2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
25 #include "xfs_trans.h"
26 #include "xfs_trans_priv.h"
30 #include "xfs_dmapi.h"
31 #include "xfs_mount.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_ialloc_btree.h"
35 #include "xfs_dir2_sf.h"
36 #include "xfs_attr_sf.h"
37 #include "xfs_dinode.h"
38 #include "xfs_inode.h"
39 #include "xfs_buf_item.h"
40 #include "xfs_inode_item.h"
41 #include "xfs_btree.h"
42 #include "xfs_alloc.h"
43 #include "xfs_ialloc.h"
46 #include "xfs_error.h"
47 #include "xfs_utils.h"
48 #include "xfs_dir2_trace.h"
49 #include "xfs_quota.h"
53 #include <linux/vs_tag.h>
55 kmem_zone_t *xfs_ifork_zone;
56 kmem_zone_t *xfs_inode_zone;
57 kmem_zone_t *xfs_chashlist_zone;
60 * Used in xfs_itruncate(). This is the maximum number of extents
61 * freed from a file in a single transaction.
63 #define XFS_ITRUNC_MAX_EXTENTS 2
65 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
66 STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
67 STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
68 STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
73 * Make sure that the extents in the given memory buffer
88 for (i = 0; i < nrecs; i++) {
89 ep = xfs_iext_get_ext(ifp, i);
90 rec.l0 = get_unaligned((__uint64_t*)&ep->l0);
91 rec.l1 = get_unaligned((__uint64_t*)&ep->l1);
93 xfs_bmbt_disk_get_all(&rec, &irec);
95 xfs_bmbt_get_all(&rec, &irec);
96 if (fmt == XFS_EXTFMT_NOSTATE)
97 ASSERT(irec.br_state == XFS_EXT_NORM);
101 #define xfs_validate_extents(ifp, nrecs, disk, fmt)
105 * Check that none of the inode's in the buffer have a next
106 * unlinked field of 0.
118 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
120 for (i = 0; i < j; i++) {
121 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
122 i * mp->m_sb.sb_inodesize);
123 if (!dip->di_next_unlinked) {
124 xfs_fs_cmn_err(CE_ALERT, mp,
125 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
127 ASSERT(dip->di_next_unlinked);
134 * This routine is called to map an inode number within a file
135 * system to the buffer containing the on-disk version of the
136 * inode. It returns a pointer to the buffer containing the
137 * on-disk inode in the bpp parameter, and in the dip parameter
138 * it returns a pointer to the on-disk inode within that buffer.
140 * If a non-zero error is returned, then the contents of bpp and
141 * dipp are undefined.
143 * Use xfs_imap() to determine the size and location of the
144 * buffer to read from disk.
162 * Call the space management code to find the location of the
166 error = xfs_imap(mp, tp, ino, &imap, XFS_IMAP_LOOKUP);
169 "xfs_inotobp: xfs_imap() returned an "
170 "error %d on %s. Returning error.", error, mp->m_fsname);
175 * If the inode number maps to a block outside the bounds of the
176 * file system then return NULL rather than calling read_buf
177 * and panicing when we get an error from the driver.
179 if ((imap.im_blkno + imap.im_len) >
180 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
182 "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
183 "of the file system %s. Returning EINVAL.",
184 (unsigned long long)imap.im_blkno,
185 imap.im_len, mp->m_fsname);
186 return XFS_ERROR(EINVAL);
190 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
191 * default to just a read_buf() call.
193 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
194 (int)imap.im_len, XFS_BUF_LOCK, &bp);
198 "xfs_inotobp: xfs_trans_read_buf() returned an "
199 "error %d on %s. Returning error.", error, mp->m_fsname);
202 dip = (xfs_dinode_t *)xfs_buf_offset(bp, 0);
204 INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
205 XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
206 if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP,
207 XFS_RANDOM_ITOBP_INOTOBP))) {
208 XFS_CORRUPTION_ERROR("xfs_inotobp", XFS_ERRLEVEL_LOW, mp, dip);
209 xfs_trans_brelse(tp, bp);
211 "xfs_inotobp: XFS_TEST_ERROR() returned an "
212 "error on %s. Returning EFSCORRUPTED.", mp->m_fsname);
213 return XFS_ERROR(EFSCORRUPTED);
216 xfs_inobp_check(mp, bp);
219 * Set *dipp to point to the on-disk inode in the buffer.
221 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
223 *offset = imap.im_boffset;
229 * This routine is called to map an inode to the buffer containing
230 * the on-disk version of the inode. It returns a pointer to the
231 * buffer containing the on-disk inode in the bpp parameter, and in
232 * the dip parameter it returns a pointer to the on-disk inode within
235 * If a non-zero error is returned, then the contents of bpp and
236 * dipp are undefined.
238 * If the inode is new and has not yet been initialized, use xfs_imap()
239 * to determine the size and location of the buffer to read from disk.
240 * If the inode has already been mapped to its buffer and read in once,
241 * then use the mapping information stored in the inode rather than
242 * calling xfs_imap(). This allows us to avoid the overhead of looking
243 * at the inode btree for small block file systems (see xfs_dilocate()).
244 * We can tell whether the inode has been mapped in before by comparing
245 * its disk block address to 0. Only uninitialized inodes will have
246 * 0 for the disk block address.
264 if (ip->i_blkno == (xfs_daddr_t)0) {
266 * Call the space management code to find the location of the
270 if ((error = xfs_imap(mp, tp, ip->i_ino, &imap,
271 XFS_IMAP_LOOKUP | imap_flags)))
275 * If the inode number maps to a block outside the bounds
276 * of the file system then return NULL rather than calling
277 * read_buf and panicing when we get an error from the
280 if ((imap.im_blkno + imap.im_len) >
281 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
283 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
284 "(imap.im_blkno (0x%llx) "
285 "+ imap.im_len (0x%llx)) > "
286 " XFS_FSB_TO_BB(mp, "
287 "mp->m_sb.sb_dblocks) (0x%llx)",
288 (unsigned long long) imap.im_blkno,
289 (unsigned long long) imap.im_len,
290 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
292 return XFS_ERROR(EINVAL);
296 * Fill in the fields in the inode that will be used to
297 * map the inode to its buffer from now on.
299 ip->i_blkno = imap.im_blkno;
300 ip->i_len = imap.im_len;
301 ip->i_boffset = imap.im_boffset;
304 * We've already mapped the inode once, so just use the
305 * mapping that we saved the first time.
307 imap.im_blkno = ip->i_blkno;
308 imap.im_len = ip->i_len;
309 imap.im_boffset = ip->i_boffset;
311 ASSERT(bno == 0 || bno == imap.im_blkno);
314 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
315 * default to just a read_buf() call.
317 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
318 (int)imap.im_len, XFS_BUF_LOCK, &bp);
321 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
322 "xfs_trans_read_buf() returned error %d, "
323 "imap.im_blkno 0x%llx, imap.im_len 0x%llx",
324 error, (unsigned long long) imap.im_blkno,
325 (unsigned long long) imap.im_len);
331 * Validate the magic number and version of every inode in the buffer
332 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
333 * No validation is done here in userspace (xfs_repair).
335 #if !defined(__KERNEL__)
338 ni = BBTOB(imap.im_len) >> mp->m_sb.sb_inodelog;
339 #else /* usual case */
343 for (i = 0; i < ni; i++) {
347 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
348 (i << mp->m_sb.sb_inodelog));
349 di_ok = INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
350 XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
351 if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
352 XFS_ERRTAG_ITOBP_INOTOBP,
353 XFS_RANDOM_ITOBP_INOTOBP))) {
354 if (imap_flags & XFS_IMAP_BULKSTAT) {
355 xfs_trans_brelse(tp, bp);
356 return XFS_ERROR(EINVAL);
360 "Device %s - bad inode magic/vsn "
361 "daddr %lld #%d (magic=%x)",
362 XFS_BUFTARG_NAME(mp->m_ddev_targp),
363 (unsigned long long)imap.im_blkno, i,
364 INT_GET(dip->di_core.di_magic, ARCH_CONVERT));
366 XFS_CORRUPTION_ERROR("xfs_itobp", XFS_ERRLEVEL_HIGH,
368 xfs_trans_brelse(tp, bp);
369 return XFS_ERROR(EFSCORRUPTED);
373 xfs_inobp_check(mp, bp);
376 * Mark the buffer as an inode buffer now that it looks good
378 XFS_BUF_SET_VTYPE(bp, B_FS_INO);
381 * Set *dipp to point to the on-disk inode in the buffer.
383 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
389 * Move inode type and inode format specific information from the
390 * on-disk inode to the in-core inode. For fifos, devs, and sockets
391 * this means set if_rdev to the proper value. For files, directories,
392 * and symlinks this means to bring in the in-line data or extent
393 * pointers. For a file in B-tree format, only the root is immediately
394 * brought in-core. The rest will be in-lined in if_extents when it
395 * is first referenced (see xfs_iread_extents()).
402 xfs_attr_shortform_t *atp;
406 ip->i_df.if_ext_max =
407 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
411 INT_GET(dip->di_core.di_nextents, ARCH_CONVERT) +
412 INT_GET(dip->di_core.di_anextents, ARCH_CONVERT) >
413 INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT))) {
414 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
415 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
416 (unsigned long long)ip->i_ino,
417 (int)(INT_GET(dip->di_core.di_nextents, ARCH_CONVERT)
418 + INT_GET(dip->di_core.di_anextents, ARCH_CONVERT)),
420 INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT));
421 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
423 return XFS_ERROR(EFSCORRUPTED);
426 if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) {
427 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
428 "corrupt dinode %Lu, forkoff = 0x%x.",
429 (unsigned long long)ip->i_ino,
430 (int)(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT)));
431 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
433 return XFS_ERROR(EFSCORRUPTED);
436 switch (ip->i_d.di_mode & S_IFMT) {
441 if (unlikely(INT_GET(dip->di_core.di_format, ARCH_CONVERT) != XFS_DINODE_FMT_DEV)) {
442 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
444 return XFS_ERROR(EFSCORRUPTED);
447 ip->i_df.if_u2.if_rdev = INT_GET(dip->di_u.di_dev, ARCH_CONVERT);
453 switch (INT_GET(dip->di_core.di_format, ARCH_CONVERT)) {
454 case XFS_DINODE_FMT_LOCAL:
456 * no local regular files yet
458 if (unlikely((INT_GET(dip->di_core.di_mode, ARCH_CONVERT) & S_IFMT) == S_IFREG)) {
459 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
461 "(local format for regular file).",
462 (unsigned long long) ip->i_ino);
463 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
466 return XFS_ERROR(EFSCORRUPTED);
469 di_size = INT_GET(dip->di_core.di_size, ARCH_CONVERT);
470 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
471 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
473 "(bad size %Ld for local inode).",
474 (unsigned long long) ip->i_ino,
475 (long long) di_size);
476 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
479 return XFS_ERROR(EFSCORRUPTED);
483 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
485 case XFS_DINODE_FMT_EXTENTS:
486 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
488 case XFS_DINODE_FMT_BTREE:
489 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
492 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
494 return XFS_ERROR(EFSCORRUPTED);
499 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
500 return XFS_ERROR(EFSCORRUPTED);
505 if (!XFS_DFORK_Q(dip))
507 ASSERT(ip->i_afp == NULL);
508 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
509 ip->i_afp->if_ext_max =
510 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
511 switch (INT_GET(dip->di_core.di_aformat, ARCH_CONVERT)) {
512 case XFS_DINODE_FMT_LOCAL:
513 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
514 size = be16_to_cpu(atp->hdr.totsize);
515 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
517 case XFS_DINODE_FMT_EXTENTS:
518 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
520 case XFS_DINODE_FMT_BTREE:
521 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
524 error = XFS_ERROR(EFSCORRUPTED);
528 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
530 xfs_idestroy_fork(ip, XFS_DATA_FORK);
536 * The file is in-lined in the on-disk inode.
537 * If it fits into if_inline_data, then copy
538 * it there, otherwise allocate a buffer for it
539 * and copy the data there. Either way, set
540 * if_data to point at the data.
541 * If we allocate a buffer for the data, make
542 * sure that its size is a multiple of 4 and
543 * record the real size in i_real_bytes.
556 * If the size is unreasonable, then something
557 * is wrong and we just bail out rather than crash in
558 * kmem_alloc() or memcpy() below.
560 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
561 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
563 "(bad size %d for local fork, size = %d).",
564 (unsigned long long) ip->i_ino, size,
565 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
566 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
568 return XFS_ERROR(EFSCORRUPTED);
570 ifp = XFS_IFORK_PTR(ip, whichfork);
573 ifp->if_u1.if_data = NULL;
574 else if (size <= sizeof(ifp->if_u2.if_inline_data))
575 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
577 real_size = roundup(size, 4);
578 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
580 ifp->if_bytes = size;
581 ifp->if_real_bytes = real_size;
583 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
584 ifp->if_flags &= ~XFS_IFEXTENTS;
585 ifp->if_flags |= XFS_IFINLINE;
590 * The file consists of a set of extents all
591 * of which fit into the on-disk inode.
592 * If there are few enough extents to fit into
593 * the if_inline_ext, then copy them there.
594 * Otherwise allocate a buffer for them and copy
595 * them into it. Either way, set if_extents
596 * to point at the extents.
604 xfs_bmbt_rec_t *ep, *dp;
610 ifp = XFS_IFORK_PTR(ip, whichfork);
611 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
612 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
615 * If the number of extents is unreasonable, then something
616 * is wrong and we just bail out rather than crash in
617 * kmem_alloc() or memcpy() below.
619 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
620 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
621 "corrupt inode %Lu ((a)extents = %d).",
622 (unsigned long long) ip->i_ino, nex);
623 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
625 return XFS_ERROR(EFSCORRUPTED);
628 ifp->if_real_bytes = 0;
630 ifp->if_u1.if_extents = NULL;
631 else if (nex <= XFS_INLINE_EXTS)
632 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
634 xfs_iext_add(ifp, 0, nex);
636 ifp->if_bytes = size;
638 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
639 xfs_validate_extents(ifp, nex, 1, XFS_EXTFMT_INODE(ip));
640 for (i = 0; i < nex; i++, dp++) {
641 ep = xfs_iext_get_ext(ifp, i);
642 ep->l0 = INT_GET(get_unaligned((__uint64_t*)&dp->l0),
644 ep->l1 = INT_GET(get_unaligned((__uint64_t*)&dp->l1),
647 xfs_bmap_trace_exlist("xfs_iformat_extents", ip, nex,
649 if (whichfork != XFS_DATA_FORK ||
650 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
651 if (unlikely(xfs_check_nostate_extents(
653 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
656 return XFS_ERROR(EFSCORRUPTED);
659 ifp->if_flags |= XFS_IFEXTENTS;
664 * The file has too many extents to fit into
665 * the inode, so they are in B-tree format.
666 * Allocate a buffer for the root of the B-tree
667 * and copy the root into it. The i_extents
668 * field will remain NULL until all of the
669 * extents are read in (when they are needed).
677 xfs_bmdr_block_t *dfp;
683 ifp = XFS_IFORK_PTR(ip, whichfork);
684 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
685 size = XFS_BMAP_BROOT_SPACE(dfp);
686 nrecs = XFS_BMAP_BROOT_NUMRECS(dfp);
689 * blow out if -- fork has less extents than can fit in
690 * fork (fork shouldn't be a btree format), root btree
691 * block has more records than can fit into the fork,
692 * or the number of extents is greater than the number of
695 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
696 || XFS_BMDR_SPACE_CALC(nrecs) >
697 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
698 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
699 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
700 "corrupt inode %Lu (btree).",
701 (unsigned long long) ip->i_ino);
702 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
704 return XFS_ERROR(EFSCORRUPTED);
707 ifp->if_broot_bytes = size;
708 ifp->if_broot = kmem_alloc(size, KM_SLEEP);
709 ASSERT(ifp->if_broot != NULL);
711 * Copy and convert from the on-disk structure
712 * to the in-memory structure.
714 xfs_bmdr_to_bmbt(dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
715 ifp->if_broot, size);
716 ifp->if_flags &= ~XFS_IFEXTENTS;
717 ifp->if_flags |= XFS_IFBROOT;
723 * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
726 * buf = on-disk representation
727 * dip = native representation
728 * dir = direction - +ve -> disk to native
729 * -ve -> native to disk
732 xfs_xlate_dinode_core(
734 xfs_dinode_core_t *dip,
737 xfs_dinode_core_t *buf_core = (xfs_dinode_core_t *)buf;
738 xfs_dinode_core_t *mem_core = (xfs_dinode_core_t *)dip;
739 xfs_arch_t arch = ARCH_CONVERT;
740 uint32_t uid = 0, gid = 0;
746 tag = mem_core->di_tag;
747 /* FIXME: supposed to use superblock flag */
748 uid = TAGINO_UID(1, mem_core->di_uid, tag);
749 gid = TAGINO_GID(1, mem_core->di_gid, tag);
750 tag = TAGINO_TAG(1, tag);
753 INT_XLATE(buf_core->di_magic, mem_core->di_magic, dir, arch);
754 INT_XLATE(buf_core->di_mode, mem_core->di_mode, dir, arch);
755 INT_XLATE(buf_core->di_version, mem_core->di_version, dir, arch);
756 INT_XLATE(buf_core->di_format, mem_core->di_format, dir, arch);
757 INT_XLATE(buf_core->di_onlink, mem_core->di_onlink, dir, arch);
758 INT_XLATE(buf_core->di_uid, uid, dir, arch);
759 INT_XLATE(buf_core->di_gid, gid, dir, arch);
760 INT_XLATE(buf_core->di_tag, tag, dir, arch);
761 INT_XLATE(buf_core->di_nlink, mem_core->di_nlink, dir, arch);
762 INT_XLATE(buf_core->di_projid, mem_core->di_projid, dir, arch);
765 /* FIXME: supposed to use superblock flag */
766 mem_core->di_uid = INOTAG_UID(1, uid, gid);
767 mem_core->di_gid = INOTAG_GID(1, uid, gid);
768 mem_core->di_tag = INOTAG_TAG(1, uid, gid, tag);
769 memcpy(mem_core->di_pad, buf_core->di_pad,
770 sizeof(buf_core->di_pad));
772 memcpy(buf_core->di_pad, mem_core->di_pad,
773 sizeof(buf_core->di_pad));
776 INT_XLATE(buf_core->di_flushiter, mem_core->di_flushiter, dir, arch);
778 INT_XLATE(buf_core->di_atime.t_sec, mem_core->di_atime.t_sec,
780 INT_XLATE(buf_core->di_atime.t_nsec, mem_core->di_atime.t_nsec,
782 INT_XLATE(buf_core->di_mtime.t_sec, mem_core->di_mtime.t_sec,
784 INT_XLATE(buf_core->di_mtime.t_nsec, mem_core->di_mtime.t_nsec,
786 INT_XLATE(buf_core->di_ctime.t_sec, mem_core->di_ctime.t_sec,
788 INT_XLATE(buf_core->di_ctime.t_nsec, mem_core->di_ctime.t_nsec,
790 INT_XLATE(buf_core->di_size, mem_core->di_size, dir, arch);
791 INT_XLATE(buf_core->di_nblocks, mem_core->di_nblocks, dir, arch);
792 INT_XLATE(buf_core->di_extsize, mem_core->di_extsize, dir, arch);
793 INT_XLATE(buf_core->di_nextents, mem_core->di_nextents, dir, arch);
794 INT_XLATE(buf_core->di_anextents, mem_core->di_anextents, dir, arch);
795 INT_XLATE(buf_core->di_forkoff, mem_core->di_forkoff, dir, arch);
796 INT_XLATE(buf_core->di_aformat, mem_core->di_aformat, dir, arch);
797 INT_XLATE(buf_core->di_dmevmask, mem_core->di_dmevmask, dir, arch);
798 INT_XLATE(buf_core->di_dmstate, mem_core->di_dmstate, dir, arch);
799 INT_XLATE(buf_core->di_flags, mem_core->di_flags, dir, arch);
800 INT_XLATE(buf_core->di_gen, mem_core->di_gen, dir, arch);
809 if (di_flags & XFS_DIFLAG_ANY) {
810 if (di_flags & XFS_DIFLAG_REALTIME)
811 flags |= XFS_XFLAG_REALTIME;
812 if (di_flags & XFS_DIFLAG_PREALLOC)
813 flags |= XFS_XFLAG_PREALLOC;
814 if (di_flags & XFS_DIFLAG_IMMUTABLE)
815 flags |= XFS_XFLAG_IMMUTABLE;
816 if (di_flags & XFS_DIFLAG_IUNLINK)
817 flags |= XFS_XFLAG_IUNLINK;
818 if (di_flags & XFS_DIFLAG_BARRIER)
819 flags |= XFS_XFLAG_BARRIER;
820 if (di_flags & XFS_DIFLAG_APPEND)
821 flags |= XFS_XFLAG_APPEND;
822 if (di_flags & XFS_DIFLAG_SYNC)
823 flags |= XFS_XFLAG_SYNC;
824 if (di_flags & XFS_DIFLAG_NOATIME)
825 flags |= XFS_XFLAG_NOATIME;
826 if (di_flags & XFS_DIFLAG_NODUMP)
827 flags |= XFS_XFLAG_NODUMP;
828 if (di_flags & XFS_DIFLAG_RTINHERIT)
829 flags |= XFS_XFLAG_RTINHERIT;
830 if (di_flags & XFS_DIFLAG_PROJINHERIT)
831 flags |= XFS_XFLAG_PROJINHERIT;
832 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
833 flags |= XFS_XFLAG_NOSYMLINKS;
834 if (di_flags & XFS_DIFLAG_EXTSIZE)
835 flags |= XFS_XFLAG_EXTSIZE;
836 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
837 flags |= XFS_XFLAG_EXTSZINHERIT;
838 if (di_flags & XFS_DIFLAG_NODEFRAG)
839 flags |= XFS_XFLAG_NODEFRAG;
849 xfs_dinode_core_t *dic = &ip->i_d;
851 return _xfs_dic2xflags(dic->di_flags) |
852 (XFS_CFORK_Q(dic) ? XFS_XFLAG_HASATTR : 0);
857 xfs_dinode_core_t *dic)
859 return _xfs_dic2xflags(INT_GET(dic->di_flags, ARCH_CONVERT)) |
860 (XFS_CFORK_Q_DISK(dic) ? XFS_XFLAG_HASATTR : 0);
864 * Given a mount structure and an inode number, return a pointer
865 * to a newly allocated in-core inode corresponding to the given
868 * Initialize the inode's attributes and extent pointers if it
869 * already has them (it will not if the inode has no links).
885 ASSERT(xfs_inode_zone != NULL);
887 ip = kmem_zone_zalloc(xfs_inode_zone, KM_SLEEP);
890 spin_lock_init(&ip->i_flags_lock);
893 * Get pointer's to the on-disk inode and the buffer containing it.
894 * If the inode number refers to a block outside the file system
895 * then xfs_itobp() will return NULL. In this case we should
896 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
897 * know that this is a new incore inode.
899 error = xfs_itobp(mp, tp, ip, &dip, &bp, bno, imap_flags);
901 kmem_zone_free(xfs_inode_zone, ip);
906 * Initialize inode's trace buffers.
907 * Do this before xfs_iformat in case it adds entries.
909 #ifdef XFS_BMAP_TRACE
910 ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_SLEEP);
912 #ifdef XFS_BMBT_TRACE
913 ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_SLEEP);
916 ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_SLEEP);
918 #ifdef XFS_ILOCK_TRACE
919 ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_SLEEP);
921 #ifdef XFS_DIR2_TRACE
922 ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_SLEEP);
926 * If we got something that isn't an inode it means someone
927 * (nfs or dmi) has a stale handle.
929 if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC) {
930 kmem_zone_free(xfs_inode_zone, ip);
931 xfs_trans_brelse(tp, bp);
933 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
934 "dip->di_core.di_magic (0x%x) != "
935 "XFS_DINODE_MAGIC (0x%x)",
936 INT_GET(dip->di_core.di_magic, ARCH_CONVERT),
939 return XFS_ERROR(EINVAL);
943 * If the on-disk inode is already linked to a directory
944 * entry, copy all of the inode into the in-core inode.
945 * xfs_iformat() handles copying in the inode format
946 * specific information.
947 * Otherwise, just get the truly permanent information.
949 if (dip->di_core.di_mode) {
950 xfs_xlate_dinode_core((xfs_caddr_t)&dip->di_core,
952 error = xfs_iformat(ip, dip);
954 kmem_zone_free(xfs_inode_zone, ip);
955 xfs_trans_brelse(tp, bp);
957 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
958 "xfs_iformat() returned error %d",
964 ip->i_d.di_magic = INT_GET(dip->di_core.di_magic, ARCH_CONVERT);
965 ip->i_d.di_version = INT_GET(dip->di_core.di_version, ARCH_CONVERT);
966 ip->i_d.di_gen = INT_GET(dip->di_core.di_gen, ARCH_CONVERT);
967 ip->i_d.di_flushiter = INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT);
969 * Make sure to pull in the mode here as well in
970 * case the inode is released without being used.
971 * This ensures that xfs_inactive() will see that
972 * the inode is already free and not try to mess
973 * with the uninitialized part of it.
977 * Initialize the per-fork minima and maxima for a new
978 * inode here. xfs_iformat will do it for old inodes.
980 ip->i_df.if_ext_max =
981 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
984 INIT_LIST_HEAD(&ip->i_reclaim);
987 * The inode format changed when we moved the link count and
988 * made it 32 bits long. If this is an old format inode,
989 * convert it in memory to look like a new one. If it gets
990 * flushed to disk we will convert back before flushing or
991 * logging it. We zero out the new projid field and the old link
992 * count field. We'll handle clearing the pad field (the remains
993 * of the old uuid field) when we actually convert the inode to
994 * the new format. We don't change the version number so that we
995 * can distinguish this from a real new format inode.
997 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
998 ip->i_d.di_nlink = ip->i_d.di_onlink;
999 ip->i_d.di_onlink = 0;
1000 ip->i_d.di_projid = 0;
1003 ip->i_delayed_blks = 0;
1006 * Mark the buffer containing the inode as something to keep
1007 * around for a while. This helps to keep recently accessed
1008 * meta-data in-core longer.
1010 XFS_BUF_SET_REF(bp, XFS_INO_REF);
1013 * Use xfs_trans_brelse() to release the buffer containing the
1014 * on-disk inode, because it was acquired with xfs_trans_read_buf()
1015 * in xfs_itobp() above. If tp is NULL, this is just a normal
1016 * brelse(). If we're within a transaction, then xfs_trans_brelse()
1017 * will only release the buffer if it is not dirty within the
1018 * transaction. It will be OK to release the buffer in this case,
1019 * because inodes on disk are never destroyed and we will be
1020 * locking the new in-core inode before putting it in the hash
1021 * table where other processes can find it. Thus we don't have
1022 * to worry about the inode being changed just because we released
1025 xfs_trans_brelse(tp, bp);
1031 * Read in extents from a btree-format inode.
1032 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1042 xfs_extnum_t nextents;
1045 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
1046 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
1048 return XFS_ERROR(EFSCORRUPTED);
1050 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
1051 size = nextents * sizeof(xfs_bmbt_rec_t);
1052 ifp = XFS_IFORK_PTR(ip, whichfork);
1055 * We know that the size is valid (it's checked in iformat_btree)
1057 ifp->if_lastex = NULLEXTNUM;
1058 ifp->if_bytes = ifp->if_real_bytes = 0;
1059 ifp->if_flags |= XFS_IFEXTENTS;
1060 xfs_iext_add(ifp, 0, nextents);
1061 error = xfs_bmap_read_extents(tp, ip, whichfork);
1063 xfs_iext_destroy(ifp);
1064 ifp->if_flags &= ~XFS_IFEXTENTS;
1067 xfs_validate_extents(ifp, nextents, 0, XFS_EXTFMT_INODE(ip));
1072 * Allocate an inode on disk and return a copy of its in-core version.
1073 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1074 * appropriately within the inode. The uid and gid for the inode are
1075 * set according to the contents of the given cred structure.
1077 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1078 * has a free inode available, call xfs_iget()
1079 * to obtain the in-core version of the allocated inode. Finally,
1080 * fill in the inode and log its initial contents. In this case,
1081 * ialloc_context would be set to NULL and call_again set to false.
1083 * If xfs_dialloc() does not have an available inode,
1084 * it will replenish its supply by doing an allocation. Since we can
1085 * only do one allocation within a transaction without deadlocks, we
1086 * must commit the current transaction before returning the inode itself.
1087 * In this case, therefore, we will set call_again to true and return.
1088 * The caller should then commit the current transaction, start a new
1089 * transaction, and call xfs_ialloc() again to actually get the inode.
1091 * To ensure that some other process does not grab the inode that
1092 * was allocated during the first call to xfs_ialloc(), this routine
1093 * also returns the [locked] bp pointing to the head of the freelist
1094 * as ialloc_context. The caller should hold this buffer across
1095 * the commit and pass it back into this routine on the second call.
1107 xfs_buf_t **ialloc_context,
1108 boolean_t *call_again,
1118 * Call the space management code to pick
1119 * the on-disk inode to be allocated.
1121 error = xfs_dialloc(tp, pip->i_ino, mode, okalloc,
1122 ialloc_context, call_again, &ino);
1126 if (*call_again || ino == NULLFSINO) {
1130 ASSERT(*ialloc_context == NULL);
1133 * Get the in-core inode with the lock held exclusively.
1134 * This is because we're setting fields here we need
1135 * to prevent others from looking at until we're done.
1137 error = xfs_trans_iget(tp->t_mountp, tp, ino,
1138 XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
1145 ip->i_d.di_mode = (__uint16_t)mode;
1146 ip->i_d.di_onlink = 0;
1147 ip->i_d.di_nlink = nlink;
1148 ASSERT(ip->i_d.di_nlink == nlink);
1149 ip->i_d.di_uid = current_fsuid(cr);
1150 ip->i_d.di_gid = current_fsgid(cr);
1151 ip->i_d.di_tag = current_fstag(cr, vp);
1152 ip->i_d.di_projid = prid;
1153 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1156 * If the superblock version is up to where we support new format
1157 * inodes and this is currently an old format inode, then change
1158 * the inode version number now. This way we only do the conversion
1159 * here rather than here and in the flush/logging code.
1161 if (XFS_SB_VERSION_HASNLINK(&tp->t_mountp->m_sb) &&
1162 ip->i_d.di_version == XFS_DINODE_VERSION_1) {
1163 ip->i_d.di_version = XFS_DINODE_VERSION_2;
1165 * We've already zeroed the old link count, the projid field,
1166 * and the pad field.
1171 * Project ids won't be stored on disk if we are using a version 1 inode.
1173 if ( (prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1))
1174 xfs_bump_ino_vers2(tp, ip);
1176 if (XFS_INHERIT_GID(pip, vp->v_vfsp)) {
1177 ip->i_d.di_gid = pip->i_d.di_gid;
1178 if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1179 ip->i_d.di_mode |= S_ISGID;
1184 * If the group ID of the new file does not match the effective group
1185 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1186 * (and only if the irix_sgid_inherit compatibility variable is set).
1188 if ((irix_sgid_inherit) &&
1189 (ip->i_d.di_mode & S_ISGID) &&
1190 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1191 ip->i_d.di_mode &= ~S_ISGID;
1194 ip->i_d.di_size = 0;
1195 ip->i_d.di_nextents = 0;
1196 ASSERT(ip->i_d.di_nblocks == 0);
1197 xfs_ichgtime(ip, XFS_ICHGTIME_CHG|XFS_ICHGTIME_ACC|XFS_ICHGTIME_MOD);
1199 * di_gen will have been taken care of in xfs_iread.
1201 ip->i_d.di_extsize = 0;
1202 ip->i_d.di_dmevmask = 0;
1203 ip->i_d.di_dmstate = 0;
1204 ip->i_d.di_flags = 0;
1205 flags = XFS_ILOG_CORE;
1206 switch (mode & S_IFMT) {
1211 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1212 ip->i_df.if_u2.if_rdev = rdev;
1213 ip->i_df.if_flags = 0;
1214 flags |= XFS_ILOG_DEV;
1218 if (unlikely(pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
1221 if ((mode & S_IFMT) == S_IFDIR) {
1222 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1223 di_flags |= XFS_DIFLAG_RTINHERIT;
1224 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1225 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1226 ip->i_d.di_extsize = pip->i_d.di_extsize;
1228 } else if ((mode & S_IFMT) == S_IFREG) {
1229 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) {
1230 di_flags |= XFS_DIFLAG_REALTIME;
1231 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
1233 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1234 di_flags |= XFS_DIFLAG_EXTSIZE;
1235 ip->i_d.di_extsize = pip->i_d.di_extsize;
1238 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1239 xfs_inherit_noatime)
1240 di_flags |= XFS_DIFLAG_NOATIME;
1241 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1243 di_flags |= XFS_DIFLAG_NODUMP;
1244 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1246 di_flags |= XFS_DIFLAG_SYNC;
1247 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1248 xfs_inherit_nosymlinks)
1249 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1250 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1251 di_flags |= XFS_DIFLAG_PROJINHERIT;
1252 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1253 xfs_inherit_nodefrag)
1254 di_flags |= XFS_DIFLAG_NODEFRAG;
1255 ip->i_d.di_flags |= di_flags;
1259 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1260 ip->i_df.if_flags = XFS_IFEXTENTS;
1261 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1262 ip->i_df.if_u1.if_extents = NULL;
1268 * Attribute fork settings for new inode.
1270 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1271 ip->i_d.di_anextents = 0;
1274 * Log the new values stuffed into the inode.
1276 xfs_trans_log_inode(tp, ip, flags);
1278 /* now that we have an i_mode we can setup inode ops and unlock */
1279 bhv_vfs_init_vnode(XFS_MTOVFS(tp->t_mountp), vp, XFS_ITOBHV(ip), 1);
1286 * Check to make sure that there are no blocks allocated to the
1287 * file beyond the size of the file. We don't check this for
1288 * files with fixed size extents or real time extents, but we
1289 * at least do it for regular files.
1298 xfs_fileoff_t map_first;
1300 xfs_bmbt_irec_t imaps[2];
1302 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1305 if (ip->i_d.di_flags & (XFS_DIFLAG_REALTIME | XFS_DIFLAG_EXTSIZE))
1309 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1311 * The filesystem could be shutting down, so bmapi may return
1314 if (xfs_bmapi(NULL, ip, map_first,
1316 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1318 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1321 ASSERT(nimaps == 1);
1322 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1327 * Calculate the last possible buffered byte in a file. This must
1328 * include data that was buffered beyond the EOF by the write code.
1329 * This also needs to deal with overflowing the xfs_fsize_t type
1330 * which can happen for sizes near the limit.
1332 * We also need to take into account any blocks beyond the EOF. It
1333 * may be the case that they were buffered by a write which failed.
1334 * In that case the pages will still be in memory, but the inode size
1335 * will never have been updated.
1342 xfs_fsize_t last_byte;
1343 xfs_fileoff_t last_block;
1344 xfs_fileoff_t size_last_block;
1347 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE | MR_ACCESS));
1351 * Only check for blocks beyond the EOF if the extents have
1352 * been read in. This eliminates the need for the inode lock,
1353 * and it also saves us from looking when it really isn't
1356 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1357 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1365 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_d.di_size);
1366 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1368 last_byte = XFS_FSB_TO_B(mp, last_block);
1369 if (last_byte < 0) {
1370 return XFS_MAXIOFFSET(mp);
1372 last_byte += (1 << mp->m_writeio_log);
1373 if (last_byte < 0) {
1374 return XFS_MAXIOFFSET(mp);
1379 #if defined(XFS_RW_TRACE)
1385 xfs_fsize_t new_size,
1386 xfs_off_t toss_start,
1387 xfs_off_t toss_finish)
1389 if (ip->i_rwtrace == NULL) {
1393 ktrace_enter(ip->i_rwtrace,
1396 (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
1397 (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
1398 (void*)((long)flag),
1399 (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
1400 (void*)(unsigned long)(new_size & 0xffffffff),
1401 (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
1402 (void*)(unsigned long)(toss_start & 0xffffffff),
1403 (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
1404 (void*)(unsigned long)(toss_finish & 0xffffffff),
1405 (void*)(unsigned long)current_cpu(),
1406 (void*)(unsigned long)current_pid(),
1412 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1416 * Start the truncation of the file to new_size. The new size
1417 * must be smaller than the current size. This routine will
1418 * clear the buffer and page caches of file data in the removed
1419 * range, and xfs_itruncate_finish() will remove the underlying
1422 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1423 * must NOT have the inode lock held at all. This is because we're
1424 * calling into the buffer/page cache code and we can't hold the
1425 * inode lock when we do so.
1427 * We need to wait for any direct I/Os in flight to complete before we
1428 * proceed with the truncate. This is needed to prevent the extents
1429 * being read or written by the direct I/Os from being removed while the
1430 * I/O is in flight as there is no other method of synchronising
1431 * direct I/O with the truncate operation. Also, because we hold
1432 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1433 * started until the truncate completes and drops the lock. Essentially,
1434 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1435 * between direct I/Os and the truncate operation.
1437 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1438 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1439 * in the case that the caller is locking things out of order and
1440 * may not be able to call xfs_itruncate_finish() with the inode lock
1441 * held without dropping the I/O lock. If the caller must drop the
1442 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1443 * must be called again with all the same restrictions as the initial
1447 xfs_itruncate_start(
1450 xfs_fsize_t new_size)
1452 xfs_fsize_t last_byte;
1453 xfs_off_t toss_start;
1457 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1458 ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
1459 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1460 (flags == XFS_ITRUNC_MAYBE));
1465 vn_iowait(vp); /* wait for the completion of any pending DIOs */
1468 * Call toss_pages or flushinval_pages to get rid of pages
1469 * overlapping the region being removed. We have to use
1470 * the less efficient flushinval_pages in the case that the
1471 * caller may not be able to finish the truncate without
1472 * dropping the inode's I/O lock. Make sure
1473 * to catch any pages brought in by buffers overlapping
1474 * the EOF by searching out beyond the isize by our
1475 * block size. We round new_size up to a block boundary
1476 * so that we don't toss things on the same block as
1477 * new_size but before it.
1479 * Before calling toss_page or flushinval_pages, make sure to
1480 * call remapf() over the same region if the file is mapped.
1481 * This frees up mapped file references to the pages in the
1482 * given range and for the flushinval_pages case it ensures
1483 * that we get the latest mapped changes flushed out.
1485 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1486 toss_start = XFS_FSB_TO_B(mp, toss_start);
1487 if (toss_start < 0) {
1489 * The place to start tossing is beyond our maximum
1490 * file size, so there is no way that the data extended
1495 last_byte = xfs_file_last_byte(ip);
1496 xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
1498 if (last_byte > toss_start) {
1499 if (flags & XFS_ITRUNC_DEFINITE) {
1500 bhv_vop_toss_pages(vp, toss_start, -1, FI_REMAPF_LOCKED);
1502 bhv_vop_flushinval_pages(vp, toss_start, -1, FI_REMAPF_LOCKED);
1507 if (new_size == 0) {
1508 ASSERT(VN_CACHED(vp) == 0);
1514 * Shrink the file to the given new_size. The new
1515 * size must be smaller than the current size.
1516 * This will free up the underlying blocks
1517 * in the removed range after a call to xfs_itruncate_start()
1518 * or xfs_atruncate_start().
1520 * The transaction passed to this routine must have made
1521 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1522 * This routine may commit the given transaction and
1523 * start new ones, so make sure everything involved in
1524 * the transaction is tidy before calling here.
1525 * Some transaction will be returned to the caller to be
1526 * committed. The incoming transaction must already include
1527 * the inode, and both inode locks must be held exclusively.
1528 * The inode must also be "held" within the transaction. On
1529 * return the inode will be "held" within the returned transaction.
1530 * This routine does NOT require any disk space to be reserved
1531 * for it within the transaction.
1533 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1534 * and it indicates the fork which is to be truncated. For the
1535 * attribute fork we only support truncation to size 0.
1537 * We use the sync parameter to indicate whether or not the first
1538 * transaction we perform might have to be synchronous. For the attr fork,
1539 * it needs to be so if the unlink of the inode is not yet known to be
1540 * permanent in the log. This keeps us from freeing and reusing the
1541 * blocks of the attribute fork before the unlink of the inode becomes
1544 * For the data fork, we normally have to run synchronously if we're
1545 * being called out of the inactive path or we're being called
1546 * out of the create path where we're truncating an existing file.
1547 * Either way, the truncate needs to be sync so blocks don't reappear
1548 * in the file with altered data in case of a crash. wsync filesystems
1549 * can run the first case async because anything that shrinks the inode
1550 * has to run sync so by the time we're called here from inactive, the
1551 * inode size is permanently set to 0.
1553 * Calls from the truncate path always need to be sync unless we're
1554 * in a wsync filesystem and the file has already been unlinked.
1556 * The caller is responsible for correctly setting the sync parameter.
1557 * It gets too hard for us to guess here which path we're being called
1558 * out of just based on inode state.
1561 xfs_itruncate_finish(
1564 xfs_fsize_t new_size,
1568 xfs_fsblock_t first_block;
1569 xfs_fileoff_t first_unmap_block;
1570 xfs_fileoff_t last_block;
1571 xfs_filblks_t unmap_len=0;
1576 xfs_bmap_free_t free_list;
1579 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1580 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
1581 ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
1582 ASSERT(*tp != NULL);
1583 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1584 ASSERT(ip->i_transp == *tp);
1585 ASSERT(ip->i_itemp != NULL);
1586 ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
1590 mp = (ntp)->t_mountp;
1591 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1594 * We only support truncating the entire attribute fork.
1596 if (fork == XFS_ATTR_FORK) {
1599 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1600 xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
1602 * The first thing we do is set the size to new_size permanently
1603 * on disk. This way we don't have to worry about anyone ever
1604 * being able to look at the data being freed even in the face
1605 * of a crash. What we're getting around here is the case where
1606 * we free a block, it is allocated to another file, it is written
1607 * to, and then we crash. If the new data gets written to the
1608 * file but the log buffers containing the free and reallocation
1609 * don't, then we'd end up with garbage in the blocks being freed.
1610 * As long as we make the new_size permanent before actually
1611 * freeing any blocks it doesn't matter if they get writtten to.
1613 * The callers must signal into us whether or not the size
1614 * setting here must be synchronous. There are a few cases
1615 * where it doesn't have to be synchronous. Those cases
1616 * occur if the file is unlinked and we know the unlink is
1617 * permanent or if the blocks being truncated are guaranteed
1618 * to be beyond the inode eof (regardless of the link count)
1619 * and the eof value is permanent. Both of these cases occur
1620 * only on wsync-mounted filesystems. In those cases, we're
1621 * guaranteed that no user will ever see the data in the blocks
1622 * that are being truncated so the truncate can run async.
1623 * In the free beyond eof case, the file may wind up with
1624 * more blocks allocated to it than it needs if we crash
1625 * and that won't get fixed until the next time the file
1626 * is re-opened and closed but that's ok as that shouldn't
1627 * be too many blocks.
1629 * However, we can't just make all wsync xactions run async
1630 * because there's one call out of the create path that needs
1631 * to run sync where it's truncating an existing file to size
1632 * 0 whose size is > 0.
1634 * It's probably possible to come up with a test in this
1635 * routine that would correctly distinguish all the above
1636 * cases from the values of the function parameters and the
1637 * inode state but for sanity's sake, I've decided to let the
1638 * layers above just tell us. It's simpler to correctly figure
1639 * out in the layer above exactly under what conditions we
1640 * can run async and I think it's easier for others read and
1641 * follow the logic in case something has to be changed.
1642 * cscope is your friend -- rcc.
1644 * The attribute fork is much simpler.
1646 * For the attribute fork we allow the caller to tell us whether
1647 * the unlink of the inode that led to this call is yet permanent
1648 * in the on disk log. If it is not and we will be freeing extents
1649 * in this inode then we make the first transaction synchronous
1650 * to make sure that the unlink is permanent by the time we free
1653 if (fork == XFS_DATA_FORK) {
1654 if (ip->i_d.di_nextents > 0) {
1655 ip->i_d.di_size = new_size;
1656 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1659 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1660 if (ip->i_d.di_anextents > 0)
1661 xfs_trans_set_sync(ntp);
1663 ASSERT(fork == XFS_DATA_FORK ||
1664 (fork == XFS_ATTR_FORK &&
1665 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1666 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1669 * Since it is possible for space to become allocated beyond
1670 * the end of the file (in a crash where the space is allocated
1671 * but the inode size is not yet updated), simply remove any
1672 * blocks which show up between the new EOF and the maximum
1673 * possible file size. If the first block to be removed is
1674 * beyond the maximum file size (ie it is the same as last_block),
1675 * then there is nothing to do.
1677 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1678 ASSERT(first_unmap_block <= last_block);
1680 if (last_block == first_unmap_block) {
1683 unmap_len = last_block - first_unmap_block + 1;
1687 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1688 * will tell us whether it freed the entire range or
1689 * not. If this is a synchronous mount (wsync),
1690 * then we can tell bunmapi to keep all the
1691 * transactions asynchronous since the unlink
1692 * transaction that made this inode inactive has
1693 * already hit the disk. There's no danger of
1694 * the freed blocks being reused, there being a
1695 * crash, and the reused blocks suddenly reappearing
1696 * in this file with garbage in them once recovery
1699 XFS_BMAP_INIT(&free_list, &first_block);
1700 error = XFS_BUNMAPI(mp, ntp, &ip->i_iocore,
1701 first_unmap_block, unmap_len,
1702 XFS_BMAPI_AFLAG(fork) |
1703 (sync ? 0 : XFS_BMAPI_ASYNC),
1704 XFS_ITRUNC_MAX_EXTENTS,
1705 &first_block, &free_list,
1709 * If the bunmapi call encounters an error,
1710 * return to the caller where the transaction
1711 * can be properly aborted. We just need to
1712 * make sure we're not holding any resources
1713 * that we were not when we came in.
1715 xfs_bmap_cancel(&free_list);
1720 * Duplicate the transaction that has the permanent
1721 * reservation and commit the old transaction.
1723 error = xfs_bmap_finish(tp, &free_list, first_block,
1728 * If the bmap finish call encounters an error,
1729 * return to the caller where the transaction
1730 * can be properly aborted. We just need to
1731 * make sure we're not holding any resources
1732 * that we were not when we came in.
1734 * Aborting from this point might lose some
1735 * blocks in the file system, but oh well.
1737 xfs_bmap_cancel(&free_list);
1740 * If the passed in transaction committed
1741 * in xfs_bmap_finish(), then we want to
1742 * add the inode to this one before returning.
1743 * This keeps things simple for the higher
1744 * level code, because it always knows that
1745 * the inode is locked and held in the
1746 * transaction that returns to it whether
1747 * errors occur or not. We don't mark the
1748 * inode dirty so that this transaction can
1749 * be easily aborted if possible.
1751 xfs_trans_ijoin(ntp, ip,
1752 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1753 xfs_trans_ihold(ntp, ip);
1760 * The first xact was committed,
1761 * so add the inode to the new one.
1762 * Mark it dirty so it will be logged
1763 * and moved forward in the log as
1764 * part of every commit.
1766 xfs_trans_ijoin(ntp, ip,
1767 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1768 xfs_trans_ihold(ntp, ip);
1769 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1771 ntp = xfs_trans_dup(ntp);
1772 (void) xfs_trans_commit(*tp, 0, NULL);
1774 error = xfs_trans_reserve(ntp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1775 XFS_TRANS_PERM_LOG_RES,
1776 XFS_ITRUNCATE_LOG_COUNT);
1778 * Add the inode being truncated to the next chained
1781 xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1782 xfs_trans_ihold(ntp, ip);
1787 * Only update the size in the case of the data fork, but
1788 * always re-log the inode so that our permanent transaction
1789 * can keep on rolling it forward in the log.
1791 if (fork == XFS_DATA_FORK) {
1792 xfs_isize_check(mp, ip, new_size);
1793 ip->i_d.di_size = new_size;
1795 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1796 ASSERT((new_size != 0) ||
1797 (fork == XFS_ATTR_FORK) ||
1798 (ip->i_delayed_blks == 0));
1799 ASSERT((new_size != 0) ||
1800 (fork == XFS_ATTR_FORK) ||
1801 (ip->i_d.di_nextents == 0));
1802 xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
1810 * Do the first part of growing a file: zero any data in the last
1811 * block that is beyond the old EOF. We need to do this before
1812 * the inode is joined to the transaction to modify the i_size.
1813 * That way we can drop the inode lock and call into the buffer
1814 * cache to get the buffer mapping the EOF.
1819 xfs_fsize_t new_size,
1824 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1825 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1826 ASSERT(new_size > ip->i_d.di_size);
1829 * Zero any pages that may have been created by
1830 * xfs_write_file() beyond the end of the file
1831 * and any blocks between the old and new file sizes.
1833 error = xfs_zero_eof(XFS_ITOV(ip), &ip->i_iocore, new_size,
1834 ip->i_d.di_size, new_size);
1841 * This routine is called to extend the size of a file.
1842 * The inode must have both the iolock and the ilock locked
1843 * for update and it must be a part of the current transaction.
1844 * The xfs_igrow_start() function must have been called previously.
1845 * If the change_flag is not zero, the inode change timestamp will
1852 xfs_fsize_t new_size,
1855 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1856 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1857 ASSERT(ip->i_transp == tp);
1858 ASSERT(new_size > ip->i_d.di_size);
1861 * Update the file size. Update the inode change timestamp
1862 * if change_flag set.
1864 ip->i_d.di_size = new_size;
1866 xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
1867 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1873 * This is called when the inode's link count goes to 0.
1874 * We place the on-disk inode on a list in the AGI. It
1875 * will be pulled from this list when the inode is freed.
1887 xfs_agnumber_t agno;
1888 xfs_daddr_t agdaddr;
1895 ASSERT(ip->i_d.di_nlink == 0);
1896 ASSERT(ip->i_d.di_mode != 0);
1897 ASSERT(ip->i_transp == tp);
1901 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1902 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1905 * Get the agi buffer first. It ensures lock ordering
1908 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1909 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1914 * Validate the magic number of the agi block.
1916 agi = XFS_BUF_TO_AGI(agibp);
1918 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1919 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
1920 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK,
1921 XFS_RANDOM_IUNLINK))) {
1922 XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi);
1923 xfs_trans_brelse(tp, agibp);
1924 return XFS_ERROR(EFSCORRUPTED);
1927 * Get the index into the agi hash table for the
1928 * list this inode will go on.
1930 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1932 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1933 ASSERT(agi->agi_unlinked[bucket_index]);
1934 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1936 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
1938 * There is already another inode in the bucket we need
1939 * to add ourselves to. Add us at the front of the list.
1940 * Here we put the head pointer into our next pointer,
1941 * and then we fall through to point the head at us.
1943 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
1947 ASSERT(INT_GET(dip->di_next_unlinked, ARCH_CONVERT) == NULLAGINO);
1948 ASSERT(dip->di_next_unlinked);
1949 /* both on-disk, don't endian flip twice */
1950 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1951 offset = ip->i_boffset +
1952 offsetof(xfs_dinode_t, di_next_unlinked);
1953 xfs_trans_inode_buf(tp, ibp);
1954 xfs_trans_log_buf(tp, ibp, offset,
1955 (offset + sizeof(xfs_agino_t) - 1));
1956 xfs_inobp_check(mp, ibp);
1960 * Point the bucket head pointer at the inode being inserted.
1963 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1964 offset = offsetof(xfs_agi_t, agi_unlinked) +
1965 (sizeof(xfs_agino_t) * bucket_index);
1966 xfs_trans_log_buf(tp, agibp, offset,
1967 (offset + sizeof(xfs_agino_t) - 1));
1972 * Pull the on-disk inode from the AGI unlinked list.
1985 xfs_agnumber_t agno;
1986 xfs_daddr_t agdaddr;
1988 xfs_agino_t next_agino;
1989 xfs_buf_t *last_ibp;
1990 xfs_dinode_t *last_dip = NULL;
1992 int offset, last_offset = 0;
1997 * First pull the on-disk inode from the AGI unlinked list.
2001 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2002 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
2005 * Get the agi buffer first. It ensures lock ordering
2008 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
2009 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
2012 "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
2013 error, mp->m_fsname);
2017 * Validate the magic number of the agi block.
2019 agi = XFS_BUF_TO_AGI(agibp);
2021 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
2022 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
2023 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE,
2024 XFS_RANDOM_IUNLINK_REMOVE))) {
2025 XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW,
2027 xfs_trans_brelse(tp, agibp);
2029 "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
2031 return XFS_ERROR(EFSCORRUPTED);
2034 * Get the index into the agi hash table for the
2035 * list this inode will go on.
2037 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2039 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2040 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
2041 ASSERT(agi->agi_unlinked[bucket_index]);
2043 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2045 * We're at the head of the list. Get the inode's
2046 * on-disk buffer to see if there is anyone after us
2047 * on the list. Only modify our next pointer if it
2048 * is not already NULLAGINO. This saves us the overhead
2049 * of dealing with the buffer when there is no need to
2052 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
2055 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2056 error, mp->m_fsname);
2059 next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
2060 ASSERT(next_agino != 0);
2061 if (next_agino != NULLAGINO) {
2062 INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
2063 offset = ip->i_boffset +
2064 offsetof(xfs_dinode_t, di_next_unlinked);
2065 xfs_trans_inode_buf(tp, ibp);
2066 xfs_trans_log_buf(tp, ibp, offset,
2067 (offset + sizeof(xfs_agino_t) - 1));
2068 xfs_inobp_check(mp, ibp);
2070 xfs_trans_brelse(tp, ibp);
2073 * Point the bucket head pointer at the next inode.
2075 ASSERT(next_agino != 0);
2076 ASSERT(next_agino != agino);
2077 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2078 offset = offsetof(xfs_agi_t, agi_unlinked) +
2079 (sizeof(xfs_agino_t) * bucket_index);
2080 xfs_trans_log_buf(tp, agibp, offset,
2081 (offset + sizeof(xfs_agino_t) - 1));
2084 * We need to search the list for the inode being freed.
2086 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2088 while (next_agino != agino) {
2090 * If the last inode wasn't the one pointing to
2091 * us, then release its buffer since we're not
2092 * going to do anything with it.
2094 if (last_ibp != NULL) {
2095 xfs_trans_brelse(tp, last_ibp);
2097 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2098 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
2099 &last_ibp, &last_offset);
2102 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
2103 error, mp->m_fsname);
2106 next_agino = INT_GET(last_dip->di_next_unlinked, ARCH_CONVERT);
2107 ASSERT(next_agino != NULLAGINO);
2108 ASSERT(next_agino != 0);
2111 * Now last_ibp points to the buffer previous to us on
2112 * the unlinked list. Pull us from the list.
2114 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
2117 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2118 error, mp->m_fsname);
2121 next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
2122 ASSERT(next_agino != 0);
2123 ASSERT(next_agino != agino);
2124 if (next_agino != NULLAGINO) {
2125 INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
2126 offset = ip->i_boffset +
2127 offsetof(xfs_dinode_t, di_next_unlinked);
2128 xfs_trans_inode_buf(tp, ibp);
2129 xfs_trans_log_buf(tp, ibp, offset,
2130 (offset + sizeof(xfs_agino_t) - 1));
2131 xfs_inobp_check(mp, ibp);
2133 xfs_trans_brelse(tp, ibp);
2136 * Point the previous inode on the list to the next inode.
2138 INT_SET(last_dip->di_next_unlinked, ARCH_CONVERT, next_agino);
2139 ASSERT(next_agino != 0);
2140 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2141 xfs_trans_inode_buf(tp, last_ibp);
2142 xfs_trans_log_buf(tp, last_ibp, offset,
2143 (offset + sizeof(xfs_agino_t) - 1));
2144 xfs_inobp_check(mp, last_ibp);
2149 static __inline__ int xfs_inode_clean(xfs_inode_t *ip)
2151 return (((ip->i_itemp == NULL) ||
2152 !(ip->i_itemp->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
2153 (ip->i_update_core == 0));
2158 xfs_inode_t *free_ip,
2162 xfs_mount_t *mp = free_ip->i_mount;
2163 int blks_per_cluster;
2166 int i, j, found, pre_flushed;
2170 xfs_inode_t *ip, **ip_found;
2171 xfs_inode_log_item_t *iip;
2172 xfs_log_item_t *lip;
2175 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
2176 blks_per_cluster = 1;
2177 ninodes = mp->m_sb.sb_inopblock;
2178 nbufs = XFS_IALLOC_BLOCKS(mp);
2180 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
2181 mp->m_sb.sb_blocksize;
2182 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
2183 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
2186 ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
2188 for (j = 0; j < nbufs; j++, inum += ninodes) {
2189 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2190 XFS_INO_TO_AGBNO(mp, inum));
2194 * Look for each inode in memory and attempt to lock it,
2195 * we can be racing with flush and tail pushing here.
2196 * any inode we get the locks on, add to an array of
2197 * inode items to process later.
2199 * The get the buffer lock, we could beat a flush
2200 * or tail pushing thread to the lock here, in which
2201 * case they will go looking for the inode buffer
2202 * and fail, we need some other form of interlock
2206 for (i = 0; i < ninodes; i++) {
2207 ih = XFS_IHASH(mp, inum + i);
2208 read_lock(&ih->ih_lock);
2209 for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
2210 if (ip->i_ino == inum + i)
2214 /* Inode not in memory or we found it already,
2217 if (!ip || xfs_iflags_test(ip, XFS_ISTALE)) {
2218 read_unlock(&ih->ih_lock);
2222 if (xfs_inode_clean(ip)) {
2223 read_unlock(&ih->ih_lock);
2227 /* If we can get the locks then add it to the
2228 * list, otherwise by the time we get the bp lock
2229 * below it will already be attached to the
2233 /* This inode will already be locked - by us, lets
2237 if (ip == free_ip) {
2238 if (xfs_iflock_nowait(ip)) {
2239 xfs_iflags_set(ip, XFS_ISTALE);
2240 if (xfs_inode_clean(ip)) {
2243 ip_found[found++] = ip;
2246 read_unlock(&ih->ih_lock);
2250 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2251 if (xfs_iflock_nowait(ip)) {
2252 xfs_iflags_set(ip, XFS_ISTALE);
2254 if (xfs_inode_clean(ip)) {
2256 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2258 ip_found[found++] = ip;
2261 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2265 read_unlock(&ih->ih_lock);
2268 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2269 mp->m_bsize * blks_per_cluster,
2273 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
2275 if (lip->li_type == XFS_LI_INODE) {
2276 iip = (xfs_inode_log_item_t *)lip;
2277 ASSERT(iip->ili_logged == 1);
2278 lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
2280 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2282 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2285 lip = lip->li_bio_list;
2288 for (i = 0; i < found; i++) {
2293 ip->i_update_core = 0;
2295 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2299 iip->ili_last_fields = iip->ili_format.ilf_fields;
2300 iip->ili_format.ilf_fields = 0;
2301 iip->ili_logged = 1;
2303 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2306 xfs_buf_attach_iodone(bp,
2307 (void(*)(xfs_buf_t*,xfs_log_item_t*))
2308 xfs_istale_done, (xfs_log_item_t *)iip);
2309 if (ip != free_ip) {
2310 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2314 if (found || pre_flushed)
2315 xfs_trans_stale_inode_buf(tp, bp);
2316 xfs_trans_binval(tp, bp);
2319 kmem_free(ip_found, ninodes * sizeof(xfs_inode_t *));
2323 * This is called to return an inode to the inode free list.
2324 * The inode should already be truncated to 0 length and have
2325 * no pages associated with it. This routine also assumes that
2326 * the inode is already a part of the transaction.
2328 * The on-disk copy of the inode will have been added to the list
2329 * of unlinked inodes in the AGI. We need to remove the inode from
2330 * that list atomically with respect to freeing it here.
2336 xfs_bmap_free_t *flist)
2340 xfs_ino_t first_ino;
2342 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2343 ASSERT(ip->i_transp == tp);
2344 ASSERT(ip->i_d.di_nlink == 0);
2345 ASSERT(ip->i_d.di_nextents == 0);
2346 ASSERT(ip->i_d.di_anextents == 0);
2347 ASSERT((ip->i_d.di_size == 0) ||
2348 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2349 ASSERT(ip->i_d.di_nblocks == 0);
2352 * Pull the on-disk inode from the AGI unlinked list.
2354 error = xfs_iunlink_remove(tp, ip);
2359 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2363 ip->i_d.di_mode = 0; /* mark incore inode as free */
2364 ip->i_d.di_flags = 0;
2365 ip->i_d.di_dmevmask = 0;
2366 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2367 ip->i_df.if_ext_max =
2368 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2369 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2370 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2372 * Bump the generation count so no one will be confused
2373 * by reincarnations of this inode.
2376 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2379 xfs_ifree_cluster(ip, tp, first_ino);
2386 * Reallocate the space for if_broot based on the number of records
2387 * being added or deleted as indicated in rec_diff. Move the records
2388 * and pointers in if_broot to fit the new size. When shrinking this
2389 * will eliminate holes between the records and pointers created by
2390 * the caller. When growing this will create holes to be filled in
2393 * The caller must not request to add more records than would fit in
2394 * the on-disk inode root. If the if_broot is currently NULL, then
2395 * if we adding records one will be allocated. The caller must also
2396 * not request that the number of records go below zero, although
2397 * it can go to zero.
2399 * ip -- the inode whose if_broot area is changing
2400 * ext_diff -- the change in the number of records, positive or negative,
2401 * requested for the if_broot array.
2411 xfs_bmbt_block_t *new_broot;
2418 * Handle the degenerate case quietly.
2420 if (rec_diff == 0) {
2424 ifp = XFS_IFORK_PTR(ip, whichfork);
2427 * If there wasn't any memory allocated before, just
2428 * allocate it now and get out.
2430 if (ifp->if_broot_bytes == 0) {
2431 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
2432 ifp->if_broot = (xfs_bmbt_block_t*)kmem_alloc(new_size,
2434 ifp->if_broot_bytes = (int)new_size;
2439 * If there is already an existing if_broot, then we need
2440 * to realloc() it and shift the pointers to their new
2441 * location. The records don't change location because
2442 * they are kept butted up against the btree block header.
2444 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2445 new_max = cur_max + rec_diff;
2446 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2447 ifp->if_broot = (xfs_bmbt_block_t *)
2448 kmem_realloc(ifp->if_broot,
2450 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2452 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2453 ifp->if_broot_bytes);
2454 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2456 ifp->if_broot_bytes = (int)new_size;
2457 ASSERT(ifp->if_broot_bytes <=
2458 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2459 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2464 * rec_diff is less than 0. In this case, we are shrinking the
2465 * if_broot buffer. It must already exist. If we go to zero
2466 * records, just get rid of the root and clear the status bit.
2468 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
2469 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2470 new_max = cur_max + rec_diff;
2471 ASSERT(new_max >= 0);
2473 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2477 new_broot = (xfs_bmbt_block_t *)kmem_alloc(new_size, KM_SLEEP);
2479 * First copy over the btree block header.
2481 memcpy(new_broot, ifp->if_broot, sizeof(xfs_bmbt_block_t));
2484 ifp->if_flags &= ~XFS_IFBROOT;
2488 * Only copy the records and pointers if there are any.
2492 * First copy the records.
2494 op = (char *)XFS_BMAP_BROOT_REC_ADDR(ifp->if_broot, 1,
2495 ifp->if_broot_bytes);
2496 np = (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot, 1,
2498 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2501 * Then copy the pointers.
2503 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2504 ifp->if_broot_bytes);
2505 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot, 1,
2507 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2509 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2510 ifp->if_broot = new_broot;
2511 ifp->if_broot_bytes = (int)new_size;
2512 ASSERT(ifp->if_broot_bytes <=
2513 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2519 * This is called when the amount of space needed for if_data
2520 * is increased or decreased. The change in size is indicated by
2521 * the number of bytes that need to be added or deleted in the
2522 * byte_diff parameter.
2524 * If the amount of space needed has decreased below the size of the
2525 * inline buffer, then switch to using the inline buffer. Otherwise,
2526 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2527 * to what is needed.
2529 * ip -- the inode whose if_data area is changing
2530 * byte_diff -- the change in the number of bytes, positive or negative,
2531 * requested for the if_data array.
2543 if (byte_diff == 0) {
2547 ifp = XFS_IFORK_PTR(ip, whichfork);
2548 new_size = (int)ifp->if_bytes + byte_diff;
2549 ASSERT(new_size >= 0);
2551 if (new_size == 0) {
2552 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2553 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2555 ifp->if_u1.if_data = NULL;
2557 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2559 * If the valid extents/data can fit in if_inline_ext/data,
2560 * copy them from the malloc'd vector and free it.
2562 if (ifp->if_u1.if_data == NULL) {
2563 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2564 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2565 ASSERT(ifp->if_real_bytes != 0);
2566 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2568 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2569 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2574 * Stuck with malloc/realloc.
2575 * For inline data, the underlying buffer must be
2576 * a multiple of 4 bytes in size so that it can be
2577 * logged and stay on word boundaries. We enforce
2580 real_size = roundup(new_size, 4);
2581 if (ifp->if_u1.if_data == NULL) {
2582 ASSERT(ifp->if_real_bytes == 0);
2583 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2584 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2586 * Only do the realloc if the underlying size
2587 * is really changing.
2589 if (ifp->if_real_bytes != real_size) {
2590 ifp->if_u1.if_data =
2591 kmem_realloc(ifp->if_u1.if_data,
2597 ASSERT(ifp->if_real_bytes == 0);
2598 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2599 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2603 ifp->if_real_bytes = real_size;
2604 ifp->if_bytes = new_size;
2605 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2612 * Map inode to disk block and offset.
2614 * mp -- the mount point structure for the current file system
2615 * tp -- the current transaction
2616 * ino -- the inode number of the inode to be located
2617 * imap -- this structure is filled in with the information necessary
2618 * to retrieve the given inode from disk
2619 * flags -- flags to pass to xfs_dilocate indicating whether or not
2620 * lookups in the inode btree were OK or not
2630 xfs_fsblock_t fsbno;
2635 fsbno = imap->im_blkno ?
2636 XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK;
2637 error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags);
2641 imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno);
2642 imap->im_len = XFS_FSB_TO_BB(mp, len);
2643 imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno);
2644 imap->im_ioffset = (ushort)off;
2645 imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog);
2656 ifp = XFS_IFORK_PTR(ip, whichfork);
2657 if (ifp->if_broot != NULL) {
2658 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2659 ifp->if_broot = NULL;
2663 * If the format is local, then we can't have an extents
2664 * array so just look for an inline data array. If we're
2665 * not local then we may or may not have an extents list,
2666 * so check and free it up if we do.
2668 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2669 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2670 (ifp->if_u1.if_data != NULL)) {
2671 ASSERT(ifp->if_real_bytes != 0);
2672 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2673 ifp->if_u1.if_data = NULL;
2674 ifp->if_real_bytes = 0;
2676 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2677 ((ifp->if_flags & XFS_IFEXTIREC) ||
2678 ((ifp->if_u1.if_extents != NULL) &&
2679 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
2680 ASSERT(ifp->if_real_bytes != 0);
2681 xfs_iext_destroy(ifp);
2683 ASSERT(ifp->if_u1.if_extents == NULL ||
2684 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2685 ASSERT(ifp->if_real_bytes == 0);
2686 if (whichfork == XFS_ATTR_FORK) {
2687 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2693 * This is called free all the memory associated with an inode.
2694 * It must free the inode itself and any buffers allocated for
2695 * if_extents/if_data and if_broot. It must also free the lock
2696 * associated with the inode.
2703 switch (ip->i_d.di_mode & S_IFMT) {
2707 xfs_idestroy_fork(ip, XFS_DATA_FORK);
2711 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
2712 mrfree(&ip->i_lock);
2713 mrfree(&ip->i_iolock);
2714 freesema(&ip->i_flock);
2715 #ifdef XFS_BMAP_TRACE
2716 ktrace_free(ip->i_xtrace);
2718 #ifdef XFS_BMBT_TRACE
2719 ktrace_free(ip->i_btrace);
2722 ktrace_free(ip->i_rwtrace);
2724 #ifdef XFS_ILOCK_TRACE
2725 ktrace_free(ip->i_lock_trace);
2727 #ifdef XFS_DIR2_TRACE
2728 ktrace_free(ip->i_dir_trace);
2731 /* XXXdpd should be able to assert this but shutdown
2732 * is leaving the AIL behind. */
2733 ASSERT(((ip->i_itemp->ili_item.li_flags & XFS_LI_IN_AIL) == 0) ||
2734 XFS_FORCED_SHUTDOWN(ip->i_mount));
2735 xfs_inode_item_destroy(ip);
2737 kmem_zone_free(xfs_inode_zone, ip);
2742 * Increment the pin count of the given buffer.
2743 * This value is protected by ipinlock spinlock in the mount structure.
2749 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2751 atomic_inc(&ip->i_pincount);
2755 * Decrement the pin count of the given inode, and wake up
2756 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2757 * inode must have been previously pinned with a call to xfs_ipin().
2763 ASSERT(atomic_read(&ip->i_pincount) > 0);
2765 if (atomic_dec_and_lock(&ip->i_pincount, &ip->i_flags_lock)) {
2768 * If the inode is currently being reclaimed, the link between
2769 * the bhv_vnode and the xfs_inode will be broken after the
2770 * XFS_IRECLAIM* flag is set. Hence, if these flags are not
2771 * set, then we can move forward and mark the linux inode dirty
2772 * knowing that it is still valid as it won't freed until after
2773 * the bhv_vnode<->xfs_inode link is broken in xfs_reclaim. The
2774 * i_flags_lock is used to synchronise the setting of the
2775 * XFS_IRECLAIM* flags and the breaking of the link, and so we
2776 * can execute atomically w.r.t to reclaim by holding this lock
2779 * However, we still need to issue the unpin wakeup call as the
2780 * inode reclaim may be blocked waiting for the inode to become
2784 if (!__xfs_iflags_test(ip, XFS_IRECLAIM|XFS_IRECLAIMABLE)) {
2785 bhv_vnode_t *vp = XFS_ITOV_NULL(ip);
2786 struct inode *inode = NULL;
2789 inode = vn_to_inode(vp);
2790 BUG_ON(inode->i_state & I_CLEAR);
2792 /* make sync come back and flush this inode */
2793 if (!(inode->i_state & (I_NEW|I_FREEING)))
2794 mark_inode_dirty_sync(inode);
2796 spin_unlock(&ip->i_flags_lock);
2797 wake_up(&ip->i_ipin_wait);
2802 * This is called to wait for the given inode to be unpinned.
2803 * It will sleep until this happens. The caller must have the
2804 * inode locked in at least shared mode so that the buffer cannot
2805 * be subsequently pinned once someone is waiting for it to be
2812 xfs_inode_log_item_t *iip;
2815 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE | MR_ACCESS));
2817 if (atomic_read(&ip->i_pincount) == 0) {
2822 if (iip && iip->ili_last_lsn) {
2823 lsn = iip->ili_last_lsn;
2829 * Give the log a push so we don't wait here too long.
2831 xfs_log_force(ip->i_mount, lsn, XFS_LOG_FORCE);
2833 wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
2838 * xfs_iextents_copy()
2840 * This is called to copy the REAL extents (as opposed to the delayed
2841 * allocation extents) from the inode into the given buffer. It
2842 * returns the number of bytes copied into the buffer.
2844 * If there are no delayed allocation extents, then we can just
2845 * memcpy() the extents into the buffer. Otherwise, we need to
2846 * examine each extent in turn and skip those which are delayed.
2851 xfs_bmbt_rec_t *buffer,
2855 xfs_bmbt_rec_t *dest_ep;
2857 #ifdef XFS_BMAP_TRACE
2858 static char fname[] = "xfs_iextents_copy";
2863 xfs_fsblock_t start_block;
2865 ifp = XFS_IFORK_PTR(ip, whichfork);
2866 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
2867 ASSERT(ifp->if_bytes > 0);
2869 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2870 xfs_bmap_trace_exlist(fname, ip, nrecs, whichfork);
2874 * There are some delayed allocation extents in the
2875 * inode, so copy the extents one at a time and skip
2876 * the delayed ones. There must be at least one
2877 * non-delayed extent.
2881 for (i = 0; i < nrecs; i++) {
2882 ep = xfs_iext_get_ext(ifp, i);
2883 start_block = xfs_bmbt_get_startblock(ep);
2884 if (ISNULLSTARTBLOCK(start_block)) {
2886 * It's a delayed allocation extent, so skip it.
2891 /* Translate to on disk format */
2892 put_unaligned(INT_GET(ep->l0, ARCH_CONVERT),
2893 (__uint64_t*)&dest_ep->l0);
2894 put_unaligned(INT_GET(ep->l1, ARCH_CONVERT),
2895 (__uint64_t*)&dest_ep->l1);
2899 ASSERT(copied != 0);
2900 xfs_validate_extents(ifp, copied, 1, XFS_EXTFMT_INODE(ip));
2902 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2906 * Each of the following cases stores data into the same region
2907 * of the on-disk inode, so only one of them can be valid at
2908 * any given time. While it is possible to have conflicting formats
2909 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2910 * in EXTENTS format, this can only happen when the fork has
2911 * changed formats after being modified but before being flushed.
2912 * In these cases, the format always takes precedence, because the
2913 * format indicates the current state of the fork.
2920 xfs_inode_log_item_t *iip,
2927 #ifdef XFS_TRANS_DEBUG
2930 static const short brootflag[2] =
2931 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2932 static const short dataflag[2] =
2933 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2934 static const short extflag[2] =
2935 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2939 ifp = XFS_IFORK_PTR(ip, whichfork);
2941 * This can happen if we gave up in iformat in an error path,
2942 * for the attribute fork.
2945 ASSERT(whichfork == XFS_ATTR_FORK);
2948 cp = XFS_DFORK_PTR(dip, whichfork);
2950 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2951 case XFS_DINODE_FMT_LOCAL:
2952 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2953 (ifp->if_bytes > 0)) {
2954 ASSERT(ifp->if_u1.if_data != NULL);
2955 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2956 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2960 case XFS_DINODE_FMT_EXTENTS:
2961 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2962 !(iip->ili_format.ilf_fields & extflag[whichfork]));
2963 ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
2964 (ifp->if_bytes == 0));
2965 ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
2966 (ifp->if_bytes > 0));
2967 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
2968 (ifp->if_bytes > 0)) {
2969 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2970 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2975 case XFS_DINODE_FMT_BTREE:
2976 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
2977 (ifp->if_broot_bytes > 0)) {
2978 ASSERT(ifp->if_broot != NULL);
2979 ASSERT(ifp->if_broot_bytes <=
2980 (XFS_IFORK_SIZE(ip, whichfork) +
2981 XFS_BROOT_SIZE_ADJ));
2982 xfs_bmbt_to_bmdr(ifp->if_broot, ifp->if_broot_bytes,
2983 (xfs_bmdr_block_t *)cp,
2984 XFS_DFORK_SIZE(dip, mp, whichfork));
2988 case XFS_DINODE_FMT_DEV:
2989 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
2990 ASSERT(whichfork == XFS_DATA_FORK);
2991 INT_SET(dip->di_u.di_dev, ARCH_CONVERT, ip->i_df.if_u2.if_rdev);
2995 case XFS_DINODE_FMT_UUID:
2996 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
2997 ASSERT(whichfork == XFS_DATA_FORK);
2998 memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid,
3012 * xfs_iflush() will write a modified inode's changes out to the
3013 * inode's on disk home. The caller must have the inode lock held
3014 * in at least shared mode and the inode flush semaphore must be
3015 * held as well. The inode lock will still be held upon return from
3016 * the call and the caller is free to unlock it.
3017 * The inode flush lock will be unlocked when the inode reaches the disk.
3018 * The flags indicate how the inode's buffer should be written out.
3025 xfs_inode_log_item_t *iip;
3033 int clcount; /* count of inodes clustered */
3035 enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
3038 XFS_STATS_INC(xs_iflush_count);
3040 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
3041 ASSERT(issemalocked(&(ip->i_flock)));
3042 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3043 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3049 * If the inode isn't dirty, then just release the inode
3050 * flush lock and do nothing.
3052 if ((ip->i_update_core == 0) &&
3053 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3054 ASSERT((iip != NULL) ?
3055 !(iip->ili_item.li_flags & XFS_LI_IN_AIL) : 1);
3061 * We can't flush the inode until it is unpinned, so
3062 * wait for it. We know noone new can pin it, because
3063 * we are holding the inode lock shared and you need
3064 * to hold it exclusively to pin the inode.
3066 xfs_iunpin_wait(ip);
3069 * This may have been unpinned because the filesystem is shutting
3070 * down forcibly. If that's the case we must not write this inode
3071 * to disk, because the log record didn't make it to disk!
3073 if (XFS_FORCED_SHUTDOWN(mp)) {
3074 ip->i_update_core = 0;
3076 iip->ili_format.ilf_fields = 0;
3078 return XFS_ERROR(EIO);
3082 * Get the buffer containing the on-disk inode.
3084 error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0, 0);
3091 * Decide how buffer will be flushed out. This is done before
3092 * the call to xfs_iflush_int because this field is zeroed by it.
3094 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3096 * Flush out the inode buffer according to the directions
3097 * of the caller. In the cases where the caller has given
3098 * us a choice choose the non-delwri case. This is because
3099 * the inode is in the AIL and we need to get it out soon.
3102 case XFS_IFLUSH_SYNC:
3103 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3106 case XFS_IFLUSH_ASYNC:
3107 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3110 case XFS_IFLUSH_DELWRI:
3120 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3121 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3122 case XFS_IFLUSH_DELWRI:
3125 case XFS_IFLUSH_ASYNC:
3128 case XFS_IFLUSH_SYNC:
3139 * First flush out the inode that xfs_iflush was called with.
3141 error = xfs_iflush_int(ip, bp);
3148 * see if other inodes can be gathered into this write
3151 ip->i_chash->chl_buf = bp;
3153 ch = XFS_CHASH(mp, ip->i_blkno);
3154 s = mutex_spinlock(&ch->ch_lock);
3157 for (iq = ip->i_cnext; iq != ip; iq = iq->i_cnext) {
3159 * Do an un-protected check to see if the inode is dirty and
3160 * is a candidate for flushing. These checks will be repeated
3161 * later after the appropriate locks are acquired.
3164 if ((iq->i_update_core == 0) &&
3166 !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
3167 xfs_ipincount(iq) == 0) {
3172 * Try to get locks. If any are unavailable,
3173 * then this inode cannot be flushed and is skipped.
3176 /* get inode locks (just i_lock) */
3177 if (xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) {
3178 /* get inode flush lock */
3179 if (xfs_iflock_nowait(iq)) {
3180 /* check if pinned */
3181 if (xfs_ipincount(iq) == 0) {
3182 /* arriving here means that
3183 * this inode can be flushed.
3184 * first re-check that it's
3188 if ((iq->i_update_core != 0)||
3190 (iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3192 error = xfs_iflush_int(iq, bp);
3196 goto cluster_corrupt_out;
3205 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3208 mutex_spinunlock(&ch->ch_lock, s);
3211 XFS_STATS_INC(xs_icluster_flushcnt);
3212 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
3216 * If the buffer is pinned then push on the log so we won't
3217 * get stuck waiting in the write for too long.
3219 if (XFS_BUF_ISPINNED(bp)){
3220 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
3223 if (flags & INT_DELWRI) {
3224 xfs_bdwrite(mp, bp);
3225 } else if (flags & INT_ASYNC) {
3226 xfs_bawrite(mp, bp);
3228 error = xfs_bwrite(mp, bp);
3234 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3235 xfs_iflush_abort(ip);
3237 * Unlocks the flush lock
3239 return XFS_ERROR(EFSCORRUPTED);
3241 cluster_corrupt_out:
3242 /* Corruption detected in the clustering loop. Invalidate the
3243 * inode buffer and shut down the filesystem.
3245 mutex_spinunlock(&ch->ch_lock, s);
3248 * Clean up the buffer. If it was B_DELWRI, just release it --
3249 * brelse can handle it with no problems. If not, shut down the
3250 * filesystem before releasing the buffer.
3252 if ((bufwasdelwri= XFS_BUF_ISDELAYWRITE(bp))) {
3256 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3260 * Just like incore_relse: if we have b_iodone functions,
3261 * mark the buffer as an error and call them. Otherwise
3262 * mark it as stale and brelse.
3264 if (XFS_BUF_IODONE_FUNC(bp)) {
3265 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
3269 XFS_BUF_ERROR(bp,EIO);
3277 xfs_iflush_abort(iq);
3279 * Unlocks the flush lock
3281 return XFS_ERROR(EFSCORRUPTED);
3290 xfs_inode_log_item_t *iip;
3293 #ifdef XFS_TRANS_DEBUG
3298 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
3299 ASSERT(issemalocked(&(ip->i_flock)));
3300 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3301 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3308 * If the inode isn't dirty, then just release the inode
3309 * flush lock and do nothing.
3311 if ((ip->i_update_core == 0) &&
3312 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3317 /* set *dip = inode's place in the buffer */
3318 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset);
3321 * Clear i_update_core before copying out the data.
3322 * This is for coordination with our timestamp updates
3323 * that don't hold the inode lock. They will always
3324 * update the timestamps BEFORE setting i_update_core,
3325 * so if we clear i_update_core after they set it we
3326 * are guaranteed to see their updates to the timestamps.
3327 * I believe that this depends on strongly ordered memory
3328 * semantics, but we have that. We use the SYNCHRONIZE
3329 * macro to make sure that the compiler does not reorder
3330 * the i_update_core access below the data copy below.
3332 ip->i_update_core = 0;
3336 * Make sure to get the latest atime from the Linux inode.
3338 xfs_synchronize_atime(ip);
3340 if (XFS_TEST_ERROR(INT_GET(dip->di_core.di_magic,ARCH_CONVERT) != XFS_DINODE_MAGIC,
3341 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3342 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3343 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3344 ip->i_ino, (int) INT_GET(dip->di_core.di_magic, ARCH_CONVERT), dip);
3347 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3348 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3349 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3350 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3351 ip->i_ino, ip, ip->i_d.di_magic);
3354 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3356 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3357 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3358 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3359 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3360 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3364 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
3366 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3367 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3368 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3369 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3370 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3371 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3376 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3377 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3378 XFS_RANDOM_IFLUSH_5)) {
3379 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3380 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3382 ip->i_d.di_nextents + ip->i_d.di_anextents,
3387 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3388 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3389 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3390 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3391 ip->i_ino, ip->i_d.di_forkoff, ip);
3395 * bump the flush iteration count, used to detect flushes which
3396 * postdate a log record during recovery.
3399 ip->i_d.di_flushiter++;
3402 * Copy the dirty parts of the inode into the on-disk
3403 * inode. We always copy out the core of the inode,
3404 * because if the inode is dirty at all the core must
3407 xfs_xlate_dinode_core((xfs_caddr_t)&(dip->di_core), &(ip->i_d), -1);
3409 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3410 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3411 ip->i_d.di_flushiter = 0;
3414 * If this is really an old format inode and the superblock version
3415 * has not been updated to support only new format inodes, then
3416 * convert back to the old inode format. If the superblock version
3417 * has been updated, then make the conversion permanent.
3419 ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 ||
3420 XFS_SB_VERSION_HASNLINK(&mp->m_sb));
3421 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
3422 if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) {
3426 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3427 INT_SET(dip->di_core.di_onlink, ARCH_CONVERT, ip->i_d.di_nlink);
3430 * The superblock version has already been bumped,
3431 * so just make the conversion to the new inode
3434 ip->i_d.di_version = XFS_DINODE_VERSION_2;
3435 INT_SET(dip->di_core.di_version, ARCH_CONVERT, XFS_DINODE_VERSION_2);
3436 ip->i_d.di_onlink = 0;
3437 dip->di_core.di_onlink = 0;
3438 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3439 memset(&(dip->di_core.di_pad[0]), 0,
3440 sizeof(dip->di_core.di_pad));
3441 ASSERT(ip->i_d.di_projid == 0);
3445 if (xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp) == EFSCORRUPTED) {
3449 if (XFS_IFORK_Q(ip)) {
3451 * The only error from xfs_iflush_fork is on the data fork.
3453 (void) xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3455 xfs_inobp_check(mp, bp);
3458 * We've recorded everything logged in the inode, so we'd
3459 * like to clear the ilf_fields bits so we don't log and
3460 * flush things unnecessarily. However, we can't stop
3461 * logging all this information until the data we've copied
3462 * into the disk buffer is written to disk. If we did we might
3463 * overwrite the copy of the inode in the log with all the
3464 * data after re-logging only part of it, and in the face of
3465 * a crash we wouldn't have all the data we need to recover.
3467 * What we do is move the bits to the ili_last_fields field.
3468 * When logging the inode, these bits are moved back to the
3469 * ilf_fields field. In the xfs_iflush_done() routine we
3470 * clear ili_last_fields, since we know that the information
3471 * those bits represent is permanently on disk. As long as
3472 * the flush completes before the inode is logged again, then
3473 * both ilf_fields and ili_last_fields will be cleared.
3475 * We can play with the ilf_fields bits here, because the inode
3476 * lock must be held exclusively in order to set bits there
3477 * and the flush lock protects the ili_last_fields bits.
3478 * Set ili_logged so the flush done
3479 * routine can tell whether or not to look in the AIL.
3480 * Also, store the current LSN of the inode so that we can tell
3481 * whether the item has moved in the AIL from xfs_iflush_done().
3482 * In order to read the lsn we need the AIL lock, because
3483 * it is a 64 bit value that cannot be read atomically.
3485 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3486 iip->ili_last_fields = iip->ili_format.ilf_fields;
3487 iip->ili_format.ilf_fields = 0;
3488 iip->ili_logged = 1;
3490 ASSERT(sizeof(xfs_lsn_t) == 8); /* don't lock if it shrinks */
3492 iip->ili_flush_lsn = iip->ili_item.li_lsn;
3496 * Attach the function xfs_iflush_done to the inode's
3497 * buffer. This will remove the inode from the AIL
3498 * and unlock the inode's flush lock when the inode is
3499 * completely written to disk.
3501 xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
3502 xfs_iflush_done, (xfs_log_item_t *)iip);
3504 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3505 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3508 * We're flushing an inode which is not in the AIL and has
3509 * not been logged but has i_update_core set. For this
3510 * case we can use a B_DELWRI flush and immediately drop
3511 * the inode flush lock because we can avoid the whole
3512 * AIL state thing. It's OK to drop the flush lock now,
3513 * because we've already locked the buffer and to do anything
3514 * you really need both.
3517 ASSERT(iip->ili_logged == 0);
3518 ASSERT(iip->ili_last_fields == 0);
3519 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3527 return XFS_ERROR(EFSCORRUPTED);
3532 * Flush all inactive inodes in mp.
3542 XFS_MOUNT_ILOCK(mp);
3548 /* Make sure we skip markers inserted by sync */
3549 if (ip->i_mount == NULL) {
3554 vp = XFS_ITOV_NULL(ip);
3556 XFS_MOUNT_IUNLOCK(mp);
3557 xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC);
3561 ASSERT(vn_count(vp) == 0);
3564 } while (ip != mp->m_inodes);
3566 XFS_MOUNT_IUNLOCK(mp);
3570 * xfs_iaccess: check accessibility of inode for mode.
3579 mode_t orgmode = mode;
3580 struct inode *inode = vn_to_inode(XFS_ITOV(ip));
3582 if (mode & S_IWUSR) {
3583 umode_t imode = inode->i_mode;
3585 if (IS_RDONLY(inode) &&
3586 (S_ISREG(imode) || S_ISDIR(imode) || S_ISLNK(imode)))
3587 return XFS_ERROR(EROFS);
3589 if (IS_IMMUTABLE(inode))
3590 return XFS_ERROR(EACCES);
3594 * If there's an Access Control List it's used instead of
3597 if ((error = _ACL_XFS_IACCESS(ip, mode, cr)) != -1)
3598 return error ? XFS_ERROR(error) : 0;
3600 if (current_fsuid(cr) != ip->i_d.di_uid) {
3602 if (!in_group_p((gid_t)ip->i_d.di_gid))
3607 * If the DACs are ok we don't need any capability check.
3609 if ((ip->i_d.di_mode & mode) == mode)
3612 * Read/write DACs are always overridable.
3613 * Executable DACs are overridable if at least one exec bit is set.
3615 if (!(orgmode & S_IXUSR) ||
3616 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
3617 if (capable_cred(cr, CAP_DAC_OVERRIDE))
3620 if ((orgmode == S_IRUSR) ||
3621 (S_ISDIR(inode->i_mode) && (!(orgmode & S_IWUSR)))) {
3622 if (capable_cred(cr, CAP_DAC_READ_SEARCH))
3625 cmn_err(CE_NOTE, "Ick: mode=%o, orgmode=%o", mode, orgmode);
3627 return XFS_ERROR(EACCES);
3629 return XFS_ERROR(EACCES);
3633 * xfs_iroundup: round up argument to next power of two
3642 if ((v & (v - 1)) == 0)
3644 ASSERT((v & 0x80000000) == 0);
3645 if ((v & (v + 1)) == 0)
3647 for (i = 0, m = 1; i < 31; i++, m <<= 1) {
3651 if ((v & (v + 1)) == 0)
3658 #ifdef XFS_ILOCK_TRACE
3659 ktrace_t *xfs_ilock_trace_buf;
3662 xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
3664 ktrace_enter(ip->i_lock_trace,
3666 (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
3667 (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
3668 (void *)ra, /* caller of ilock */
3669 (void *)(unsigned long)current_cpu(),
3670 (void *)(unsigned long)current_pid(),
3671 NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL);
3676 * Return a pointer to the extent record at file index idx.
3680 xfs_ifork_t *ifp, /* inode fork pointer */
3681 xfs_extnum_t idx) /* index of target extent */
3684 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
3685 return ifp->if_u1.if_ext_irec->er_extbuf;
3686 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3687 xfs_ext_irec_t *erp; /* irec pointer */
3688 int erp_idx = 0; /* irec index */
3689 xfs_extnum_t page_idx = idx; /* ext index in target list */
3691 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3692 return &erp->er_extbuf[page_idx];
3693 } else if (ifp->if_bytes) {
3694 return &ifp->if_u1.if_extents[idx];
3701 * Insert new item(s) into the extent records for incore inode
3702 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3706 xfs_ifork_t *ifp, /* inode fork pointer */
3707 xfs_extnum_t idx, /* starting index of new items */
3708 xfs_extnum_t count, /* number of inserted items */
3709 xfs_bmbt_irec_t *new) /* items to insert */
3711 xfs_bmbt_rec_t *ep; /* extent record pointer */
3712 xfs_extnum_t i; /* extent record index */
3714 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3715 xfs_iext_add(ifp, idx, count);
3716 for (i = idx; i < idx + count; i++, new++) {
3717 ep = xfs_iext_get_ext(ifp, i);
3718 xfs_bmbt_set_all(ep, new);
3723 * This is called when the amount of space required for incore file
3724 * extents needs to be increased. The ext_diff parameter stores the
3725 * number of new extents being added and the idx parameter contains
3726 * the extent index where the new extents will be added. If the new
3727 * extents are being appended, then we just need to (re)allocate and
3728 * initialize the space. Otherwise, if the new extents are being
3729 * inserted into the middle of the existing entries, a bit more work
3730 * is required to make room for the new extents to be inserted. The
3731 * caller is responsible for filling in the new extent entries upon
3736 xfs_ifork_t *ifp, /* inode fork pointer */
3737 xfs_extnum_t idx, /* index to begin adding exts */
3738 int ext_diff) /* number of extents to add */
3740 int byte_diff; /* new bytes being added */
3741 int new_size; /* size of extents after adding */
3742 xfs_extnum_t nextents; /* number of extents in file */
3744 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3745 ASSERT((idx >= 0) && (idx <= nextents));
3746 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
3747 new_size = ifp->if_bytes + byte_diff;
3749 * If the new number of extents (nextents + ext_diff)
3750 * fits inside the inode, then continue to use the inline
3753 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
3754 if (idx < nextents) {
3755 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
3756 &ifp->if_u2.if_inline_ext[idx],
3757 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3758 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
3760 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3761 ifp->if_real_bytes = 0;
3762 ifp->if_lastex = nextents + ext_diff;
3765 * Otherwise use a linear (direct) extent list.
3766 * If the extents are currently inside the inode,
3767 * xfs_iext_realloc_direct will switch us from
3768 * inline to direct extent allocation mode.
3770 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
3771 xfs_iext_realloc_direct(ifp, new_size);
3772 if (idx < nextents) {
3773 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
3774 &ifp->if_u1.if_extents[idx],
3775 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3776 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
3779 /* Indirection array */
3781 xfs_ext_irec_t *erp;
3785 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
3786 if (ifp->if_flags & XFS_IFEXTIREC) {
3787 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3789 xfs_iext_irec_init(ifp);
3790 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3791 erp = ifp->if_u1.if_ext_irec;
3793 /* Extents fit in target extent page */
3794 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3795 if (page_idx < erp->er_extcount) {
3796 memmove(&erp->er_extbuf[page_idx + ext_diff],
3797 &erp->er_extbuf[page_idx],
3798 (erp->er_extcount - page_idx) *
3799 sizeof(xfs_bmbt_rec_t));
3800 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3802 erp->er_extcount += ext_diff;
3803 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3805 /* Insert a new extent page */
3807 xfs_iext_add_indirect_multi(ifp,
3808 erp_idx, page_idx, ext_diff);
3811 * If extent(s) are being appended to the last page in
3812 * the indirection array and the new extent(s) don't fit
3813 * in the page, then erp is NULL and erp_idx is set to
3814 * the next index needed in the indirection array.
3817 int count = ext_diff;
3820 erp = xfs_iext_irec_new(ifp, erp_idx);
3821 erp->er_extcount = count;
3822 count -= MIN(count, (int)XFS_LINEAR_EXTS);
3829 ifp->if_bytes = new_size;
3833 * This is called when incore extents are being added to the indirection
3834 * array and the new extents do not fit in the target extent list. The
3835 * erp_idx parameter contains the irec index for the target extent list
3836 * in the indirection array, and the idx parameter contains the extent
3837 * index within the list. The number of extents being added is stored
3838 * in the count parameter.
3840 * |-------| |-------|
3841 * | | | | idx - number of extents before idx
3843 * | | | | count - number of extents being inserted at idx
3844 * |-------| |-------|
3845 * | count | | nex2 | nex2 - number of extents after idx + count
3846 * |-------| |-------|
3849 xfs_iext_add_indirect_multi(
3850 xfs_ifork_t *ifp, /* inode fork pointer */
3851 int erp_idx, /* target extent irec index */
3852 xfs_extnum_t idx, /* index within target list */
3853 int count) /* new extents being added */
3855 int byte_diff; /* new bytes being added */
3856 xfs_ext_irec_t *erp; /* pointer to irec entry */
3857 xfs_extnum_t ext_diff; /* number of extents to add */
3858 xfs_extnum_t ext_cnt; /* new extents still needed */
3859 xfs_extnum_t nex2; /* extents after idx + count */
3860 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3861 int nlists; /* number of irec's (lists) */
3863 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3864 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3865 nex2 = erp->er_extcount - idx;
3866 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3869 * Save second part of target extent list
3870 * (all extents past */
3872 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3873 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_SLEEP);
3874 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3875 erp->er_extcount -= nex2;
3876 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3877 memset(&erp->er_extbuf[idx], 0, byte_diff);
3881 * Add the new extents to the end of the target
3882 * list, then allocate new irec record(s) and
3883 * extent buffer(s) as needed to store the rest
3884 * of the new extents.
3887 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3889 erp->er_extcount += ext_diff;
3890 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3891 ext_cnt -= ext_diff;
3895 erp = xfs_iext_irec_new(ifp, erp_idx);
3896 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3897 erp->er_extcount = ext_diff;
3898 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3899 ext_cnt -= ext_diff;
3902 /* Add nex2 extents back to indirection array */
3904 xfs_extnum_t ext_avail;
3907 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3908 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3911 * If nex2 extents fit in the current page, append
3912 * nex2_ep after the new extents.
3914 if (nex2 <= ext_avail) {
3915 i = erp->er_extcount;
3918 * Otherwise, check if space is available in the
3921 else if ((erp_idx < nlists - 1) &&
3922 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3923 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3926 /* Create a hole for nex2 extents */
3927 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3928 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3931 * Final choice, create a new extent page for
3936 erp = xfs_iext_irec_new(ifp, erp_idx);
3938 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
3939 kmem_free(nex2_ep, byte_diff);
3940 erp->er_extcount += nex2;
3941 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3946 * This is called when the amount of space required for incore file
3947 * extents needs to be decreased. The ext_diff parameter stores the
3948 * number of extents to be removed and the idx parameter contains
3949 * the extent index where the extents will be removed from.
3951 * If the amount of space needed has decreased below the linear
3952 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3953 * extent array. Otherwise, use kmem_realloc() to adjust the
3954 * size to what is needed.
3958 xfs_ifork_t *ifp, /* inode fork pointer */
3959 xfs_extnum_t idx, /* index to begin removing exts */
3960 int ext_diff) /* number of extents to remove */
3962 xfs_extnum_t nextents; /* number of extents in file */
3963 int new_size; /* size of extents after removal */
3965 ASSERT(ext_diff > 0);
3966 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3967 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3969 if (new_size == 0) {
3970 xfs_iext_destroy(ifp);
3971 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3972 xfs_iext_remove_indirect(ifp, idx, ext_diff);
3973 } else if (ifp->if_real_bytes) {
3974 xfs_iext_remove_direct(ifp, idx, ext_diff);
3976 xfs_iext_remove_inline(ifp, idx, ext_diff);
3978 ifp->if_bytes = new_size;
3982 * This removes ext_diff extents from the inline buffer, beginning
3983 * at extent index idx.
3986 xfs_iext_remove_inline(
3987 xfs_ifork_t *ifp, /* inode fork pointer */
3988 xfs_extnum_t idx, /* index to begin removing exts */
3989 int ext_diff) /* number of extents to remove */
3991 int nextents; /* number of extents in file */
3993 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3994 ASSERT(idx < XFS_INLINE_EXTS);
3995 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3996 ASSERT(((nextents - ext_diff) > 0) &&
3997 (nextents - ext_diff) < XFS_INLINE_EXTS);
3999 if (idx + ext_diff < nextents) {
4000 memmove(&ifp->if_u2.if_inline_ext[idx],
4001 &ifp->if_u2.if_inline_ext[idx + ext_diff],
4002 (nextents - (idx + ext_diff)) *
4003 sizeof(xfs_bmbt_rec_t));
4004 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
4005 0, ext_diff * sizeof(xfs_bmbt_rec_t));
4007 memset(&ifp->if_u2.if_inline_ext[idx], 0,
4008 ext_diff * sizeof(xfs_bmbt_rec_t));
4013 * This removes ext_diff extents from a linear (direct) extent list,
4014 * beginning at extent index idx. If the extents are being removed
4015 * from the end of the list (ie. truncate) then we just need to re-
4016 * allocate the list to remove the extra space. Otherwise, if the
4017 * extents are being removed from the middle of the existing extent
4018 * entries, then we first need to move the extent records beginning
4019 * at idx + ext_diff up in the list to overwrite the records being
4020 * removed, then remove the extra space via kmem_realloc.
4023 xfs_iext_remove_direct(
4024 xfs_ifork_t *ifp, /* inode fork pointer */
4025 xfs_extnum_t idx, /* index to begin removing exts */
4026 int ext_diff) /* number of extents to remove */
4028 xfs_extnum_t nextents; /* number of extents in file */
4029 int new_size; /* size of extents after removal */
4031 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4032 new_size = ifp->if_bytes -
4033 (ext_diff * sizeof(xfs_bmbt_rec_t));
4034 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4036 if (new_size == 0) {
4037 xfs_iext_destroy(ifp);
4040 /* Move extents up in the list (if needed) */
4041 if (idx + ext_diff < nextents) {
4042 memmove(&ifp->if_u1.if_extents[idx],
4043 &ifp->if_u1.if_extents[idx + ext_diff],
4044 (nextents - (idx + ext_diff)) *
4045 sizeof(xfs_bmbt_rec_t));
4047 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
4048 0, ext_diff * sizeof(xfs_bmbt_rec_t));
4050 * Reallocate the direct extent list. If the extents
4051 * will fit inside the inode then xfs_iext_realloc_direct
4052 * will switch from direct to inline extent allocation
4055 xfs_iext_realloc_direct(ifp, new_size);
4056 ifp->if_bytes = new_size;
4060 * This is called when incore extents are being removed from the
4061 * indirection array and the extents being removed span multiple extent
4062 * buffers. The idx parameter contains the file extent index where we
4063 * want to begin removing extents, and the count parameter contains
4064 * how many extents need to be removed.
4066 * |-------| |-------|
4067 * | nex1 | | | nex1 - number of extents before idx
4068 * |-------| | count |
4069 * | | | | count - number of extents being removed at idx
4070 * | count | |-------|
4071 * | | | nex2 | nex2 - number of extents after idx + count
4072 * |-------| |-------|
4075 xfs_iext_remove_indirect(
4076 xfs_ifork_t *ifp, /* inode fork pointer */
4077 xfs_extnum_t idx, /* index to begin removing extents */
4078 int count) /* number of extents to remove */
4080 xfs_ext_irec_t *erp; /* indirection array pointer */
4081 int erp_idx = 0; /* indirection array index */
4082 xfs_extnum_t ext_cnt; /* extents left to remove */
4083 xfs_extnum_t ext_diff; /* extents to remove in current list */
4084 xfs_extnum_t nex1; /* number of extents before idx */
4085 xfs_extnum_t nex2; /* extents after idx + count */
4086 int nlists; /* entries in indirection array */
4087 int page_idx = idx; /* index in target extent list */
4089 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4090 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
4091 ASSERT(erp != NULL);
4092 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4096 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
4097 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
4099 * Check for deletion of entire list;
4100 * xfs_iext_irec_remove() updates extent offsets.
4102 if (ext_diff == erp->er_extcount) {
4103 xfs_iext_irec_remove(ifp, erp_idx);
4104 ext_cnt -= ext_diff;
4107 ASSERT(erp_idx < ifp->if_real_bytes /
4109 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4116 /* Move extents up (if needed) */
4118 memmove(&erp->er_extbuf[nex1],
4119 &erp->er_extbuf[nex1 + ext_diff],
4120 nex2 * sizeof(xfs_bmbt_rec_t));
4122 /* Zero out rest of page */
4123 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
4124 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
4125 /* Update remaining counters */
4126 erp->er_extcount -= ext_diff;
4127 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
4128 ext_cnt -= ext_diff;
4133 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
4134 xfs_iext_irec_compact(ifp);
4138 * Create, destroy, or resize a linear (direct) block of extents.
4141 xfs_iext_realloc_direct(
4142 xfs_ifork_t *ifp, /* inode fork pointer */
4143 int new_size) /* new size of extents */
4145 int rnew_size; /* real new size of extents */
4147 rnew_size = new_size;
4149 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
4150 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
4151 (new_size != ifp->if_real_bytes)));
4153 /* Free extent records */
4154 if (new_size == 0) {
4155 xfs_iext_destroy(ifp);
4157 /* Resize direct extent list and zero any new bytes */
4158 else if (ifp->if_real_bytes) {
4159 /* Check if extents will fit inside the inode */
4160 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
4161 xfs_iext_direct_to_inline(ifp, new_size /
4162 (uint)sizeof(xfs_bmbt_rec_t));
4163 ifp->if_bytes = new_size;
4166 if ((new_size & (new_size - 1)) != 0) {
4167 rnew_size = xfs_iroundup(new_size);
4169 if (rnew_size != ifp->if_real_bytes) {
4170 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
4171 kmem_realloc(ifp->if_u1.if_extents,
4176 if (rnew_size > ifp->if_real_bytes) {
4177 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
4178 (uint)sizeof(xfs_bmbt_rec_t)], 0,
4179 rnew_size - ifp->if_real_bytes);
4183 * Switch from the inline extent buffer to a direct
4184 * extent list. Be sure to include the inline extent
4185 * bytes in new_size.
4188 new_size += ifp->if_bytes;
4189 if ((new_size & (new_size - 1)) != 0) {
4190 rnew_size = xfs_iroundup(new_size);
4192 xfs_iext_inline_to_direct(ifp, rnew_size);
4194 ifp->if_real_bytes = rnew_size;
4195 ifp->if_bytes = new_size;
4199 * Switch from linear (direct) extent records to inline buffer.
4202 xfs_iext_direct_to_inline(
4203 xfs_ifork_t *ifp, /* inode fork pointer */
4204 xfs_extnum_t nextents) /* number of extents in file */
4206 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4207 ASSERT(nextents <= XFS_INLINE_EXTS);
4209 * The inline buffer was zeroed when we switched
4210 * from inline to direct extent allocation mode,
4211 * so we don't need to clear it here.
4213 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
4214 nextents * sizeof(xfs_bmbt_rec_t));
4215 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
4216 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
4217 ifp->if_real_bytes = 0;
4221 * Switch from inline buffer to linear (direct) extent records.
4222 * new_size should already be rounded up to the next power of 2
4223 * by the caller (when appropriate), so use new_size as it is.
4224 * However, since new_size may be rounded up, we can't update
4225 * if_bytes here. It is the caller's responsibility to update
4226 * if_bytes upon return.
4229 xfs_iext_inline_to_direct(
4230 xfs_ifork_t *ifp, /* inode fork pointer */
4231 int new_size) /* number of extents in file */
4233 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
4234 kmem_alloc(new_size, KM_SLEEP);
4235 memset(ifp->if_u1.if_extents, 0, new_size);
4236 if (ifp->if_bytes) {
4237 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
4239 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4240 sizeof(xfs_bmbt_rec_t));
4242 ifp->if_real_bytes = new_size;
4246 * Resize an extent indirection array to new_size bytes.
4249 xfs_iext_realloc_indirect(
4250 xfs_ifork_t *ifp, /* inode fork pointer */
4251 int new_size) /* new indirection array size */
4253 int nlists; /* number of irec's (ex lists) */
4254 int size; /* current indirection array size */
4256 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4257 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4258 size = nlists * sizeof(xfs_ext_irec_t);
4259 ASSERT(ifp->if_real_bytes);
4260 ASSERT((new_size >= 0) && (new_size != size));
4261 if (new_size == 0) {
4262 xfs_iext_destroy(ifp);
4264 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
4265 kmem_realloc(ifp->if_u1.if_ext_irec,
4266 new_size, size, KM_SLEEP);
4271 * Switch from indirection array to linear (direct) extent allocations.
4274 xfs_iext_indirect_to_direct(
4275 xfs_ifork_t *ifp) /* inode fork pointer */
4277 xfs_bmbt_rec_t *ep; /* extent record pointer */
4278 xfs_extnum_t nextents; /* number of extents in file */
4279 int size; /* size of file extents */
4281 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4282 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4283 ASSERT(nextents <= XFS_LINEAR_EXTS);
4284 size = nextents * sizeof(xfs_bmbt_rec_t);
4286 xfs_iext_irec_compact_full(ifp);
4287 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
4289 ep = ifp->if_u1.if_ext_irec->er_extbuf;
4290 kmem_free(ifp->if_u1.if_ext_irec, sizeof(xfs_ext_irec_t));
4291 ifp->if_flags &= ~XFS_IFEXTIREC;
4292 ifp->if_u1.if_extents = ep;
4293 ifp->if_bytes = size;
4294 if (nextents < XFS_LINEAR_EXTS) {
4295 xfs_iext_realloc_direct(ifp, size);
4300 * Free incore file extents.
4304 xfs_ifork_t *ifp) /* inode fork pointer */
4306 if (ifp->if_flags & XFS_IFEXTIREC) {
4310 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4311 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
4312 xfs_iext_irec_remove(ifp, erp_idx);
4314 ifp->if_flags &= ~XFS_IFEXTIREC;
4315 } else if (ifp->if_real_bytes) {
4316 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
4317 } else if (ifp->if_bytes) {
4318 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4319 sizeof(xfs_bmbt_rec_t));
4321 ifp->if_u1.if_extents = NULL;
4322 ifp->if_real_bytes = 0;
4327 * Return a pointer to the extent record for file system block bno.
4329 xfs_bmbt_rec_t * /* pointer to found extent record */
4330 xfs_iext_bno_to_ext(
4331 xfs_ifork_t *ifp, /* inode fork pointer */
4332 xfs_fileoff_t bno, /* block number to search for */
4333 xfs_extnum_t *idxp) /* index of target extent */
4335 xfs_bmbt_rec_t *base; /* pointer to first extent */
4336 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
4337 xfs_bmbt_rec_t *ep = NULL; /* pointer to target extent */
4338 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4339 int high; /* upper boundary in search */
4340 xfs_extnum_t idx = 0; /* index of target extent */
4341 int low; /* lower boundary in search */
4342 xfs_extnum_t nextents; /* number of file extents */
4343 xfs_fileoff_t startoff = 0; /* start offset of extent */
4345 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4346 if (nextents == 0) {
4351 if (ifp->if_flags & XFS_IFEXTIREC) {
4352 /* Find target extent list */
4354 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
4355 base = erp->er_extbuf;
4356 high = erp->er_extcount - 1;
4358 base = ifp->if_u1.if_extents;
4359 high = nextents - 1;
4361 /* Binary search extent records */
4362 while (low <= high) {
4363 idx = (low + high) >> 1;
4365 startoff = xfs_bmbt_get_startoff(ep);
4366 blockcount = xfs_bmbt_get_blockcount(ep);
4367 if (bno < startoff) {
4369 } else if (bno >= startoff + blockcount) {
4372 /* Convert back to file-based extent index */
4373 if (ifp->if_flags & XFS_IFEXTIREC) {
4374 idx += erp->er_extoff;
4380 /* Convert back to file-based extent index */
4381 if (ifp->if_flags & XFS_IFEXTIREC) {
4382 idx += erp->er_extoff;
4384 if (bno >= startoff + blockcount) {
4385 if (++idx == nextents) {
4388 ep = xfs_iext_get_ext(ifp, idx);
4396 * Return a pointer to the indirection array entry containing the
4397 * extent record for filesystem block bno. Store the index of the
4398 * target irec in *erp_idxp.
4400 xfs_ext_irec_t * /* pointer to found extent record */
4401 xfs_iext_bno_to_irec(
4402 xfs_ifork_t *ifp, /* inode fork pointer */
4403 xfs_fileoff_t bno, /* block number to search for */
4404 int *erp_idxp) /* irec index of target ext list */
4406 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4407 xfs_ext_irec_t *erp_next; /* next indirection array entry */
4408 int erp_idx; /* indirection array index */
4409 int nlists; /* number of extent irec's (lists) */
4410 int high; /* binary search upper limit */
4411 int low; /* binary search lower limit */
4413 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4414 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4418 while (low <= high) {
4419 erp_idx = (low + high) >> 1;
4420 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4421 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
4422 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
4424 } else if (erp_next && bno >=
4425 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
4431 *erp_idxp = erp_idx;
4436 * Return a pointer to the indirection array entry containing the
4437 * extent record at file extent index *idxp. Store the index of the
4438 * target irec in *erp_idxp and store the page index of the target
4439 * extent record in *idxp.
4442 xfs_iext_idx_to_irec(
4443 xfs_ifork_t *ifp, /* inode fork pointer */
4444 xfs_extnum_t *idxp, /* extent index (file -> page) */
4445 int *erp_idxp, /* pointer to target irec */
4446 int realloc) /* new bytes were just added */
4448 xfs_ext_irec_t *prev; /* pointer to previous irec */
4449 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
4450 int erp_idx; /* indirection array index */
4451 int nlists; /* number of irec's (ex lists) */
4452 int high; /* binary search upper limit */
4453 int low; /* binary search lower limit */
4454 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
4456 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4457 ASSERT(page_idx >= 0 && page_idx <=
4458 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
4459 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4464 /* Binary search extent irec's */
4465 while (low <= high) {
4466 erp_idx = (low + high) >> 1;
4467 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4468 prev = erp_idx > 0 ? erp - 1 : NULL;
4469 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
4470 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
4472 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
4473 (page_idx == erp->er_extoff + erp->er_extcount &&
4476 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
4477 erp->er_extcount == XFS_LINEAR_EXTS) {
4481 erp = erp_idx < nlists ? erp + 1 : NULL;
4484 page_idx -= erp->er_extoff;
4489 *erp_idxp = erp_idx;
4494 * Allocate and initialize an indirection array once the space needed
4495 * for incore extents increases above XFS_IEXT_BUFSZ.
4499 xfs_ifork_t *ifp) /* inode fork pointer */
4501 xfs_ext_irec_t *erp; /* indirection array pointer */
4502 xfs_extnum_t nextents; /* number of extents in file */
4504 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4505 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4506 ASSERT(nextents <= XFS_LINEAR_EXTS);
4508 erp = (xfs_ext_irec_t *)
4509 kmem_alloc(sizeof(xfs_ext_irec_t), KM_SLEEP);
4511 if (nextents == 0) {
4512 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
4513 kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
4514 } else if (!ifp->if_real_bytes) {
4515 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
4516 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
4517 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
4519 erp->er_extbuf = ifp->if_u1.if_extents;
4520 erp->er_extcount = nextents;
4523 ifp->if_flags |= XFS_IFEXTIREC;
4524 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
4525 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
4526 ifp->if_u1.if_ext_irec = erp;
4532 * Allocate and initialize a new entry in the indirection array.
4536 xfs_ifork_t *ifp, /* inode fork pointer */
4537 int erp_idx) /* index for new irec */
4539 xfs_ext_irec_t *erp; /* indirection array pointer */
4540 int i; /* loop counter */
4541 int nlists; /* number of irec's (ex lists) */
4543 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4544 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4546 /* Resize indirection array */
4547 xfs_iext_realloc_indirect(ifp, ++nlists *
4548 sizeof(xfs_ext_irec_t));
4550 * Move records down in the array so the
4551 * new page can use erp_idx.
4553 erp = ifp->if_u1.if_ext_irec;
4554 for (i = nlists - 1; i > erp_idx; i--) {
4555 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
4557 ASSERT(i == erp_idx);
4559 /* Initialize new extent record */
4560 erp = ifp->if_u1.if_ext_irec;
4561 erp[erp_idx].er_extbuf = (xfs_bmbt_rec_t *)
4562 kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
4563 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4564 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
4565 erp[erp_idx].er_extcount = 0;
4566 erp[erp_idx].er_extoff = erp_idx > 0 ?
4567 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
4568 return (&erp[erp_idx]);
4572 * Remove a record from the indirection array.
4575 xfs_iext_irec_remove(
4576 xfs_ifork_t *ifp, /* inode fork pointer */
4577 int erp_idx) /* irec index to remove */
4579 xfs_ext_irec_t *erp; /* indirection array pointer */
4580 int i; /* loop counter */
4581 int nlists; /* number of irec's (ex lists) */
4583 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4584 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4585 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4586 if (erp->er_extbuf) {
4587 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
4589 kmem_free(erp->er_extbuf, XFS_IEXT_BUFSZ);
4591 /* Compact extent records */
4592 erp = ifp->if_u1.if_ext_irec;
4593 for (i = erp_idx; i < nlists - 1; i++) {
4594 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
4597 * Manually free the last extent record from the indirection
4598 * array. A call to xfs_iext_realloc_indirect() with a size
4599 * of zero would result in a call to xfs_iext_destroy() which
4600 * would in turn call this function again, creating a nasty
4604 xfs_iext_realloc_indirect(ifp,
4605 nlists * sizeof(xfs_ext_irec_t));
4607 kmem_free(ifp->if_u1.if_ext_irec,
4608 sizeof(xfs_ext_irec_t));
4610 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4614 * This is called to clean up large amounts of unused memory allocated
4615 * by the indirection array. Before compacting anything though, verify
4616 * that the indirection array is still needed and switch back to the
4617 * linear extent list (or even the inline buffer) if possible. The
4618 * compaction policy is as follows:
4620 * Full Compaction: Extents fit into a single page (or inline buffer)
4621 * Full Compaction: Extents occupy less than 10% of allocated space
4622 * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
4623 * No Compaction: Extents occupy at least 50% of allocated space
4626 xfs_iext_irec_compact(
4627 xfs_ifork_t *ifp) /* inode fork pointer */
4629 xfs_extnum_t nextents; /* number of extents in file */
4630 int nlists; /* number of irec's (ex lists) */
4632 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4633 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4634 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4636 if (nextents == 0) {
4637 xfs_iext_destroy(ifp);
4638 } else if (nextents <= XFS_INLINE_EXTS) {
4639 xfs_iext_indirect_to_direct(ifp);
4640 xfs_iext_direct_to_inline(ifp, nextents);
4641 } else if (nextents <= XFS_LINEAR_EXTS) {
4642 xfs_iext_indirect_to_direct(ifp);
4643 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 3) {
4644 xfs_iext_irec_compact_full(ifp);
4645 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
4646 xfs_iext_irec_compact_pages(ifp);
4651 * Combine extents from neighboring extent pages.
4654 xfs_iext_irec_compact_pages(
4655 xfs_ifork_t *ifp) /* inode fork pointer */
4657 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
4658 int erp_idx = 0; /* indirection array index */
4659 int nlists; /* number of irec's (ex lists) */
4661 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4662 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4663 while (erp_idx < nlists - 1) {
4664 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4666 if (erp_next->er_extcount <=
4667 (XFS_LINEAR_EXTS - erp->er_extcount)) {
4668 memmove(&erp->er_extbuf[erp->er_extcount],
4669 erp_next->er_extbuf, erp_next->er_extcount *
4670 sizeof(xfs_bmbt_rec_t));
4671 erp->er_extcount += erp_next->er_extcount;
4673 * Free page before removing extent record
4674 * so er_extoffs don't get modified in
4675 * xfs_iext_irec_remove.
4677 kmem_free(erp_next->er_extbuf, XFS_IEXT_BUFSZ);
4678 erp_next->er_extbuf = NULL;
4679 xfs_iext_irec_remove(ifp, erp_idx + 1);
4680 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4688 * Fully compact the extent records managed by the indirection array.
4691 xfs_iext_irec_compact_full(
4692 xfs_ifork_t *ifp) /* inode fork pointer */
4694 xfs_bmbt_rec_t *ep, *ep_next; /* extent record pointers */
4695 xfs_ext_irec_t *erp, *erp_next; /* extent irec pointers */
4696 int erp_idx = 0; /* extent irec index */
4697 int ext_avail; /* empty entries in ex list */
4698 int ext_diff; /* number of exts to add */
4699 int nlists; /* number of irec's (ex lists) */
4701 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4702 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4703 erp = ifp->if_u1.if_ext_irec;
4704 ep = &erp->er_extbuf[erp->er_extcount];
4706 ep_next = erp_next->er_extbuf;
4707 while (erp_idx < nlists - 1) {
4708 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
4709 ext_diff = MIN(ext_avail, erp_next->er_extcount);
4710 memcpy(ep, ep_next, ext_diff * sizeof(xfs_bmbt_rec_t));
4711 erp->er_extcount += ext_diff;
4712 erp_next->er_extcount -= ext_diff;
4713 /* Remove next page */
4714 if (erp_next->er_extcount == 0) {
4716 * Free page before removing extent record
4717 * so er_extoffs don't get modified in
4718 * xfs_iext_irec_remove.
4720 kmem_free(erp_next->er_extbuf,
4721 erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
4722 erp_next->er_extbuf = NULL;
4723 xfs_iext_irec_remove(ifp, erp_idx + 1);
4724 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4725 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4726 /* Update next page */
4728 /* Move rest of page up to become next new page */
4729 memmove(erp_next->er_extbuf, ep_next,
4730 erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
4731 ep_next = erp_next->er_extbuf;
4732 memset(&ep_next[erp_next->er_extcount], 0,
4733 (XFS_LINEAR_EXTS - erp_next->er_extcount) *
4734 sizeof(xfs_bmbt_rec_t));
4736 if (erp->er_extcount == XFS_LINEAR_EXTS) {
4738 if (erp_idx < nlists)
4739 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4743 ep = &erp->er_extbuf[erp->er_extcount];
4745 ep_next = erp_next->er_extbuf;
4750 * This is called to update the er_extoff field in the indirection
4751 * array when extents have been added or removed from one of the
4752 * extent lists. erp_idx contains the irec index to begin updating
4753 * at and ext_diff contains the number of extents that were added
4757 xfs_iext_irec_update_extoffs(
4758 xfs_ifork_t *ifp, /* inode fork pointer */
4759 int erp_idx, /* irec index to update */
4760 int ext_diff) /* number of new extents */
4762 int i; /* loop counter */
4763 int nlists; /* number of irec's (ex lists */
4765 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4766 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4767 for (i = erp_idx; i < nlists; i++) {
4768 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;