
4147 else if (ifp->if_real_bytes) {
4148 /* Check if extents will fit inside the inode */
4149 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
4150 xfs_iext_direct_to_inline(ifp, new_size /
4151 (uint)sizeof(xfs_bmbt_rec_t));
4152 ifp->if_bytes = new_size;
4155 if ((new_size & (new_size - 1)) != 0) {
4156 rnew_size = xfs_iroundup(new_size);
4158 if (rnew_size != ifp->if_real_bytes) {
4159 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
4160 kmem_realloc(ifp->if_u1.if_extents,
4165 if (rnew_size > ifp->if_real_bytes) {
4166 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
4167 (uint)sizeof(xfs_bmbt_rec_t)], 0,
4168 rnew_size - ifp->if_real_bytes);
4172 * Switch from the inline extent buffer to a direct
4173 * extent list. Be sure to include the inline extent
4174 * bytes in new_size.
4177 new_size += ifp->if_bytes;
4178 if ((new_size & (new_size - 1)) != 0) {
4179 rnew_size = xfs_iroundup(new_size);
4181 xfs_iext_inline_to_direct(ifp, rnew_size);
4183 ifp->if_real_bytes = rnew_size;
4184 ifp->if_bytes = new_size;
4188 * Switch from linear (direct) extent records to inline buffer.
4191 xfs_iext_direct_to_inline(
4192 xfs_ifork_t *ifp, /* inode fork pointer */
4193 xfs_extnum_t nextents) /* number of extents in file */
4195 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4196 ASSERT(nextents <= XFS_INLINE_EXTS);
4198 * The inline buffer was zeroed when we switched
4199 * from inline to direct extent allocation mode,
4200 * so we don't need to clear it here.
4202 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
4203 nextents * sizeof(xfs_bmbt_rec_t));
4204 kmem_free(ifp->if_u1.if_extents, KM_SLEEP);
4205 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
4206 ifp->if_real_bytes = 0;
4210 * Switch from inline buffer to linear (direct) extent records.
4211 * new_size should already be rounded up to the next power of 2
4212 * by the caller (when appropriate), so use new_size as it is.
4213 * However, since new_size may be rounded up, we can't update
4214 * if_bytes here. It is the caller's responsibility to update
4215 * if_bytes upon return.
4218 xfs_iext_inline_to_direct(
4219 xfs_ifork_t *ifp, /* inode fork pointer */
4220 int new_size) /* number of extents in file */
4222 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
4223 kmem_alloc(new_size, KM_SLEEP);
4224 memset(ifp->if_u1.if_extents, 0, new_size);
4225 if (ifp->if_bytes) {
4226 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
4228 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4229 sizeof(xfs_bmbt_rec_t));
4231 ifp->if_real_bytes = new_size;
4235 * Resize an extent indirection array to new_size bytes.
4238 xfs_iext_realloc_indirect(
4239 xfs_ifork_t *ifp, /* inode fork pointer */
4240 int new_size) /* new indirection array size */
4242 int nlists; /* number of irec's (ex lists) */
4243 int size; /* current indirection array size */
4245 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4246 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4247 size = nlists * sizeof(xfs_ext_irec_t);
4248 ASSERT(ifp->if_real_bytes);
4249 ASSERT((new_size >= 0) && (new_size != size));
4250 if (new_size == 0) {
4251 xfs_iext_destroy(ifp);
4253 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
4254 kmem_realloc(ifp->if_u1.if_ext_irec,
4255 new_size, size, KM_SLEEP);
4260 * Switch from indirection array to linear (direct) extent allocations.
4263 xfs_iext_indirect_to_direct(
4264 xfs_ifork_t *ifp) /* inode fork pointer */
4266 xfs_bmbt_rec_t *ep; /* extent record pointer */
4267 xfs_extnum_t nextents; /* number of extents in file */
4268 int size; /* size of file extents */
4270 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4271 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4272 ASSERT(nextents <= XFS_LINEAR_EXTS);
4273 size = nextents * sizeof(xfs_bmbt_rec_t);
4275 xfs_iext_irec_compact_full(ifp);
4276 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
4278 ep = ifp->if_u1.if_ext_irec->er_extbuf;
4279 kmem_free(ifp->if_u1.if_ext_irec, sizeof(xfs_ext_irec_t));
4280 ifp->if_flags &= ~XFS_IFEXTIREC;
4281 ifp->if_u1.if_extents = ep;
4282 ifp->if_bytes = size;
4283 if (nextents < XFS_LINEAR_EXTS) {
4284 xfs_iext_realloc_direct(ifp, size);
4289 * Free incore file extents.
4293 xfs_ifork_t *ifp) /* inode fork pointer */
4295 if (ifp->if_flags & XFS_IFEXTIREC) {
4299 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4300 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
4301 xfs_iext_irec_remove(ifp, erp_idx);
4303 ifp->if_flags &= ~XFS_IFEXTIREC;
4304 } else if (ifp->if_real_bytes) {
4305 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
4306 } else if (ifp->if_bytes) {
4307 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4308 sizeof(xfs_bmbt_rec_t));
4310 ifp->if_u1.if_extents = NULL;
4311 ifp->if_real_bytes = 0;
4316 * Return a pointer to the extent record for file system block bno.
4318 xfs_bmbt_rec_t * /* pointer to found extent record */
4319 xfs_iext_bno_to_ext(
4320 xfs_ifork_t *ifp, /* inode fork pointer */
4321 xfs_fileoff_t bno, /* block number to search for */
4322 xfs_extnum_t *idxp) /* index of target extent */
4324 xfs_bmbt_rec_t *base; /* pointer to first extent */
4325 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
4326 xfs_bmbt_rec_t *ep = NULL; /* pointer to target extent */
4327 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4328 int high; /* upper boundary in search */
4329 xfs_extnum_t idx = 0; /* index of target extent */
4330 int low; /* lower boundary in search */
4331 xfs_extnum_t nextents; /* number of file extents */
4332 xfs_fileoff_t startoff = 0; /* start offset of extent */
4334 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4335 if (nextents == 0) {
4340 if (ifp->if_flags & XFS_IFEXTIREC) {
4341 /* Find target extent list */
4343 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
4344 base = erp->er_extbuf;
4345 high = erp->er_extcount - 1;
4347 base = ifp->if_u1.if_extents;
4348 high = nextents - 1;
4350 /* Binary search extent records */
4351 while (low <= high) {
4352 idx = (low + high) >> 1;
4354 startoff = xfs_bmbt_get_startoff(ep);
4355 blockcount = xfs_bmbt_get_blockcount(ep);
4356 if (bno < startoff) {
4358 } else if (bno >= startoff + blockcount) {
4361 /* Convert back to file-based extent index */
4362 if (ifp->if_flags & XFS_IFEXTIREC) {
4363 idx += erp->er_extoff;
4369 /* Convert back to file-based extent index */
4370 if (ifp->if_flags & XFS_IFEXTIREC) {
4371 idx += erp->er_extoff;
4373 if (bno >= startoff + blockcount) {
4374 if (++idx == nextents) {
4377 ep = xfs_iext_get_ext(ifp, idx);
4385 * Return a pointer to the indirection array entry containing the
4386 * extent record for filesystem block bno. Store the index of the
4387 * target irec in *erp_idxp.
4389 xfs_ext_irec_t * /* pointer to found extent record */
4390 xfs_iext_bno_to_irec(
4391 xfs_ifork_t *ifp, /* inode fork pointer */
4392 xfs_fileoff_t bno, /* block number to search for */
4393 int *erp_idxp) /* irec index of target ext list */
4395 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4396 xfs_ext_irec_t *erp_next; /* next indirection array entry */
4397 int erp_idx; /* indirection array index */
4398 int nlists; /* number of extent irec's (lists) */
4399 int high; /* binary search upper limit */
4400 int low; /* binary search lower limit */
4402 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4403 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4407 while (low <= high) {
4408 erp_idx = (low + high) >> 1;
4409 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4410 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
4411 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
4413 } else if (erp_next && bno >=
4414 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
4420 *erp_idxp = erp_idx;
4425 * Return a pointer to the indirection array entry containing the
4426 * extent record at file extent index *idxp. Store the index of the
4427 * target irec in *erp_idxp and store the page index of the target
4428 * extent record in *idxp.
4431 xfs_iext_idx_to_irec(
4432 xfs_ifork_t *ifp, /* inode fork pointer */
4433 xfs_extnum_t *idxp, /* extent index (file -> page) */
4434 int *erp_idxp, /* pointer to target irec */
4435 int realloc) /* new bytes were just added */
4437 xfs_ext_irec_t *prev; /* pointer to previous irec */
4438 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
4439 int erp_idx; /* indirection array index */
4440 int nlists; /* number of irec's (ex lists) */
4441 int high; /* binary search upper limit */
4442 int low; /* binary search lower limit */
4443 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
4445 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4446 ASSERT(page_idx >= 0 && page_idx <=
4447 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
4448 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4453 /* Binary search extent irec's */
4454 while (low <= high) {
4455 erp_idx = (low + high) >> 1;
4456 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4457 prev = erp_idx > 0 ? erp - 1 : NULL;
4458 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
4459 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
4461 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
4462 (page_idx == erp->er_extoff + erp->er_extcount &&
4465 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
4466 erp->er_extcount == XFS_LINEAR_EXTS) {
4470 erp = erp_idx < nlists ? erp + 1 : NULL;
4473 page_idx -= erp->er_extoff;
4478 *erp_idxp = erp_idx;
4483 * Allocate and initialize an indirection array once the space needed
4484 * for incore extents increases above XFS_IEXT_BUFSZ.
4488 xfs_ifork_t *ifp) /* inode fork pointer */
4490 xfs_ext_irec_t *erp; /* indirection array pointer */
4491 xfs_extnum_t nextents; /* number of extents in file */
4493 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4494 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4495 ASSERT(nextents <= XFS_LINEAR_EXTS);
4497 erp = (xfs_ext_irec_t *)
4498 kmem_alloc(sizeof(xfs_ext_irec_t), KM_SLEEP);
4500 if (nextents == 0) {
4501 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
4502 kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
4503 } else if (!ifp->if_real_bytes) {
4504 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
4505 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
4506 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
4508 erp->er_extbuf = ifp->if_u1.if_extents;
4509 erp->er_extcount = nextents;
4512 ifp->if_flags |= XFS_IFEXTIREC;
4513 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
4514 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
4515 ifp->if_u1.if_ext_irec = erp;
4521 * Allocate and initialize a new entry in the indirection array.
4525 xfs_ifork_t *ifp, /* inode fork pointer */
4526 int erp_idx) /* index for new irec */
4528 xfs_ext_irec_t *erp; /* indirection array pointer */
4529 int i; /* loop counter */
4530 int nlists; /* number of irec's (ex lists) */
4532 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4533 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4535 /* Resize indirection array */
4536 xfs_iext_realloc_indirect(ifp, ++nlists *
4537 sizeof(xfs_ext_irec_t));
4539 * Move records down in the array so the
4540 * new page can use erp_idx.
4542 erp = ifp->if_u1.if_ext_irec;
4543 for (i = nlists - 1; i > erp_idx; i--) {
4544 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
4546 ASSERT(i == erp_idx);
4548 /* Initialize new extent record */
4549 erp = ifp->if_u1.if_ext_irec;
4550 erp[erp_idx].er_extbuf = (xfs_bmbt_rec_t *)
4551 kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
4552 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4553 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
4554 erp[erp_idx].er_extcount = 0;
4555 erp[erp_idx].er_extoff = erp_idx > 0 ?
4556 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
4557 return (&erp[erp_idx]);
4561 * Remove a record from the indirection array.
4564 xfs_iext_irec_remove(
4565 xfs_ifork_t *ifp, /* inode fork pointer */
4566 int erp_idx) /* irec index to remove */
4568 xfs_ext_irec_t *erp; /* indirection array pointer */
4569 int i; /* loop counter */
4570 int nlists; /* number of irec's (ex lists) */
4572 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4573 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4574 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4575 if (erp->er_extbuf) {
4576 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
4578 kmem_free(erp->er_extbuf, XFS_IEXT_BUFSZ);
4580 /* Compact extent records */
4581 erp = ifp->if_u1.if_ext_irec;
4582 for (i = erp_idx; i < nlists - 1; i++) {
4583 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
4586 * Manually free the last extent record from the indirection
4587 * array. A call to xfs_iext_realloc_indirect() with a size
4588 * of zero would result in a call to xfs_iext_destroy() which
4589 * would in turn call this function again, creating a nasty
4593 xfs_iext_realloc_indirect(ifp,
4594 nlists * sizeof(xfs_ext_irec_t));
4596 kmem_free(ifp->if_u1.if_ext_irec,
4597 sizeof(xfs_ext_irec_t));
4599 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4603 * This is called to clean up large amounts of unused memory allocated
4604 * by the indirection array. Before compacting anything though, verify
4605 * that the indirection array is still needed and switch back to the
4606 * linear extent list (or even the inline buffer) if possible. The
4607 * compaction policy is as follows:
4609 * Full Compaction: Extents fit into a single page (or inline buffer)
4610 * Full Compaction: Extents occupy less than 10% of allocated space
4611 * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
4612 * No Compaction: Extents occupy at least 50% of allocated space
4615 xfs_iext_irec_compact(
4616 xfs_ifork_t *ifp) /* inode fork pointer */
4618 xfs_extnum_t nextents; /* number of extents in file */
4619 int nlists; /* number of irec's (ex lists) */
4621 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4622 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4623 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4625 if (nextents == 0) {
4626 xfs_iext_destroy(ifp);
4627 } else if (nextents <= XFS_INLINE_EXTS) {
4628 xfs_iext_indirect_to_direct(ifp);
4629 xfs_iext_direct_to_inline(ifp, nextents);
4630 } else if (nextents <= XFS_LINEAR_EXTS) {
4631 xfs_iext_indirect_to_direct(ifp);
4632 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 3) {
4633 xfs_iext_irec_compact_full(ifp);
4634 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
4635 xfs_iext_irec_compact_pages(ifp);
4640 * Combine extents from neighboring extent pages.
4643 xfs_iext_irec_compact_pages(
4644 xfs_ifork_t *ifp) /* inode fork pointer */
4646 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
4647 int erp_idx = 0; /* indirection array index */
4648 int nlists; /* number of irec's (ex lists) */
4650 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4651 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4652 while (erp_idx < nlists - 1) {
4653 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4655 if (erp_next->er_extcount <=
4656 (XFS_LINEAR_EXTS - erp->er_extcount)) {
4657 memmove(&erp->er_extbuf[erp->er_extcount],
4658 erp_next->er_extbuf, erp_next->er_extcount *
4659 sizeof(xfs_bmbt_rec_t));
4660 erp->er_extcount += erp_next->er_extcount;
4662 * Free page before removing extent record
4663 * so er_extoffs don't get modified in
4664 * xfs_iext_irec_remove.
4666 kmem_free(erp_next->er_extbuf, XFS_IEXT_BUFSZ);
4667 erp_next->er_extbuf = NULL;
4668 xfs_iext_irec_remove(ifp, erp_idx + 1);
4669 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4677 * Fully compact the extent records managed by the indirection array.
4680 xfs_iext_irec_compact_full(
4681 xfs_ifork_t *ifp) /* inode fork pointer */
4683 xfs_bmbt_rec_t *ep, *ep_next; /* extent record pointers */
4684 xfs_ext_irec_t *erp, *erp_next; /* extent irec pointers */
4685 int erp_idx = 0; /* extent irec index */
4686 int ext_avail; /* empty entries in ex list */
4687 int ext_diff; /* number of exts to add */
4688 int nlists; /* number of irec's (ex lists) */
4690 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4691 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4692 erp = ifp->if_u1.if_ext_irec;
4693 ep = &erp->er_extbuf[erp->er_extcount];
4695 ep_next = erp_next->er_extbuf;
4696 while (erp_idx < nlists - 1) {
4697 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
4698 ext_diff = MIN(ext_avail, erp_next->er_extcount);
4699 memcpy(ep, ep_next, ext_diff * sizeof(xfs_bmbt_rec_t));
4700 erp->er_extcount += ext_diff;
4701 erp_next->er_extcount -= ext_diff;
4702 /* Remove next page */
4703 if (erp_next->er_extcount == 0) {
4705 * Free page before removing extent record
4706 * so er_extoffs don't get modified in
4707 * xfs_iext_irec_remove.
4709 kmem_free(erp_next->er_extbuf,
4710 erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
4711 erp_next->er_extbuf = NULL;
4712 xfs_iext_irec_remove(ifp, erp_idx + 1);
4713 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4714 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4715 /* Update next page */
4717 /* Move rest of page up to become next new page */
4718 memmove(erp_next->er_extbuf, ep_next,
4719 erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
4720 ep_next = erp_next->er_extbuf;
4721 memset(&ep_next[erp_next->er_extcount], 0,
4722 (XFS_LINEAR_EXTS - erp_next->er_extcount) *
4723 sizeof(xfs_bmbt_rec_t));
4725 if (erp->er_extcount == XFS_LINEAR_EXTS) {
4727 if (erp_idx < nlists)
4728 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4732 ep = &erp->er_extbuf[erp->er_extcount];
4734 ep_next = erp_next->er_extbuf;
4739 * This is called to update the er_extoff field in the indirection
4740 * array when extents have been added or removed from one of the
4741 * extent lists. erp_idx contains the irec index to begin updating
4742 * at and ext_diff contains the number of extents that were added
4746 xfs_iext_irec_update_extoffs(
4747 xfs_ifork_t *ifp, /* inode fork pointer */
4748 int erp_idx, /* irec index to update */
4749 int ext_diff) /* number of new extents */
4751 int i; /* loop counter */
4752 int nlists; /* number of irec's (ex lists */
4754 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4755 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4756 for (i = erp_idx; i < nlists; i++) {
4757 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;